Linux Audio

Check our new training course

In-person Linux kernel drivers training

Jun 16-20, 2025
Register
Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/ext4/inode.c
   4 *
   5 * Copyright (C) 1992, 1993, 1994, 1995
   6 * Remy Card (card@masi.ibp.fr)
   7 * Laboratoire MASI - Institut Blaise Pascal
   8 * Universite Pierre et Marie Curie (Paris VI)
   9 *
  10 *  from
  11 *
  12 *  linux/fs/minix/inode.c
  13 *
  14 *  Copyright (C) 1991, 1992  Linus Torvalds
  15 *
  16 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  17 *	(jj@sunsite.ms.mff.cuni.cz)
  18 *
  19 *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  20 */
  21
  22#include <linux/fs.h>
  23#include <linux/mount.h>
  24#include <linux/time.h>
  25#include <linux/highuid.h>
  26#include <linux/pagemap.h>
  27#include <linux/dax.h>
  28#include <linux/quotaops.h>
  29#include <linux/string.h>
  30#include <linux/buffer_head.h>
  31#include <linux/writeback.h>
  32#include <linux/pagevec.h>
  33#include <linux/mpage.h>
  34#include <linux/namei.h>
  35#include <linux/uio.h>
  36#include <linux/bio.h>
  37#include <linux/workqueue.h>
  38#include <linux/kernel.h>
  39#include <linux/printk.h>
  40#include <linux/slab.h>
  41#include <linux/bitops.h>
  42#include <linux/iomap.h>
  43#include <linux/iversion.h>
  44
  45#include "ext4_jbd2.h"
  46#include "xattr.h"
  47#include "acl.h"
  48#include "truncate.h"
  49
  50#include <trace/events/ext4.h>
  51
  52static void ext4_journalled_zero_new_buffers(handle_t *handle,
  53					    struct inode *inode,
  54					    struct folio *folio,
  55					    unsigned from, unsigned to);
  56
  57static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
  58			      struct ext4_inode_info *ei)
  59{
  60	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  61	__u32 csum;
  62	__u16 dummy_csum = 0;
  63	int offset = offsetof(struct ext4_inode, i_checksum_lo);
  64	unsigned int csum_size = sizeof(dummy_csum);
  65
  66	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
  67	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
  68	offset += csum_size;
  69	csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
  70			   EXT4_GOOD_OLD_INODE_SIZE - offset);
  71
  72	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
  73		offset = offsetof(struct ext4_inode, i_checksum_hi);
  74		csum = ext4_chksum(sbi, csum, (__u8 *)raw +
  75				   EXT4_GOOD_OLD_INODE_SIZE,
  76				   offset - EXT4_GOOD_OLD_INODE_SIZE);
  77		if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
  78			csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
  79					   csum_size);
  80			offset += csum_size;
  81		}
  82		csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
  83				   EXT4_INODE_SIZE(inode->i_sb) - offset);
  84	}
  85
  86	return csum;
  87}
  88
  89static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
  90				  struct ext4_inode_info *ei)
  91{
  92	__u32 provided, calculated;
  93
  94	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  95	    cpu_to_le32(EXT4_OS_LINUX) ||
  96	    !ext4_has_metadata_csum(inode->i_sb))
  97		return 1;
  98
  99	provided = le16_to_cpu(raw->i_checksum_lo);
 100	calculated = ext4_inode_csum(inode, raw, ei);
 101	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
 102	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
 103		provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
 104	else
 105		calculated &= 0xFFFF;
 106
 107	return provided == calculated;
 108}
 109
 110void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
 111			 struct ext4_inode_info *ei)
 112{
 113	__u32 csum;
 114
 115	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
 116	    cpu_to_le32(EXT4_OS_LINUX) ||
 117	    !ext4_has_metadata_csum(inode->i_sb))
 118		return;
 119
 120	csum = ext4_inode_csum(inode, raw, ei);
 121	raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
 122	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
 123	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
 124		raw->i_checksum_hi = cpu_to_le16(csum >> 16);
 125}
 126
 127static inline int ext4_begin_ordered_truncate(struct inode *inode,
 128					      loff_t new_size)
 129{
 130	trace_ext4_begin_ordered_truncate(inode, new_size);
 131	/*
 132	 * If jinode is zero, then we never opened the file for
 133	 * writing, so there's no need to call
 134	 * jbd2_journal_begin_ordered_truncate() since there's no
 135	 * outstanding writes we need to flush.
 136	 */
 137	if (!EXT4_I(inode)->jinode)
 138		return 0;
 139	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
 140						   EXT4_I(inode)->jinode,
 141						   new_size);
 142}
 143
 
 144static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
 145				  int pextents);
 146
 147/*
 148 * Test whether an inode is a fast symlink.
 149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
 150 */
 151int ext4_inode_is_fast_symlink(struct inode *inode)
 152{
 153	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 154		int ea_blocks = EXT4_I(inode)->i_file_acl ?
 155				EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
 156
 157		if (ext4_has_inline_data(inode))
 158			return 0;
 159
 160		return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
 161	}
 162	return S_ISLNK(inode->i_mode) && inode->i_size &&
 163	       (inode->i_size < EXT4_N_BLOCKS * 4);
 164}
 165
 166/*
 167 * Called at the last iput() if i_nlink is zero.
 168 */
 169void ext4_evict_inode(struct inode *inode)
 170{
 171	handle_t *handle;
 172	int err;
 173	/*
 174	 * Credits for final inode cleanup and freeing:
 175	 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
 176	 * (xattr block freeing), bitmap, group descriptor (inode freeing)
 177	 */
 178	int extra_credits = 6;
 179	struct ext4_xattr_inode_array *ea_inode_array = NULL;
 180	bool freeze_protected = false;
 181
 182	trace_ext4_evict_inode(inode);
 183
 184	if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
 185		ext4_evict_ea_inode(inode);
 186	if (inode->i_nlink) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 187		truncate_inode_pages_final(&inode->i_data);
 188
 189		goto no_delete;
 190	}
 191
 192	if (is_bad_inode(inode))
 193		goto no_delete;
 194	dquot_initialize(inode);
 195
 196	if (ext4_should_order_data(inode))
 197		ext4_begin_ordered_truncate(inode, 0);
 198	truncate_inode_pages_final(&inode->i_data);
 199
 200	/*
 201	 * For inodes with journalled data, transaction commit could have
 202	 * dirtied the inode. And for inodes with dioread_nolock, unwritten
 203	 * extents converting worker could merge extents and also have dirtied
 204	 * the inode. Flush worker is ignoring it because of I_FREEING flag but
 205	 * we still need to remove the inode from the writeback lists.
 206	 */
 207	if (!list_empty_careful(&inode->i_io_list))
 208		inode_io_list_del(inode);
 209
 210	/*
 211	 * Protect us against freezing - iput() caller didn't have to have any
 212	 * protection against it. When we are in a running transaction though,
 213	 * we are already protected against freezing and we cannot grab further
 214	 * protection due to lock ordering constraints.
 215	 */
 216	if (!ext4_journal_current_handle()) {
 217		sb_start_intwrite(inode->i_sb);
 218		freeze_protected = true;
 219	}
 220
 221	if (!IS_NOQUOTA(inode))
 222		extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
 223
 224	/*
 225	 * Block bitmap, group descriptor, and inode are accounted in both
 226	 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
 227	 */
 228	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
 229			 ext4_blocks_for_truncate(inode) + extra_credits - 3);
 230	if (IS_ERR(handle)) {
 231		ext4_std_error(inode->i_sb, PTR_ERR(handle));
 232		/*
 233		 * If we're going to skip the normal cleanup, we still need to
 234		 * make sure that the in-core orphan linked list is properly
 235		 * cleaned up.
 236		 */
 237		ext4_orphan_del(NULL, inode);
 238		if (freeze_protected)
 239			sb_end_intwrite(inode->i_sb);
 240		goto no_delete;
 241	}
 242
 243	if (IS_SYNC(inode))
 244		ext4_handle_sync(handle);
 245
 246	/*
 247	 * Set inode->i_size to 0 before calling ext4_truncate(). We need
 248	 * special handling of symlinks here because i_size is used to
 249	 * determine whether ext4_inode_info->i_data contains symlink data or
 250	 * block mappings. Setting i_size to 0 will remove its fast symlink
 251	 * status. Erase i_data so that it becomes a valid empty block map.
 252	 */
 253	if (ext4_inode_is_fast_symlink(inode))
 254		memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
 255	inode->i_size = 0;
 256	err = ext4_mark_inode_dirty(handle, inode);
 257	if (err) {
 258		ext4_warning(inode->i_sb,
 259			     "couldn't mark inode dirty (err %d)", err);
 260		goto stop_handle;
 261	}
 262	if (inode->i_blocks) {
 263		err = ext4_truncate(inode);
 264		if (err) {
 265			ext4_error_err(inode->i_sb, -err,
 266				       "couldn't truncate inode %lu (err %d)",
 267				       inode->i_ino, err);
 268			goto stop_handle;
 269		}
 270	}
 271
 272	/* Remove xattr references. */
 273	err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
 274				      extra_credits);
 275	if (err) {
 276		ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
 277stop_handle:
 278		ext4_journal_stop(handle);
 279		ext4_orphan_del(NULL, inode);
 280		if (freeze_protected)
 281			sb_end_intwrite(inode->i_sb);
 282		ext4_xattr_inode_array_free(ea_inode_array);
 283		goto no_delete;
 284	}
 285
 286	/*
 287	 * Kill off the orphan record which ext4_truncate created.
 288	 * AKPM: I think this can be inside the above `if'.
 289	 * Note that ext4_orphan_del() has to be able to cope with the
 290	 * deletion of a non-existent orphan - this is because we don't
 291	 * know if ext4_truncate() actually created an orphan record.
 292	 * (Well, we could do this if we need to, but heck - it works)
 293	 */
 294	ext4_orphan_del(handle, inode);
 295	EXT4_I(inode)->i_dtime	= (__u32)ktime_get_real_seconds();
 296
 297	/*
 298	 * One subtle ordering requirement: if anything has gone wrong
 299	 * (transaction abort, IO errors, whatever), then we can still
 300	 * do these next steps (the fs will already have been marked as
 301	 * having errors), but we can't free the inode if the mark_dirty
 302	 * fails.
 303	 */
 304	if (ext4_mark_inode_dirty(handle, inode))
 305		/* If that failed, just do the required in-core inode clear. */
 306		ext4_clear_inode(inode);
 307	else
 308		ext4_free_inode(handle, inode);
 309	ext4_journal_stop(handle);
 310	if (freeze_protected)
 311		sb_end_intwrite(inode->i_sb);
 312	ext4_xattr_inode_array_free(ea_inode_array);
 313	return;
 314no_delete:
 315	/*
 316	 * Check out some where else accidentally dirty the evicting inode,
 317	 * which may probably cause inode use-after-free issues later.
 318	 */
 319	WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
 320
 321	if (!list_empty(&EXT4_I(inode)->i_fc_list))
 322		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
 323	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
 324}
 325
 326#ifdef CONFIG_QUOTA
 327qsize_t *ext4_get_reserved_space(struct inode *inode)
 328{
 329	return &EXT4_I(inode)->i_reserved_quota;
 330}
 331#endif
 332
 333/*
 334 * Called with i_data_sem down, which is important since we can call
 335 * ext4_discard_preallocations() from here.
 336 */
 337void ext4_da_update_reserve_space(struct inode *inode,
 338					int used, int quota_claim)
 339{
 340	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 341	struct ext4_inode_info *ei = EXT4_I(inode);
 342
 343	spin_lock(&ei->i_block_reservation_lock);
 344	trace_ext4_da_update_reserve_space(inode, used, quota_claim);
 345	if (unlikely(used > ei->i_reserved_data_blocks)) {
 346		ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
 347			 "with only %d reserved data blocks",
 348			 __func__, inode->i_ino, used,
 349			 ei->i_reserved_data_blocks);
 350		WARN_ON(1);
 351		used = ei->i_reserved_data_blocks;
 352	}
 353
 354	/* Update per-inode reservations */
 355	ei->i_reserved_data_blocks -= used;
 356	percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
 357
 358	spin_unlock(&ei->i_block_reservation_lock);
 359
 360	/* Update quota subsystem for data blocks */
 361	if (quota_claim)
 362		dquot_claim_block(inode, EXT4_C2B(sbi, used));
 363	else {
 364		/*
 365		 * We did fallocate with an offset that is already delayed
 366		 * allocated. So on delayed allocated writeback we should
 367		 * not re-claim the quota for fallocated blocks.
 368		 */
 369		dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
 370	}
 371
 372	/*
 373	 * If we have done all the pending block allocations and if
 374	 * there aren't any writers on the inode, we can discard the
 375	 * inode's preallocations.
 376	 */
 377	if ((ei->i_reserved_data_blocks == 0) &&
 378	    !inode_is_open_for_write(inode))
 379		ext4_discard_preallocations(inode);
 380}
 381
 382static int __check_block_validity(struct inode *inode, const char *func,
 383				unsigned int line,
 384				struct ext4_map_blocks *map)
 385{
 386	if (ext4_has_feature_journal(inode->i_sb) &&
 387	    (inode->i_ino ==
 388	     le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
 389		return 0;
 390	if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
 391		ext4_error_inode(inode, func, line, map->m_pblk,
 392				 "lblock %lu mapped to illegal pblock %llu "
 393				 "(length %d)", (unsigned long) map->m_lblk,
 394				 map->m_pblk, map->m_len);
 395		return -EFSCORRUPTED;
 396	}
 397	return 0;
 398}
 399
 400int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
 401		       ext4_lblk_t len)
 402{
 403	int ret;
 404
 405	if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
 406		return fscrypt_zeroout_range(inode, lblk, pblk, len);
 407
 408	ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
 409	if (ret > 0)
 410		ret = 0;
 411
 412	return ret;
 413}
 414
 415#define check_block_validity(inode, map)	\
 416	__check_block_validity((inode), __func__, __LINE__, (map))
 417
 418#ifdef ES_AGGRESSIVE_TEST
 419static void ext4_map_blocks_es_recheck(handle_t *handle,
 420				       struct inode *inode,
 421				       struct ext4_map_blocks *es_map,
 422				       struct ext4_map_blocks *map,
 423				       int flags)
 424{
 425	int retval;
 426
 427	map->m_flags = 0;
 428	/*
 429	 * There is a race window that the result is not the same.
 430	 * e.g. xfstests #223 when dioread_nolock enables.  The reason
 431	 * is that we lookup a block mapping in extent status tree with
 432	 * out taking i_data_sem.  So at the time the unwritten extent
 433	 * could be converted.
 434	 */
 435	down_read(&EXT4_I(inode)->i_data_sem);
 436	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 437		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 438	} else {
 439		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 440	}
 441	up_read((&EXT4_I(inode)->i_data_sem));
 442
 443	/*
 444	 * We don't check m_len because extent will be collpased in status
 445	 * tree.  So the m_len might not equal.
 446	 */
 447	if (es_map->m_lblk != map->m_lblk ||
 448	    es_map->m_flags != map->m_flags ||
 449	    es_map->m_pblk != map->m_pblk) {
 450		printk("ES cache assertion failed for inode: %lu "
 451		       "es_cached ex [%d/%d/%llu/%x] != "
 452		       "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
 453		       inode->i_ino, es_map->m_lblk, es_map->m_len,
 454		       es_map->m_pblk, es_map->m_flags, map->m_lblk,
 455		       map->m_len, map->m_pblk, map->m_flags,
 456		       retval, flags);
 457	}
 458}
 459#endif /* ES_AGGRESSIVE_TEST */
 460
 461static int ext4_map_query_blocks(handle_t *handle, struct inode *inode,
 462				 struct ext4_map_blocks *map)
 463{
 464	unsigned int status;
 465	int retval;
 466
 467	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 468		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 469	else
 470		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 471
 472	if (retval <= 0)
 473		return retval;
 474
 475	if (unlikely(retval != map->m_len)) {
 476		ext4_warning(inode->i_sb,
 477			     "ES len assertion failed for inode "
 478			     "%lu: retval %d != map->m_len %d",
 479			     inode->i_ino, retval, map->m_len);
 480		WARN_ON(1);
 481	}
 482
 483	status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 484			EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 485	ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
 486			      map->m_pblk, status, false);
 487	return retval;
 488}
 489
 490static int ext4_map_create_blocks(handle_t *handle, struct inode *inode,
 491				  struct ext4_map_blocks *map, int flags)
 492{
 493	struct extent_status es;
 494	unsigned int status;
 495	int err, retval = 0;
 496
 497	/*
 498	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE
 499	 * indicates that the blocks and quotas has already been
 500	 * checked when the data was copied into the page cache.
 501	 */
 502	if (map->m_flags & EXT4_MAP_DELAYED)
 503		flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
 504
 505	/*
 506	 * Here we clear m_flags because after allocating an new extent,
 507	 * it will be set again.
 508	 */
 509	map->m_flags &= ~EXT4_MAP_FLAGS;
 510
 511	/*
 512	 * We need to check for EXT4 here because migrate could have
 513	 * changed the inode type in between.
 514	 */
 515	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 516		retval = ext4_ext_map_blocks(handle, inode, map, flags);
 517	} else {
 518		retval = ext4_ind_map_blocks(handle, inode, map, flags);
 519
 520		/*
 521		 * We allocated new blocks which will result in i_data's
 522		 * format changing. Force the migrate to fail by clearing
 523		 * migrate flags.
 524		 */
 525		if (retval > 0 && map->m_flags & EXT4_MAP_NEW)
 526			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 527	}
 528	if (retval <= 0)
 529		return retval;
 530
 531	if (unlikely(retval != map->m_len)) {
 532		ext4_warning(inode->i_sb,
 533			     "ES len assertion failed for inode %lu: "
 534			     "retval %d != map->m_len %d",
 535			     inode->i_ino, retval, map->m_len);
 536		WARN_ON(1);
 537	}
 538
 539	/*
 540	 * We have to zeroout blocks before inserting them into extent
 541	 * status tree. Otherwise someone could look them up there and
 542	 * use them before they are really zeroed. We also have to
 543	 * unmap metadata before zeroing as otherwise writeback can
 544	 * overwrite zeros with stale data from block device.
 545	 */
 546	if (flags & EXT4_GET_BLOCKS_ZERO &&
 547	    map->m_flags & EXT4_MAP_MAPPED && map->m_flags & EXT4_MAP_NEW) {
 548		err = ext4_issue_zeroout(inode, map->m_lblk, map->m_pblk,
 549					 map->m_len);
 550		if (err)
 551			return err;
 552	}
 553
 554	/*
 555	 * If the extent has been zeroed out, we don't need to update
 556	 * extent status tree.
 557	 */
 558	if (flags & EXT4_GET_BLOCKS_PRE_IO &&
 559	    ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
 560		if (ext4_es_is_written(&es))
 561			return retval;
 562	}
 563
 564	status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 565			EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 566	ext4_es_insert_extent(inode, map->m_lblk, map->m_len, map->m_pblk,
 567			      status, flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE);
 568
 569	return retval;
 570}
 571
 572/*
 573 * The ext4_map_blocks() function tries to look up the requested blocks,
 574 * and returns if the blocks are already mapped.
 575 *
 576 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 577 * and store the allocated blocks in the result buffer head and mark it
 578 * mapped.
 579 *
 580 * If file type is extents based, it will call ext4_ext_map_blocks(),
 581 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 582 * based files
 583 *
 584 * On success, it returns the number of blocks being mapped or allocated.
 585 * If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are
 586 * pre-allocated and unwritten, the resulting @map is marked as unwritten.
 587 * If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped.
 588 *
 589 * It returns 0 if plain look up failed (blocks have not been allocated), in
 590 * that case, @map is returned as unmapped but we still do fill map->m_len to
 591 * indicate the length of a hole starting at map->m_lblk.
 592 *
 593 * It returns the error in case of allocation failure.
 594 */
 595int ext4_map_blocks(handle_t *handle, struct inode *inode,
 596		    struct ext4_map_blocks *map, int flags)
 597{
 598	struct extent_status es;
 599	int retval;
 600	int ret = 0;
 601#ifdef ES_AGGRESSIVE_TEST
 602	struct ext4_map_blocks orig_map;
 603
 604	memcpy(&orig_map, map, sizeof(*map));
 605#endif
 606
 607	map->m_flags = 0;
 608	ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
 609		  flags, map->m_len, (unsigned long) map->m_lblk);
 610
 611	/*
 612	 * ext4_map_blocks returns an int, and m_len is an unsigned int
 613	 */
 614	if (unlikely(map->m_len > INT_MAX))
 615		map->m_len = INT_MAX;
 616
 617	/* We can handle the block number less than EXT_MAX_BLOCKS */
 618	if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
 619		return -EFSCORRUPTED;
 620
 621	/* Lookup extent status tree firstly */
 622	if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
 623	    ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
 624		if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
 625			map->m_pblk = ext4_es_pblock(&es) +
 626					map->m_lblk - es.es_lblk;
 627			map->m_flags |= ext4_es_is_written(&es) ?
 628					EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
 629			retval = es.es_len - (map->m_lblk - es.es_lblk);
 630			if (retval > map->m_len)
 631				retval = map->m_len;
 632			map->m_len = retval;
 633		} else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
 634			map->m_pblk = 0;
 635			map->m_flags |= ext4_es_is_delayed(&es) ?
 636					EXT4_MAP_DELAYED : 0;
 637			retval = es.es_len - (map->m_lblk - es.es_lblk);
 638			if (retval > map->m_len)
 639				retval = map->m_len;
 640			map->m_len = retval;
 641			retval = 0;
 642		} else {
 643			BUG();
 644		}
 645
 646		if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
 647			return retval;
 648#ifdef ES_AGGRESSIVE_TEST
 649		ext4_map_blocks_es_recheck(handle, inode, map,
 650					   &orig_map, flags);
 651#endif
 652		goto found;
 653	}
 654	/*
 655	 * In the query cache no-wait mode, nothing we can do more if we
 656	 * cannot find extent in the cache.
 657	 */
 658	if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
 659		return 0;
 660
 661	/*
 662	 * Try to see if we can get the block without requesting a new
 663	 * file system block.
 664	 */
 665	down_read(&EXT4_I(inode)->i_data_sem);
 666	retval = ext4_map_query_blocks(handle, inode, map);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 667	up_read((&EXT4_I(inode)->i_data_sem));
 668
 669found:
 670	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 671		ret = check_block_validity(inode, map);
 672		if (ret != 0)
 673			return ret;
 674	}
 675
 676	/* If it is only a block(s) look up */
 677	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 678		return retval;
 679
 680	/*
 681	 * Returns if the blocks have already allocated
 682	 *
 683	 * Note that if blocks have been preallocated
 684	 * ext4_ext_map_blocks() returns with buffer head unmapped
 
 685	 */
 686	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 687		/*
 688		 * If we need to convert extent to unwritten
 689		 * we continue and do the actual work in
 690		 * ext4_ext_map_blocks()
 691		 */
 692		if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
 693			return retval;
 694
 695	/*
 
 
 
 
 
 
 696	 * New blocks allocate and/or writing to unwritten extent
 697	 * will possibly result in updating i_data, so we take
 698	 * the write lock of i_data_sem, and call get_block()
 699	 * with create == 1 flag.
 700	 */
 701	down_write(&EXT4_I(inode)->i_data_sem);
 702	retval = ext4_map_create_blocks(handle, inode, map, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 703	up_write((&EXT4_I(inode)->i_data_sem));
 704	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 705		ret = check_block_validity(inode, map);
 706		if (ret != 0)
 707			return ret;
 708
 709		/*
 710		 * Inodes with freshly allocated blocks where contents will be
 711		 * visible after transaction commit must be on transaction's
 712		 * ordered data list.
 713		 */
 714		if (map->m_flags & EXT4_MAP_NEW &&
 715		    !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
 716		    !(flags & EXT4_GET_BLOCKS_ZERO) &&
 717		    !ext4_is_quota_file(inode) &&
 718		    ext4_should_order_data(inode)) {
 719			loff_t start_byte =
 720				(loff_t)map->m_lblk << inode->i_blkbits;
 721			loff_t length = (loff_t)map->m_len << inode->i_blkbits;
 722
 723			if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
 724				ret = ext4_jbd2_inode_add_wait(handle, inode,
 725						start_byte, length);
 726			else
 727				ret = ext4_jbd2_inode_add_write(handle, inode,
 728						start_byte, length);
 729			if (ret)
 730				return ret;
 731		}
 732	}
 733	if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
 734				map->m_flags & EXT4_MAP_MAPPED))
 735		ext4_fc_track_range(handle, inode, map->m_lblk,
 736					map->m_lblk + map->m_len - 1);
 737	if (retval < 0)
 738		ext_debug(inode, "failed with err %d\n", retval);
 739	return retval;
 740}
 741
 742/*
 743 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
 744 * we have to be careful as someone else may be manipulating b_state as well.
 745 */
 746static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
 747{
 748	unsigned long old_state;
 749	unsigned long new_state;
 750
 751	flags &= EXT4_MAP_FLAGS;
 752
 753	/* Dummy buffer_head? Set non-atomically. */
 754	if (!bh->b_page) {
 755		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
 756		return;
 757	}
 758	/*
 759	 * Someone else may be modifying b_state. Be careful! This is ugly but
 760	 * once we get rid of using bh as a container for mapping information
 761	 * to pass to / from get_block functions, this can go away.
 762	 */
 763	old_state = READ_ONCE(bh->b_state);
 764	do {
 
 765		new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
 766	} while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
 
 767}
 768
 769static int _ext4_get_block(struct inode *inode, sector_t iblock,
 770			   struct buffer_head *bh, int flags)
 771{
 772	struct ext4_map_blocks map;
 773	int ret = 0;
 774
 775	if (ext4_has_inline_data(inode))
 776		return -ERANGE;
 777
 778	map.m_lblk = iblock;
 779	map.m_len = bh->b_size >> inode->i_blkbits;
 780
 781	ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
 782			      flags);
 783	if (ret > 0) {
 784		map_bh(bh, inode->i_sb, map.m_pblk);
 785		ext4_update_bh_state(bh, map.m_flags);
 786		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 787		ret = 0;
 788	} else if (ret == 0) {
 789		/* hole case, need to fill in bh->b_size */
 790		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 791	}
 792	return ret;
 793}
 794
 795int ext4_get_block(struct inode *inode, sector_t iblock,
 796		   struct buffer_head *bh, int create)
 797{
 798	return _ext4_get_block(inode, iblock, bh,
 799			       create ? EXT4_GET_BLOCKS_CREATE : 0);
 800}
 801
 802/*
 803 * Get block function used when preparing for buffered write if we require
 804 * creating an unwritten extent if blocks haven't been allocated.  The extent
 805 * will be converted to written after the IO is complete.
 806 */
 807int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
 808			     struct buffer_head *bh_result, int create)
 809{
 810	int ret = 0;
 811
 812	ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
 813		   inode->i_ino, create);
 814	ret = _ext4_get_block(inode, iblock, bh_result,
 815			       EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
 816
 817	/*
 818	 * If the buffer is marked unwritten, mark it as new to make sure it is
 819	 * zeroed out correctly in case of partial writes. Otherwise, there is
 820	 * a chance of stale data getting exposed.
 821	 */
 822	if (ret == 0 && buffer_unwritten(bh_result))
 823		set_buffer_new(bh_result);
 824
 825	return ret;
 826}
 827
 828/* Maximum number of blocks we map for direct IO at once. */
 829#define DIO_MAX_BLOCKS 4096
 830
 831/*
 832 * `handle' can be NULL if create is zero
 833 */
 834struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 835				ext4_lblk_t block, int map_flags)
 836{
 837	struct ext4_map_blocks map;
 838	struct buffer_head *bh;
 839	int create = map_flags & EXT4_GET_BLOCKS_CREATE;
 840	bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
 841	int err;
 842
 843	ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 844		    || handle != NULL || create == 0);
 845	ASSERT(create == 0 || !nowait);
 846
 847	map.m_lblk = block;
 848	map.m_len = 1;
 849	err = ext4_map_blocks(handle, inode, &map, map_flags);
 850
 851	if (err == 0)
 852		return create ? ERR_PTR(-ENOSPC) : NULL;
 853	if (err < 0)
 854		return ERR_PTR(err);
 855
 856	if (nowait)
 857		return sb_find_get_block(inode->i_sb, map.m_pblk);
 858
 859	/*
 860	 * Since bh could introduce extra ref count such as referred by
 861	 * journal_head etc. Try to avoid using __GFP_MOVABLE here
 862	 * as it may fail the migration when journal_head remains.
 863	 */
 864	bh = getblk_unmovable(inode->i_sb->s_bdev, map.m_pblk,
 865				inode->i_sb->s_blocksize);
 866
 867	if (unlikely(!bh))
 868		return ERR_PTR(-ENOMEM);
 869	if (map.m_flags & EXT4_MAP_NEW) {
 870		ASSERT(create != 0);
 871		ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 872			    || (handle != NULL));
 873
 874		/*
 875		 * Now that we do not always journal data, we should
 876		 * keep in mind whether this should always journal the
 877		 * new buffer as metadata.  For now, regular file
 878		 * writes use ext4_get_block instead, so it's not a
 879		 * problem.
 880		 */
 881		lock_buffer(bh);
 882		BUFFER_TRACE(bh, "call get_create_access");
 883		err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
 884						     EXT4_JTR_NONE);
 885		if (unlikely(err)) {
 886			unlock_buffer(bh);
 887			goto errout;
 888		}
 889		if (!buffer_uptodate(bh)) {
 890			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 891			set_buffer_uptodate(bh);
 892		}
 893		unlock_buffer(bh);
 894		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 895		err = ext4_handle_dirty_metadata(handle, inode, bh);
 896		if (unlikely(err))
 897			goto errout;
 898	} else
 899		BUFFER_TRACE(bh, "not a new buffer");
 900	return bh;
 901errout:
 902	brelse(bh);
 903	return ERR_PTR(err);
 904}
 905
 906struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 907			       ext4_lblk_t block, int map_flags)
 908{
 909	struct buffer_head *bh;
 910	int ret;
 911
 912	bh = ext4_getblk(handle, inode, block, map_flags);
 913	if (IS_ERR(bh))
 914		return bh;
 915	if (!bh || ext4_buffer_uptodate(bh))
 916		return bh;
 917
 918	ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
 919	if (ret) {
 920		put_bh(bh);
 921		return ERR_PTR(ret);
 922	}
 923	return bh;
 924}
 925
 926/* Read a contiguous batch of blocks. */
 927int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
 928		     bool wait, struct buffer_head **bhs)
 929{
 930	int i, err;
 931
 932	for (i = 0; i < bh_count; i++) {
 933		bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
 934		if (IS_ERR(bhs[i])) {
 935			err = PTR_ERR(bhs[i]);
 936			bh_count = i;
 937			goto out_brelse;
 938		}
 939	}
 940
 941	for (i = 0; i < bh_count; i++)
 942		/* Note that NULL bhs[i] is valid because of holes. */
 943		if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
 944			ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
 945
 946	if (!wait)
 947		return 0;
 948
 949	for (i = 0; i < bh_count; i++)
 950		if (bhs[i])
 951			wait_on_buffer(bhs[i]);
 952
 953	for (i = 0; i < bh_count; i++) {
 954		if (bhs[i] && !buffer_uptodate(bhs[i])) {
 955			err = -EIO;
 956			goto out_brelse;
 957		}
 958	}
 959	return 0;
 960
 961out_brelse:
 962	for (i = 0; i < bh_count; i++) {
 963		brelse(bhs[i]);
 964		bhs[i] = NULL;
 965	}
 966	return err;
 967}
 968
 969int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
 970			   struct buffer_head *head,
 971			   unsigned from,
 972			   unsigned to,
 973			   int *partial,
 974			   int (*fn)(handle_t *handle, struct inode *inode,
 975				     struct buffer_head *bh))
 976{
 977	struct buffer_head *bh;
 978	unsigned block_start, block_end;
 979	unsigned blocksize = head->b_size;
 980	int err, ret = 0;
 981	struct buffer_head *next;
 982
 983	for (bh = head, block_start = 0;
 984	     ret == 0 && (bh != head || !block_start);
 985	     block_start = block_end, bh = next) {
 986		next = bh->b_this_page;
 987		block_end = block_start + blocksize;
 988		if (block_end <= from || block_start >= to) {
 989			if (partial && !buffer_uptodate(bh))
 990				*partial = 1;
 991			continue;
 992		}
 993		err = (*fn)(handle, inode, bh);
 994		if (!ret)
 995			ret = err;
 996	}
 997	return ret;
 998}
 999
1000/*
1001 * Helper for handling dirtying of journalled data. We also mark the folio as
1002 * dirty so that writeback code knows about this page (and inode) contains
1003 * dirty data. ext4_writepages() then commits appropriate transaction to
1004 * make data stable.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1005 */
1006static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
1007{
1008	folio_mark_dirty(bh->b_folio);
1009	return ext4_handle_dirty_metadata(handle, NULL, bh);
1010}
1011
1012int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1013				struct buffer_head *bh)
1014{
 
 
 
1015	if (!buffer_mapped(bh) || buffer_freed(bh))
1016		return 0;
 
 
 
 
 
 
 
 
 
 
1017	BUFFER_TRACE(bh, "get write access");
1018	return ext4_journal_get_write_access(handle, inode->i_sb, bh,
1019					    EXT4_JTR_NONE);
 
 
 
1020}
1021
1022int ext4_block_write_begin(handle_t *handle, struct folio *folio,
1023			   loff_t pos, unsigned len,
1024			   get_block_t *get_block)
1025{
1026	unsigned from = pos & (PAGE_SIZE - 1);
1027	unsigned to = from + len;
1028	struct inode *inode = folio->mapping->host;
1029	unsigned block_start, block_end;
1030	sector_t block;
1031	int err = 0;
1032	unsigned blocksize = inode->i_sb->s_blocksize;
1033	unsigned bbits;
1034	struct buffer_head *bh, *head, *wait[2];
1035	int nr_wait = 0;
1036	int i;
1037	bool should_journal_data = ext4_should_journal_data(inode);
1038
1039	BUG_ON(!folio_test_locked(folio));
1040	BUG_ON(from > PAGE_SIZE);
1041	BUG_ON(to > PAGE_SIZE);
1042	BUG_ON(from > to);
1043
1044	head = folio_buffers(folio);
1045	if (!head)
1046		head = create_empty_buffers(folio, blocksize, 0);
1047	bbits = ilog2(blocksize);
1048	block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1049
1050	for (bh = head, block_start = 0; bh != head || !block_start;
1051	    block++, block_start = block_end, bh = bh->b_this_page) {
1052		block_end = block_start + blocksize;
1053		if (block_end <= from || block_start >= to) {
1054			if (folio_test_uptodate(folio)) {
1055				set_buffer_uptodate(bh);
1056			}
1057			continue;
1058		}
1059		if (buffer_new(bh))
1060			clear_buffer_new(bh);
1061		if (!buffer_mapped(bh)) {
1062			WARN_ON(bh->b_size != blocksize);
1063			err = get_block(inode, block, bh, 1);
1064			if (err)
1065				break;
1066			if (buffer_new(bh)) {
1067				/*
1068				 * We may be zeroing partial buffers or all new
1069				 * buffers in case of failure. Prepare JBD2 for
1070				 * that.
1071				 */
1072				if (should_journal_data)
1073					do_journal_get_write_access(handle,
1074								    inode, bh);
1075				if (folio_test_uptodate(folio)) {
1076					/*
1077					 * Unlike __block_write_begin() we leave
1078					 * dirtying of new uptodate buffers to
1079					 * ->write_end() time or
1080					 * folio_zero_new_buffers().
1081					 */
1082					set_buffer_uptodate(bh);
 
1083					continue;
1084				}
1085				if (block_end > to || block_start < from)
1086					folio_zero_segments(folio, to,
1087							    block_end,
1088							    block_start, from);
1089				continue;
1090			}
1091		}
1092		if (folio_test_uptodate(folio)) {
1093			set_buffer_uptodate(bh);
1094			continue;
1095		}
1096		if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1097		    !buffer_unwritten(bh) &&
1098		    (block_start < from || block_end > to)) {
1099			ext4_read_bh_lock(bh, 0, false);
1100			wait[nr_wait++] = bh;
1101		}
1102	}
1103	/*
1104	 * If we issued read requests, let them complete.
1105	 */
1106	for (i = 0; i < nr_wait; i++) {
1107		wait_on_buffer(wait[i]);
1108		if (!buffer_uptodate(wait[i]))
1109			err = -EIO;
1110	}
1111	if (unlikely(err)) {
1112		if (should_journal_data)
1113			ext4_journalled_zero_new_buffers(handle, inode, folio,
1114							 from, to);
1115		else
1116			folio_zero_new_buffers(folio, from, to);
1117	} else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1118		for (i = 0; i < nr_wait; i++) {
1119			int err2;
1120
1121			err2 = fscrypt_decrypt_pagecache_blocks(folio,
1122						blocksize, bh_offset(wait[i]));
1123			if (err2) {
1124				clear_buffer_uptodate(wait[i]);
1125				err = err2;
1126			}
1127		}
1128	}
1129
1130	return err;
1131}
 
1132
1133/*
1134 * To preserve ordering, it is essential that the hole instantiation and
1135 * the data write be encapsulated in a single transaction.  We cannot
1136 * close off a transaction and start a new one between the ext4_get_block()
1137 * and the ext4_write_end().  So doing the jbd2_journal_start at the start of
1138 * ext4_write_begin() is the right place.
1139 */
1140static int ext4_write_begin(struct file *file, struct address_space *mapping,
1141			    loff_t pos, unsigned len,
1142			    struct folio **foliop, void **fsdata)
1143{
1144	struct inode *inode = mapping->host;
1145	int ret, needed_blocks;
1146	handle_t *handle;
1147	int retries = 0;
1148	struct folio *folio;
1149	pgoff_t index;
1150	unsigned from, to;
1151
1152	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
1153		return -EIO;
1154
1155	trace_ext4_write_begin(inode, pos, len);
1156	/*
1157	 * Reserve one block more for addition to orphan list in case
1158	 * we allocate blocks but write fails for some reason
1159	 */
1160	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1161	index = pos >> PAGE_SHIFT;
1162	from = pos & (PAGE_SIZE - 1);
1163	to = from + len;
1164
1165	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1166		ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1167						    foliop);
1168		if (ret < 0)
1169			return ret;
1170		if (ret == 1)
1171			return 0;
1172	}
1173
1174	/*
1175	 * __filemap_get_folio() can take a long time if the
1176	 * system is thrashing due to memory pressure, or if the folio
1177	 * is being written back.  So grab it first before we start
1178	 * the transaction handle.  This also allows us to allocate
1179	 * the folio (if needed) without using GFP_NOFS.
1180	 */
1181retry_grab:
1182	folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
1183					mapping_gfp_mask(mapping));
1184	if (IS_ERR(folio))
1185		return PTR_ERR(folio);
1186	/*
1187	 * The same as page allocation, we prealloc buffer heads before
1188	 * starting the handle.
1189	 */
1190	if (!folio_buffers(folio))
1191		create_empty_buffers(folio, inode->i_sb->s_blocksize, 0);
1192
1193	folio_unlock(folio);
1194
1195retry_journal:
1196	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1197	if (IS_ERR(handle)) {
1198		folio_put(folio);
1199		return PTR_ERR(handle);
1200	}
1201
1202	folio_lock(folio);
1203	if (folio->mapping != mapping) {
1204		/* The folio got truncated from under us */
1205		folio_unlock(folio);
1206		folio_put(folio);
1207		ext4_journal_stop(handle);
1208		goto retry_grab;
1209	}
1210	/* In case writeback began while the folio was unlocked */
1211	folio_wait_stable(folio);
1212
 
1213	if (ext4_should_dioread_nolock(inode))
1214		ret = ext4_block_write_begin(handle, folio, pos, len,
1215					     ext4_get_block_unwritten);
1216	else
1217		ret = ext4_block_write_begin(handle, folio, pos, len,
1218					     ext4_get_block);
 
 
 
 
 
 
 
1219	if (!ret && ext4_should_journal_data(inode)) {
1220		ret = ext4_walk_page_buffers(handle, inode,
1221					     folio_buffers(folio), from, to,
1222					     NULL, do_journal_get_write_access);
1223	}
1224
1225	if (ret) {
1226		bool extended = (pos + len > inode->i_size) &&
1227				!ext4_verity_in_progress(inode);
1228
1229		folio_unlock(folio);
1230		/*
1231		 * ext4_block_write_begin may have instantiated a few blocks
1232		 * outside i_size.  Trim these off again. Don't need
1233		 * i_size_read because we hold i_rwsem.
1234		 *
1235		 * Add inode to orphan list in case we crash before
1236		 * truncate finishes
1237		 */
1238		if (extended && ext4_can_truncate(inode))
1239			ext4_orphan_add(handle, inode);
1240
1241		ext4_journal_stop(handle);
1242		if (extended) {
1243			ext4_truncate_failed_write(inode);
1244			/*
1245			 * If truncate failed early the inode might
1246			 * still be on the orphan list; we need to
1247			 * make sure the inode is removed from the
1248			 * orphan list in that case.
1249			 */
1250			if (inode->i_nlink)
1251				ext4_orphan_del(NULL, inode);
1252		}
1253
1254		if (ret == -ENOSPC &&
1255		    ext4_should_retry_alloc(inode->i_sb, &retries))
1256			goto retry_journal;
1257		folio_put(folio);
1258		return ret;
1259	}
1260	*foliop = folio;
1261	return ret;
1262}
1263
1264/* For write_end() in data=journal mode */
1265static int write_end_fn(handle_t *handle, struct inode *inode,
1266			struct buffer_head *bh)
1267{
1268	int ret;
1269	if (!buffer_mapped(bh) || buffer_freed(bh))
1270		return 0;
1271	set_buffer_uptodate(bh);
1272	ret = ext4_dirty_journalled_data(handle, bh);
1273	clear_buffer_meta(bh);
1274	clear_buffer_prio(bh);
1275	return ret;
1276}
1277
1278/*
1279 * We need to pick up the new inode size which generic_commit_write gave us
1280 * `file' can be NULL - eg, when called from page_symlink().
1281 *
1282 * ext4 never places buffers on inode->i_mapping->i_private_list.  metadata
1283 * buffers are managed internally.
1284 */
1285static int ext4_write_end(struct file *file,
1286			  struct address_space *mapping,
1287			  loff_t pos, unsigned len, unsigned copied,
1288			  struct folio *folio, void *fsdata)
1289{
1290	handle_t *handle = ext4_journal_current_handle();
1291	struct inode *inode = mapping->host;
1292	loff_t old_size = inode->i_size;
1293	int ret = 0, ret2;
1294	int i_size_changed = 0;
1295	bool verity = ext4_verity_in_progress(inode);
1296
1297	trace_ext4_write_end(inode, pos, len, copied);
1298
1299	if (ext4_has_inline_data(inode) &&
1300	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1301		return ext4_write_inline_data_end(inode, pos, len, copied,
1302						  folio);
1303
1304	copied = block_write_end(file, mapping, pos, len, copied, folio, fsdata);
1305	/*
1306	 * it's important to update i_size while still holding folio lock:
1307	 * page writeout could otherwise come in and zero beyond i_size.
1308	 *
1309	 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1310	 * blocks are being written past EOF, so skip the i_size update.
1311	 */
1312	if (!verity)
1313		i_size_changed = ext4_update_inode_size(inode, pos + copied);
1314	folio_unlock(folio);
1315	folio_put(folio);
1316
1317	if (old_size < pos && !verity) {
1318		pagecache_isize_extended(inode, old_size, pos);
1319		ext4_zero_partial_blocks(handle, inode, old_size, pos - old_size);
1320	}
1321	/*
1322	 * Don't mark the inode dirty under folio lock. First, it unnecessarily
1323	 * makes the holding time of folio lock longer. Second, it forces lock
1324	 * ordering of folio lock and transaction start for journaling
1325	 * filesystems.
1326	 */
1327	if (i_size_changed)
1328		ret = ext4_mark_inode_dirty(handle, inode);
1329
1330	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1331		/* if we have allocated more blocks and copied
1332		 * less. We will have blocks allocated outside
1333		 * inode->i_size. So truncate them
1334		 */
1335		ext4_orphan_add(handle, inode);
1336
1337	ret2 = ext4_journal_stop(handle);
1338	if (!ret)
1339		ret = ret2;
1340
1341	if (pos + len > inode->i_size && !verity) {
1342		ext4_truncate_failed_write(inode);
1343		/*
1344		 * If truncate failed early the inode might still be
1345		 * on the orphan list; we need to make sure the inode
1346		 * is removed from the orphan list in that case.
1347		 */
1348		if (inode->i_nlink)
1349			ext4_orphan_del(NULL, inode);
1350	}
1351
1352	return ret ? ret : copied;
1353}
1354
1355/*
1356 * This is a private version of folio_zero_new_buffers() which doesn't
1357 * set the buffer to be dirty, since in data=journalled mode we need
1358 * to call ext4_dirty_journalled_data() instead.
1359 */
1360static void ext4_journalled_zero_new_buffers(handle_t *handle,
1361					    struct inode *inode,
1362					    struct folio *folio,
1363					    unsigned from, unsigned to)
1364{
1365	unsigned int block_start = 0, block_end;
1366	struct buffer_head *head, *bh;
1367
1368	bh = head = folio_buffers(folio);
1369	do {
1370		block_end = block_start + bh->b_size;
1371		if (buffer_new(bh)) {
1372			if (block_end > from && block_start < to) {
1373				if (!folio_test_uptodate(folio)) {
1374					unsigned start, size;
1375
1376					start = max(from, block_start);
1377					size = min(to, block_end) - start;
1378
1379					folio_zero_range(folio, start, size);
 
1380				}
1381				clear_buffer_new(bh);
1382				write_end_fn(handle, inode, bh);
1383			}
1384		}
1385		block_start = block_end;
1386		bh = bh->b_this_page;
1387	} while (bh != head);
1388}
1389
1390static int ext4_journalled_write_end(struct file *file,
1391				     struct address_space *mapping,
1392				     loff_t pos, unsigned len, unsigned copied,
1393				     struct folio *folio, void *fsdata)
1394{
1395	handle_t *handle = ext4_journal_current_handle();
1396	struct inode *inode = mapping->host;
1397	loff_t old_size = inode->i_size;
1398	int ret = 0, ret2;
1399	int partial = 0;
1400	unsigned from, to;
1401	int size_changed = 0;
1402	bool verity = ext4_verity_in_progress(inode);
1403
1404	trace_ext4_journalled_write_end(inode, pos, len, copied);
1405	from = pos & (PAGE_SIZE - 1);
1406	to = from + len;
1407
1408	BUG_ON(!ext4_handle_valid(handle));
1409
1410	if (ext4_has_inline_data(inode))
1411		return ext4_write_inline_data_end(inode, pos, len, copied,
1412						  folio);
1413
1414	if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1415		copied = 0;
1416		ext4_journalled_zero_new_buffers(handle, inode, folio,
1417						 from, to);
1418	} else {
1419		if (unlikely(copied < len))
1420			ext4_journalled_zero_new_buffers(handle, inode, folio,
1421							 from + copied, to);
1422		ret = ext4_walk_page_buffers(handle, inode,
1423					     folio_buffers(folio),
1424					     from, from + copied, &partial,
1425					     write_end_fn);
1426		if (!partial)
1427			folio_mark_uptodate(folio);
1428	}
1429	if (!verity)
1430		size_changed = ext4_update_inode_size(inode, pos + copied);
 
1431	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1432	folio_unlock(folio);
1433	folio_put(folio);
1434
1435	if (old_size < pos && !verity) {
1436		pagecache_isize_extended(inode, old_size, pos);
1437		ext4_zero_partial_blocks(handle, inode, old_size, pos - old_size);
1438	}
1439
1440	if (size_changed) {
1441		ret2 = ext4_mark_inode_dirty(handle, inode);
1442		if (!ret)
1443			ret = ret2;
1444	}
1445
1446	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1447		/* if we have allocated more blocks and copied
1448		 * less. We will have blocks allocated outside
1449		 * inode->i_size. So truncate them
1450		 */
1451		ext4_orphan_add(handle, inode);
1452
1453	ret2 = ext4_journal_stop(handle);
1454	if (!ret)
1455		ret = ret2;
1456	if (pos + len > inode->i_size && !verity) {
1457		ext4_truncate_failed_write(inode);
1458		/*
1459		 * If truncate failed early the inode might still be
1460		 * on the orphan list; we need to make sure the inode
1461		 * is removed from the orphan list in that case.
1462		 */
1463		if (inode->i_nlink)
1464			ext4_orphan_del(NULL, inode);
1465	}
1466
1467	return ret ? ret : copied;
1468}
1469
1470/*
1471 * Reserve space for 'nr_resv' clusters
1472 */
1473static int ext4_da_reserve_space(struct inode *inode, int nr_resv)
1474{
1475	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1476	struct ext4_inode_info *ei = EXT4_I(inode);
1477	int ret;
1478
1479	/*
1480	 * We will charge metadata quota at writeout time; this saves
1481	 * us from metadata over-estimation, though we may go over by
1482	 * a small amount in the end.  Here we just reserve for data.
1483	 */
1484	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, nr_resv));
1485	if (ret)
1486		return ret;
1487
1488	spin_lock(&ei->i_block_reservation_lock);
1489	if (ext4_claim_free_clusters(sbi, nr_resv, 0)) {
1490		spin_unlock(&ei->i_block_reservation_lock);
1491		dquot_release_reservation_block(inode, EXT4_C2B(sbi, nr_resv));
1492		return -ENOSPC;
1493	}
1494	ei->i_reserved_data_blocks += nr_resv;
1495	trace_ext4_da_reserve_space(inode, nr_resv);
1496	spin_unlock(&ei->i_block_reservation_lock);
1497
1498	return 0;       /* success */
1499}
1500
1501void ext4_da_release_space(struct inode *inode, int to_free)
1502{
1503	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1504	struct ext4_inode_info *ei = EXT4_I(inode);
1505
1506	if (!to_free)
1507		return;		/* Nothing to release, exit */
1508
1509	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1510
1511	trace_ext4_da_release_space(inode, to_free);
1512	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1513		/*
1514		 * if there aren't enough reserved blocks, then the
1515		 * counter is messed up somewhere.  Since this
1516		 * function is called from invalidate page, it's
1517		 * harmless to return without any action.
1518		 */
1519		ext4_warning(inode->i_sb, "ext4_da_release_space: "
1520			 "ino %lu, to_free %d with only %d reserved "
1521			 "data blocks", inode->i_ino, to_free,
1522			 ei->i_reserved_data_blocks);
1523		WARN_ON(1);
1524		to_free = ei->i_reserved_data_blocks;
1525	}
1526	ei->i_reserved_data_blocks -= to_free;
1527
1528	/* update fs dirty data blocks counter */
1529	percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1530
1531	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1532
1533	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1534}
1535
1536/*
1537 * Delayed allocation stuff
1538 */
1539
1540struct mpage_da_data {
1541	/* These are input fields for ext4_do_writepages() */
1542	struct inode *inode;
1543	struct writeback_control *wbc;
1544	unsigned int can_map:1;	/* Can writepages call map blocks? */
1545
1546	/* These are internal state of ext4_do_writepages() */
1547	pgoff_t first_page;	/* The first page to write */
1548	pgoff_t next_page;	/* Current page to examine */
1549	pgoff_t last_page;	/* Last page to examine */
1550	/*
1551	 * Extent to map - this can be after first_page because that can be
1552	 * fully mapped. We somewhat abuse m_flags to store whether the extent
1553	 * is delalloc or unwritten.
1554	 */
1555	struct ext4_map_blocks map;
1556	struct ext4_io_submit io_submit;	/* IO submission data */
1557	unsigned int do_map:1;
1558	unsigned int scanned_until_end:1;
1559	unsigned int journalled_more_data:1;
1560};
1561
1562static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1563				       bool invalidate)
1564{
1565	unsigned nr, i;
1566	pgoff_t index, end;
1567	struct folio_batch fbatch;
1568	struct inode *inode = mpd->inode;
1569	struct address_space *mapping = inode->i_mapping;
1570
1571	/* This is necessary when next_page == 0. */
1572	if (mpd->first_page >= mpd->next_page)
1573		return;
1574
1575	mpd->scanned_until_end = 0;
1576	index = mpd->first_page;
1577	end   = mpd->next_page - 1;
1578	if (invalidate) {
1579		ext4_lblk_t start, last;
1580		start = index << (PAGE_SHIFT - inode->i_blkbits);
1581		last = end << (PAGE_SHIFT - inode->i_blkbits);
1582
1583		/*
1584		 * avoid racing with extent status tree scans made by
1585		 * ext4_insert_delayed_block()
1586		 */
1587		down_write(&EXT4_I(inode)->i_data_sem);
1588		ext4_es_remove_extent(inode, start, last - start + 1);
1589		up_write(&EXT4_I(inode)->i_data_sem);
1590	}
1591
1592	folio_batch_init(&fbatch);
1593	while (index <= end) {
1594		nr = filemap_get_folios(mapping, &index, end, &fbatch);
1595		if (nr == 0)
1596			break;
1597		for (i = 0; i < nr; i++) {
1598			struct folio *folio = fbatch.folios[i];
1599
1600			if (folio->index < mpd->first_page)
1601				continue;
1602			if (folio_next_index(folio) - 1 > end)
1603				continue;
1604			BUG_ON(!folio_test_locked(folio));
1605			BUG_ON(folio_test_writeback(folio));
1606			if (invalidate) {
1607				if (folio_mapped(folio))
1608					folio_clear_dirty_for_io(folio);
1609				block_invalidate_folio(folio, 0,
1610						folio_size(folio));
1611				folio_clear_uptodate(folio);
1612			}
1613			folio_unlock(folio);
1614		}
1615		folio_batch_release(&fbatch);
1616	}
1617}
1618
1619static void ext4_print_free_blocks(struct inode *inode)
1620{
1621	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1622	struct super_block *sb = inode->i_sb;
1623	struct ext4_inode_info *ei = EXT4_I(inode);
1624
1625	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1626	       EXT4_C2B(EXT4_SB(inode->i_sb),
1627			ext4_count_free_clusters(sb)));
1628	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1629	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1630	       (long long) EXT4_C2B(EXT4_SB(sb),
1631		percpu_counter_sum(&sbi->s_freeclusters_counter)));
1632	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1633	       (long long) EXT4_C2B(EXT4_SB(sb),
1634		percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1635	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1636	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1637		 ei->i_reserved_data_blocks);
1638	return;
1639}
1640
1641/*
1642 * Check whether the cluster containing lblk has been allocated or has
1643 * delalloc reservation.
1644 *
1645 * Returns 0 if the cluster doesn't have either, 1 if it has delalloc
1646 * reservation, 2 if it's already been allocated, negative error code on
1647 * failure.
1648 */
1649static int ext4_clu_alloc_state(struct inode *inode, ext4_lblk_t lblk)
1650{
1651	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1652	int ret;
1653
1654	/* Has delalloc reservation? */
1655	if (ext4_es_scan_clu(inode, &ext4_es_is_delayed, lblk))
1656		return 1;
1657
1658	/* Already been allocated? */
1659	if (ext4_es_scan_clu(inode, &ext4_es_is_mapped, lblk))
1660		return 2;
1661	ret = ext4_clu_mapped(inode, EXT4_B2C(sbi, lblk));
1662	if (ret < 0)
1663		return ret;
1664	if (ret > 0)
1665		return 2;
1666
1667	return 0;
1668}
1669
1670/*
1671 * ext4_insert_delayed_blocks - adds a multiple delayed blocks to the extents
1672 *                              status tree, incrementing the reserved
1673 *                              cluster/block count or making pending
1674 *                              reservations where needed
1675 *
1676 * @inode - file containing the newly added block
1677 * @lblk - start logical block to be added
1678 * @len - length of blocks to be added
1679 *
1680 * Returns 0 on success, negative error code on failure.
1681 */
1682static int ext4_insert_delayed_blocks(struct inode *inode, ext4_lblk_t lblk,
1683				      ext4_lblk_t len)
1684{
1685	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1686	int ret;
1687	bool lclu_allocated = false;
1688	bool end_allocated = false;
1689	ext4_lblk_t resv_clu;
1690	ext4_lblk_t end = lblk + len - 1;
1691
1692	/*
1693	 * If the cluster containing lblk or end is shared with a delayed,
1694	 * written, or unwritten extent in a bigalloc file system, it's
1695	 * already been accounted for and does not need to be reserved.
1696	 * A pending reservation must be made for the cluster if it's
1697	 * shared with a written or unwritten extent and doesn't already
1698	 * have one.  Written and unwritten extents can be purged from the
1699	 * extents status tree if the system is under memory pressure, so
1700	 * it's necessary to examine the extent tree if a search of the
1701	 * extents status tree doesn't get a match.
1702	 */
1703	if (sbi->s_cluster_ratio == 1) {
1704		ret = ext4_da_reserve_space(inode, len);
1705		if (ret != 0)   /* ENOSPC */
1706			return ret;
 
1707	} else {   /* bigalloc */
1708		resv_clu = EXT4_B2C(sbi, end) - EXT4_B2C(sbi, lblk) + 1;
1709
1710		ret = ext4_clu_alloc_state(inode, lblk);
1711		if (ret < 0)
1712			return ret;
1713		if (ret > 0) {
1714			resv_clu--;
1715			lclu_allocated = (ret == 2);
1716		}
1717
1718		if (EXT4_B2C(sbi, lblk) != EXT4_B2C(sbi, end)) {
1719			ret = ext4_clu_alloc_state(inode, end);
1720			if (ret < 0)
1721				return ret;
1722			if (ret > 0) {
1723				resv_clu--;
1724				end_allocated = (ret == 2);
1725			}
1726		}
1727
1728		if (resv_clu) {
1729			ret = ext4_da_reserve_space(inode, resv_clu);
1730			if (ret != 0)   /* ENOSPC */
1731				return ret;
1732		}
1733	}
1734
1735	ext4_es_insert_delayed_extent(inode, lblk, len, lclu_allocated,
1736				      end_allocated);
1737	return 0;
 
 
 
1738}
1739
1740/*
1741 * Looks up the requested blocks and sets the delalloc extent map.
1742 * First try to look up for the extent entry that contains the requested
1743 * blocks in the extent status tree without i_data_sem, then try to look
1744 * up for the ondisk extent mapping with i_data_sem in read mode,
1745 * finally hold i_data_sem in write mode, looks up again and add a
1746 * delalloc extent entry if it still couldn't find any extent. Pass out
1747 * the mapped extent through @map and return 0 on success.
1748 */
1749static int ext4_da_map_blocks(struct inode *inode, struct ext4_map_blocks *map)
1750{
1751	struct extent_status es;
1752	int retval;
 
1753#ifdef ES_AGGRESSIVE_TEST
1754	struct ext4_map_blocks orig_map;
1755
1756	memcpy(&orig_map, map, sizeof(*map));
1757#endif
1758
 
 
 
1759	map->m_flags = 0;
1760	ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1761		  (unsigned long) map->m_lblk);
1762
1763	/* Lookup extent status tree firstly */
1764	if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
1765		map->m_len = min_t(unsigned int, map->m_len,
1766				   es.es_len - (map->m_lblk - es.es_lblk));
1767
1768		if (ext4_es_is_hole(&es))
1769			goto add_delayed;
 
1770
1771found:
1772		/*
1773		 * Delayed extent could be allocated by fallocate.
1774		 * So we need to check it.
1775		 */
1776		if (ext4_es_is_delayed(&es)) {
1777			map->m_flags |= EXT4_MAP_DELAYED;
 
 
1778			return 0;
1779		}
1780
1781		map->m_pblk = ext4_es_pblock(&es) + map->m_lblk - es.es_lblk;
 
 
 
 
1782		if (ext4_es_is_written(&es))
1783			map->m_flags |= EXT4_MAP_MAPPED;
1784		else if (ext4_es_is_unwritten(&es))
1785			map->m_flags |= EXT4_MAP_UNWRITTEN;
1786		else
1787			BUG();
1788
1789#ifdef ES_AGGRESSIVE_TEST
1790		ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1791#endif
1792		return 0;
1793	}
1794
1795	/*
1796	 * Try to see if we can get the block without requesting a new
1797	 * file system block.
1798	 */
1799	down_read(&EXT4_I(inode)->i_data_sem);
1800	if (ext4_has_inline_data(inode))
1801		retval = 0;
 
 
1802	else
1803		retval = ext4_map_query_blocks(NULL, inode, map);
1804	up_read(&EXT4_I(inode)->i_data_sem);
1805	if (retval)
1806		return retval < 0 ? retval : 0;
1807
1808add_delayed:
1809	down_write(&EXT4_I(inode)->i_data_sem);
1810	/*
1811	 * Page fault path (ext4_page_mkwrite does not take i_rwsem)
1812	 * and fallocate path (no folio lock) can race. Make sure we
1813	 * lookup the extent status tree here again while i_data_sem
1814	 * is held in write mode, before inserting a new da entry in
1815	 * the extent status tree.
1816	 */
1817	if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
1818		map->m_len = min_t(unsigned int, map->m_len,
1819				   es.es_len - (map->m_lblk - es.es_lblk));
1820
1821		if (!ext4_es_is_hole(&es)) {
1822			up_write(&EXT4_I(inode)->i_data_sem);
1823			goto found;
1824		}
1825	} else if (!ext4_has_inline_data(inode)) {
1826		retval = ext4_map_query_blocks(NULL, inode, map);
1827		if (retval) {
1828			up_write(&EXT4_I(inode)->i_data_sem);
1829			return retval < 0 ? retval : 0;
1830		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1831	}
1832
1833	map->m_flags |= EXT4_MAP_DELAYED;
1834	retval = ext4_insert_delayed_blocks(inode, map->m_lblk, map->m_len);
1835	up_write(&EXT4_I(inode)->i_data_sem);
1836
1837	return retval;
1838}
1839
1840/*
1841 * This is a special get_block_t callback which is used by
1842 * ext4_da_write_begin().  It will either return mapped block or
1843 * reserve space for a single block.
1844 *
1845 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1846 * We also have b_blocknr = -1 and b_bdev initialized properly
1847 *
1848 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1849 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1850 * initialized properly.
1851 */
1852int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1853			   struct buffer_head *bh, int create)
1854{
1855	struct ext4_map_blocks map;
1856	sector_t invalid_block = ~((sector_t) 0xffff);
1857	int ret = 0;
1858
1859	BUG_ON(create == 0);
1860	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1861
1862	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1863		invalid_block = ~0;
1864
1865	map.m_lblk = iblock;
1866	map.m_len = 1;
1867
1868	/*
1869	 * first, we need to know whether the block is allocated already
1870	 * preallocated blocks are unmapped but should treated
1871	 * the same as allocated blocks.
1872	 */
1873	ret = ext4_da_map_blocks(inode, &map);
1874	if (ret < 0)
1875		return ret;
1876
1877	if (map.m_flags & EXT4_MAP_DELAYED) {
1878		map_bh(bh, inode->i_sb, invalid_block);
1879		set_buffer_new(bh);
1880		set_buffer_delay(bh);
1881		return 0;
1882	}
1883
1884	map_bh(bh, inode->i_sb, map.m_pblk);
1885	ext4_update_bh_state(bh, map.m_flags);
1886
1887	if (buffer_unwritten(bh)) {
1888		/* A delayed write to unwritten bh should be marked
1889		 * new and mapped.  Mapped ensures that we don't do
1890		 * get_block multiple times when we write to the same
1891		 * offset and new ensures that we do proper zero out
1892		 * for partial write.
1893		 */
1894		set_buffer_new(bh);
1895		set_buffer_mapped(bh);
1896	}
1897	return 0;
1898}
1899
1900static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
 
1901{
1902	mpd->first_page += folio_nr_pages(folio);
1903	folio_unlock(folio);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1904}
1905
1906static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
1907{
1908	size_t len;
1909	loff_t size;
1910	int err;
1911
1912	BUG_ON(folio->index != mpd->first_page);
1913	folio_clear_dirty_for_io(folio);
1914	/*
1915	 * We have to be very careful here!  Nothing protects writeback path
1916	 * against i_size changes and the page can be writeably mapped into
1917	 * page tables. So an application can be growing i_size and writing
1918	 * data through mmap while writeback runs. folio_clear_dirty_for_io()
1919	 * write-protects our page in page tables and the page cannot get
1920	 * written to again until we release folio lock. So only after
1921	 * folio_clear_dirty_for_io() we are safe to sample i_size for
1922	 * ext4_bio_write_folio() to zero-out tail of the written page. We rely
1923	 * on the barrier provided by folio_test_clear_dirty() in
1924	 * folio_clear_dirty_for_io() to make sure i_size is really sampled only
1925	 * after page tables are updated.
1926	 */
1927	size = i_size_read(mpd->inode);
1928	len = folio_size(folio);
1929	if (folio_pos(folio) + len > size &&
1930	    !ext4_verity_in_progress(mpd->inode))
1931		len = size & (len - 1);
1932	err = ext4_bio_write_folio(&mpd->io_submit, folio, len);
 
 
1933	if (!err)
1934		mpd->wbc->nr_to_write--;
 
1935
1936	return err;
1937}
1938
1939#define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
1940
1941/*
1942 * mballoc gives us at most this number of blocks...
1943 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1944 * The rest of mballoc seems to handle chunks up to full group size.
1945 */
1946#define MAX_WRITEPAGES_EXTENT_LEN 2048
1947
1948/*
1949 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1950 *
1951 * @mpd - extent of blocks
1952 * @lblk - logical number of the block in the file
1953 * @bh - buffer head we want to add to the extent
1954 *
1955 * The function is used to collect contig. blocks in the same state. If the
1956 * buffer doesn't require mapping for writeback and we haven't started the
1957 * extent of buffers to map yet, the function returns 'true' immediately - the
1958 * caller can write the buffer right away. Otherwise the function returns true
1959 * if the block has been added to the extent, false if the block couldn't be
1960 * added.
1961 */
1962static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1963				   struct buffer_head *bh)
1964{
1965	struct ext4_map_blocks *map = &mpd->map;
1966
1967	/* Buffer that doesn't need mapping for writeback? */
1968	if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1969	    (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1970		/* So far no extent to map => we write the buffer right away */
1971		if (map->m_len == 0)
1972			return true;
1973		return false;
1974	}
1975
1976	/* First block in the extent? */
1977	if (map->m_len == 0) {
1978		/* We cannot map unless handle is started... */
1979		if (!mpd->do_map)
1980			return false;
1981		map->m_lblk = lblk;
1982		map->m_len = 1;
1983		map->m_flags = bh->b_state & BH_FLAGS;
1984		return true;
1985	}
1986
1987	/* Don't go larger than mballoc is willing to allocate */
1988	if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1989		return false;
1990
1991	/* Can we merge the block to our big extent? */
1992	if (lblk == map->m_lblk + map->m_len &&
1993	    (bh->b_state & BH_FLAGS) == map->m_flags) {
1994		map->m_len++;
1995		return true;
1996	}
1997	return false;
1998}
1999
2000/*
2001 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2002 *
2003 * @mpd - extent of blocks for mapping
2004 * @head - the first buffer in the page
2005 * @bh - buffer we should start processing from
2006 * @lblk - logical number of the block in the file corresponding to @bh
2007 *
2008 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2009 * the page for IO if all buffers in this page were mapped and there's no
2010 * accumulated extent of buffers to map or add buffers in the page to the
2011 * extent of buffers to map. The function returns 1 if the caller can continue
2012 * by processing the next page, 0 if it should stop adding buffers to the
2013 * extent to map because we cannot extend it anymore. It can also return value
2014 * < 0 in case of error during IO submission.
2015 */
2016static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2017				   struct buffer_head *head,
2018				   struct buffer_head *bh,
2019				   ext4_lblk_t lblk)
2020{
2021	struct inode *inode = mpd->inode;
2022	int err;
2023	ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2024							>> inode->i_blkbits;
2025
2026	if (ext4_verity_in_progress(inode))
2027		blocks = EXT_MAX_BLOCKS;
2028
2029	do {
2030		BUG_ON(buffer_locked(bh));
2031
2032		if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2033			/* Found extent to map? */
2034			if (mpd->map.m_len)
2035				return 0;
2036			/* Buffer needs mapping and handle is not started? */
2037			if (!mpd->do_map)
2038				return 0;
2039			/* Everything mapped so far and we hit EOF */
2040			break;
2041		}
2042	} while (lblk++, (bh = bh->b_this_page) != head);
2043	/* So far everything mapped? Submit the page for IO. */
2044	if (mpd->map.m_len == 0) {
2045		err = mpage_submit_folio(mpd, head->b_folio);
2046		if (err < 0)
2047			return err;
2048		mpage_folio_done(mpd, head->b_folio);
2049	}
2050	if (lblk >= blocks) {
2051		mpd->scanned_until_end = 1;
2052		return 0;
2053	}
2054	return 1;
2055}
2056
2057/*
2058 * mpage_process_folio - update folio buffers corresponding to changed extent
2059 *			 and may submit fully mapped page for IO
2060 * @mpd: description of extent to map, on return next extent to map
2061 * @folio: Contains these buffers.
2062 * @m_lblk: logical block mapping.
2063 * @m_pblk: corresponding physical mapping.
2064 * @map_bh: determines on return whether this page requires any further
2065 *		  mapping or not.
2066 *
2067 * Scan given folio buffers corresponding to changed extent and update buffer
 
 
 
 
 
2068 * state according to new extent state.
2069 * We map delalloc buffers to their physical location, clear unwritten bits.
2070 * If the given folio is not fully mapped, we update @mpd to the next extent in
2071 * the given folio that needs mapping & return @map_bh as true.
2072 */
2073static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2074			      ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2075			      bool *map_bh)
2076{
2077	struct buffer_head *head, *bh;
2078	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2079	ext4_lblk_t lblk = *m_lblk;
2080	ext4_fsblk_t pblock = *m_pblk;
2081	int err = 0;
2082	int blkbits = mpd->inode->i_blkbits;
2083	ssize_t io_end_size = 0;
2084	struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2085
2086	bh = head = folio_buffers(folio);
2087	do {
2088		if (lblk < mpd->map.m_lblk)
2089			continue;
2090		if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2091			/*
2092			 * Buffer after end of mapped extent.
2093			 * Find next buffer in the folio to map.
2094			 */
2095			mpd->map.m_len = 0;
2096			mpd->map.m_flags = 0;
2097			io_end_vec->size += io_end_size;
2098
2099			err = mpage_process_page_bufs(mpd, head, bh, lblk);
2100			if (err > 0)
2101				err = 0;
2102			if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2103				io_end_vec = ext4_alloc_io_end_vec(io_end);
2104				if (IS_ERR(io_end_vec)) {
2105					err = PTR_ERR(io_end_vec);
2106					goto out;
2107				}
2108				io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2109			}
2110			*map_bh = true;
2111			goto out;
2112		}
2113		if (buffer_delay(bh)) {
2114			clear_buffer_delay(bh);
2115			bh->b_blocknr = pblock++;
2116		}
2117		clear_buffer_unwritten(bh);
2118		io_end_size += (1 << blkbits);
2119	} while (lblk++, (bh = bh->b_this_page) != head);
2120
2121	io_end_vec->size += io_end_size;
2122	*map_bh = false;
2123out:
2124	*m_lblk = lblk;
2125	*m_pblk = pblock;
2126	return err;
2127}
2128
2129/*
2130 * mpage_map_buffers - update buffers corresponding to changed extent and
2131 *		       submit fully mapped pages for IO
2132 *
2133 * @mpd - description of extent to map, on return next extent to map
2134 *
2135 * Scan buffers corresponding to changed extent (we expect corresponding pages
2136 * to be already locked) and update buffer state according to new extent state.
2137 * We map delalloc buffers to their physical location, clear unwritten bits,
2138 * and mark buffers as uninit when we perform writes to unwritten extents
2139 * and do extent conversion after IO is finished. If the last page is not fully
2140 * mapped, we update @map to the next extent in the last page that needs
2141 * mapping. Otherwise we submit the page for IO.
2142 */
2143static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2144{
2145	struct folio_batch fbatch;
2146	unsigned nr, i;
2147	struct inode *inode = mpd->inode;
2148	int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2149	pgoff_t start, end;
2150	ext4_lblk_t lblk;
2151	ext4_fsblk_t pblock;
2152	int err;
2153	bool map_bh = false;
2154
2155	start = mpd->map.m_lblk >> bpp_bits;
2156	end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2157	lblk = start << bpp_bits;
2158	pblock = mpd->map.m_pblk;
2159
2160	folio_batch_init(&fbatch);
2161	while (start <= end) {
2162		nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2163		if (nr == 0)
2164			break;
2165		for (i = 0; i < nr; i++) {
2166			struct folio *folio = fbatch.folios[i];
2167
2168			err = mpage_process_folio(mpd, folio, &lblk, &pblock,
2169						 &map_bh);
2170			/*
2171			 * If map_bh is true, means page may require further bh
2172			 * mapping, or maybe the page was submitted for IO.
2173			 * So we return to call further extent mapping.
2174			 */
2175			if (err < 0 || map_bh)
2176				goto out;
2177			/* Page fully mapped - let IO run! */
2178			err = mpage_submit_folio(mpd, folio);
2179			if (err < 0)
2180				goto out;
2181			mpage_folio_done(mpd, folio);
2182		}
2183		folio_batch_release(&fbatch);
2184	}
2185	/* Extent fully mapped and matches with page boundary. We are done. */
2186	mpd->map.m_len = 0;
2187	mpd->map.m_flags = 0;
2188	return 0;
2189out:
2190	folio_batch_release(&fbatch);
2191	return err;
2192}
2193
2194static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2195{
2196	struct inode *inode = mpd->inode;
2197	struct ext4_map_blocks *map = &mpd->map;
2198	int get_blocks_flags;
2199	int err, dioread_nolock;
2200
2201	trace_ext4_da_write_pages_extent(inode, map);
2202	/*
2203	 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2204	 * to convert an unwritten extent to be initialized (in the case
2205	 * where we have written into one or more preallocated blocks).  It is
2206	 * possible that we're going to need more metadata blocks than
2207	 * previously reserved. However we must not fail because we're in
2208	 * writeback and there is nothing we can do about it so it might result
2209	 * in data loss.  So use reserved blocks to allocate metadata if
2210	 * possible.
 
 
 
 
 
2211	 */
2212	get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2213			   EXT4_GET_BLOCKS_METADATA_NOFAIL |
2214			   EXT4_GET_BLOCKS_IO_SUBMIT;
2215	dioread_nolock = ext4_should_dioread_nolock(inode);
2216	if (dioread_nolock)
2217		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
 
 
2218
2219	err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2220	if (err < 0)
2221		return err;
2222	if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2223		if (!mpd->io_submit.io_end->handle &&
2224		    ext4_handle_valid(handle)) {
2225			mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2226			handle->h_rsv_handle = NULL;
2227		}
2228		ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2229	}
2230
2231	BUG_ON(map->m_len == 0);
2232	return 0;
2233}
2234
2235/*
2236 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2237 *				 mpd->len and submit pages underlying it for IO
2238 *
2239 * @handle - handle for journal operations
2240 * @mpd - extent to map
2241 * @give_up_on_write - we set this to true iff there is a fatal error and there
2242 *                     is no hope of writing the data. The caller should discard
2243 *                     dirty pages to avoid infinite loops.
2244 *
2245 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2246 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2247 * them to initialized or split the described range from larger unwritten
2248 * extent. Note that we need not map all the described range since allocation
2249 * can return less blocks or the range is covered by more unwritten extents. We
2250 * cannot map more because we are limited by reserved transaction credits. On
2251 * the other hand we always make sure that the last touched page is fully
2252 * mapped so that it can be written out (and thus forward progress is
2253 * guaranteed). After mapping we submit all mapped pages for IO.
2254 */
2255static int mpage_map_and_submit_extent(handle_t *handle,
2256				       struct mpage_da_data *mpd,
2257				       bool *give_up_on_write)
2258{
2259	struct inode *inode = mpd->inode;
2260	struct ext4_map_blocks *map = &mpd->map;
2261	int err;
2262	loff_t disksize;
2263	int progress = 0;
2264	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2265	struct ext4_io_end_vec *io_end_vec;
2266
2267	io_end_vec = ext4_alloc_io_end_vec(io_end);
2268	if (IS_ERR(io_end_vec))
2269		return PTR_ERR(io_end_vec);
2270	io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2271	do {
2272		err = mpage_map_one_extent(handle, mpd);
2273		if (err < 0) {
2274			struct super_block *sb = inode->i_sb;
2275
2276			if (ext4_forced_shutdown(sb))
 
2277				goto invalidate_dirty_pages;
2278			/*
2279			 * Let the uper layers retry transient errors.
2280			 * In the case of ENOSPC, if ext4_count_free_blocks()
2281			 * is non-zero, a commit should free up blocks.
2282			 */
2283			if ((err == -ENOMEM) ||
2284			    (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2285				if (progress)
2286					goto update_disksize;
2287				return err;
2288			}
2289			ext4_msg(sb, KERN_CRIT,
2290				 "Delayed block allocation failed for "
2291				 "inode %lu at logical offset %llu with"
2292				 " max blocks %u with error %d",
2293				 inode->i_ino,
2294				 (unsigned long long)map->m_lblk,
2295				 (unsigned)map->m_len, -err);
2296			ext4_msg(sb, KERN_CRIT,
2297				 "This should not happen!! Data will "
2298				 "be lost\n");
2299			if (err == -ENOSPC)
2300				ext4_print_free_blocks(inode);
2301		invalidate_dirty_pages:
2302			*give_up_on_write = true;
2303			return err;
2304		}
2305		progress = 1;
2306		/*
2307		 * Update buffer state, submit mapped pages, and get us new
2308		 * extent to map
2309		 */
2310		err = mpage_map_and_submit_buffers(mpd);
2311		if (err < 0)
2312			goto update_disksize;
2313	} while (map->m_len);
2314
2315update_disksize:
2316	/*
2317	 * Update on-disk size after IO is submitted.  Races with
2318	 * truncate are avoided by checking i_size under i_data_sem.
2319	 */
2320	disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2321	if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2322		int err2;
2323		loff_t i_size;
2324
2325		down_write(&EXT4_I(inode)->i_data_sem);
2326		i_size = i_size_read(inode);
2327		if (disksize > i_size)
2328			disksize = i_size;
2329		if (disksize > EXT4_I(inode)->i_disksize)
2330			EXT4_I(inode)->i_disksize = disksize;
2331		up_write(&EXT4_I(inode)->i_data_sem);
2332		err2 = ext4_mark_inode_dirty(handle, inode);
2333		if (err2) {
2334			ext4_error_err(inode->i_sb, -err2,
2335				       "Failed to mark inode %lu dirty",
2336				       inode->i_ino);
2337		}
2338		if (!err)
2339			err = err2;
2340	}
2341	return err;
2342}
2343
2344/*
2345 * Calculate the total number of credits to reserve for one writepages
2346 * iteration. This is called from ext4_writepages(). We map an extent of
2347 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2348 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2349 * bpp - 1 blocks in bpp different extents.
2350 */
2351static int ext4_da_writepages_trans_blocks(struct inode *inode)
2352{
2353	int bpp = ext4_journal_blocks_per_page(inode);
2354
2355	return ext4_meta_trans_blocks(inode,
2356				MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2357}
2358
2359static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio,
2360				     size_t len)
2361{
2362	struct buffer_head *page_bufs = folio_buffers(folio);
2363	struct inode *inode = folio->mapping->host;
2364	int ret, err;
2365
2366	ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2367				     NULL, do_journal_get_write_access);
2368	err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2369				     NULL, write_end_fn);
2370	if (ret == 0)
2371		ret = err;
2372	err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len);
2373	if (ret == 0)
2374		ret = err;
2375	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2376
2377	return ret;
2378}
2379
2380static int mpage_journal_page_buffers(handle_t *handle,
2381				      struct mpage_da_data *mpd,
2382				      struct folio *folio)
2383{
2384	struct inode *inode = mpd->inode;
2385	loff_t size = i_size_read(inode);
2386	size_t len = folio_size(folio);
2387
2388	folio_clear_checked(folio);
2389	mpd->wbc->nr_to_write--;
2390
2391	if (folio_pos(folio) + len > size &&
2392	    !ext4_verity_in_progress(inode))
2393		len = size & (len - 1);
2394
2395	return ext4_journal_folio_buffers(handle, folio, len);
 
 
 
 
 
2396}
2397
2398/*
2399 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2400 * 				 needing mapping, submit mapped pages
2401 *
2402 * @mpd - where to look for pages
2403 *
2404 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2405 * IO immediately. If we cannot map blocks, we submit just already mapped
2406 * buffers in the page for IO and keep page dirty. When we can map blocks and
2407 * we find a page which isn't mapped we start accumulating extent of buffers
2408 * underlying these pages that needs mapping (formed by either delayed or
2409 * unwritten buffers). We also lock the pages containing these buffers. The
2410 * extent found is returned in @mpd structure (starting at mpd->lblk with
2411 * length mpd->len blocks).
2412 *
2413 * Note that this function can attach bios to one io_end structure which are
2414 * neither logically nor physically contiguous. Although it may seem as an
2415 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2416 * case as we need to track IO to all buffers underlying a page in one io_end.
2417 */
2418static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2419{
2420	struct address_space *mapping = mpd->inode->i_mapping;
2421	struct folio_batch fbatch;
2422	unsigned int nr_folios;
 
2423	pgoff_t index = mpd->first_page;
2424	pgoff_t end = mpd->last_page;
2425	xa_mark_t tag;
2426	int i, err = 0;
2427	int blkbits = mpd->inode->i_blkbits;
2428	ext4_lblk_t lblk;
2429	struct buffer_head *head;
2430	handle_t *handle = NULL;
2431	int bpp = ext4_journal_blocks_per_page(mpd->inode);
2432
2433	if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2434		tag = PAGECACHE_TAG_TOWRITE;
2435	else
2436		tag = PAGECACHE_TAG_DIRTY;
2437
 
2438	mpd->map.m_len = 0;
2439	mpd->next_page = index;
2440	if (ext4_should_journal_data(mpd->inode)) {
2441		handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2442					    bpp);
2443		if (IS_ERR(handle))
2444			return PTR_ERR(handle);
2445	}
2446	folio_batch_init(&fbatch);
2447	while (index <= end) {
2448		nr_folios = filemap_get_folios_tag(mapping, &index, end,
2449				tag, &fbatch);
2450		if (nr_folios == 0)
2451			break;
2452
2453		for (i = 0; i < nr_folios; i++) {
2454			struct folio *folio = fbatch.folios[i];
2455
2456			/*
2457			 * Accumulated enough dirty pages? This doesn't apply
2458			 * to WB_SYNC_ALL mode. For integrity sync we have to
2459			 * keep going because someone may be concurrently
2460			 * dirtying pages, and we might have synced a lot of
2461			 * newly appeared dirty pages, but have not synced all
2462			 * of the old dirty pages.
2463			 */
2464			if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2465			    mpd->wbc->nr_to_write <=
2466			    mpd->map.m_len >> (PAGE_SHIFT - blkbits))
2467				goto out;
2468
2469			/* If we can't merge this page, we are done. */
2470			if (mpd->map.m_len > 0 && mpd->next_page != folio->index)
2471				goto out;
2472
2473			if (handle) {
2474				err = ext4_journal_ensure_credits(handle, bpp,
2475								  0);
2476				if (err < 0)
2477					goto out;
2478			}
2479
2480			folio_lock(folio);
2481			/*
2482			 * If the page is no longer dirty, or its mapping no
2483			 * longer corresponds to inode we are writing (which
2484			 * means it has been truncated or invalidated), or the
2485			 * page is already under writeback and we are not doing
2486			 * a data integrity writeback, skip the page
2487			 */
2488			if (!folio_test_dirty(folio) ||
2489			    (folio_test_writeback(folio) &&
2490			     (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2491			    unlikely(folio->mapping != mapping)) {
2492				folio_unlock(folio);
2493				continue;
2494			}
2495
2496			folio_wait_writeback(folio);
2497			BUG_ON(folio_test_writeback(folio));
2498
2499			/*
2500			 * Should never happen but for buggy code in
2501			 * other subsystems that call
2502			 * set_page_dirty() without properly warning
2503			 * the file system first.  See [1] for more
2504			 * information.
2505			 *
2506			 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2507			 */
2508			if (!folio_buffers(folio)) {
2509				ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2510				folio_clear_dirty(folio);
2511				folio_unlock(folio);
2512				continue;
2513			}
2514
2515			if (mpd->map.m_len == 0)
2516				mpd->first_page = folio->index;
2517			mpd->next_page = folio_next_index(folio);
2518			/*
2519			 * Writeout when we cannot modify metadata is simple.
2520			 * Just submit the page. For data=journal mode we
2521			 * first handle writeout of the page for checkpoint and
2522			 * only after that handle delayed page dirtying. This
2523			 * makes sure current data is checkpointed to the final
2524			 * location before possibly journalling it again which
2525			 * is desirable when the page is frequently dirtied
2526			 * through a pin.
2527			 */
2528			if (!mpd->can_map) {
2529				err = mpage_submit_folio(mpd, folio);
2530				if (err < 0)
2531					goto out;
2532				/* Pending dirtying of journalled data? */
2533				if (folio_test_checked(folio)) {
2534					err = mpage_journal_page_buffers(handle,
2535						mpd, folio);
2536					if (err < 0)
2537						goto out;
2538					mpd->journalled_more_data = 1;
 
 
2539				}
2540				mpage_folio_done(mpd, folio);
2541			} else {
2542				/* Add all dirty buffers to mpd */
2543				lblk = ((ext4_lblk_t)folio->index) <<
2544					(PAGE_SHIFT - blkbits);
2545				head = folio_buffers(folio);
2546				err = mpage_process_page_bufs(mpd, head, head,
2547						lblk);
2548				if (err <= 0)
2549					goto out;
2550				err = 0;
2551			}
 
2552		}
2553		folio_batch_release(&fbatch);
2554		cond_resched();
2555	}
2556	mpd->scanned_until_end = 1;
2557	if (handle)
2558		ext4_journal_stop(handle);
2559	return 0;
2560out:
2561	folio_batch_release(&fbatch);
2562	if (handle)
2563		ext4_journal_stop(handle);
2564	return err;
2565}
2566
 
 
 
 
 
 
2567static int ext4_do_writepages(struct mpage_da_data *mpd)
2568{
2569	struct writeback_control *wbc = mpd->wbc;
2570	pgoff_t	writeback_index = 0;
2571	long nr_to_write = wbc->nr_to_write;
2572	int range_whole = 0;
2573	int cycled = 1;
2574	handle_t *handle = NULL;
2575	struct inode *inode = mpd->inode;
2576	struct address_space *mapping = inode->i_mapping;
2577	int needed_blocks, rsv_blocks = 0, ret = 0;
2578	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2579	struct blk_plug plug;
2580	bool give_up_on_write = false;
2581
2582	trace_ext4_writepages(inode, wbc);
2583
2584	/*
2585	 * No pages to write? This is mainly a kludge to avoid starting
2586	 * a transaction for special inodes like journal inode on last iput()
2587	 * because that could violate lock ordering on umount
2588	 */
2589	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2590		goto out_writepages;
2591
 
 
 
 
 
 
 
2592	/*
2593	 * If the filesystem has aborted, it is read-only, so return
2594	 * right away instead of dumping stack traces later on that
2595	 * will obscure the real source of the problem.  We test
2596	 * fs shutdown state instead of sb->s_flag's SB_RDONLY because
2597	 * the latter could be true if the filesystem is mounted
2598	 * read-only, and in that case, ext4_writepages should
2599	 * *never* be called, so if that ever happens, we would want
2600	 * the stack trace.
2601	 */
2602	if (unlikely(ext4_forced_shutdown(mapping->host->i_sb))) {
 
2603		ret = -EROFS;
2604		goto out_writepages;
2605	}
2606
2607	/*
2608	 * If we have inline data and arrive here, it means that
2609	 * we will soon create the block for the 1st page, so
2610	 * we'd better clear the inline data here.
2611	 */
2612	if (ext4_has_inline_data(inode)) {
2613		/* Just inode will be modified... */
2614		handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2615		if (IS_ERR(handle)) {
2616			ret = PTR_ERR(handle);
2617			goto out_writepages;
2618		}
2619		BUG_ON(ext4_test_inode_state(inode,
2620				EXT4_STATE_MAY_INLINE_DATA));
2621		ext4_destroy_inline_data(handle, inode);
2622		ext4_journal_stop(handle);
2623	}
2624
2625	/*
2626	 * data=journal mode does not do delalloc so we just need to writeout /
2627	 * journal already mapped buffers. On the other hand we need to commit
2628	 * transaction to make data stable. We expect all the data to be
2629	 * already in the journal (the only exception are DMA pinned pages
2630	 * dirtied behind our back) so we commit transaction here and run the
2631	 * writeback loop to checkpoint them. The checkpointing is not actually
2632	 * necessary to make data persistent *but* quite a few places (extent
2633	 * shifting operations, fsverity, ...) depend on being able to drop
2634	 * pagecache pages after calling filemap_write_and_wait() and for that
2635	 * checkpointing needs to happen.
2636	 */
2637	if (ext4_should_journal_data(inode)) {
2638		mpd->can_map = 0;
2639		if (wbc->sync_mode == WB_SYNC_ALL)
2640			ext4_fc_commit(sbi->s_journal,
2641				       EXT4_I(inode)->i_datasync_tid);
2642	}
2643	mpd->journalled_more_data = 0;
2644
2645	if (ext4_should_dioread_nolock(inode)) {
2646		/*
2647		 * We may need to convert up to one extent per block in
2648		 * the page and we may dirty the inode.
2649		 */
2650		rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2651						PAGE_SIZE >> inode->i_blkbits);
2652	}
2653
2654	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2655		range_whole = 1;
2656
2657	if (wbc->range_cyclic) {
2658		writeback_index = mapping->writeback_index;
2659		if (writeback_index)
2660			cycled = 0;
2661		mpd->first_page = writeback_index;
2662		mpd->last_page = -1;
2663	} else {
2664		mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2665		mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2666	}
2667
2668	ext4_io_submit_init(&mpd->io_submit, wbc);
2669retry:
2670	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2671		tag_pages_for_writeback(mapping, mpd->first_page,
2672					mpd->last_page);
2673	blk_start_plug(&plug);
2674
2675	/*
2676	 * First writeback pages that don't need mapping - we can avoid
2677	 * starting a transaction unnecessarily and also avoid being blocked
2678	 * in the block layer on device congestion while having transaction
2679	 * started.
2680	 */
2681	mpd->do_map = 0;
2682	mpd->scanned_until_end = 0;
2683	mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2684	if (!mpd->io_submit.io_end) {
2685		ret = -ENOMEM;
2686		goto unplug;
2687	}
2688	ret = mpage_prepare_extent_to_map(mpd);
2689	/* Unlock pages we didn't use */
2690	mpage_release_unused_pages(mpd, false);
2691	/* Submit prepared bio */
2692	ext4_io_submit(&mpd->io_submit);
2693	ext4_put_io_end_defer(mpd->io_submit.io_end);
2694	mpd->io_submit.io_end = NULL;
2695	if (ret < 0)
2696		goto unplug;
2697
2698	while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2699		/* For each extent of pages we use new io_end */
2700		mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2701		if (!mpd->io_submit.io_end) {
2702			ret = -ENOMEM;
2703			break;
2704		}
2705
2706		WARN_ON_ONCE(!mpd->can_map);
2707		/*
2708		 * We have two constraints: We find one extent to map and we
2709		 * must always write out whole page (makes a difference when
2710		 * blocksize < pagesize) so that we don't block on IO when we
2711		 * try to write out the rest of the page. Journalled mode is
2712		 * not supported by delalloc.
2713		 */
2714		BUG_ON(ext4_should_journal_data(inode));
2715		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2716
2717		/* start a new transaction */
2718		handle = ext4_journal_start_with_reserve(inode,
2719				EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2720		if (IS_ERR(handle)) {
2721			ret = PTR_ERR(handle);
2722			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2723			       "%ld pages, ino %lu; err %d", __func__,
2724				wbc->nr_to_write, inode->i_ino, ret);
2725			/* Release allocated io_end */
2726			ext4_put_io_end(mpd->io_submit.io_end);
2727			mpd->io_submit.io_end = NULL;
2728			break;
2729		}
2730		mpd->do_map = 1;
2731
2732		trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2733		ret = mpage_prepare_extent_to_map(mpd);
2734		if (!ret && mpd->map.m_len)
2735			ret = mpage_map_and_submit_extent(handle, mpd,
2736					&give_up_on_write);
2737		/*
2738		 * Caution: If the handle is synchronous,
2739		 * ext4_journal_stop() can wait for transaction commit
2740		 * to finish which may depend on writeback of pages to
2741		 * complete or on page lock to be released.  In that
2742		 * case, we have to wait until after we have
2743		 * submitted all the IO, released page locks we hold,
2744		 * and dropped io_end reference (for extent conversion
2745		 * to be able to complete) before stopping the handle.
2746		 */
2747		if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2748			ext4_journal_stop(handle);
2749			handle = NULL;
2750			mpd->do_map = 0;
2751		}
2752		/* Unlock pages we didn't use */
2753		mpage_release_unused_pages(mpd, give_up_on_write);
2754		/* Submit prepared bio */
2755		ext4_io_submit(&mpd->io_submit);
2756
2757		/*
2758		 * Drop our io_end reference we got from init. We have
2759		 * to be careful and use deferred io_end finishing if
2760		 * we are still holding the transaction as we can
2761		 * release the last reference to io_end which may end
2762		 * up doing unwritten extent conversion.
2763		 */
2764		if (handle) {
2765			ext4_put_io_end_defer(mpd->io_submit.io_end);
2766			ext4_journal_stop(handle);
2767		} else
2768			ext4_put_io_end(mpd->io_submit.io_end);
2769		mpd->io_submit.io_end = NULL;
2770
2771		if (ret == -ENOSPC && sbi->s_journal) {
2772			/*
2773			 * Commit the transaction which would
2774			 * free blocks released in the transaction
2775			 * and try again
2776			 */
2777			jbd2_journal_force_commit_nested(sbi->s_journal);
2778			ret = 0;
2779			continue;
2780		}
2781		/* Fatal error - ENOMEM, EIO... */
2782		if (ret)
2783			break;
2784	}
2785unplug:
2786	blk_finish_plug(&plug);
2787	if (!ret && !cycled && wbc->nr_to_write > 0) {
2788		cycled = 1;
2789		mpd->last_page = writeback_index - 1;
2790		mpd->first_page = 0;
2791		goto retry;
2792	}
2793
2794	/* Update index */
2795	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2796		/*
2797		 * Set the writeback_index so that range_cyclic
2798		 * mode will write it back later
2799		 */
2800		mapping->writeback_index = mpd->first_page;
2801
2802out_writepages:
2803	trace_ext4_writepages_result(inode, wbc, ret,
2804				     nr_to_write - wbc->nr_to_write);
2805	return ret;
2806}
2807
2808static int ext4_writepages(struct address_space *mapping,
2809			   struct writeback_control *wbc)
2810{
2811	struct super_block *sb = mapping->host->i_sb;
2812	struct mpage_da_data mpd = {
2813		.inode = mapping->host,
2814		.wbc = wbc,
2815		.can_map = 1,
2816	};
2817	int ret;
2818	int alloc_ctx;
2819
2820	if (unlikely(ext4_forced_shutdown(sb)))
2821		return -EIO;
2822
2823	alloc_ctx = ext4_writepages_down_read(sb);
2824	ret = ext4_do_writepages(&mpd);
2825	/*
2826	 * For data=journal writeback we could have come across pages marked
2827	 * for delayed dirtying (PageChecked) which were just added to the
2828	 * running transaction. Try once more to get them to stable storage.
2829	 */
2830	if (!ret && mpd.journalled_more_data)
2831		ret = ext4_do_writepages(&mpd);
2832	ext4_writepages_up_read(sb, alloc_ctx);
2833
2834	return ret;
2835}
2836
2837int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2838{
2839	struct writeback_control wbc = {
2840		.sync_mode = WB_SYNC_ALL,
2841		.nr_to_write = LONG_MAX,
2842		.range_start = jinode->i_dirty_start,
2843		.range_end = jinode->i_dirty_end,
2844	};
2845	struct mpage_da_data mpd = {
2846		.inode = jinode->i_vfs_inode,
2847		.wbc = &wbc,
2848		.can_map = 0,
2849	};
2850	return ext4_do_writepages(&mpd);
2851}
2852
2853static int ext4_dax_writepages(struct address_space *mapping,
2854			       struct writeback_control *wbc)
2855{
2856	int ret;
2857	long nr_to_write = wbc->nr_to_write;
2858	struct inode *inode = mapping->host;
2859	int alloc_ctx;
2860
2861	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2862		return -EIO;
2863
2864	alloc_ctx = ext4_writepages_down_read(inode->i_sb);
2865	trace_ext4_writepages(inode, wbc);
2866
2867	ret = dax_writeback_mapping_range(mapping,
2868					  EXT4_SB(inode->i_sb)->s_daxdev, wbc);
2869	trace_ext4_writepages_result(inode, wbc, ret,
2870				     nr_to_write - wbc->nr_to_write);
2871	ext4_writepages_up_read(inode->i_sb, alloc_ctx);
2872	return ret;
2873}
2874
2875static int ext4_nonda_switch(struct super_block *sb)
2876{
2877	s64 free_clusters, dirty_clusters;
2878	struct ext4_sb_info *sbi = EXT4_SB(sb);
2879
2880	/*
2881	 * switch to non delalloc mode if we are running low
2882	 * on free block. The free block accounting via percpu
2883	 * counters can get slightly wrong with percpu_counter_batch getting
2884	 * accumulated on each CPU without updating global counters
2885	 * Delalloc need an accurate free block accounting. So switch
2886	 * to non delalloc when we are near to error range.
2887	 */
2888	free_clusters =
2889		percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2890	dirty_clusters =
2891		percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2892	/*
2893	 * Start pushing delalloc when 1/2 of free blocks are dirty.
2894	 */
2895	if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2896		try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2897
2898	if (2 * free_clusters < 3 * dirty_clusters ||
2899	    free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2900		/*
2901		 * free block count is less than 150% of dirty blocks
2902		 * or free blocks is less than watermark
2903		 */
2904		return 1;
2905	}
2906	return 0;
2907}
2908
2909static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2910			       loff_t pos, unsigned len,
2911			       struct folio **foliop, void **fsdata)
2912{
2913	int ret, retries = 0;
2914	struct folio *folio;
2915	pgoff_t index;
2916	struct inode *inode = mapping->host;
2917
2918	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2919		return -EIO;
2920
2921	index = pos >> PAGE_SHIFT;
2922
2923	if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2924		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2925		return ext4_write_begin(file, mapping, pos,
2926					len, foliop, fsdata);
2927	}
2928	*fsdata = (void *)0;
2929	trace_ext4_da_write_begin(inode, pos, len);
2930
2931	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2932		ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
2933						      foliop, fsdata);
2934		if (ret < 0)
2935			return ret;
2936		if (ret == 1)
2937			return 0;
2938	}
2939
2940retry:
2941	folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
2942			mapping_gfp_mask(mapping));
2943	if (IS_ERR(folio))
2944		return PTR_ERR(folio);
 
 
2945
2946	ret = ext4_block_write_begin(NULL, folio, pos, len,
 
2947				     ext4_da_get_block_prep);
 
 
 
2948	if (ret < 0) {
2949		folio_unlock(folio);
2950		folio_put(folio);
2951		/*
2952		 * block_write_begin may have instantiated a few blocks
2953		 * outside i_size.  Trim these off again. Don't need
2954		 * i_size_read because we hold inode lock.
2955		 */
2956		if (pos + len > inode->i_size)
2957			ext4_truncate_failed_write(inode);
2958
2959		if (ret == -ENOSPC &&
2960		    ext4_should_retry_alloc(inode->i_sb, &retries))
2961			goto retry;
2962		return ret;
2963	}
2964
2965	*foliop = folio;
2966	return ret;
2967}
2968
2969/*
2970 * Check if we should update i_disksize
2971 * when write to the end of file but not require block allocation
2972 */
2973static int ext4_da_should_update_i_disksize(struct folio *folio,
2974					    unsigned long offset)
2975{
2976	struct buffer_head *bh;
2977	struct inode *inode = folio->mapping->host;
2978	unsigned int idx;
2979	int i;
2980
2981	bh = folio_buffers(folio);
2982	idx = offset >> inode->i_blkbits;
2983
2984	for (i = 0; i < idx; i++)
2985		bh = bh->b_this_page;
2986
2987	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2988		return 0;
2989	return 1;
2990}
2991
2992static int ext4_da_do_write_end(struct address_space *mapping,
2993			loff_t pos, unsigned len, unsigned copied,
2994			struct folio *folio)
2995{
2996	struct inode *inode = mapping->host;
2997	loff_t old_size = inode->i_size;
2998	bool disksize_changed = false;
2999	loff_t new_i_size, zero_len = 0;
3000	handle_t *handle;
3001
3002	if (unlikely(!folio_buffers(folio))) {
3003		folio_unlock(folio);
3004		folio_put(folio);
3005		return -EIO;
3006	}
3007	/*
3008	 * block_write_end() will mark the inode as dirty with I_DIRTY_PAGES
3009	 * flag, which all that's needed to trigger page writeback.
3010	 */
3011	copied = block_write_end(NULL, mapping, pos, len, copied,
3012			folio, NULL);
3013	new_i_size = pos + copied;
3014
3015	/*
3016	 * It's important to update i_size while still holding folio lock,
3017	 * because folio writeout could otherwise come in and zero beyond
3018	 * i_size.
3019	 *
3020	 * Since we are holding inode lock, we are sure i_disksize <=
3021	 * i_size. We also know that if i_disksize < i_size, there are
3022	 * delalloc writes pending in the range up to i_size. If the end of
3023	 * the current write is <= i_size, there's no need to touch
3024	 * i_disksize since writeback will push i_disksize up to i_size
3025	 * eventually. If the end of the current write is > i_size and
3026	 * inside an allocated block which ext4_da_should_update_i_disksize()
3027	 * checked, we need to update i_disksize here as certain
3028	 * ext4_writepages() paths not allocating blocks and update i_disksize.
3029	 */
3030	if (new_i_size > inode->i_size) {
3031		unsigned long end;
3032
3033		i_size_write(inode, new_i_size);
3034		end = (new_i_size - 1) & (PAGE_SIZE - 1);
3035		if (copied && ext4_da_should_update_i_disksize(folio, end)) {
3036			ext4_update_i_disksize(inode, new_i_size);
3037			disksize_changed = true;
3038		}
3039	}
3040
3041	folio_unlock(folio);
3042	folio_put(folio);
3043
3044	if (pos > old_size) {
3045		pagecache_isize_extended(inode, old_size, pos);
3046		zero_len = pos - old_size;
3047	}
3048
3049	if (!disksize_changed && !zero_len)
3050		return copied;
3051
3052	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3053	if (IS_ERR(handle))
3054		return PTR_ERR(handle);
3055	if (zero_len)
3056		ext4_zero_partial_blocks(handle, inode, old_size, zero_len);
3057	ext4_mark_inode_dirty(handle, inode);
3058	ext4_journal_stop(handle);
3059
3060	return copied;
3061}
3062
3063static int ext4_da_write_end(struct file *file,
3064			     struct address_space *mapping,
3065			     loff_t pos, unsigned len, unsigned copied,
3066			     struct folio *folio, void *fsdata)
3067{
3068	struct inode *inode = mapping->host;
 
 
3069	int write_mode = (int)(unsigned long)fsdata;
3070
3071	if (write_mode == FALL_BACK_TO_NONDELALLOC)
3072		return ext4_write_end(file, mapping, pos,
3073				      len, copied, folio, fsdata);
3074
3075	trace_ext4_da_write_end(inode, pos, len, copied);
3076
3077	if (write_mode != CONVERT_INLINE_DATA &&
3078	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3079	    ext4_has_inline_data(inode))
3080		return ext4_write_inline_data_end(inode, pos, len, copied,
3081						  folio);
3082
3083	if (unlikely(copied < len) && !folio_test_uptodate(folio))
3084		copied = 0;
3085
3086	return ext4_da_do_write_end(mapping, pos, len, copied, folio);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3087}
3088
3089/*
3090 * Force all delayed allocation blocks to be allocated for a given inode.
3091 */
3092int ext4_alloc_da_blocks(struct inode *inode)
3093{
3094	trace_ext4_alloc_da_blocks(inode);
3095
3096	if (!EXT4_I(inode)->i_reserved_data_blocks)
3097		return 0;
3098
3099	/*
3100	 * We do something simple for now.  The filemap_flush() will
3101	 * also start triggering a write of the data blocks, which is
3102	 * not strictly speaking necessary (and for users of
3103	 * laptop_mode, not even desirable).  However, to do otherwise
3104	 * would require replicating code paths in:
3105	 *
3106	 * ext4_writepages() ->
3107	 *    write_cache_pages() ---> (via passed in callback function)
3108	 *        __mpage_da_writepage() -->
3109	 *           mpage_add_bh_to_extent()
3110	 *           mpage_da_map_blocks()
3111	 *
3112	 * The problem is that write_cache_pages(), located in
3113	 * mm/page-writeback.c, marks pages clean in preparation for
3114	 * doing I/O, which is not desirable if we're not planning on
3115	 * doing I/O at all.
3116	 *
3117	 * We could call write_cache_pages(), and then redirty all of
3118	 * the pages by calling redirty_page_for_writepage() but that
3119	 * would be ugly in the extreme.  So instead we would need to
3120	 * replicate parts of the code in the above functions,
3121	 * simplifying them because we wouldn't actually intend to
3122	 * write out the pages, but rather only collect contiguous
3123	 * logical block extents, call the multi-block allocator, and
3124	 * then update the buffer heads with the block allocations.
3125	 *
3126	 * For now, though, we'll cheat by calling filemap_flush(),
3127	 * which will map the blocks, and start the I/O, but not
3128	 * actually wait for the I/O to complete.
3129	 */
3130	return filemap_flush(inode->i_mapping);
3131}
3132
3133/*
3134 * bmap() is special.  It gets used by applications such as lilo and by
3135 * the swapper to find the on-disk block of a specific piece of data.
3136 *
3137 * Naturally, this is dangerous if the block concerned is still in the
3138 * journal.  If somebody makes a swapfile on an ext4 data-journaling
3139 * filesystem and enables swap, then they may get a nasty shock when the
3140 * data getting swapped to that swapfile suddenly gets overwritten by
3141 * the original zero's written out previously to the journal and
3142 * awaiting writeback in the kernel's buffer cache.
3143 *
3144 * So, if we see any bmap calls here on a modified, data-journaled file,
3145 * take extra steps to flush any blocks which might be in the cache.
3146 */
3147static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3148{
3149	struct inode *inode = mapping->host;
 
3150	sector_t ret = 0;
 
3151
3152	inode_lock_shared(inode);
3153	/*
3154	 * We can get here for an inline file via the FIBMAP ioctl
3155	 */
3156	if (ext4_has_inline_data(inode))
3157		goto out;
3158
3159	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3160	    (test_opt(inode->i_sb, DELALLOC) ||
3161	     ext4_should_journal_data(inode))) {
3162		/*
3163		 * With delalloc or journalled data we want to sync the file so
3164		 * that we can make sure we allocate blocks for file and data
3165		 * is in place for the user to see it
3166		 */
3167		filemap_write_and_wait(mapping);
3168	}
3169
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3170	ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3171
3172out:
3173	inode_unlock_shared(inode);
3174	return ret;
3175}
3176
3177static int ext4_read_folio(struct file *file, struct folio *folio)
3178{
 
3179	int ret = -EAGAIN;
3180	struct inode *inode = folio->mapping->host;
3181
3182	trace_ext4_read_folio(inode, folio);
3183
3184	if (ext4_has_inline_data(inode))
3185		ret = ext4_readpage_inline(inode, folio);
3186
3187	if (ret == -EAGAIN)
3188		return ext4_mpage_readpages(inode, NULL, folio);
3189
3190	return ret;
3191}
3192
3193static void ext4_readahead(struct readahead_control *rac)
3194{
3195	struct inode *inode = rac->mapping->host;
3196
3197	/* If the file has inline data, no need to do readahead. */
3198	if (ext4_has_inline_data(inode))
3199		return;
3200
3201	ext4_mpage_readpages(inode, rac, NULL);
3202}
3203
3204static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3205				size_t length)
3206{
3207	trace_ext4_invalidate_folio(folio, offset, length);
3208
3209	/* No journalling happens on data buffers when this function is used */
3210	WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3211
3212	block_invalidate_folio(folio, offset, length);
3213}
3214
3215static int __ext4_journalled_invalidate_folio(struct folio *folio,
3216					    size_t offset, size_t length)
3217{
3218	journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3219
3220	trace_ext4_journalled_invalidate_folio(folio, offset, length);
3221
3222	/*
3223	 * If it's a full truncate we just forget about the pending dirtying
3224	 */
3225	if (offset == 0 && length == folio_size(folio))
3226		folio_clear_checked(folio);
3227
3228	return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3229}
3230
3231/* Wrapper for aops... */
3232static void ext4_journalled_invalidate_folio(struct folio *folio,
3233					   size_t offset,
3234					   size_t length)
3235{
3236	WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3237}
3238
3239static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3240{
3241	struct inode *inode = folio->mapping->host;
3242	journal_t *journal = EXT4_JOURNAL(inode);
3243
3244	trace_ext4_release_folio(inode, folio);
3245
3246	/* Page has dirty journalled data -> cannot release */
3247	if (folio_test_checked(folio))
3248		return false;
3249	if (journal)
3250		return jbd2_journal_try_to_free_buffers(journal, folio);
3251	else
3252		return try_to_free_buffers(folio);
3253}
3254
3255static bool ext4_inode_datasync_dirty(struct inode *inode)
3256{
3257	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3258
3259	if (journal) {
3260		if (jbd2_transaction_committed(journal,
3261			EXT4_I(inode)->i_datasync_tid))
3262			return false;
3263		if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3264			return !list_empty(&EXT4_I(inode)->i_fc_list);
3265		return true;
3266	}
3267
3268	/* Any metadata buffers to write? */
3269	if (!list_empty(&inode->i_mapping->i_private_list))
3270		return true;
3271	return inode->i_state & I_DIRTY_DATASYNC;
3272}
3273
3274static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3275			   struct ext4_map_blocks *map, loff_t offset,
3276			   loff_t length, unsigned int flags)
3277{
3278	u8 blkbits = inode->i_blkbits;
3279
3280	/*
3281	 * Writes that span EOF might trigger an I/O size update on completion,
3282	 * so consider them to be dirty for the purpose of O_DSYNC, even if
3283	 * there is no other metadata changes being made or are pending.
3284	 */
3285	iomap->flags = 0;
3286	if (ext4_inode_datasync_dirty(inode) ||
3287	    offset + length > i_size_read(inode))
3288		iomap->flags |= IOMAP_F_DIRTY;
3289
3290	if (map->m_flags & EXT4_MAP_NEW)
3291		iomap->flags |= IOMAP_F_NEW;
3292
3293	if (flags & IOMAP_DAX)
3294		iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3295	else
3296		iomap->bdev = inode->i_sb->s_bdev;
3297	iomap->offset = (u64) map->m_lblk << blkbits;
3298	iomap->length = (u64) map->m_len << blkbits;
3299
3300	if ((map->m_flags & EXT4_MAP_MAPPED) &&
3301	    !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3302		iomap->flags |= IOMAP_F_MERGED;
3303
3304	/*
3305	 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3306	 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3307	 * set. In order for any allocated unwritten extents to be converted
3308	 * into written extents correctly within the ->end_io() handler, we
3309	 * need to ensure that the iomap->type is set appropriately. Hence, the
3310	 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3311	 * been set first.
3312	 */
3313	if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3314		iomap->type = IOMAP_UNWRITTEN;
3315		iomap->addr = (u64) map->m_pblk << blkbits;
3316		if (flags & IOMAP_DAX)
3317			iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3318	} else if (map->m_flags & EXT4_MAP_MAPPED) {
3319		iomap->type = IOMAP_MAPPED;
3320		iomap->addr = (u64) map->m_pblk << blkbits;
3321		if (flags & IOMAP_DAX)
3322			iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3323	} else if (map->m_flags & EXT4_MAP_DELAYED) {
3324		iomap->type = IOMAP_DELALLOC;
3325		iomap->addr = IOMAP_NULL_ADDR;
3326	} else {
3327		iomap->type = IOMAP_HOLE;
3328		iomap->addr = IOMAP_NULL_ADDR;
3329	}
3330}
3331
3332static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3333			    unsigned int flags)
3334{
3335	handle_t *handle;
3336	u8 blkbits = inode->i_blkbits;
3337	int ret, dio_credits, m_flags = 0, retries = 0;
3338
3339	/*
3340	 * Trim the mapping request to the maximum value that we can map at
3341	 * once for direct I/O.
3342	 */
3343	if (map->m_len > DIO_MAX_BLOCKS)
3344		map->m_len = DIO_MAX_BLOCKS;
3345	dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3346
3347retry:
3348	/*
3349	 * Either we allocate blocks and then don't get an unwritten extent, so
3350	 * in that case we have reserved enough credits. Or, the blocks are
3351	 * already allocated and unwritten. In that case, the extent conversion
3352	 * fits into the credits as well.
3353	 */
3354	handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3355	if (IS_ERR(handle))
3356		return PTR_ERR(handle);
3357
3358	/*
3359	 * DAX and direct I/O are the only two operations that are currently
3360	 * supported with IOMAP_WRITE.
3361	 */
3362	WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3363	if (flags & IOMAP_DAX)
3364		m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3365	/*
3366	 * We use i_size instead of i_disksize here because delalloc writeback
3367	 * can complete at any point during the I/O and subsequently push the
3368	 * i_disksize out to i_size. This could be beyond where direct I/O is
3369	 * happening and thus expose allocated blocks to direct I/O reads.
3370	 */
3371	else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3372		m_flags = EXT4_GET_BLOCKS_CREATE;
3373	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3374		m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3375
3376	ret = ext4_map_blocks(handle, inode, map, m_flags);
3377
3378	/*
3379	 * We cannot fill holes in indirect tree based inodes as that could
3380	 * expose stale data in the case of a crash. Use the magic error code
3381	 * to fallback to buffered I/O.
3382	 */
3383	if (!m_flags && !ret)
3384		ret = -ENOTBLK;
3385
3386	ext4_journal_stop(handle);
3387	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3388		goto retry;
3389
3390	return ret;
3391}
3392
3393
3394static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3395		unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3396{
3397	int ret;
3398	struct ext4_map_blocks map;
3399	u8 blkbits = inode->i_blkbits;
3400
3401	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3402		return -EINVAL;
3403
3404	if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3405		return -ERANGE;
3406
3407	/*
3408	 * Calculate the first and last logical blocks respectively.
3409	 */
3410	map.m_lblk = offset >> blkbits;
3411	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3412			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3413
3414	if (flags & IOMAP_WRITE) {
3415		/*
3416		 * We check here if the blocks are already allocated, then we
3417		 * don't need to start a journal txn and we can directly return
3418		 * the mapping information. This could boost performance
3419		 * especially in multi-threaded overwrite requests.
3420		 */
3421		if (offset + length <= i_size_read(inode)) {
3422			ret = ext4_map_blocks(NULL, inode, &map, 0);
3423			if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3424				goto out;
3425		}
3426		ret = ext4_iomap_alloc(inode, &map, flags);
3427	} else {
3428		ret = ext4_map_blocks(NULL, inode, &map, 0);
3429	}
3430
3431	if (ret < 0)
3432		return ret;
3433out:
3434	/*
3435	 * When inline encryption is enabled, sometimes I/O to an encrypted file
3436	 * has to be broken up to guarantee DUN contiguity.  Handle this by
3437	 * limiting the length of the mapping returned.
3438	 */
3439	map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3440
3441	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3442
3443	return 0;
3444}
3445
3446static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3447		loff_t length, unsigned flags, struct iomap *iomap,
3448		struct iomap *srcmap)
3449{
3450	int ret;
3451
3452	/*
3453	 * Even for writes we don't need to allocate blocks, so just pretend
3454	 * we are reading to save overhead of starting a transaction.
3455	 */
3456	flags &= ~IOMAP_WRITE;
3457	ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3458	WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
3459	return ret;
3460}
3461
3462static inline bool ext4_want_directio_fallback(unsigned flags, ssize_t written)
3463{
3464	/* must be a directio to fall back to buffered */
3465	if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) !=
3466		    (IOMAP_WRITE | IOMAP_DIRECT))
3467		return false;
3468
3469	/* atomic writes are all-or-nothing */
3470	if (flags & IOMAP_ATOMIC)
3471		return false;
3472
3473	/* can only try again if we wrote nothing */
3474	return written == 0;
3475}
3476
3477static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3478			  ssize_t written, unsigned flags, struct iomap *iomap)
3479{
3480	/*
3481	 * Check to see whether an error occurred while writing out the data to
3482	 * the allocated blocks. If so, return the magic error code for
3483	 * non-atomic write so that we fallback to buffered I/O and attempt to
3484	 * complete the remainder of the I/O.
3485	 * For non-atomic writes, any blocks that may have been
3486	 * allocated in preparation for the direct I/O will be reused during
3487	 * buffered I/O. For atomic write, we never fallback to buffered-io.
3488	 */
3489	if (ext4_want_directio_fallback(flags, written))
3490		return -ENOTBLK;
3491
3492	return 0;
3493}
3494
3495const struct iomap_ops ext4_iomap_ops = {
3496	.iomap_begin		= ext4_iomap_begin,
3497	.iomap_end		= ext4_iomap_end,
3498};
3499
3500const struct iomap_ops ext4_iomap_overwrite_ops = {
3501	.iomap_begin		= ext4_iomap_overwrite_begin,
3502	.iomap_end		= ext4_iomap_end,
3503};
3504
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3505static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3506				   loff_t length, unsigned int flags,
3507				   struct iomap *iomap, struct iomap *srcmap)
3508{
3509	int ret;
 
3510	struct ext4_map_blocks map;
3511	u8 blkbits = inode->i_blkbits;
3512
3513	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3514		return -EINVAL;
3515
3516	if (ext4_has_inline_data(inode)) {
3517		ret = ext4_inline_data_iomap(inode, iomap);
3518		if (ret != -EAGAIN) {
3519			if (ret == 0 && offset >= iomap->length)
3520				ret = -ENOENT;
3521			return ret;
3522		}
3523	}
3524
3525	/*
3526	 * Calculate the first and last logical block respectively.
3527	 */
3528	map.m_lblk = offset >> blkbits;
3529	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3530			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3531
3532	/*
3533	 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3534	 * So handle it here itself instead of querying ext4_map_blocks().
3535	 * Since ext4_map_blocks() will warn about it and will return
3536	 * -EIO error.
3537	 */
3538	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3539		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3540
3541		if (offset >= sbi->s_bitmap_maxbytes) {
3542			map.m_flags = 0;
3543			goto set_iomap;
3544		}
3545	}
3546
3547	ret = ext4_map_blocks(NULL, inode, &map, 0);
3548	if (ret < 0)
3549		return ret;
 
 
 
3550set_iomap:
3551	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
 
 
3552
3553	return 0;
3554}
3555
3556const struct iomap_ops ext4_iomap_report_ops = {
3557	.iomap_begin = ext4_iomap_begin_report,
3558};
3559
3560/*
3561 * For data=journal mode, folio should be marked dirty only when it was
3562 * writeably mapped. When that happens, it was already attached to the
3563 * transaction and marked as jbddirty (we take care of this in
3564 * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3565 * so we should have nothing to do here, except for the case when someone
3566 * had the page pinned and dirtied the page through this pin (e.g. by doing
3567 * direct IO to it). In that case we'd need to attach buffers here to the
3568 * transaction but we cannot due to lock ordering.  We cannot just dirty the
3569 * folio and leave attached buffers clean, because the buffers' dirty state is
3570 * "definitive".  We cannot just set the buffers dirty or jbddirty because all
3571 * the journalling code will explode.  So what we do is to mark the folio
3572 * "pending dirty" and next time ext4_writepages() is called, attach buffers
3573 * to the transaction appropriately.
3574 */
3575static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3576		struct folio *folio)
3577{
3578	WARN_ON_ONCE(!folio_buffers(folio));
3579	if (folio_maybe_dma_pinned(folio))
3580		folio_set_checked(folio);
3581	return filemap_dirty_folio(mapping, folio);
3582}
3583
3584static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3585{
3586	WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3587	WARN_ON_ONCE(!folio_buffers(folio));
3588	return block_dirty_folio(mapping, folio);
3589}
3590
3591static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3592				    struct file *file, sector_t *span)
3593{
3594	return iomap_swapfile_activate(sis, file, span,
3595				       &ext4_iomap_report_ops);
3596}
3597
3598static const struct address_space_operations ext4_aops = {
3599	.read_folio		= ext4_read_folio,
3600	.readahead		= ext4_readahead,
3601	.writepages		= ext4_writepages,
3602	.write_begin		= ext4_write_begin,
3603	.write_end		= ext4_write_end,
3604	.dirty_folio		= ext4_dirty_folio,
3605	.bmap			= ext4_bmap,
3606	.invalidate_folio	= ext4_invalidate_folio,
3607	.release_folio		= ext4_release_folio,
 
3608	.migrate_folio		= buffer_migrate_folio,
3609	.is_partially_uptodate  = block_is_partially_uptodate,
3610	.error_remove_folio	= generic_error_remove_folio,
3611	.swap_activate		= ext4_iomap_swap_activate,
3612};
3613
3614static const struct address_space_operations ext4_journalled_aops = {
3615	.read_folio		= ext4_read_folio,
3616	.readahead		= ext4_readahead,
3617	.writepages		= ext4_writepages,
3618	.write_begin		= ext4_write_begin,
3619	.write_end		= ext4_journalled_write_end,
3620	.dirty_folio		= ext4_journalled_dirty_folio,
3621	.bmap			= ext4_bmap,
3622	.invalidate_folio	= ext4_journalled_invalidate_folio,
3623	.release_folio		= ext4_release_folio,
 
3624	.migrate_folio		= buffer_migrate_folio_norefs,
3625	.is_partially_uptodate  = block_is_partially_uptodate,
3626	.error_remove_folio	= generic_error_remove_folio,
3627	.swap_activate		= ext4_iomap_swap_activate,
3628};
3629
3630static const struct address_space_operations ext4_da_aops = {
3631	.read_folio		= ext4_read_folio,
3632	.readahead		= ext4_readahead,
3633	.writepages		= ext4_writepages,
3634	.write_begin		= ext4_da_write_begin,
3635	.write_end		= ext4_da_write_end,
3636	.dirty_folio		= ext4_dirty_folio,
3637	.bmap			= ext4_bmap,
3638	.invalidate_folio	= ext4_invalidate_folio,
3639	.release_folio		= ext4_release_folio,
 
3640	.migrate_folio		= buffer_migrate_folio,
3641	.is_partially_uptodate  = block_is_partially_uptodate,
3642	.error_remove_folio	= generic_error_remove_folio,
3643	.swap_activate		= ext4_iomap_swap_activate,
3644};
3645
3646static const struct address_space_operations ext4_dax_aops = {
3647	.writepages		= ext4_dax_writepages,
 
3648	.dirty_folio		= noop_dirty_folio,
3649	.bmap			= ext4_bmap,
3650	.swap_activate		= ext4_iomap_swap_activate,
3651};
3652
3653void ext4_set_aops(struct inode *inode)
3654{
3655	switch (ext4_inode_journal_mode(inode)) {
3656	case EXT4_INODE_ORDERED_DATA_MODE:
3657	case EXT4_INODE_WRITEBACK_DATA_MODE:
3658		break;
3659	case EXT4_INODE_JOURNAL_DATA_MODE:
3660		inode->i_mapping->a_ops = &ext4_journalled_aops;
3661		return;
3662	default:
3663		BUG();
3664	}
3665	if (IS_DAX(inode))
3666		inode->i_mapping->a_ops = &ext4_dax_aops;
3667	else if (test_opt(inode->i_sb, DELALLOC))
3668		inode->i_mapping->a_ops = &ext4_da_aops;
3669	else
3670		inode->i_mapping->a_ops = &ext4_aops;
3671}
3672
3673/*
3674 * Here we can't skip an unwritten buffer even though it usually reads zero
3675 * because it might have data in pagecache (eg, if called from ext4_zero_range,
3676 * ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a
3677 * racing writeback can come later and flush the stale pagecache to disk.
3678 */
3679static int __ext4_block_zero_page_range(handle_t *handle,
3680		struct address_space *mapping, loff_t from, loff_t length)
3681{
3682	ext4_fsblk_t index = from >> PAGE_SHIFT;
3683	unsigned offset = from & (PAGE_SIZE-1);
3684	unsigned blocksize, pos;
3685	ext4_lblk_t iblock;
3686	struct inode *inode = mapping->host;
3687	struct buffer_head *bh;
3688	struct folio *folio;
3689	int err = 0;
3690
3691	folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT,
3692				    FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3693				    mapping_gfp_constraint(mapping, ~__GFP_FS));
3694	if (IS_ERR(folio))
3695		return PTR_ERR(folio);
3696
3697	blocksize = inode->i_sb->s_blocksize;
3698
3699	iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3700
3701	bh = folio_buffers(folio);
3702	if (!bh)
3703		bh = create_empty_buffers(folio, blocksize, 0);
3704
3705	/* Find the buffer that contains "offset" */
 
3706	pos = blocksize;
3707	while (offset >= pos) {
3708		bh = bh->b_this_page;
3709		iblock++;
3710		pos += blocksize;
3711	}
3712	if (buffer_freed(bh)) {
3713		BUFFER_TRACE(bh, "freed: skip");
3714		goto unlock;
3715	}
3716	if (!buffer_mapped(bh)) {
3717		BUFFER_TRACE(bh, "unmapped");
3718		ext4_get_block(inode, iblock, bh, 0);
3719		/* unmapped? It's a hole - nothing to do */
3720		if (!buffer_mapped(bh)) {
3721			BUFFER_TRACE(bh, "still unmapped");
3722			goto unlock;
3723		}
3724	}
3725
3726	/* Ok, it's mapped. Make sure it's up-to-date */
3727	if (folio_test_uptodate(folio))
3728		set_buffer_uptodate(bh);
3729
3730	if (!buffer_uptodate(bh)) {
3731		err = ext4_read_bh_lock(bh, 0, true);
3732		if (err)
3733			goto unlock;
3734		if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3735			/* We expect the key to be set. */
3736			BUG_ON(!fscrypt_has_encryption_key(inode));
3737			err = fscrypt_decrypt_pagecache_blocks(folio,
3738							       blocksize,
3739							       bh_offset(bh));
3740			if (err) {
3741				clear_buffer_uptodate(bh);
3742				goto unlock;
3743			}
3744		}
3745	}
3746	if (ext4_should_journal_data(inode)) {
3747		BUFFER_TRACE(bh, "get write access");
3748		err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3749						    EXT4_JTR_NONE);
3750		if (err)
3751			goto unlock;
3752	}
3753	folio_zero_range(folio, offset, length);
3754	BUFFER_TRACE(bh, "zeroed end of block");
3755
3756	if (ext4_should_journal_data(inode)) {
3757		err = ext4_dirty_journalled_data(handle, bh);
3758	} else {
3759		err = 0;
3760		mark_buffer_dirty(bh);
3761		if (ext4_should_order_data(inode))
3762			err = ext4_jbd2_inode_add_write(handle, inode, from,
3763					length);
3764	}
3765
3766unlock:
3767	folio_unlock(folio);
3768	folio_put(folio);
3769	return err;
3770}
3771
3772/*
3773 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3774 * starting from file offset 'from'.  The range to be zero'd must
3775 * be contained with in one block.  If the specified range exceeds
3776 * the end of the block it will be shortened to end of the block
3777 * that corresponds to 'from'
3778 */
3779static int ext4_block_zero_page_range(handle_t *handle,
3780		struct address_space *mapping, loff_t from, loff_t length)
3781{
3782	struct inode *inode = mapping->host;
3783	unsigned offset = from & (PAGE_SIZE-1);
3784	unsigned blocksize = inode->i_sb->s_blocksize;
3785	unsigned max = blocksize - (offset & (blocksize - 1));
3786
3787	/*
3788	 * correct length if it does not fall between
3789	 * 'from' and the end of the block
3790	 */
3791	if (length > max || length < 0)
3792		length = max;
3793
3794	if (IS_DAX(inode)) {
3795		return dax_zero_range(inode, from, length, NULL,
3796				      &ext4_iomap_ops);
3797	}
3798	return __ext4_block_zero_page_range(handle, mapping, from, length);
3799}
3800
3801/*
3802 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3803 * up to the end of the block which corresponds to `from'.
3804 * This required during truncate. We need to physically zero the tail end
3805 * of that block so it doesn't yield old data if the file is later grown.
3806 */
3807static int ext4_block_truncate_page(handle_t *handle,
3808		struct address_space *mapping, loff_t from)
3809{
3810	unsigned offset = from & (PAGE_SIZE-1);
3811	unsigned length;
3812	unsigned blocksize;
3813	struct inode *inode = mapping->host;
3814
3815	/* If we are processing an encrypted inode during orphan list handling */
3816	if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3817		return 0;
3818
3819	blocksize = inode->i_sb->s_blocksize;
3820	length = blocksize - (offset & (blocksize - 1));
3821
3822	return ext4_block_zero_page_range(handle, mapping, from, length);
3823}
3824
3825int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3826			     loff_t lstart, loff_t length)
3827{
3828	struct super_block *sb = inode->i_sb;
3829	struct address_space *mapping = inode->i_mapping;
3830	unsigned partial_start, partial_end;
3831	ext4_fsblk_t start, end;
3832	loff_t byte_end = (lstart + length - 1);
3833	int err = 0;
3834
3835	partial_start = lstart & (sb->s_blocksize - 1);
3836	partial_end = byte_end & (sb->s_blocksize - 1);
3837
3838	start = lstart >> sb->s_blocksize_bits;
3839	end = byte_end >> sb->s_blocksize_bits;
3840
3841	/* Handle partial zero within the single block */
3842	if (start == end &&
3843	    (partial_start || (partial_end != sb->s_blocksize - 1))) {
3844		err = ext4_block_zero_page_range(handle, mapping,
3845						 lstart, length);
3846		return err;
3847	}
3848	/* Handle partial zero out on the start of the range */
3849	if (partial_start) {
3850		err = ext4_block_zero_page_range(handle, mapping,
3851						 lstart, sb->s_blocksize);
3852		if (err)
3853			return err;
3854	}
3855	/* Handle partial zero out on the end of the range */
3856	if (partial_end != sb->s_blocksize - 1)
3857		err = ext4_block_zero_page_range(handle, mapping,
3858						 byte_end - partial_end,
3859						 partial_end + 1);
3860	return err;
3861}
3862
3863int ext4_can_truncate(struct inode *inode)
3864{
3865	if (S_ISREG(inode->i_mode))
3866		return 1;
3867	if (S_ISDIR(inode->i_mode))
3868		return 1;
3869	if (S_ISLNK(inode->i_mode))
3870		return !ext4_inode_is_fast_symlink(inode);
3871	return 0;
3872}
3873
3874/*
3875 * We have to make sure i_disksize gets properly updated before we truncate
3876 * page cache due to hole punching or zero range. Otherwise i_disksize update
3877 * can get lost as it may have been postponed to submission of writeback but
3878 * that will never happen after we truncate page cache.
3879 */
3880int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3881				      loff_t len)
3882{
3883	handle_t *handle;
3884	int ret;
3885
3886	loff_t size = i_size_read(inode);
3887
3888	WARN_ON(!inode_is_locked(inode));
3889	if (offset > size || offset + len < size)
3890		return 0;
3891
3892	if (EXT4_I(inode)->i_disksize >= size)
3893		return 0;
3894
3895	handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3896	if (IS_ERR(handle))
3897		return PTR_ERR(handle);
3898	ext4_update_i_disksize(inode, size);
3899	ret = ext4_mark_inode_dirty(handle, inode);
3900	ext4_journal_stop(handle);
3901
3902	return ret;
3903}
3904
3905static void ext4_wait_dax_page(struct inode *inode)
3906{
3907	filemap_invalidate_unlock(inode->i_mapping);
3908	schedule();
3909	filemap_invalidate_lock(inode->i_mapping);
3910}
3911
3912int ext4_break_layouts(struct inode *inode)
3913{
3914	struct page *page;
3915	int error;
3916
3917	if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3918		return -EINVAL;
3919
3920	do {
3921		page = dax_layout_busy_page(inode->i_mapping);
3922		if (!page)
3923			return 0;
3924
3925		error = ___wait_var_event(&page->_refcount,
3926				atomic_read(&page->_refcount) == 1,
3927				TASK_INTERRUPTIBLE, 0, 0,
3928				ext4_wait_dax_page(inode));
3929	} while (error == 0);
3930
3931	return error;
3932}
3933
3934/*
3935 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3936 * associated with the given offset and length
3937 *
3938 * @inode:  File inode
3939 * @offset: The offset where the hole will begin
3940 * @len:    The length of the hole
3941 *
3942 * Returns: 0 on success or negative on failure
3943 */
3944
3945int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3946{
3947	struct inode *inode = file_inode(file);
3948	struct super_block *sb = inode->i_sb;
3949	ext4_lblk_t first_block, stop_block;
3950	struct address_space *mapping = inode->i_mapping;
3951	loff_t first_block_offset, last_block_offset, max_length;
3952	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3953	handle_t *handle;
3954	unsigned int credits;
3955	int ret = 0, ret2 = 0;
3956
3957	trace_ext4_punch_hole(inode, offset, length, 0);
3958
3959	/*
3960	 * Write out all dirty pages to avoid race conditions
3961	 * Then release them.
3962	 */
3963	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3964		ret = filemap_write_and_wait_range(mapping, offset,
3965						   offset + length - 1);
3966		if (ret)
3967			return ret;
3968	}
3969
3970	inode_lock(inode);
3971
3972	/* No need to punch hole beyond i_size */
3973	if (offset >= inode->i_size)
3974		goto out_mutex;
3975
3976	/*
3977	 * If the hole extends beyond i_size, set the hole
3978	 * to end after the page that contains i_size
3979	 */
3980	if (offset + length > inode->i_size) {
3981		length = inode->i_size +
3982		   PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3983		   offset;
3984	}
3985
3986	/*
3987	 * For punch hole the length + offset needs to be within one block
3988	 * before last range. Adjust the length if it goes beyond that limit.
3989	 */
3990	max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
3991	if (offset + length > max_length)
3992		length = max_length - offset;
3993
3994	if (offset & (sb->s_blocksize - 1) ||
3995	    (offset + length) & (sb->s_blocksize - 1)) {
3996		/*
3997		 * Attach jinode to inode for jbd2 if we do any zeroing of
3998		 * partial block
3999		 */
4000		ret = ext4_inode_attach_jinode(inode);
4001		if (ret < 0)
4002			goto out_mutex;
4003
4004	}
4005
4006	/* Wait all existing dio workers, newcomers will block on i_rwsem */
4007	inode_dio_wait(inode);
4008
4009	ret = file_modified(file);
4010	if (ret)
4011		goto out_mutex;
4012
4013	/*
4014	 * Prevent page faults from reinstantiating pages we have released from
4015	 * page cache.
4016	 */
4017	filemap_invalidate_lock(mapping);
4018
4019	ret = ext4_break_layouts(inode);
4020	if (ret)
4021		goto out_dio;
4022
4023	first_block_offset = round_up(offset, sb->s_blocksize);
4024	last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4025
4026	/* Now release the pages and zero block aligned part of pages*/
4027	if (last_block_offset > first_block_offset) {
4028		ret = ext4_update_disksize_before_punch(inode, offset, length);
4029		if (ret)
4030			goto out_dio;
4031		truncate_pagecache_range(inode, first_block_offset,
4032					 last_block_offset);
4033	}
4034
4035	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4036		credits = ext4_writepage_trans_blocks(inode);
4037	else
4038		credits = ext4_blocks_for_truncate(inode);
4039	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4040	if (IS_ERR(handle)) {
4041		ret = PTR_ERR(handle);
4042		ext4_std_error(sb, ret);
4043		goto out_dio;
4044	}
4045
4046	ret = ext4_zero_partial_blocks(handle, inode, offset,
4047				       length);
4048	if (ret)
4049		goto out_stop;
4050
4051	first_block = (offset + sb->s_blocksize - 1) >>
4052		EXT4_BLOCK_SIZE_BITS(sb);
4053	stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4054
4055	/* If there are blocks to remove, do it */
4056	if (stop_block > first_block) {
4057		ext4_lblk_t hole_len = stop_block - first_block;
4058
4059		down_write(&EXT4_I(inode)->i_data_sem);
4060		ext4_discard_preallocations(inode);
4061
4062		ext4_es_remove_extent(inode, first_block, hole_len);
 
 
 
 
 
4063
4064		if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4065			ret = ext4_ext_remove_space(inode, first_block,
4066						    stop_block - 1);
4067		else
4068			ret = ext4_ind_remove_space(handle, inode, first_block,
4069						    stop_block);
4070
4071		ext4_es_insert_extent(inode, first_block, hole_len, ~0,
4072				      EXTENT_STATUS_HOLE, 0);
4073		up_write(&EXT4_I(inode)->i_data_sem);
4074	}
4075	ext4_fc_track_range(handle, inode, first_block, stop_block);
4076	if (IS_SYNC(inode))
4077		ext4_handle_sync(handle);
4078
4079	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4080	ret2 = ext4_mark_inode_dirty(handle, inode);
4081	if (unlikely(ret2))
4082		ret = ret2;
4083	if (ret >= 0)
4084		ext4_update_inode_fsync_trans(handle, inode, 1);
4085out_stop:
4086	ext4_journal_stop(handle);
4087out_dio:
4088	filemap_invalidate_unlock(mapping);
4089out_mutex:
4090	inode_unlock(inode);
4091	return ret;
4092}
4093
4094int ext4_inode_attach_jinode(struct inode *inode)
4095{
4096	struct ext4_inode_info *ei = EXT4_I(inode);
4097	struct jbd2_inode *jinode;
4098
4099	if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4100		return 0;
4101
4102	jinode = jbd2_alloc_inode(GFP_KERNEL);
4103	spin_lock(&inode->i_lock);
4104	if (!ei->jinode) {
4105		if (!jinode) {
4106			spin_unlock(&inode->i_lock);
4107			return -ENOMEM;
4108		}
4109		ei->jinode = jinode;
4110		jbd2_journal_init_jbd_inode(ei->jinode, inode);
4111		jinode = NULL;
4112	}
4113	spin_unlock(&inode->i_lock);
4114	if (unlikely(jinode != NULL))
4115		jbd2_free_inode(jinode);
4116	return 0;
4117}
4118
4119/*
4120 * ext4_truncate()
4121 *
4122 * We block out ext4_get_block() block instantiations across the entire
4123 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4124 * simultaneously on behalf of the same inode.
4125 *
4126 * As we work through the truncate and commit bits of it to the journal there
4127 * is one core, guiding principle: the file's tree must always be consistent on
4128 * disk.  We must be able to restart the truncate after a crash.
4129 *
4130 * The file's tree may be transiently inconsistent in memory (although it
4131 * probably isn't), but whenever we close off and commit a journal transaction,
4132 * the contents of (the filesystem + the journal) must be consistent and
4133 * restartable.  It's pretty simple, really: bottom up, right to left (although
4134 * left-to-right works OK too).
4135 *
4136 * Note that at recovery time, journal replay occurs *before* the restart of
4137 * truncate against the orphan inode list.
4138 *
4139 * The committed inode has the new, desired i_size (which is the same as
4140 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4141 * that this inode's truncate did not complete and it will again call
4142 * ext4_truncate() to have another go.  So there will be instantiated blocks
4143 * to the right of the truncation point in a crashed ext4 filesystem.  But
4144 * that's fine - as long as they are linked from the inode, the post-crash
4145 * ext4_truncate() run will find them and release them.
4146 */
4147int ext4_truncate(struct inode *inode)
4148{
4149	struct ext4_inode_info *ei = EXT4_I(inode);
4150	unsigned int credits;
4151	int err = 0, err2;
4152	handle_t *handle;
4153	struct address_space *mapping = inode->i_mapping;
4154
4155	/*
4156	 * There is a possibility that we're either freeing the inode
4157	 * or it's a completely new inode. In those cases we might not
4158	 * have i_rwsem locked because it's not necessary.
4159	 */
4160	if (!(inode->i_state & (I_NEW|I_FREEING)))
4161		WARN_ON(!inode_is_locked(inode));
4162	trace_ext4_truncate_enter(inode);
4163
4164	if (!ext4_can_truncate(inode))
4165		goto out_trace;
4166
4167	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4168		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4169
4170	if (ext4_has_inline_data(inode)) {
4171		int has_inline = 1;
4172
4173		err = ext4_inline_data_truncate(inode, &has_inline);
4174		if (err || has_inline)
4175			goto out_trace;
4176	}
4177
4178	/* If we zero-out tail of the page, we have to create jinode for jbd2 */
4179	if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4180		err = ext4_inode_attach_jinode(inode);
4181		if (err)
4182			goto out_trace;
4183	}
4184
4185	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4186		credits = ext4_writepage_trans_blocks(inode);
4187	else
4188		credits = ext4_blocks_for_truncate(inode);
4189
4190	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4191	if (IS_ERR(handle)) {
4192		err = PTR_ERR(handle);
4193		goto out_trace;
4194	}
4195
4196	if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4197		ext4_block_truncate_page(handle, mapping, inode->i_size);
4198
4199	/*
4200	 * We add the inode to the orphan list, so that if this
4201	 * truncate spans multiple transactions, and we crash, we will
4202	 * resume the truncate when the filesystem recovers.  It also
4203	 * marks the inode dirty, to catch the new size.
4204	 *
4205	 * Implication: the file must always be in a sane, consistent
4206	 * truncatable state while each transaction commits.
4207	 */
4208	err = ext4_orphan_add(handle, inode);
4209	if (err)
4210		goto out_stop;
4211
4212	down_write(&EXT4_I(inode)->i_data_sem);
4213
4214	ext4_discard_preallocations(inode);
4215
4216	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4217		err = ext4_ext_truncate(handle, inode);
4218	else
4219		ext4_ind_truncate(handle, inode);
4220
4221	up_write(&ei->i_data_sem);
4222	if (err)
4223		goto out_stop;
4224
4225	if (IS_SYNC(inode))
4226		ext4_handle_sync(handle);
4227
4228out_stop:
4229	/*
4230	 * If this was a simple ftruncate() and the file will remain alive,
4231	 * then we need to clear up the orphan record which we created above.
4232	 * However, if this was a real unlink then we were called by
4233	 * ext4_evict_inode(), and we allow that function to clean up the
4234	 * orphan info for us.
4235	 */
4236	if (inode->i_nlink)
4237		ext4_orphan_del(handle, inode);
4238
4239	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4240	err2 = ext4_mark_inode_dirty(handle, inode);
4241	if (unlikely(err2 && !err))
4242		err = err2;
4243	ext4_journal_stop(handle);
4244
4245out_trace:
4246	trace_ext4_truncate_exit(inode);
4247	return err;
4248}
4249
4250static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4251{
4252	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4253		return inode_peek_iversion_raw(inode);
4254	else
4255		return inode_peek_iversion(inode);
4256}
4257
4258static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4259				 struct ext4_inode_info *ei)
4260{
4261	struct inode *inode = &(ei->vfs_inode);
4262	u64 i_blocks = READ_ONCE(inode->i_blocks);
4263	struct super_block *sb = inode->i_sb;
4264
4265	if (i_blocks <= ~0U) {
4266		/*
4267		 * i_blocks can be represented in a 32 bit variable
4268		 * as multiple of 512 bytes
4269		 */
4270		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4271		raw_inode->i_blocks_high = 0;
4272		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4273		return 0;
4274	}
4275
4276	/*
4277	 * This should never happen since sb->s_maxbytes should not have
4278	 * allowed this, sb->s_maxbytes was set according to the huge_file
4279	 * feature in ext4_fill_super().
4280	 */
4281	if (!ext4_has_feature_huge_file(sb))
4282		return -EFSCORRUPTED;
4283
4284	if (i_blocks <= 0xffffffffffffULL) {
4285		/*
4286		 * i_blocks can be represented in a 48 bit variable
4287		 * as multiple of 512 bytes
4288		 */
4289		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4290		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4291		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4292	} else {
4293		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4294		/* i_block is stored in file system block size */
4295		i_blocks = i_blocks >> (inode->i_blkbits - 9);
4296		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4297		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4298	}
4299	return 0;
4300}
4301
4302static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4303{
4304	struct ext4_inode_info *ei = EXT4_I(inode);
4305	uid_t i_uid;
4306	gid_t i_gid;
4307	projid_t i_projid;
4308	int block;
4309	int err;
4310
4311	err = ext4_inode_blocks_set(raw_inode, ei);
4312
4313	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4314	i_uid = i_uid_read(inode);
4315	i_gid = i_gid_read(inode);
4316	i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4317	if (!(test_opt(inode->i_sb, NO_UID32))) {
4318		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4319		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4320		/*
4321		 * Fix up interoperability with old kernels. Otherwise,
4322		 * old inodes get re-used with the upper 16 bits of the
4323		 * uid/gid intact.
4324		 */
4325		if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4326			raw_inode->i_uid_high = 0;
4327			raw_inode->i_gid_high = 0;
4328		} else {
4329			raw_inode->i_uid_high =
4330				cpu_to_le16(high_16_bits(i_uid));
4331			raw_inode->i_gid_high =
4332				cpu_to_le16(high_16_bits(i_gid));
4333		}
4334	} else {
4335		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4336		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4337		raw_inode->i_uid_high = 0;
4338		raw_inode->i_gid_high = 0;
4339	}
4340	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4341
4342	EXT4_INODE_SET_CTIME(inode, raw_inode);
4343	EXT4_INODE_SET_MTIME(inode, raw_inode);
4344	EXT4_INODE_SET_ATIME(inode, raw_inode);
4345	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4346
4347	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4348	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4349	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4350		raw_inode->i_file_acl_high =
4351			cpu_to_le16(ei->i_file_acl >> 32);
4352	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4353	ext4_isize_set(raw_inode, ei->i_disksize);
4354
4355	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4356	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4357		if (old_valid_dev(inode->i_rdev)) {
4358			raw_inode->i_block[0] =
4359				cpu_to_le32(old_encode_dev(inode->i_rdev));
4360			raw_inode->i_block[1] = 0;
4361		} else {
4362			raw_inode->i_block[0] = 0;
4363			raw_inode->i_block[1] =
4364				cpu_to_le32(new_encode_dev(inode->i_rdev));
4365			raw_inode->i_block[2] = 0;
4366		}
4367	} else if (!ext4_has_inline_data(inode)) {
4368		for (block = 0; block < EXT4_N_BLOCKS; block++)
4369			raw_inode->i_block[block] = ei->i_data[block];
4370	}
4371
4372	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4373		u64 ivers = ext4_inode_peek_iversion(inode);
4374
4375		raw_inode->i_disk_version = cpu_to_le32(ivers);
4376		if (ei->i_extra_isize) {
4377			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4378				raw_inode->i_version_hi =
4379					cpu_to_le32(ivers >> 32);
4380			raw_inode->i_extra_isize =
4381				cpu_to_le16(ei->i_extra_isize);
4382		}
4383	}
4384
4385	if (i_projid != EXT4_DEF_PROJID &&
4386	    !ext4_has_feature_project(inode->i_sb))
4387		err = err ?: -EFSCORRUPTED;
4388
4389	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4390	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4391		raw_inode->i_projid = cpu_to_le32(i_projid);
4392
4393	ext4_inode_csum_set(inode, raw_inode, ei);
4394	return err;
4395}
4396
4397/*
4398 * ext4_get_inode_loc returns with an extra refcount against the inode's
4399 * underlying buffer_head on success. If we pass 'inode' and it does not
4400 * have in-inode xattr, we have all inode data in memory that is needed
4401 * to recreate the on-disk version of this inode.
4402 */
4403static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4404				struct inode *inode, struct ext4_iloc *iloc,
4405				ext4_fsblk_t *ret_block)
4406{
4407	struct ext4_group_desc	*gdp;
4408	struct buffer_head	*bh;
4409	ext4_fsblk_t		block;
4410	struct blk_plug		plug;
4411	int			inodes_per_block, inode_offset;
4412
4413	iloc->bh = NULL;
4414	if (ino < EXT4_ROOT_INO ||
4415	    ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4416		return -EFSCORRUPTED;
4417
4418	iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4419	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4420	if (!gdp)
4421		return -EIO;
4422
4423	/*
4424	 * Figure out the offset within the block group inode table
4425	 */
4426	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4427	inode_offset = ((ino - 1) %
4428			EXT4_INODES_PER_GROUP(sb));
4429	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4430
4431	block = ext4_inode_table(sb, gdp);
4432	if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4433	    (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4434		ext4_error(sb, "Invalid inode table block %llu in "
4435			   "block_group %u", block, iloc->block_group);
4436		return -EFSCORRUPTED;
4437	}
4438	block += (inode_offset / inodes_per_block);
4439
4440	bh = sb_getblk(sb, block);
4441	if (unlikely(!bh))
4442		return -ENOMEM;
4443	if (ext4_buffer_uptodate(bh))
4444		goto has_buffer;
4445
4446	lock_buffer(bh);
4447	if (ext4_buffer_uptodate(bh)) {
4448		/* Someone brought it uptodate while we waited */
4449		unlock_buffer(bh);
4450		goto has_buffer;
4451	}
4452
4453	/*
4454	 * If we have all information of the inode in memory and this
4455	 * is the only valid inode in the block, we need not read the
4456	 * block.
4457	 */
4458	if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4459		struct buffer_head *bitmap_bh;
4460		int i, start;
4461
4462		start = inode_offset & ~(inodes_per_block - 1);
4463
4464		/* Is the inode bitmap in cache? */
4465		bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4466		if (unlikely(!bitmap_bh))
4467			goto make_io;
4468
4469		/*
4470		 * If the inode bitmap isn't in cache then the
4471		 * optimisation may end up performing two reads instead
4472		 * of one, so skip it.
4473		 */
4474		if (!buffer_uptodate(bitmap_bh)) {
4475			brelse(bitmap_bh);
4476			goto make_io;
4477		}
4478		for (i = start; i < start + inodes_per_block; i++) {
4479			if (i == inode_offset)
4480				continue;
4481			if (ext4_test_bit(i, bitmap_bh->b_data))
4482				break;
4483		}
4484		brelse(bitmap_bh);
4485		if (i == start + inodes_per_block) {
4486			struct ext4_inode *raw_inode =
4487				(struct ext4_inode *) (bh->b_data + iloc->offset);
4488
4489			/* all other inodes are free, so skip I/O */
4490			memset(bh->b_data, 0, bh->b_size);
4491			if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4492				ext4_fill_raw_inode(inode, raw_inode);
4493			set_buffer_uptodate(bh);
4494			unlock_buffer(bh);
4495			goto has_buffer;
4496		}
4497	}
4498
4499make_io:
4500	/*
4501	 * If we need to do any I/O, try to pre-readahead extra
4502	 * blocks from the inode table.
4503	 */
4504	blk_start_plug(&plug);
4505	if (EXT4_SB(sb)->s_inode_readahead_blks) {
4506		ext4_fsblk_t b, end, table;
4507		unsigned num;
4508		__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4509
4510		table = ext4_inode_table(sb, gdp);
4511		/* s_inode_readahead_blks is always a power of 2 */
4512		b = block & ~((ext4_fsblk_t) ra_blks - 1);
4513		if (table > b)
4514			b = table;
4515		end = b + ra_blks;
4516		num = EXT4_INODES_PER_GROUP(sb);
4517		if (ext4_has_group_desc_csum(sb))
4518			num -= ext4_itable_unused_count(sb, gdp);
4519		table += num / inodes_per_block;
4520		if (end > table)
4521			end = table;
4522		while (b <= end)
4523			ext4_sb_breadahead_unmovable(sb, b++);
4524	}
4525
4526	/*
4527	 * There are other valid inodes in the buffer, this inode
4528	 * has in-inode xattrs, or we don't have this inode in memory.
4529	 * Read the block from disk.
4530	 */
4531	trace_ext4_load_inode(sb, ino);
4532	ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL,
4533			    ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO));
4534	blk_finish_plug(&plug);
4535	wait_on_buffer(bh);
 
4536	if (!buffer_uptodate(bh)) {
4537		if (ret_block)
4538			*ret_block = block;
4539		brelse(bh);
4540		return -EIO;
4541	}
4542has_buffer:
4543	iloc->bh = bh;
4544	return 0;
4545}
4546
4547static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4548					struct ext4_iloc *iloc)
4549{
4550	ext4_fsblk_t err_blk = 0;
4551	int ret;
4552
4553	ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4554					&err_blk);
4555
4556	if (ret == -EIO)
4557		ext4_error_inode_block(inode, err_blk, EIO,
4558					"unable to read itable block");
4559
4560	return ret;
4561}
4562
4563int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4564{
4565	ext4_fsblk_t err_blk = 0;
4566	int ret;
4567
4568	ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4569					&err_blk);
4570
4571	if (ret == -EIO)
4572		ext4_error_inode_block(inode, err_blk, EIO,
4573					"unable to read itable block");
4574
4575	return ret;
4576}
4577
4578
4579int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4580			  struct ext4_iloc *iloc)
4581{
4582	return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4583}
4584
4585static bool ext4_should_enable_dax(struct inode *inode)
4586{
4587	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4588
4589	if (test_opt2(inode->i_sb, DAX_NEVER))
4590		return false;
4591	if (!S_ISREG(inode->i_mode))
4592		return false;
4593	if (ext4_should_journal_data(inode))
4594		return false;
4595	if (ext4_has_inline_data(inode))
4596		return false;
4597	if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4598		return false;
4599	if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4600		return false;
4601	if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4602		return false;
4603	if (test_opt(inode->i_sb, DAX_ALWAYS))
4604		return true;
4605
4606	return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4607}
4608
4609void ext4_set_inode_flags(struct inode *inode, bool init)
4610{
4611	unsigned int flags = EXT4_I(inode)->i_flags;
4612	unsigned int new_fl = 0;
4613
4614	WARN_ON_ONCE(IS_DAX(inode) && init);
4615
4616	if (flags & EXT4_SYNC_FL)
4617		new_fl |= S_SYNC;
4618	if (flags & EXT4_APPEND_FL)
4619		new_fl |= S_APPEND;
4620	if (flags & EXT4_IMMUTABLE_FL)
4621		new_fl |= S_IMMUTABLE;
4622	if (flags & EXT4_NOATIME_FL)
4623		new_fl |= S_NOATIME;
4624	if (flags & EXT4_DIRSYNC_FL)
4625		new_fl |= S_DIRSYNC;
4626
4627	/* Because of the way inode_set_flags() works we must preserve S_DAX
4628	 * here if already set. */
4629	new_fl |= (inode->i_flags & S_DAX);
4630	if (init && ext4_should_enable_dax(inode))
4631		new_fl |= S_DAX;
4632
4633	if (flags & EXT4_ENCRYPT_FL)
4634		new_fl |= S_ENCRYPTED;
4635	if (flags & EXT4_CASEFOLD_FL)
4636		new_fl |= S_CASEFOLD;
4637	if (flags & EXT4_VERITY_FL)
4638		new_fl |= S_VERITY;
4639	inode_set_flags(inode, new_fl,
4640			S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4641			S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4642}
4643
4644static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4645				  struct ext4_inode_info *ei)
4646{
4647	blkcnt_t i_blocks ;
4648	struct inode *inode = &(ei->vfs_inode);
4649	struct super_block *sb = inode->i_sb;
4650
4651	if (ext4_has_feature_huge_file(sb)) {
4652		/* we are using combined 48 bit field */
4653		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4654					le32_to_cpu(raw_inode->i_blocks_lo);
4655		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4656			/* i_blocks represent file system block size */
4657			return i_blocks  << (inode->i_blkbits - 9);
4658		} else {
4659			return i_blocks;
4660		}
4661	} else {
4662		return le32_to_cpu(raw_inode->i_blocks_lo);
4663	}
4664}
4665
4666static inline int ext4_iget_extra_inode(struct inode *inode,
4667					 struct ext4_inode *raw_inode,
4668					 struct ext4_inode_info *ei)
4669{
4670	__le32 *magic = (void *)raw_inode +
4671			EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4672
4673	if (EXT4_INODE_HAS_XATTR_SPACE(inode)  &&
4674	    *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4675		int err;
4676
4677		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4678		err = ext4_find_inline_data_nolock(inode);
4679		if (!err && ext4_has_inline_data(inode))
4680			ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4681		return err;
4682	} else
4683		EXT4_I(inode)->i_inline_off = 0;
4684	return 0;
4685}
4686
4687int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4688{
4689	if (!ext4_has_feature_project(inode->i_sb))
4690		return -EOPNOTSUPP;
4691	*projid = EXT4_I(inode)->i_projid;
4692	return 0;
4693}
4694
4695/*
4696 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4697 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4698 * set.
4699 */
4700static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4701{
4702	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4703		inode_set_iversion_raw(inode, val);
4704	else
4705		inode_set_iversion_queried(inode, val);
4706}
4707
4708static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags)
4709
4710{
4711	if (flags & EXT4_IGET_EA_INODE) {
4712		if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4713			return "missing EA_INODE flag";
4714		if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4715		    EXT4_I(inode)->i_file_acl)
4716			return "ea_inode with extended attributes";
4717	} else {
4718		if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4719			return "unexpected EA_INODE flag";
4720	}
4721	if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD))
4722		return "unexpected bad inode w/o EXT4_IGET_BAD";
4723	return NULL;
4724}
4725
4726struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4727			  ext4_iget_flags flags, const char *function,
4728			  unsigned int line)
4729{
4730	struct ext4_iloc iloc;
4731	struct ext4_inode *raw_inode;
4732	struct ext4_inode_info *ei;
4733	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4734	struct inode *inode;
4735	const char *err_str;
4736	journal_t *journal = EXT4_SB(sb)->s_journal;
4737	long ret;
4738	loff_t size;
4739	int block;
4740	uid_t i_uid;
4741	gid_t i_gid;
4742	projid_t i_projid;
4743
4744	if ((!(flags & EXT4_IGET_SPECIAL) &&
4745	     ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4746	      ino == le32_to_cpu(es->s_usr_quota_inum) ||
4747	      ino == le32_to_cpu(es->s_grp_quota_inum) ||
4748	      ino == le32_to_cpu(es->s_prj_quota_inum) ||
4749	      ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4750	    (ino < EXT4_ROOT_INO) ||
4751	    (ino > le32_to_cpu(es->s_inodes_count))) {
4752		if (flags & EXT4_IGET_HANDLE)
4753			return ERR_PTR(-ESTALE);
4754		__ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4755			     "inode #%lu: comm %s: iget: illegal inode #",
4756			     ino, current->comm);
4757		return ERR_PTR(-EFSCORRUPTED);
4758	}
4759
4760	inode = iget_locked(sb, ino);
4761	if (!inode)
4762		return ERR_PTR(-ENOMEM);
4763	if (!(inode->i_state & I_NEW)) {
4764		if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4765			ext4_error_inode(inode, function, line, 0, err_str);
4766			iput(inode);
4767			return ERR_PTR(-EFSCORRUPTED);
4768		}
4769		return inode;
4770	}
4771
4772	ei = EXT4_I(inode);
4773	iloc.bh = NULL;
4774
4775	ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4776	if (ret < 0)
4777		goto bad_inode;
4778	raw_inode = ext4_raw_inode(&iloc);
4779
 
 
 
 
 
 
 
4780	if ((flags & EXT4_IGET_HANDLE) &&
4781	    (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4782		ret = -ESTALE;
4783		goto bad_inode;
4784	}
4785
4786	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4787		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4788		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4789			EXT4_INODE_SIZE(inode->i_sb) ||
4790		    (ei->i_extra_isize & 3)) {
4791			ext4_error_inode(inode, function, line, 0,
4792					 "iget: bad extra_isize %u "
4793					 "(inode size %u)",
4794					 ei->i_extra_isize,
4795					 EXT4_INODE_SIZE(inode->i_sb));
4796			ret = -EFSCORRUPTED;
4797			goto bad_inode;
4798		}
4799	} else
4800		ei->i_extra_isize = 0;
4801
4802	/* Precompute checksum seed for inode metadata */
4803	if (ext4_has_metadata_csum(sb)) {
4804		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4805		__u32 csum;
4806		__le32 inum = cpu_to_le32(inode->i_ino);
4807		__le32 gen = raw_inode->i_generation;
4808		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4809				   sizeof(inum));
4810		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4811					      sizeof(gen));
4812	}
4813
4814	if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4815	    ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4816	     (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4817		ext4_error_inode_err(inode, function, line, 0,
4818				EFSBADCRC, "iget: checksum invalid");
4819		ret = -EFSBADCRC;
4820		goto bad_inode;
4821	}
4822
4823	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4824	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4825	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4826	if (ext4_has_feature_project(sb) &&
4827	    EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4828	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4829		i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4830	else
4831		i_projid = EXT4_DEF_PROJID;
4832
4833	if (!(test_opt(inode->i_sb, NO_UID32))) {
4834		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4835		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4836	}
4837	i_uid_write(inode, i_uid);
4838	i_gid_write(inode, i_gid);
4839	ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4840	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4841
4842	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
4843	ei->i_inline_off = 0;
4844	ei->i_dir_start_lookup = 0;
4845	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4846	/* We now have enough fields to check if the inode was active or not.
4847	 * This is needed because nfsd might try to access dead inodes
4848	 * the test is that same one that e2fsck uses
4849	 * NeilBrown 1999oct15
4850	 */
4851	if (inode->i_nlink == 0) {
4852		if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4853		     !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4854		    ino != EXT4_BOOT_LOADER_INO) {
4855			/* this inode is deleted or unallocated */
4856			if (flags & EXT4_IGET_SPECIAL) {
4857				ext4_error_inode(inode, function, line, 0,
4858						 "iget: special inode unallocated");
4859				ret = -EFSCORRUPTED;
4860			} else
4861				ret = -ESTALE;
4862			goto bad_inode;
4863		}
4864		/* The only unlinked inodes we let through here have
4865		 * valid i_mode and are being read by the orphan
4866		 * recovery code: that's fine, we're about to complete
4867		 * the process of deleting those.
4868		 * OR it is the EXT4_BOOT_LOADER_INO which is
4869		 * not initialized on a new filesystem. */
4870	}
4871	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4872	ext4_set_inode_flags(inode, true);
4873	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4874	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4875	if (ext4_has_feature_64bit(sb))
4876		ei->i_file_acl |=
4877			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4878	inode->i_size = ext4_isize(sb, raw_inode);
4879	if ((size = i_size_read(inode)) < 0) {
4880		ext4_error_inode(inode, function, line, 0,
4881				 "iget: bad i_size value: %lld", size);
4882		ret = -EFSCORRUPTED;
4883		goto bad_inode;
4884	}
4885	/*
4886	 * If dir_index is not enabled but there's dir with INDEX flag set,
4887	 * we'd normally treat htree data as empty space. But with metadata
4888	 * checksumming that corrupts checksums so forbid that.
4889	 */
4890	if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4891	    ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4892		ext4_error_inode(inode, function, line, 0,
4893			 "iget: Dir with htree data on filesystem without dir_index feature.");
4894		ret = -EFSCORRUPTED;
4895		goto bad_inode;
4896	}
4897	ei->i_disksize = inode->i_size;
4898#ifdef CONFIG_QUOTA
4899	ei->i_reserved_quota = 0;
4900#endif
4901	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4902	ei->i_block_group = iloc.block_group;
4903	ei->i_last_alloc_group = ~0;
4904	/*
4905	 * NOTE! The in-memory inode i_data array is in little-endian order
4906	 * even on big-endian machines: we do NOT byteswap the block numbers!
4907	 */
4908	for (block = 0; block < EXT4_N_BLOCKS; block++)
4909		ei->i_data[block] = raw_inode->i_block[block];
4910	INIT_LIST_HEAD(&ei->i_orphan);
4911	ext4_fc_init_inode(&ei->vfs_inode);
4912
4913	/*
4914	 * Set transaction id's of transactions that have to be committed
4915	 * to finish f[data]sync. We set them to currently running transaction
4916	 * as we cannot be sure that the inode or some of its metadata isn't
4917	 * part of the transaction - the inode could have been reclaimed and
4918	 * now it is reread from disk.
4919	 */
4920	if (journal) {
4921		transaction_t *transaction;
4922		tid_t tid;
4923
4924		read_lock(&journal->j_state_lock);
4925		if (journal->j_running_transaction)
4926			transaction = journal->j_running_transaction;
4927		else
4928			transaction = journal->j_committing_transaction;
4929		if (transaction)
4930			tid = transaction->t_tid;
4931		else
4932			tid = journal->j_commit_sequence;
4933		read_unlock(&journal->j_state_lock);
4934		ei->i_sync_tid = tid;
4935		ei->i_datasync_tid = tid;
4936	}
4937
4938	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4939		if (ei->i_extra_isize == 0) {
4940			/* The extra space is currently unused. Use it. */
4941			BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4942			ei->i_extra_isize = sizeof(struct ext4_inode) -
4943					    EXT4_GOOD_OLD_INODE_SIZE;
4944		} else {
4945			ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4946			if (ret)
4947				goto bad_inode;
4948		}
4949	}
4950
4951	EXT4_INODE_GET_CTIME(inode, raw_inode);
4952	EXT4_INODE_GET_ATIME(inode, raw_inode);
4953	EXT4_INODE_GET_MTIME(inode, raw_inode);
4954	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4955
4956	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4957		u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4958
4959		if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4960			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4961				ivers |=
4962		    (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4963		}
4964		ext4_inode_set_iversion_queried(inode, ivers);
4965	}
4966
4967	ret = 0;
4968	if (ei->i_file_acl &&
4969	    !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4970		ext4_error_inode(inode, function, line, 0,
4971				 "iget: bad extended attribute block %llu",
4972				 ei->i_file_acl);
4973		ret = -EFSCORRUPTED;
4974		goto bad_inode;
4975	} else if (!ext4_has_inline_data(inode)) {
4976		/* validate the block references in the inode */
4977		if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4978			(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4979			(S_ISLNK(inode->i_mode) &&
4980			!ext4_inode_is_fast_symlink(inode)))) {
4981			if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4982				ret = ext4_ext_check_inode(inode);
4983			else
4984				ret = ext4_ind_check_inode(inode);
4985		}
4986	}
4987	if (ret)
4988		goto bad_inode;
4989
4990	if (S_ISREG(inode->i_mode)) {
4991		inode->i_op = &ext4_file_inode_operations;
4992		inode->i_fop = &ext4_file_operations;
4993		ext4_set_aops(inode);
4994	} else if (S_ISDIR(inode->i_mode)) {
4995		inode->i_op = &ext4_dir_inode_operations;
4996		inode->i_fop = &ext4_dir_operations;
4997	} else if (S_ISLNK(inode->i_mode)) {
4998		/* VFS does not allow setting these so must be corruption */
4999		if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5000			ext4_error_inode(inode, function, line, 0,
5001					 "iget: immutable or append flags "
5002					 "not allowed on symlinks");
5003			ret = -EFSCORRUPTED;
5004			goto bad_inode;
5005		}
5006		if (IS_ENCRYPTED(inode)) {
5007			inode->i_op = &ext4_encrypted_symlink_inode_operations;
5008		} else if (ext4_inode_is_fast_symlink(inode)) {
5009			inode->i_link = (char *)ei->i_data;
5010			inode->i_op = &ext4_fast_symlink_inode_operations;
5011			nd_terminate_link(ei->i_data, inode->i_size,
5012				sizeof(ei->i_data) - 1);
5013		} else {
5014			inode->i_op = &ext4_symlink_inode_operations;
5015		}
5016	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5017	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5018		inode->i_op = &ext4_special_inode_operations;
5019		if (raw_inode->i_block[0])
5020			init_special_inode(inode, inode->i_mode,
5021			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5022		else
5023			init_special_inode(inode, inode->i_mode,
5024			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5025	} else if (ino == EXT4_BOOT_LOADER_INO) {
5026		make_bad_inode(inode);
5027	} else {
5028		ret = -EFSCORRUPTED;
5029		ext4_error_inode(inode, function, line, 0,
5030				 "iget: bogus i_mode (%o)", inode->i_mode);
5031		goto bad_inode;
5032	}
5033	if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) {
5034		ext4_error_inode(inode, function, line, 0,
5035				 "casefold flag without casefold feature");
5036		ret = -EFSCORRUPTED;
5037		goto bad_inode;
5038	}
5039	if ((err_str = check_igot_inode(inode, flags)) != NULL) {
5040		ext4_error_inode(inode, function, line, 0, err_str);
5041		ret = -EFSCORRUPTED;
5042		goto bad_inode;
5043	}
5044
5045	brelse(iloc.bh);
5046	unlock_new_inode(inode);
5047	return inode;
5048
5049bad_inode:
5050	brelse(iloc.bh);
5051	iget_failed(inode);
5052	return ERR_PTR(ret);
5053}
5054
5055static void __ext4_update_other_inode_time(struct super_block *sb,
5056					   unsigned long orig_ino,
5057					   unsigned long ino,
5058					   struct ext4_inode *raw_inode)
5059{
5060	struct inode *inode;
5061
5062	inode = find_inode_by_ino_rcu(sb, ino);
5063	if (!inode)
5064		return;
5065
5066	if (!inode_is_dirtytime_only(inode))
5067		return;
5068
5069	spin_lock(&inode->i_lock);
5070	if (inode_is_dirtytime_only(inode)) {
5071		struct ext4_inode_info	*ei = EXT4_I(inode);
5072
5073		inode->i_state &= ~I_DIRTY_TIME;
5074		spin_unlock(&inode->i_lock);
5075
5076		spin_lock(&ei->i_raw_lock);
5077		EXT4_INODE_SET_CTIME(inode, raw_inode);
5078		EXT4_INODE_SET_MTIME(inode, raw_inode);
5079		EXT4_INODE_SET_ATIME(inode, raw_inode);
5080		ext4_inode_csum_set(inode, raw_inode, ei);
5081		spin_unlock(&ei->i_raw_lock);
5082		trace_ext4_other_inode_update_time(inode, orig_ino);
5083		return;
5084	}
5085	spin_unlock(&inode->i_lock);
5086}
5087
5088/*
5089 * Opportunistically update the other time fields for other inodes in
5090 * the same inode table block.
5091 */
5092static void ext4_update_other_inodes_time(struct super_block *sb,
5093					  unsigned long orig_ino, char *buf)
5094{
5095	unsigned long ino;
5096	int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5097	int inode_size = EXT4_INODE_SIZE(sb);
5098
5099	/*
5100	 * Calculate the first inode in the inode table block.  Inode
5101	 * numbers are one-based.  That is, the first inode in a block
5102	 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5103	 */
5104	ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5105	rcu_read_lock();
5106	for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5107		if (ino == orig_ino)
5108			continue;
5109		__ext4_update_other_inode_time(sb, orig_ino, ino,
5110					       (struct ext4_inode *)buf);
5111	}
5112	rcu_read_unlock();
5113}
5114
5115/*
5116 * Post the struct inode info into an on-disk inode location in the
5117 * buffer-cache.  This gobbles the caller's reference to the
5118 * buffer_head in the inode location struct.
5119 *
5120 * The caller must have write access to iloc->bh.
5121 */
5122static int ext4_do_update_inode(handle_t *handle,
5123				struct inode *inode,
5124				struct ext4_iloc *iloc)
5125{
5126	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5127	struct ext4_inode_info *ei = EXT4_I(inode);
5128	struct buffer_head *bh = iloc->bh;
5129	struct super_block *sb = inode->i_sb;
5130	int err;
5131	int need_datasync = 0, set_large_file = 0;
5132
5133	spin_lock(&ei->i_raw_lock);
5134
5135	/*
5136	 * For fields not tracked in the in-memory inode, initialise them
5137	 * to zero for new inodes.
5138	 */
5139	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5140		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5141
5142	if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5143		need_datasync = 1;
5144	if (ei->i_disksize > 0x7fffffffULL) {
5145		if (!ext4_has_feature_large_file(sb) ||
5146		    EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5147			set_large_file = 1;
5148	}
5149
5150	err = ext4_fill_raw_inode(inode, raw_inode);
5151	spin_unlock(&ei->i_raw_lock);
5152	if (err) {
5153		EXT4_ERROR_INODE(inode, "corrupted inode contents");
5154		goto out_brelse;
5155	}
5156
5157	if (inode->i_sb->s_flags & SB_LAZYTIME)
5158		ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5159					      bh->b_data);
5160
5161	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5162	err = ext4_handle_dirty_metadata(handle, NULL, bh);
5163	if (err)
5164		goto out_error;
5165	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5166	if (set_large_file) {
5167		BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5168		err = ext4_journal_get_write_access(handle, sb,
5169						    EXT4_SB(sb)->s_sbh,
5170						    EXT4_JTR_NONE);
5171		if (err)
5172			goto out_error;
5173		lock_buffer(EXT4_SB(sb)->s_sbh);
5174		ext4_set_feature_large_file(sb);
5175		ext4_superblock_csum_set(sb);
5176		unlock_buffer(EXT4_SB(sb)->s_sbh);
5177		ext4_handle_sync(handle);
5178		err = ext4_handle_dirty_metadata(handle, NULL,
5179						 EXT4_SB(sb)->s_sbh);
5180	}
5181	ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5182out_error:
5183	ext4_std_error(inode->i_sb, err);
5184out_brelse:
5185	brelse(bh);
5186	return err;
5187}
5188
5189/*
5190 * ext4_write_inode()
5191 *
5192 * We are called from a few places:
5193 *
5194 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5195 *   Here, there will be no transaction running. We wait for any running
5196 *   transaction to commit.
5197 *
5198 * - Within flush work (sys_sync(), kupdate and such).
5199 *   We wait on commit, if told to.
5200 *
5201 * - Within iput_final() -> write_inode_now()
5202 *   We wait on commit, if told to.
5203 *
5204 * In all cases it is actually safe for us to return without doing anything,
5205 * because the inode has been copied into a raw inode buffer in
5206 * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
5207 * writeback.
5208 *
5209 * Note that we are absolutely dependent upon all inode dirtiers doing the
5210 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5211 * which we are interested.
5212 *
5213 * It would be a bug for them to not do this.  The code:
5214 *
5215 *	mark_inode_dirty(inode)
5216 *	stuff();
5217 *	inode->i_size = expr;
5218 *
5219 * is in error because write_inode() could occur while `stuff()' is running,
5220 * and the new i_size will be lost.  Plus the inode will no longer be on the
5221 * superblock's dirty inode list.
5222 */
5223int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5224{
5225	int err;
5226
5227	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
 
5228		return 0;
5229
5230	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5231		return -EIO;
5232
5233	if (EXT4_SB(inode->i_sb)->s_journal) {
5234		if (ext4_journal_current_handle()) {
5235			ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5236			dump_stack();
5237			return -EIO;
5238		}
5239
5240		/*
5241		 * No need to force transaction in WB_SYNC_NONE mode. Also
5242		 * ext4_sync_fs() will force the commit after everything is
5243		 * written.
5244		 */
5245		if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5246			return 0;
5247
5248		err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5249						EXT4_I(inode)->i_sync_tid);
5250	} else {
5251		struct ext4_iloc iloc;
5252
5253		err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5254		if (err)
5255			return err;
5256		/*
5257		 * sync(2) will flush the whole buffer cache. No need to do
5258		 * it here separately for each inode.
5259		 */
5260		if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5261			sync_dirty_buffer(iloc.bh);
5262		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5263			ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5264					       "IO error syncing inode");
5265			err = -EIO;
5266		}
5267		brelse(iloc.bh);
5268	}
5269	return err;
5270}
5271
5272/*
5273 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5274 * buffers that are attached to a folio straddling i_size and are undergoing
5275 * commit. In that case we have to wait for commit to finish and try again.
5276 */
5277static void ext4_wait_for_tail_page_commit(struct inode *inode)
5278{
5279	unsigned offset;
5280	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5281	tid_t commit_tid;
5282	int ret;
5283	bool has_transaction;
5284
5285	offset = inode->i_size & (PAGE_SIZE - 1);
5286	/*
5287	 * If the folio is fully truncated, we don't need to wait for any commit
5288	 * (and we even should not as __ext4_journalled_invalidate_folio() may
5289	 * strip all buffers from the folio but keep the folio dirty which can then
5290	 * confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5291	 * buffers). Also we don't need to wait for any commit if all buffers in
5292	 * the folio remain valid. This is most beneficial for the common case of
5293	 * blocksize == PAGESIZE.
5294	 */
5295	if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5296		return;
5297	while (1) {
5298		struct folio *folio = filemap_lock_folio(inode->i_mapping,
5299				      inode->i_size >> PAGE_SHIFT);
5300		if (IS_ERR(folio))
5301			return;
5302		ret = __ext4_journalled_invalidate_folio(folio, offset,
5303						folio_size(folio) - offset);
5304		folio_unlock(folio);
5305		folio_put(folio);
5306		if (ret != -EBUSY)
5307			return;
5308		has_transaction = false;
5309		read_lock(&journal->j_state_lock);
5310		if (journal->j_committing_transaction) {
5311			commit_tid = journal->j_committing_transaction->t_tid;
5312			has_transaction = true;
5313		}
5314		read_unlock(&journal->j_state_lock);
5315		if (has_transaction)
5316			jbd2_log_wait_commit(journal, commit_tid);
5317	}
5318}
5319
5320/*
5321 * ext4_setattr()
5322 *
5323 * Called from notify_change.
5324 *
5325 * We want to trap VFS attempts to truncate the file as soon as
5326 * possible.  In particular, we want to make sure that when the VFS
5327 * shrinks i_size, we put the inode on the orphan list and modify
5328 * i_disksize immediately, so that during the subsequent flushing of
5329 * dirty pages and freeing of disk blocks, we can guarantee that any
5330 * commit will leave the blocks being flushed in an unused state on
5331 * disk.  (On recovery, the inode will get truncated and the blocks will
5332 * be freed, so we have a strong guarantee that no future commit will
5333 * leave these blocks visible to the user.)
5334 *
5335 * Another thing we have to assure is that if we are in ordered mode
5336 * and inode is still attached to the committing transaction, we must
5337 * we start writeout of all the dirty pages which are being truncated.
5338 * This way we are sure that all the data written in the previous
5339 * transaction are already on disk (truncate waits for pages under
5340 * writeback).
5341 *
5342 * Called with inode->i_rwsem down.
5343 */
5344int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5345		 struct iattr *attr)
5346{
5347	struct inode *inode = d_inode(dentry);
5348	int error, rc = 0;
5349	int orphan = 0;
5350	const unsigned int ia_valid = attr->ia_valid;
5351	bool inc_ivers = true;
5352
5353	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5354		return -EIO;
5355
5356	if (unlikely(IS_IMMUTABLE(inode)))
5357		return -EPERM;
5358
5359	if (unlikely(IS_APPEND(inode) &&
5360		     (ia_valid & (ATTR_MODE | ATTR_UID |
5361				  ATTR_GID | ATTR_TIMES_SET))))
5362		return -EPERM;
5363
5364	error = setattr_prepare(idmap, dentry, attr);
5365	if (error)
5366		return error;
5367
5368	error = fscrypt_prepare_setattr(dentry, attr);
5369	if (error)
5370		return error;
5371
5372	error = fsverity_prepare_setattr(dentry, attr);
5373	if (error)
5374		return error;
5375
5376	if (is_quota_modification(idmap, inode, attr)) {
5377		error = dquot_initialize(inode);
5378		if (error)
5379			return error;
5380	}
5381
5382	if (i_uid_needs_update(idmap, attr, inode) ||
5383	    i_gid_needs_update(idmap, attr, inode)) {
5384		handle_t *handle;
5385
5386		/* (user+group)*(old+new) structure, inode write (sb,
5387		 * inode block, ? - but truncate inode update has it) */
5388		handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5389			(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5390			 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5391		if (IS_ERR(handle)) {
5392			error = PTR_ERR(handle);
5393			goto err_out;
5394		}
5395
5396		/* dquot_transfer() calls back ext4_get_inode_usage() which
5397		 * counts xattr inode references.
5398		 */
5399		down_read(&EXT4_I(inode)->xattr_sem);
5400		error = dquot_transfer(idmap, inode, attr);
5401		up_read(&EXT4_I(inode)->xattr_sem);
5402
5403		if (error) {
5404			ext4_journal_stop(handle);
5405			return error;
5406		}
5407		/* Update corresponding info in inode so that everything is in
5408		 * one transaction */
5409		i_uid_update(idmap, attr, inode);
5410		i_gid_update(idmap, attr, inode);
5411		error = ext4_mark_inode_dirty(handle, inode);
5412		ext4_journal_stop(handle);
5413		if (unlikely(error)) {
5414			return error;
5415		}
5416	}
5417
5418	if (attr->ia_valid & ATTR_SIZE) {
5419		handle_t *handle;
5420		loff_t oldsize = inode->i_size;
5421		loff_t old_disksize;
5422		int shrink = (attr->ia_size < inode->i_size);
5423
5424		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5425			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5426
5427			if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5428				return -EFBIG;
5429			}
5430		}
5431		if (!S_ISREG(inode->i_mode)) {
5432			return -EINVAL;
5433		}
5434
5435		if (attr->ia_size == inode->i_size)
5436			inc_ivers = false;
5437
5438		if (shrink) {
5439			if (ext4_should_order_data(inode)) {
5440				error = ext4_begin_ordered_truncate(inode,
5441							    attr->ia_size);
5442				if (error)
5443					goto err_out;
5444			}
5445			/*
5446			 * Blocks are going to be removed from the inode. Wait
5447			 * for dio in flight.
5448			 */
5449			inode_dio_wait(inode);
5450		}
5451
5452		filemap_invalidate_lock(inode->i_mapping);
5453
5454		rc = ext4_break_layouts(inode);
5455		if (rc) {
5456			filemap_invalidate_unlock(inode->i_mapping);
5457			goto err_out;
5458		}
5459
5460		if (attr->ia_size != inode->i_size) {
5461			/* attach jbd2 jinode for EOF folio tail zeroing */
5462			if (attr->ia_size & (inode->i_sb->s_blocksize - 1) ||
5463			    oldsize & (inode->i_sb->s_blocksize - 1)) {
5464				error = ext4_inode_attach_jinode(inode);
5465				if (error)
5466					goto err_out;
5467			}
5468
5469			handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5470			if (IS_ERR(handle)) {
5471				error = PTR_ERR(handle);
5472				goto out_mmap_sem;
5473			}
5474			if (ext4_handle_valid(handle) && shrink) {
5475				error = ext4_orphan_add(handle, inode);
5476				orphan = 1;
5477			}
5478			/*
5479			 * Update c/mtime and tail zero the EOF folio on
5480			 * truncate up. ext4_truncate() handles the shrink case
5481			 * below.
5482			 */
5483			if (!shrink) {
5484				inode_set_mtime_to_ts(inode,
5485						      inode_set_ctime_current(inode));
5486				if (oldsize & (inode->i_sb->s_blocksize - 1))
5487					ext4_block_truncate_page(handle,
5488							inode->i_mapping, oldsize);
5489			}
5490
5491			if (shrink)
5492				ext4_fc_track_range(handle, inode,
5493					(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5494					inode->i_sb->s_blocksize_bits,
5495					EXT_MAX_BLOCKS - 1);
5496			else
5497				ext4_fc_track_range(
5498					handle, inode,
5499					(oldsize > 0 ? oldsize - 1 : oldsize) >>
5500					inode->i_sb->s_blocksize_bits,
5501					(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5502					inode->i_sb->s_blocksize_bits);
5503
5504			down_write(&EXT4_I(inode)->i_data_sem);
5505			old_disksize = EXT4_I(inode)->i_disksize;
5506			EXT4_I(inode)->i_disksize = attr->ia_size;
5507			rc = ext4_mark_inode_dirty(handle, inode);
5508			if (!error)
5509				error = rc;
5510			/*
5511			 * We have to update i_size under i_data_sem together
5512			 * with i_disksize to avoid races with writeback code
5513			 * running ext4_wb_update_i_disksize().
5514			 */
5515			if (!error)
5516				i_size_write(inode, attr->ia_size);
5517			else
5518				EXT4_I(inode)->i_disksize = old_disksize;
5519			up_write(&EXT4_I(inode)->i_data_sem);
5520			ext4_journal_stop(handle);
5521			if (error)
5522				goto out_mmap_sem;
5523			if (!shrink) {
5524				pagecache_isize_extended(inode, oldsize,
5525							 inode->i_size);
5526			} else if (ext4_should_journal_data(inode)) {
5527				ext4_wait_for_tail_page_commit(inode);
5528			}
5529		}
5530
5531		/*
5532		 * Truncate pagecache after we've waited for commit
5533		 * in data=journal mode to make pages freeable.
5534		 */
5535		truncate_pagecache(inode, inode->i_size);
5536		/*
5537		 * Call ext4_truncate() even if i_size didn't change to
5538		 * truncate possible preallocated blocks.
5539		 */
5540		if (attr->ia_size <= oldsize) {
5541			rc = ext4_truncate(inode);
5542			if (rc)
5543				error = rc;
5544		}
5545out_mmap_sem:
5546		filemap_invalidate_unlock(inode->i_mapping);
5547	}
5548
5549	if (!error) {
5550		if (inc_ivers)
5551			inode_inc_iversion(inode);
5552		setattr_copy(idmap, inode, attr);
5553		mark_inode_dirty(inode);
5554	}
5555
5556	/*
5557	 * If the call to ext4_truncate failed to get a transaction handle at
5558	 * all, we need to clean up the in-core orphan list manually.
5559	 */
5560	if (orphan && inode->i_nlink)
5561		ext4_orphan_del(NULL, inode);
5562
5563	if (!error && (ia_valid & ATTR_MODE))
5564		rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
5565
5566err_out:
5567	if  (error)
5568		ext4_std_error(inode->i_sb, error);
5569	if (!error)
5570		error = rc;
5571	return error;
5572}
5573
5574u32 ext4_dio_alignment(struct inode *inode)
5575{
5576	if (fsverity_active(inode))
5577		return 0;
5578	if (ext4_should_journal_data(inode))
5579		return 0;
5580	if (ext4_has_inline_data(inode))
5581		return 0;
5582	if (IS_ENCRYPTED(inode)) {
5583		if (!fscrypt_dio_supported(inode))
5584			return 0;
5585		return i_blocksize(inode);
5586	}
5587	return 1; /* use the iomap defaults */
5588}
5589
5590int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
5591		 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5592{
5593	struct inode *inode = d_inode(path->dentry);
5594	struct ext4_inode *raw_inode;
5595	struct ext4_inode_info *ei = EXT4_I(inode);
5596	unsigned int flags;
5597
5598	if ((request_mask & STATX_BTIME) &&
5599	    EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5600		stat->result_mask |= STATX_BTIME;
5601		stat->btime.tv_sec = ei->i_crtime.tv_sec;
5602		stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5603	}
5604
5605	/*
5606	 * Return the DIO alignment restrictions if requested.  We only return
5607	 * this information when requested, since on encrypted files it might
5608	 * take a fair bit of work to get if the file wasn't opened recently.
5609	 */
5610	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5611		u32 dio_align = ext4_dio_alignment(inode);
5612
5613		stat->result_mask |= STATX_DIOALIGN;
5614		if (dio_align == 1) {
5615			struct block_device *bdev = inode->i_sb->s_bdev;
5616
5617			/* iomap defaults */
5618			stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5619			stat->dio_offset_align = bdev_logical_block_size(bdev);
5620		} else {
5621			stat->dio_mem_align = dio_align;
5622			stat->dio_offset_align = dio_align;
5623		}
5624	}
5625
5626	if ((request_mask & STATX_WRITE_ATOMIC) && S_ISREG(inode->i_mode)) {
5627		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5628		unsigned int awu_min = 0, awu_max = 0;
5629
5630		if (ext4_inode_can_atomic_write(inode)) {
5631			awu_min = sbi->s_awu_min;
5632			awu_max = sbi->s_awu_max;
5633		}
5634
5635		generic_fill_statx_atomic_writes(stat, awu_min, awu_max);
5636	}
5637
5638	flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5639	if (flags & EXT4_APPEND_FL)
5640		stat->attributes |= STATX_ATTR_APPEND;
5641	if (flags & EXT4_COMPR_FL)
5642		stat->attributes |= STATX_ATTR_COMPRESSED;
5643	if (flags & EXT4_ENCRYPT_FL)
5644		stat->attributes |= STATX_ATTR_ENCRYPTED;
5645	if (flags & EXT4_IMMUTABLE_FL)
5646		stat->attributes |= STATX_ATTR_IMMUTABLE;
5647	if (flags & EXT4_NODUMP_FL)
5648		stat->attributes |= STATX_ATTR_NODUMP;
5649	if (flags & EXT4_VERITY_FL)
5650		stat->attributes |= STATX_ATTR_VERITY;
5651
5652	stat->attributes_mask |= (STATX_ATTR_APPEND |
5653				  STATX_ATTR_COMPRESSED |
5654				  STATX_ATTR_ENCRYPTED |
5655				  STATX_ATTR_IMMUTABLE |
5656				  STATX_ATTR_NODUMP |
5657				  STATX_ATTR_VERITY);
5658
5659	generic_fillattr(idmap, request_mask, inode, stat);
5660	return 0;
5661}
5662
5663int ext4_file_getattr(struct mnt_idmap *idmap,
5664		      const struct path *path, struct kstat *stat,
5665		      u32 request_mask, unsigned int query_flags)
5666{
5667	struct inode *inode = d_inode(path->dentry);
5668	u64 delalloc_blocks;
5669
5670	ext4_getattr(idmap, path, stat, request_mask, query_flags);
5671
5672	/*
5673	 * If there is inline data in the inode, the inode will normally not
5674	 * have data blocks allocated (it may have an external xattr block).
5675	 * Report at least one sector for such files, so tools like tar, rsync,
5676	 * others don't incorrectly think the file is completely sparse.
5677	 */
5678	if (unlikely(ext4_has_inline_data(inode)))
5679		stat->blocks += (stat->size + 511) >> 9;
5680
5681	/*
5682	 * We can't update i_blocks if the block allocation is delayed
5683	 * otherwise in the case of system crash before the real block
5684	 * allocation is done, we will have i_blocks inconsistent with
5685	 * on-disk file blocks.
5686	 * We always keep i_blocks updated together with real
5687	 * allocation. But to not confuse with user, stat
5688	 * will return the blocks that include the delayed allocation
5689	 * blocks for this file.
5690	 */
5691	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5692				   EXT4_I(inode)->i_reserved_data_blocks);
5693	stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5694	return 0;
5695}
5696
5697static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5698				   int pextents)
5699{
5700	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5701		return ext4_ind_trans_blocks(inode, lblocks);
5702	return ext4_ext_index_trans_blocks(inode, pextents);
5703}
5704
5705/*
5706 * Account for index blocks, block groups bitmaps and block group
5707 * descriptor blocks if modify datablocks and index blocks
5708 * worse case, the indexs blocks spread over different block groups
5709 *
5710 * If datablocks are discontiguous, they are possible to spread over
5711 * different block groups too. If they are contiguous, with flexbg,
5712 * they could still across block group boundary.
5713 *
5714 * Also account for superblock, inode, quota and xattr blocks
5715 */
5716static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5717				  int pextents)
5718{
5719	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5720	int gdpblocks;
5721	int idxblocks;
5722	int ret;
5723
5724	/*
5725	 * How many index blocks need to touch to map @lblocks logical blocks
5726	 * to @pextents physical extents?
5727	 */
5728	idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5729
5730	ret = idxblocks;
5731
5732	/*
5733	 * Now let's see how many group bitmaps and group descriptors need
5734	 * to account
5735	 */
5736	groups = idxblocks + pextents;
5737	gdpblocks = groups;
5738	if (groups > ngroups)
5739		groups = ngroups;
5740	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5741		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5742
5743	/* bitmaps and block group descriptor blocks */
5744	ret += groups + gdpblocks;
5745
5746	/* Blocks for super block, inode, quota and xattr blocks */
5747	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5748
5749	return ret;
5750}
5751
5752/*
5753 * Calculate the total number of credits to reserve to fit
5754 * the modification of a single pages into a single transaction,
5755 * which may include multiple chunks of block allocations.
5756 *
5757 * This could be called via ext4_write_begin()
5758 *
5759 * We need to consider the worse case, when
5760 * one new block per extent.
5761 */
5762int ext4_writepage_trans_blocks(struct inode *inode)
5763{
5764	int bpp = ext4_journal_blocks_per_page(inode);
5765	int ret;
5766
5767	ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5768
5769	/* Account for data blocks for journalled mode */
5770	if (ext4_should_journal_data(inode))
5771		ret += bpp;
5772	return ret;
5773}
5774
5775/*
5776 * Calculate the journal credits for a chunk of data modification.
5777 *
5778 * This is called from DIO, fallocate or whoever calling
5779 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5780 *
5781 * journal buffers for data blocks are not included here, as DIO
5782 * and fallocate do no need to journal data buffers.
5783 */
5784int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5785{
5786	return ext4_meta_trans_blocks(inode, nrblocks, 1);
5787}
5788
5789/*
5790 * The caller must have previously called ext4_reserve_inode_write().
5791 * Give this, we know that the caller already has write access to iloc->bh.
5792 */
5793int ext4_mark_iloc_dirty(handle_t *handle,
5794			 struct inode *inode, struct ext4_iloc *iloc)
5795{
5796	int err = 0;
5797
5798	if (unlikely(ext4_forced_shutdown(inode->i_sb))) {
5799		put_bh(iloc->bh);
5800		return -EIO;
5801	}
5802	ext4_fc_track_inode(handle, inode);
5803
5804	/* the do_update_inode consumes one bh->b_count */
5805	get_bh(iloc->bh);
5806
5807	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5808	err = ext4_do_update_inode(handle, inode, iloc);
5809	put_bh(iloc->bh);
5810	return err;
5811}
5812
5813/*
5814 * On success, We end up with an outstanding reference count against
5815 * iloc->bh.  This _must_ be cleaned up later.
5816 */
5817
5818int
5819ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5820			 struct ext4_iloc *iloc)
5821{
5822	int err;
5823
5824	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5825		return -EIO;
5826
5827	err = ext4_get_inode_loc(inode, iloc);
5828	if (!err) {
5829		BUFFER_TRACE(iloc->bh, "get_write_access");
5830		err = ext4_journal_get_write_access(handle, inode->i_sb,
5831						    iloc->bh, EXT4_JTR_NONE);
5832		if (err) {
5833			brelse(iloc->bh);
5834			iloc->bh = NULL;
5835		}
5836	}
5837	ext4_std_error(inode->i_sb, err);
5838	return err;
5839}
5840
5841static int __ext4_expand_extra_isize(struct inode *inode,
5842				     unsigned int new_extra_isize,
5843				     struct ext4_iloc *iloc,
5844				     handle_t *handle, int *no_expand)
5845{
5846	struct ext4_inode *raw_inode;
5847	struct ext4_xattr_ibody_header *header;
5848	unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5849	struct ext4_inode_info *ei = EXT4_I(inode);
5850	int error;
5851
5852	/* this was checked at iget time, but double check for good measure */
5853	if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5854	    (ei->i_extra_isize & 3)) {
5855		EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5856				 ei->i_extra_isize,
5857				 EXT4_INODE_SIZE(inode->i_sb));
5858		return -EFSCORRUPTED;
5859	}
5860	if ((new_extra_isize < ei->i_extra_isize) ||
5861	    (new_extra_isize < 4) ||
5862	    (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5863		return -EINVAL;	/* Should never happen */
5864
5865	raw_inode = ext4_raw_inode(iloc);
5866
5867	header = IHDR(inode, raw_inode);
5868
5869	/* No extended attributes present */
5870	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5871	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5872		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5873		       EXT4_I(inode)->i_extra_isize, 0,
5874		       new_extra_isize - EXT4_I(inode)->i_extra_isize);
5875		EXT4_I(inode)->i_extra_isize = new_extra_isize;
5876		return 0;
5877	}
5878
5879	/*
5880	 * We may need to allocate external xattr block so we need quotas
5881	 * initialized. Here we can be called with various locks held so we
5882	 * cannot affort to initialize quotas ourselves. So just bail.
5883	 */
5884	if (dquot_initialize_needed(inode))
5885		return -EAGAIN;
5886
5887	/* try to expand with EAs present */
5888	error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5889					   raw_inode, handle);
5890	if (error) {
5891		/*
5892		 * Inode size expansion failed; don't try again
5893		 */
5894		*no_expand = 1;
5895	}
5896
5897	return error;
5898}
5899
5900/*
5901 * Expand an inode by new_extra_isize bytes.
5902 * Returns 0 on success or negative error number on failure.
5903 */
5904static int ext4_try_to_expand_extra_isize(struct inode *inode,
5905					  unsigned int new_extra_isize,
5906					  struct ext4_iloc iloc,
5907					  handle_t *handle)
5908{
5909	int no_expand;
5910	int error;
5911
5912	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5913		return -EOVERFLOW;
5914
5915	/*
5916	 * In nojournal mode, we can immediately attempt to expand
5917	 * the inode.  When journaled, we first need to obtain extra
5918	 * buffer credits since we may write into the EA block
5919	 * with this same handle. If journal_extend fails, then it will
5920	 * only result in a minor loss of functionality for that inode.
5921	 * If this is felt to be critical, then e2fsck should be run to
5922	 * force a large enough s_min_extra_isize.
5923	 */
5924	if (ext4_journal_extend(handle,
5925				EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5926		return -ENOSPC;
5927
5928	if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5929		return -EBUSY;
5930
5931	error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5932					  handle, &no_expand);
5933	ext4_write_unlock_xattr(inode, &no_expand);
5934
5935	return error;
5936}
5937
5938int ext4_expand_extra_isize(struct inode *inode,
5939			    unsigned int new_extra_isize,
5940			    struct ext4_iloc *iloc)
5941{
5942	handle_t *handle;
5943	int no_expand;
5944	int error, rc;
5945
5946	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5947		brelse(iloc->bh);
5948		return -EOVERFLOW;
5949	}
5950
5951	handle = ext4_journal_start(inode, EXT4_HT_INODE,
5952				    EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5953	if (IS_ERR(handle)) {
5954		error = PTR_ERR(handle);
5955		brelse(iloc->bh);
5956		return error;
5957	}
5958
5959	ext4_write_lock_xattr(inode, &no_expand);
5960
5961	BUFFER_TRACE(iloc->bh, "get_write_access");
5962	error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5963					      EXT4_JTR_NONE);
5964	if (error) {
5965		brelse(iloc->bh);
5966		goto out_unlock;
5967	}
5968
5969	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5970					  handle, &no_expand);
5971
5972	rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5973	if (!error)
5974		error = rc;
5975
5976out_unlock:
5977	ext4_write_unlock_xattr(inode, &no_expand);
5978	ext4_journal_stop(handle);
5979	return error;
5980}
5981
5982/*
5983 * What we do here is to mark the in-core inode as clean with respect to inode
5984 * dirtiness (it may still be data-dirty).
5985 * This means that the in-core inode may be reaped by prune_icache
5986 * without having to perform any I/O.  This is a very good thing,
5987 * because *any* task may call prune_icache - even ones which
5988 * have a transaction open against a different journal.
5989 *
5990 * Is this cheating?  Not really.  Sure, we haven't written the
5991 * inode out, but prune_icache isn't a user-visible syncing function.
5992 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5993 * we start and wait on commits.
5994 */
5995int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5996				const char *func, unsigned int line)
5997{
5998	struct ext4_iloc iloc;
5999	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6000	int err;
6001
6002	might_sleep();
6003	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6004	err = ext4_reserve_inode_write(handle, inode, &iloc);
6005	if (err)
6006		goto out;
6007
6008	if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6009		ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6010					       iloc, handle);
6011
6012	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
6013out:
6014	if (unlikely(err))
6015		ext4_error_inode_err(inode, func, line, 0, err,
6016					"mark_inode_dirty error");
6017	return err;
6018}
6019
6020/*
6021 * ext4_dirty_inode() is called from __mark_inode_dirty()
6022 *
6023 * We're really interested in the case where a file is being extended.
6024 * i_size has been changed by generic_commit_write() and we thus need
6025 * to include the updated inode in the current transaction.
6026 *
6027 * Also, dquot_alloc_block() will always dirty the inode when blocks
6028 * are allocated to the file.
6029 *
6030 * If the inode is marked synchronous, we don't honour that here - doing
6031 * so would cause a commit on atime updates, which we don't bother doing.
6032 * We handle synchronous inodes at the highest possible level.
6033 */
6034void ext4_dirty_inode(struct inode *inode, int flags)
6035{
6036	handle_t *handle;
6037
6038	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6039	if (IS_ERR(handle))
6040		return;
6041	ext4_mark_inode_dirty(handle, inode);
6042	ext4_journal_stop(handle);
6043}
6044
6045int ext4_change_inode_journal_flag(struct inode *inode, int val)
6046{
6047	journal_t *journal;
6048	handle_t *handle;
6049	int err;
6050	int alloc_ctx;
6051
6052	/*
6053	 * We have to be very careful here: changing a data block's
6054	 * journaling status dynamically is dangerous.  If we write a
6055	 * data block to the journal, change the status and then delete
6056	 * that block, we risk forgetting to revoke the old log record
6057	 * from the journal and so a subsequent replay can corrupt data.
6058	 * So, first we make sure that the journal is empty and that
6059	 * nobody is changing anything.
6060	 */
6061
6062	journal = EXT4_JOURNAL(inode);
6063	if (!journal)
6064		return 0;
6065	if (is_journal_aborted(journal))
6066		return -EROFS;
6067
6068	/* Wait for all existing dio workers */
6069	inode_dio_wait(inode);
6070
6071	/*
6072	 * Before flushing the journal and switching inode's aops, we have
6073	 * to flush all dirty data the inode has. There can be outstanding
6074	 * delayed allocations, there can be unwritten extents created by
6075	 * fallocate or buffered writes in dioread_nolock mode covered by
6076	 * dirty data which can be converted only after flushing the dirty
6077	 * data (and journalled aops don't know how to handle these cases).
6078	 */
6079	if (val) {
6080		filemap_invalidate_lock(inode->i_mapping);
6081		err = filemap_write_and_wait(inode->i_mapping);
6082		if (err < 0) {
6083			filemap_invalidate_unlock(inode->i_mapping);
6084			return err;
6085		}
6086	}
6087
6088	alloc_ctx = ext4_writepages_down_write(inode->i_sb);
6089	jbd2_journal_lock_updates(journal);
6090
6091	/*
6092	 * OK, there are no updates running now, and all cached data is
6093	 * synced to disk.  We are now in a completely consistent state
6094	 * which doesn't have anything in the journal, and we know that
6095	 * no filesystem updates are running, so it is safe to modify
6096	 * the inode's in-core data-journaling state flag now.
6097	 */
6098
6099	if (val)
6100		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6101	else {
6102		err = jbd2_journal_flush(journal, 0);
6103		if (err < 0) {
6104			jbd2_journal_unlock_updates(journal);
6105			ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6106			return err;
6107		}
6108		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6109	}
6110	ext4_set_aops(inode);
6111
6112	jbd2_journal_unlock_updates(journal);
6113	ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6114
6115	if (val)
6116		filemap_invalidate_unlock(inode->i_mapping);
6117
6118	/* Finally we can mark the inode as dirty. */
6119
6120	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6121	if (IS_ERR(handle))
6122		return PTR_ERR(handle);
6123
6124	ext4_fc_mark_ineligible(inode->i_sb,
6125		EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6126	err = ext4_mark_inode_dirty(handle, inode);
6127	ext4_handle_sync(handle);
6128	ext4_journal_stop(handle);
6129	ext4_std_error(inode->i_sb, err);
6130
6131	return err;
6132}
6133
6134static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6135			    struct buffer_head *bh)
6136{
6137	return !buffer_mapped(bh);
6138}
6139
6140vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6141{
6142	struct vm_area_struct *vma = vmf->vma;
6143	struct folio *folio = page_folio(vmf->page);
6144	loff_t size;
6145	unsigned long len;
6146	int err;
6147	vm_fault_t ret;
6148	struct file *file = vma->vm_file;
6149	struct inode *inode = file_inode(file);
6150	struct address_space *mapping = inode->i_mapping;
6151	handle_t *handle;
6152	get_block_t *get_block;
6153	int retries = 0;
6154
6155	if (unlikely(IS_IMMUTABLE(inode)))
6156		return VM_FAULT_SIGBUS;
6157
6158	sb_start_pagefault(inode->i_sb);
6159	file_update_time(vma->vm_file);
6160
6161	filemap_invalidate_lock_shared(mapping);
6162
6163	err = ext4_convert_inline_data(inode);
6164	if (err)
6165		goto out_ret;
6166
6167	/*
6168	 * On data journalling we skip straight to the transaction handle:
6169	 * there's no delalloc; page truncated will be checked later; the
6170	 * early return w/ all buffers mapped (calculates size/len) can't
6171	 * be used; and there's no dioread_nolock, so only ext4_get_block.
6172	 */
6173	if (ext4_should_journal_data(inode))
6174		goto retry_alloc;
6175
6176	/* Delalloc case is easy... */
6177	if (test_opt(inode->i_sb, DELALLOC) &&
6178	    !ext4_nonda_switch(inode->i_sb)) {
6179		do {
6180			err = block_page_mkwrite(vma, vmf,
6181						   ext4_da_get_block_prep);
6182		} while (err == -ENOSPC &&
6183		       ext4_should_retry_alloc(inode->i_sb, &retries));
6184		goto out_ret;
6185	}
6186
6187	folio_lock(folio);
6188	size = i_size_read(inode);
6189	/* Page got truncated from under us? */
6190	if (folio->mapping != mapping || folio_pos(folio) > size) {
6191		folio_unlock(folio);
6192		ret = VM_FAULT_NOPAGE;
6193		goto out;
6194	}
6195
6196	len = folio_size(folio);
6197	if (folio_pos(folio) + len > size)
6198		len = size - folio_pos(folio);
 
6199	/*
6200	 * Return if we have all the buffers mapped. This avoids the need to do
6201	 * journal_start/journal_stop which can block and take a long time
6202	 *
6203	 * This cannot be done for data journalling, as we have to add the
6204	 * inode to the transaction's list to writeprotect pages on commit.
6205	 */
6206	if (folio_buffers(folio)) {
6207		if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6208					    0, len, NULL,
6209					    ext4_bh_unmapped)) {
6210			/* Wait so that we don't change page under IO */
6211			folio_wait_stable(folio);
6212			ret = VM_FAULT_LOCKED;
6213			goto out;
6214		}
6215	}
6216	folio_unlock(folio);
6217	/* OK, we need to fill the hole... */
6218	if (ext4_should_dioread_nolock(inode))
6219		get_block = ext4_get_block_unwritten;
6220	else
6221		get_block = ext4_get_block;
6222retry_alloc:
6223	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6224				    ext4_writepage_trans_blocks(inode));
6225	if (IS_ERR(handle)) {
6226		ret = VM_FAULT_SIGBUS;
6227		goto out;
6228	}
6229	/*
6230	 * Data journalling can't use block_page_mkwrite() because it
6231	 * will set_buffer_dirty() before do_journal_get_write_access()
6232	 * thus might hit warning messages for dirty metadata buffers.
6233	 */
6234	if (!ext4_should_journal_data(inode)) {
6235		err = block_page_mkwrite(vma, vmf, get_block);
6236	} else {
6237		folio_lock(folio);
6238		size = i_size_read(inode);
6239		/* Page got truncated from under us? */
6240		if (folio->mapping != mapping || folio_pos(folio) > size) {
6241			ret = VM_FAULT_NOPAGE;
6242			goto out_error;
6243		}
6244
6245		len = folio_size(folio);
6246		if (folio_pos(folio) + len > size)
6247			len = size - folio_pos(folio);
 
6248
6249		err = ext4_block_write_begin(handle, folio, 0, len,
6250					     ext4_get_block);
6251		if (!err) {
6252			ret = VM_FAULT_SIGBUS;
6253			if (ext4_journal_folio_buffers(handle, folio, len))
 
 
6254				goto out_error;
 
 
 
 
 
 
 
 
6255		} else {
6256			folio_unlock(folio);
6257		}
6258	}
6259	ext4_journal_stop(handle);
6260	if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6261		goto retry_alloc;
6262out_ret:
6263	ret = vmf_fs_error(err);
6264out:
6265	filemap_invalidate_unlock_shared(mapping);
6266	sb_end_pagefault(inode->i_sb);
6267	return ret;
6268out_error:
6269	folio_unlock(folio);
6270	ext4_journal_stop(handle);
6271	goto out;
6272}
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/ext4/inode.c
   4 *
   5 * Copyright (C) 1992, 1993, 1994, 1995
   6 * Remy Card (card@masi.ibp.fr)
   7 * Laboratoire MASI - Institut Blaise Pascal
   8 * Universite Pierre et Marie Curie (Paris VI)
   9 *
  10 *  from
  11 *
  12 *  linux/fs/minix/inode.c
  13 *
  14 *  Copyright (C) 1991, 1992  Linus Torvalds
  15 *
  16 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  17 *	(jj@sunsite.ms.mff.cuni.cz)
  18 *
  19 *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  20 */
  21
  22#include <linux/fs.h>
  23#include <linux/mount.h>
  24#include <linux/time.h>
  25#include <linux/highuid.h>
  26#include <linux/pagemap.h>
  27#include <linux/dax.h>
  28#include <linux/quotaops.h>
  29#include <linux/string.h>
  30#include <linux/buffer_head.h>
  31#include <linux/writeback.h>
  32#include <linux/pagevec.h>
  33#include <linux/mpage.h>
  34#include <linux/namei.h>
  35#include <linux/uio.h>
  36#include <linux/bio.h>
  37#include <linux/workqueue.h>
  38#include <linux/kernel.h>
  39#include <linux/printk.h>
  40#include <linux/slab.h>
  41#include <linux/bitops.h>
  42#include <linux/iomap.h>
  43#include <linux/iversion.h>
  44
  45#include "ext4_jbd2.h"
  46#include "xattr.h"
  47#include "acl.h"
  48#include "truncate.h"
  49
  50#include <trace/events/ext4.h>
  51
 
 
 
 
 
  52static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
  53			      struct ext4_inode_info *ei)
  54{
  55	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  56	__u32 csum;
  57	__u16 dummy_csum = 0;
  58	int offset = offsetof(struct ext4_inode, i_checksum_lo);
  59	unsigned int csum_size = sizeof(dummy_csum);
  60
  61	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
  62	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
  63	offset += csum_size;
  64	csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
  65			   EXT4_GOOD_OLD_INODE_SIZE - offset);
  66
  67	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
  68		offset = offsetof(struct ext4_inode, i_checksum_hi);
  69		csum = ext4_chksum(sbi, csum, (__u8 *)raw +
  70				   EXT4_GOOD_OLD_INODE_SIZE,
  71				   offset - EXT4_GOOD_OLD_INODE_SIZE);
  72		if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
  73			csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
  74					   csum_size);
  75			offset += csum_size;
  76		}
  77		csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
  78				   EXT4_INODE_SIZE(inode->i_sb) - offset);
  79	}
  80
  81	return csum;
  82}
  83
  84static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
  85				  struct ext4_inode_info *ei)
  86{
  87	__u32 provided, calculated;
  88
  89	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  90	    cpu_to_le32(EXT4_OS_LINUX) ||
  91	    !ext4_has_metadata_csum(inode->i_sb))
  92		return 1;
  93
  94	provided = le16_to_cpu(raw->i_checksum_lo);
  95	calculated = ext4_inode_csum(inode, raw, ei);
  96	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  97	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
  98		provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
  99	else
 100		calculated &= 0xFFFF;
 101
 102	return provided == calculated;
 103}
 104
 105void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
 106			 struct ext4_inode_info *ei)
 107{
 108	__u32 csum;
 109
 110	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
 111	    cpu_to_le32(EXT4_OS_LINUX) ||
 112	    !ext4_has_metadata_csum(inode->i_sb))
 113		return;
 114
 115	csum = ext4_inode_csum(inode, raw, ei);
 116	raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
 117	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
 118	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
 119		raw->i_checksum_hi = cpu_to_le16(csum >> 16);
 120}
 121
 122static inline int ext4_begin_ordered_truncate(struct inode *inode,
 123					      loff_t new_size)
 124{
 125	trace_ext4_begin_ordered_truncate(inode, new_size);
 126	/*
 127	 * If jinode is zero, then we never opened the file for
 128	 * writing, so there's no need to call
 129	 * jbd2_journal_begin_ordered_truncate() since there's no
 130	 * outstanding writes we need to flush.
 131	 */
 132	if (!EXT4_I(inode)->jinode)
 133		return 0;
 134	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
 135						   EXT4_I(inode)->jinode,
 136						   new_size);
 137}
 138
 139static int __ext4_journalled_writepage(struct page *page, unsigned int len);
 140static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
 141				  int pextents);
 142
 143/*
 144 * Test whether an inode is a fast symlink.
 145 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
 146 */
 147int ext4_inode_is_fast_symlink(struct inode *inode)
 148{
 149	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 150		int ea_blocks = EXT4_I(inode)->i_file_acl ?
 151				EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
 152
 153		if (ext4_has_inline_data(inode))
 154			return 0;
 155
 156		return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
 157	}
 158	return S_ISLNK(inode->i_mode) && inode->i_size &&
 159	       (inode->i_size < EXT4_N_BLOCKS * 4);
 160}
 161
 162/*
 163 * Called at the last iput() if i_nlink is zero.
 164 */
 165void ext4_evict_inode(struct inode *inode)
 166{
 167	handle_t *handle;
 168	int err;
 169	/*
 170	 * Credits for final inode cleanup and freeing:
 171	 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
 172	 * (xattr block freeing), bitmap, group descriptor (inode freeing)
 173	 */
 174	int extra_credits = 6;
 175	struct ext4_xattr_inode_array *ea_inode_array = NULL;
 176	bool freeze_protected = false;
 177
 178	trace_ext4_evict_inode(inode);
 179
 180	if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
 181		ext4_evict_ea_inode(inode);
 182	if (inode->i_nlink) {
 183		/*
 184		 * When journalling data dirty buffers are tracked only in the
 185		 * journal. So although mm thinks everything is clean and
 186		 * ready for reaping the inode might still have some pages to
 187		 * write in the running transaction or waiting to be
 188		 * checkpointed. Thus calling jbd2_journal_invalidate_folio()
 189		 * (via truncate_inode_pages()) to discard these buffers can
 190		 * cause data loss. Also even if we did not discard these
 191		 * buffers, we would have no way to find them after the inode
 192		 * is reaped and thus user could see stale data if he tries to
 193		 * read them before the transaction is checkpointed. So be
 194		 * careful and force everything to disk here... We use
 195		 * ei->i_datasync_tid to store the newest transaction
 196		 * containing inode's data.
 197		 *
 198		 * Note that directories do not have this problem because they
 199		 * don't use page cache.
 200		 */
 201		if (inode->i_ino != EXT4_JOURNAL_INO &&
 202		    ext4_should_journal_data(inode) &&
 203		    S_ISREG(inode->i_mode) && inode->i_data.nrpages) {
 204			journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 205			tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
 206
 207			jbd2_complete_transaction(journal, commit_tid);
 208			filemap_write_and_wait(&inode->i_data);
 209		}
 210		truncate_inode_pages_final(&inode->i_data);
 211
 212		goto no_delete;
 213	}
 214
 215	if (is_bad_inode(inode))
 216		goto no_delete;
 217	dquot_initialize(inode);
 218
 219	if (ext4_should_order_data(inode))
 220		ext4_begin_ordered_truncate(inode, 0);
 221	truncate_inode_pages_final(&inode->i_data);
 222
 223	/*
 224	 * For inodes with journalled data, transaction commit could have
 225	 * dirtied the inode. And for inodes with dioread_nolock, unwritten
 226	 * extents converting worker could merge extents and also have dirtied
 227	 * the inode. Flush worker is ignoring it because of I_FREEING flag but
 228	 * we still need to remove the inode from the writeback lists.
 229	 */
 230	if (!list_empty_careful(&inode->i_io_list))
 231		inode_io_list_del(inode);
 232
 233	/*
 234	 * Protect us against freezing - iput() caller didn't have to have any
 235	 * protection against it. When we are in a running transaction though,
 236	 * we are already protected against freezing and we cannot grab further
 237	 * protection due to lock ordering constraints.
 238	 */
 239	if (!ext4_journal_current_handle()) {
 240		sb_start_intwrite(inode->i_sb);
 241		freeze_protected = true;
 242	}
 243
 244	if (!IS_NOQUOTA(inode))
 245		extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
 246
 247	/*
 248	 * Block bitmap, group descriptor, and inode are accounted in both
 249	 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
 250	 */
 251	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
 252			 ext4_blocks_for_truncate(inode) + extra_credits - 3);
 253	if (IS_ERR(handle)) {
 254		ext4_std_error(inode->i_sb, PTR_ERR(handle));
 255		/*
 256		 * If we're going to skip the normal cleanup, we still need to
 257		 * make sure that the in-core orphan linked list is properly
 258		 * cleaned up.
 259		 */
 260		ext4_orphan_del(NULL, inode);
 261		if (freeze_protected)
 262			sb_end_intwrite(inode->i_sb);
 263		goto no_delete;
 264	}
 265
 266	if (IS_SYNC(inode))
 267		ext4_handle_sync(handle);
 268
 269	/*
 270	 * Set inode->i_size to 0 before calling ext4_truncate(). We need
 271	 * special handling of symlinks here because i_size is used to
 272	 * determine whether ext4_inode_info->i_data contains symlink data or
 273	 * block mappings. Setting i_size to 0 will remove its fast symlink
 274	 * status. Erase i_data so that it becomes a valid empty block map.
 275	 */
 276	if (ext4_inode_is_fast_symlink(inode))
 277		memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
 278	inode->i_size = 0;
 279	err = ext4_mark_inode_dirty(handle, inode);
 280	if (err) {
 281		ext4_warning(inode->i_sb,
 282			     "couldn't mark inode dirty (err %d)", err);
 283		goto stop_handle;
 284	}
 285	if (inode->i_blocks) {
 286		err = ext4_truncate(inode);
 287		if (err) {
 288			ext4_error_err(inode->i_sb, -err,
 289				       "couldn't truncate inode %lu (err %d)",
 290				       inode->i_ino, err);
 291			goto stop_handle;
 292		}
 293	}
 294
 295	/* Remove xattr references. */
 296	err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
 297				      extra_credits);
 298	if (err) {
 299		ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
 300stop_handle:
 301		ext4_journal_stop(handle);
 302		ext4_orphan_del(NULL, inode);
 303		if (freeze_protected)
 304			sb_end_intwrite(inode->i_sb);
 305		ext4_xattr_inode_array_free(ea_inode_array);
 306		goto no_delete;
 307	}
 308
 309	/*
 310	 * Kill off the orphan record which ext4_truncate created.
 311	 * AKPM: I think this can be inside the above `if'.
 312	 * Note that ext4_orphan_del() has to be able to cope with the
 313	 * deletion of a non-existent orphan - this is because we don't
 314	 * know if ext4_truncate() actually created an orphan record.
 315	 * (Well, we could do this if we need to, but heck - it works)
 316	 */
 317	ext4_orphan_del(handle, inode);
 318	EXT4_I(inode)->i_dtime	= (__u32)ktime_get_real_seconds();
 319
 320	/*
 321	 * One subtle ordering requirement: if anything has gone wrong
 322	 * (transaction abort, IO errors, whatever), then we can still
 323	 * do these next steps (the fs will already have been marked as
 324	 * having errors), but we can't free the inode if the mark_dirty
 325	 * fails.
 326	 */
 327	if (ext4_mark_inode_dirty(handle, inode))
 328		/* If that failed, just do the required in-core inode clear. */
 329		ext4_clear_inode(inode);
 330	else
 331		ext4_free_inode(handle, inode);
 332	ext4_journal_stop(handle);
 333	if (freeze_protected)
 334		sb_end_intwrite(inode->i_sb);
 335	ext4_xattr_inode_array_free(ea_inode_array);
 336	return;
 337no_delete:
 338	/*
 339	 * Check out some where else accidentally dirty the evicting inode,
 340	 * which may probably cause inode use-after-free issues later.
 341	 */
 342	WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
 343
 344	if (!list_empty(&EXT4_I(inode)->i_fc_list))
 345		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
 346	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
 347}
 348
 349#ifdef CONFIG_QUOTA
 350qsize_t *ext4_get_reserved_space(struct inode *inode)
 351{
 352	return &EXT4_I(inode)->i_reserved_quota;
 353}
 354#endif
 355
 356/*
 357 * Called with i_data_sem down, which is important since we can call
 358 * ext4_discard_preallocations() from here.
 359 */
 360void ext4_da_update_reserve_space(struct inode *inode,
 361					int used, int quota_claim)
 362{
 363	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 364	struct ext4_inode_info *ei = EXT4_I(inode);
 365
 366	spin_lock(&ei->i_block_reservation_lock);
 367	trace_ext4_da_update_reserve_space(inode, used, quota_claim);
 368	if (unlikely(used > ei->i_reserved_data_blocks)) {
 369		ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
 370			 "with only %d reserved data blocks",
 371			 __func__, inode->i_ino, used,
 372			 ei->i_reserved_data_blocks);
 373		WARN_ON(1);
 374		used = ei->i_reserved_data_blocks;
 375	}
 376
 377	/* Update per-inode reservations */
 378	ei->i_reserved_data_blocks -= used;
 379	percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
 380
 381	spin_unlock(&ei->i_block_reservation_lock);
 382
 383	/* Update quota subsystem for data blocks */
 384	if (quota_claim)
 385		dquot_claim_block(inode, EXT4_C2B(sbi, used));
 386	else {
 387		/*
 388		 * We did fallocate with an offset that is already delayed
 389		 * allocated. So on delayed allocated writeback we should
 390		 * not re-claim the quota for fallocated blocks.
 391		 */
 392		dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
 393	}
 394
 395	/*
 396	 * If we have done all the pending block allocations and if
 397	 * there aren't any writers on the inode, we can discard the
 398	 * inode's preallocations.
 399	 */
 400	if ((ei->i_reserved_data_blocks == 0) &&
 401	    !inode_is_open_for_write(inode))
 402		ext4_discard_preallocations(inode, 0);
 403}
 404
 405static int __check_block_validity(struct inode *inode, const char *func,
 406				unsigned int line,
 407				struct ext4_map_blocks *map)
 408{
 409	if (ext4_has_feature_journal(inode->i_sb) &&
 410	    (inode->i_ino ==
 411	     le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
 412		return 0;
 413	if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
 414		ext4_error_inode(inode, func, line, map->m_pblk,
 415				 "lblock %lu mapped to illegal pblock %llu "
 416				 "(length %d)", (unsigned long) map->m_lblk,
 417				 map->m_pblk, map->m_len);
 418		return -EFSCORRUPTED;
 419	}
 420	return 0;
 421}
 422
 423int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
 424		       ext4_lblk_t len)
 425{
 426	int ret;
 427
 428	if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
 429		return fscrypt_zeroout_range(inode, lblk, pblk, len);
 430
 431	ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
 432	if (ret > 0)
 433		ret = 0;
 434
 435	return ret;
 436}
 437
 438#define check_block_validity(inode, map)	\
 439	__check_block_validity((inode), __func__, __LINE__, (map))
 440
 441#ifdef ES_AGGRESSIVE_TEST
 442static void ext4_map_blocks_es_recheck(handle_t *handle,
 443				       struct inode *inode,
 444				       struct ext4_map_blocks *es_map,
 445				       struct ext4_map_blocks *map,
 446				       int flags)
 447{
 448	int retval;
 449
 450	map->m_flags = 0;
 451	/*
 452	 * There is a race window that the result is not the same.
 453	 * e.g. xfstests #223 when dioread_nolock enables.  The reason
 454	 * is that we lookup a block mapping in extent status tree with
 455	 * out taking i_data_sem.  So at the time the unwritten extent
 456	 * could be converted.
 457	 */
 458	down_read(&EXT4_I(inode)->i_data_sem);
 459	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 460		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 461	} else {
 462		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 463	}
 464	up_read((&EXT4_I(inode)->i_data_sem));
 465
 466	/*
 467	 * We don't check m_len because extent will be collpased in status
 468	 * tree.  So the m_len might not equal.
 469	 */
 470	if (es_map->m_lblk != map->m_lblk ||
 471	    es_map->m_flags != map->m_flags ||
 472	    es_map->m_pblk != map->m_pblk) {
 473		printk("ES cache assertion failed for inode: %lu "
 474		       "es_cached ex [%d/%d/%llu/%x] != "
 475		       "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
 476		       inode->i_ino, es_map->m_lblk, es_map->m_len,
 477		       es_map->m_pblk, es_map->m_flags, map->m_lblk,
 478		       map->m_len, map->m_pblk, map->m_flags,
 479		       retval, flags);
 480	}
 481}
 482#endif /* ES_AGGRESSIVE_TEST */
 483
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 484/*
 485 * The ext4_map_blocks() function tries to look up the requested blocks,
 486 * and returns if the blocks are already mapped.
 487 *
 488 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 489 * and store the allocated blocks in the result buffer head and mark it
 490 * mapped.
 491 *
 492 * If file type is extents based, it will call ext4_ext_map_blocks(),
 493 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 494 * based files
 495 *
 496 * On success, it returns the number of blocks being mapped or allocated.  if
 497 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
 498 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
 
 499 *
 500 * It returns 0 if plain look up failed (blocks have not been allocated), in
 501 * that case, @map is returned as unmapped but we still do fill map->m_len to
 502 * indicate the length of a hole starting at map->m_lblk.
 503 *
 504 * It returns the error in case of allocation failure.
 505 */
 506int ext4_map_blocks(handle_t *handle, struct inode *inode,
 507		    struct ext4_map_blocks *map, int flags)
 508{
 509	struct extent_status es;
 510	int retval;
 511	int ret = 0;
 512#ifdef ES_AGGRESSIVE_TEST
 513	struct ext4_map_blocks orig_map;
 514
 515	memcpy(&orig_map, map, sizeof(*map));
 516#endif
 517
 518	map->m_flags = 0;
 519	ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
 520		  flags, map->m_len, (unsigned long) map->m_lblk);
 521
 522	/*
 523	 * ext4_map_blocks returns an int, and m_len is an unsigned int
 524	 */
 525	if (unlikely(map->m_len > INT_MAX))
 526		map->m_len = INT_MAX;
 527
 528	/* We can handle the block number less than EXT_MAX_BLOCKS */
 529	if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
 530		return -EFSCORRUPTED;
 531
 532	/* Lookup extent status tree firstly */
 533	if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
 534	    ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
 535		if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
 536			map->m_pblk = ext4_es_pblock(&es) +
 537					map->m_lblk - es.es_lblk;
 538			map->m_flags |= ext4_es_is_written(&es) ?
 539					EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
 540			retval = es.es_len - (map->m_lblk - es.es_lblk);
 541			if (retval > map->m_len)
 542				retval = map->m_len;
 543			map->m_len = retval;
 544		} else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
 545			map->m_pblk = 0;
 
 
 546			retval = es.es_len - (map->m_lblk - es.es_lblk);
 547			if (retval > map->m_len)
 548				retval = map->m_len;
 549			map->m_len = retval;
 550			retval = 0;
 551		} else {
 552			BUG();
 553		}
 554
 555		if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
 556			return retval;
 557#ifdef ES_AGGRESSIVE_TEST
 558		ext4_map_blocks_es_recheck(handle, inode, map,
 559					   &orig_map, flags);
 560#endif
 561		goto found;
 562	}
 563	/*
 564	 * In the query cache no-wait mode, nothing we can do more if we
 565	 * cannot find extent in the cache.
 566	 */
 567	if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
 568		return 0;
 569
 570	/*
 571	 * Try to see if we can get the block without requesting a new
 572	 * file system block.
 573	 */
 574	down_read(&EXT4_I(inode)->i_data_sem);
 575	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 576		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 577	} else {
 578		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 579	}
 580	if (retval > 0) {
 581		unsigned int status;
 582
 583		if (unlikely(retval != map->m_len)) {
 584			ext4_warning(inode->i_sb,
 585				     "ES len assertion failed for inode "
 586				     "%lu: retval %d != map->m_len %d",
 587				     inode->i_ino, retval, map->m_len);
 588			WARN_ON(1);
 589		}
 590
 591		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 592				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 593		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
 594		    !(status & EXTENT_STATUS_WRITTEN) &&
 595		    ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
 596				       map->m_lblk + map->m_len - 1))
 597			status |= EXTENT_STATUS_DELAYED;
 598		ret = ext4_es_insert_extent(inode, map->m_lblk,
 599					    map->m_len, map->m_pblk, status);
 600		if (ret < 0)
 601			retval = ret;
 602	}
 603	up_read((&EXT4_I(inode)->i_data_sem));
 604
 605found:
 606	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 607		ret = check_block_validity(inode, map);
 608		if (ret != 0)
 609			return ret;
 610	}
 611
 612	/* If it is only a block(s) look up */
 613	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 614		return retval;
 615
 616	/*
 617	 * Returns if the blocks have already allocated
 618	 *
 619	 * Note that if blocks have been preallocated
 620	 * ext4_ext_get_block() returns the create = 0
 621	 * with buffer head unmapped.
 622	 */
 623	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 624		/*
 625		 * If we need to convert extent to unwritten
 626		 * we continue and do the actual work in
 627		 * ext4_ext_map_blocks()
 628		 */
 629		if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
 630			return retval;
 631
 632	/*
 633	 * Here we clear m_flags because after allocating an new extent,
 634	 * it will be set again.
 635	 */
 636	map->m_flags &= ~EXT4_MAP_FLAGS;
 637
 638	/*
 639	 * New blocks allocate and/or writing to unwritten extent
 640	 * will possibly result in updating i_data, so we take
 641	 * the write lock of i_data_sem, and call get_block()
 642	 * with create == 1 flag.
 643	 */
 644	down_write(&EXT4_I(inode)->i_data_sem);
 645
 646	/*
 647	 * We need to check for EXT4 here because migrate
 648	 * could have changed the inode type in between
 649	 */
 650	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 651		retval = ext4_ext_map_blocks(handle, inode, map, flags);
 652	} else {
 653		retval = ext4_ind_map_blocks(handle, inode, map, flags);
 654
 655		if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
 656			/*
 657			 * We allocated new blocks which will result in
 658			 * i_data's format changing.  Force the migrate
 659			 * to fail by clearing migrate flags
 660			 */
 661			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 662		}
 663
 664		/*
 665		 * Update reserved blocks/metadata blocks after successful
 666		 * block allocation which had been deferred till now. We don't
 667		 * support fallocate for non extent files. So we can update
 668		 * reserve space here.
 669		 */
 670		if ((retval > 0) &&
 671			(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
 672			ext4_da_update_reserve_space(inode, retval, 1);
 673	}
 674
 675	if (retval > 0) {
 676		unsigned int status;
 677
 678		if (unlikely(retval != map->m_len)) {
 679			ext4_warning(inode->i_sb,
 680				     "ES len assertion failed for inode "
 681				     "%lu: retval %d != map->m_len %d",
 682				     inode->i_ino, retval, map->m_len);
 683			WARN_ON(1);
 684		}
 685
 686		/*
 687		 * We have to zeroout blocks before inserting them into extent
 688		 * status tree. Otherwise someone could look them up there and
 689		 * use them before they are really zeroed. We also have to
 690		 * unmap metadata before zeroing as otherwise writeback can
 691		 * overwrite zeros with stale data from block device.
 692		 */
 693		if (flags & EXT4_GET_BLOCKS_ZERO &&
 694		    map->m_flags & EXT4_MAP_MAPPED &&
 695		    map->m_flags & EXT4_MAP_NEW) {
 696			ret = ext4_issue_zeroout(inode, map->m_lblk,
 697						 map->m_pblk, map->m_len);
 698			if (ret) {
 699				retval = ret;
 700				goto out_sem;
 701			}
 702		}
 703
 704		/*
 705		 * If the extent has been zeroed out, we don't need to update
 706		 * extent status tree.
 707		 */
 708		if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
 709		    ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
 710			if (ext4_es_is_written(&es))
 711				goto out_sem;
 712		}
 713		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 714				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 715		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
 716		    !(status & EXTENT_STATUS_WRITTEN) &&
 717		    ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
 718				       map->m_lblk + map->m_len - 1))
 719			status |= EXTENT_STATUS_DELAYED;
 720		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
 721					    map->m_pblk, status);
 722		if (ret < 0) {
 723			retval = ret;
 724			goto out_sem;
 725		}
 726	}
 727
 728out_sem:
 729	up_write((&EXT4_I(inode)->i_data_sem));
 730	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 731		ret = check_block_validity(inode, map);
 732		if (ret != 0)
 733			return ret;
 734
 735		/*
 736		 * Inodes with freshly allocated blocks where contents will be
 737		 * visible after transaction commit must be on transaction's
 738		 * ordered data list.
 739		 */
 740		if (map->m_flags & EXT4_MAP_NEW &&
 741		    !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
 742		    !(flags & EXT4_GET_BLOCKS_ZERO) &&
 743		    !ext4_is_quota_file(inode) &&
 744		    ext4_should_order_data(inode)) {
 745			loff_t start_byte =
 746				(loff_t)map->m_lblk << inode->i_blkbits;
 747			loff_t length = (loff_t)map->m_len << inode->i_blkbits;
 748
 749			if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
 750				ret = ext4_jbd2_inode_add_wait(handle, inode,
 751						start_byte, length);
 752			else
 753				ret = ext4_jbd2_inode_add_write(handle, inode,
 754						start_byte, length);
 755			if (ret)
 756				return ret;
 757		}
 758	}
 759	if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
 760				map->m_flags & EXT4_MAP_MAPPED))
 761		ext4_fc_track_range(handle, inode, map->m_lblk,
 762					map->m_lblk + map->m_len - 1);
 763	if (retval < 0)
 764		ext_debug(inode, "failed with err %d\n", retval);
 765	return retval;
 766}
 767
 768/*
 769 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
 770 * we have to be careful as someone else may be manipulating b_state as well.
 771 */
 772static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
 773{
 774	unsigned long old_state;
 775	unsigned long new_state;
 776
 777	flags &= EXT4_MAP_FLAGS;
 778
 779	/* Dummy buffer_head? Set non-atomically. */
 780	if (!bh->b_page) {
 781		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
 782		return;
 783	}
 784	/*
 785	 * Someone else may be modifying b_state. Be careful! This is ugly but
 786	 * once we get rid of using bh as a container for mapping information
 787	 * to pass to / from get_block functions, this can go away.
 788	 */
 
 789	do {
 790		old_state = READ_ONCE(bh->b_state);
 791		new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
 792	} while (unlikely(
 793		 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
 794}
 795
 796static int _ext4_get_block(struct inode *inode, sector_t iblock,
 797			   struct buffer_head *bh, int flags)
 798{
 799	struct ext4_map_blocks map;
 800	int ret = 0;
 801
 802	if (ext4_has_inline_data(inode))
 803		return -ERANGE;
 804
 805	map.m_lblk = iblock;
 806	map.m_len = bh->b_size >> inode->i_blkbits;
 807
 808	ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
 809			      flags);
 810	if (ret > 0) {
 811		map_bh(bh, inode->i_sb, map.m_pblk);
 812		ext4_update_bh_state(bh, map.m_flags);
 813		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 814		ret = 0;
 815	} else if (ret == 0) {
 816		/* hole case, need to fill in bh->b_size */
 817		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 818	}
 819	return ret;
 820}
 821
 822int ext4_get_block(struct inode *inode, sector_t iblock,
 823		   struct buffer_head *bh, int create)
 824{
 825	return _ext4_get_block(inode, iblock, bh,
 826			       create ? EXT4_GET_BLOCKS_CREATE : 0);
 827}
 828
 829/*
 830 * Get block function used when preparing for buffered write if we require
 831 * creating an unwritten extent if blocks haven't been allocated.  The extent
 832 * will be converted to written after the IO is complete.
 833 */
 834int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
 835			     struct buffer_head *bh_result, int create)
 836{
 
 
 837	ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
 838		   inode->i_ino, create);
 839	return _ext4_get_block(inode, iblock, bh_result,
 840			       EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
 
 
 
 
 
 
 
 
 
 
 841}
 842
 843/* Maximum number of blocks we map for direct IO at once. */
 844#define DIO_MAX_BLOCKS 4096
 845
 846/*
 847 * `handle' can be NULL if create is zero
 848 */
 849struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 850				ext4_lblk_t block, int map_flags)
 851{
 852	struct ext4_map_blocks map;
 853	struct buffer_head *bh;
 854	int create = map_flags & EXT4_GET_BLOCKS_CREATE;
 855	bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
 856	int err;
 857
 858	ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 859		    || handle != NULL || create == 0);
 860	ASSERT(create == 0 || !nowait);
 861
 862	map.m_lblk = block;
 863	map.m_len = 1;
 864	err = ext4_map_blocks(handle, inode, &map, map_flags);
 865
 866	if (err == 0)
 867		return create ? ERR_PTR(-ENOSPC) : NULL;
 868	if (err < 0)
 869		return ERR_PTR(err);
 870
 871	if (nowait)
 872		return sb_find_get_block(inode->i_sb, map.m_pblk);
 873
 874	bh = sb_getblk(inode->i_sb, map.m_pblk);
 
 
 
 
 
 
 
 875	if (unlikely(!bh))
 876		return ERR_PTR(-ENOMEM);
 877	if (map.m_flags & EXT4_MAP_NEW) {
 878		ASSERT(create != 0);
 879		ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 880			    || (handle != NULL));
 881
 882		/*
 883		 * Now that we do not always journal data, we should
 884		 * keep in mind whether this should always journal the
 885		 * new buffer as metadata.  For now, regular file
 886		 * writes use ext4_get_block instead, so it's not a
 887		 * problem.
 888		 */
 889		lock_buffer(bh);
 890		BUFFER_TRACE(bh, "call get_create_access");
 891		err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
 892						     EXT4_JTR_NONE);
 893		if (unlikely(err)) {
 894			unlock_buffer(bh);
 895			goto errout;
 896		}
 897		if (!buffer_uptodate(bh)) {
 898			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 899			set_buffer_uptodate(bh);
 900		}
 901		unlock_buffer(bh);
 902		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 903		err = ext4_handle_dirty_metadata(handle, inode, bh);
 904		if (unlikely(err))
 905			goto errout;
 906	} else
 907		BUFFER_TRACE(bh, "not a new buffer");
 908	return bh;
 909errout:
 910	brelse(bh);
 911	return ERR_PTR(err);
 912}
 913
 914struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 915			       ext4_lblk_t block, int map_flags)
 916{
 917	struct buffer_head *bh;
 918	int ret;
 919
 920	bh = ext4_getblk(handle, inode, block, map_flags);
 921	if (IS_ERR(bh))
 922		return bh;
 923	if (!bh || ext4_buffer_uptodate(bh))
 924		return bh;
 925
 926	ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
 927	if (ret) {
 928		put_bh(bh);
 929		return ERR_PTR(ret);
 930	}
 931	return bh;
 932}
 933
 934/* Read a contiguous batch of blocks. */
 935int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
 936		     bool wait, struct buffer_head **bhs)
 937{
 938	int i, err;
 939
 940	for (i = 0; i < bh_count; i++) {
 941		bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
 942		if (IS_ERR(bhs[i])) {
 943			err = PTR_ERR(bhs[i]);
 944			bh_count = i;
 945			goto out_brelse;
 946		}
 947	}
 948
 949	for (i = 0; i < bh_count; i++)
 950		/* Note that NULL bhs[i] is valid because of holes. */
 951		if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
 952			ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
 953
 954	if (!wait)
 955		return 0;
 956
 957	for (i = 0; i < bh_count; i++)
 958		if (bhs[i])
 959			wait_on_buffer(bhs[i]);
 960
 961	for (i = 0; i < bh_count; i++) {
 962		if (bhs[i] && !buffer_uptodate(bhs[i])) {
 963			err = -EIO;
 964			goto out_brelse;
 965		}
 966	}
 967	return 0;
 968
 969out_brelse:
 970	for (i = 0; i < bh_count; i++) {
 971		brelse(bhs[i]);
 972		bhs[i] = NULL;
 973	}
 974	return err;
 975}
 976
 977int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
 978			   struct buffer_head *head,
 979			   unsigned from,
 980			   unsigned to,
 981			   int *partial,
 982			   int (*fn)(handle_t *handle, struct inode *inode,
 983				     struct buffer_head *bh))
 984{
 985	struct buffer_head *bh;
 986	unsigned block_start, block_end;
 987	unsigned blocksize = head->b_size;
 988	int err, ret = 0;
 989	struct buffer_head *next;
 990
 991	for (bh = head, block_start = 0;
 992	     ret == 0 && (bh != head || !block_start);
 993	     block_start = block_end, bh = next) {
 994		next = bh->b_this_page;
 995		block_end = block_start + blocksize;
 996		if (block_end <= from || block_start >= to) {
 997			if (partial && !buffer_uptodate(bh))
 998				*partial = 1;
 999			continue;
1000		}
1001		err = (*fn)(handle, inode, bh);
1002		if (!ret)
1003			ret = err;
1004	}
1005	return ret;
1006}
1007
1008/*
1009 * To preserve ordering, it is essential that the hole instantiation and
1010 * the data write be encapsulated in a single transaction.  We cannot
1011 * close off a transaction and start a new one between the ext4_get_block()
1012 * and the commit_write().  So doing the jbd2_journal_start at the start of
1013 * prepare_write() is the right place.
1014 *
1015 * Also, this function can nest inside ext4_writepage().  In that case, we
1016 * *know* that ext4_writepage() has generated enough buffer credits to do the
1017 * whole page.  So we won't block on the journal in that case, which is good,
1018 * because the caller may be PF_MEMALLOC.
1019 *
1020 * By accident, ext4 can be reentered when a transaction is open via
1021 * quota file writes.  If we were to commit the transaction while thus
1022 * reentered, there can be a deadlock - we would be holding a quota
1023 * lock, and the commit would never complete if another thread had a
1024 * transaction open and was blocking on the quota lock - a ranking
1025 * violation.
1026 *
1027 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1028 * will _not_ run commit under these circumstances because handle->h_ref
1029 * is elevated.  We'll still have enough credits for the tiny quotafile
1030 * write.
1031 */
 
 
 
 
 
 
1032int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1033				struct buffer_head *bh)
1034{
1035	int dirty = buffer_dirty(bh);
1036	int ret;
1037
1038	if (!buffer_mapped(bh) || buffer_freed(bh))
1039		return 0;
1040	/*
1041	 * __block_write_begin() could have dirtied some buffers. Clean
1042	 * the dirty bit as jbd2_journal_get_write_access() could complain
1043	 * otherwise about fs integrity issues. Setting of the dirty bit
1044	 * by __block_write_begin() isn't a real problem here as we clear
1045	 * the bit before releasing a page lock and thus writeback cannot
1046	 * ever write the buffer.
1047	 */
1048	if (dirty)
1049		clear_buffer_dirty(bh);
1050	BUFFER_TRACE(bh, "get write access");
1051	ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1052					    EXT4_JTR_NONE);
1053	if (!ret && dirty)
1054		ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1055	return ret;
1056}
1057
1058#ifdef CONFIG_FS_ENCRYPTION
1059static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1060				  get_block_t *get_block)
1061{
1062	unsigned from = pos & (PAGE_SIZE - 1);
1063	unsigned to = from + len;
1064	struct inode *inode = page->mapping->host;
1065	unsigned block_start, block_end;
1066	sector_t block;
1067	int err = 0;
1068	unsigned blocksize = inode->i_sb->s_blocksize;
1069	unsigned bbits;
1070	struct buffer_head *bh, *head, *wait[2];
1071	int nr_wait = 0;
1072	int i;
 
1073
1074	BUG_ON(!PageLocked(page));
1075	BUG_ON(from > PAGE_SIZE);
1076	BUG_ON(to > PAGE_SIZE);
1077	BUG_ON(from > to);
1078
1079	if (!page_has_buffers(page))
1080		create_empty_buffers(page, blocksize, 0);
1081	head = page_buffers(page);
1082	bbits = ilog2(blocksize);
1083	block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1084
1085	for (bh = head, block_start = 0; bh != head || !block_start;
1086	    block++, block_start = block_end, bh = bh->b_this_page) {
1087		block_end = block_start + blocksize;
1088		if (block_end <= from || block_start >= to) {
1089			if (PageUptodate(page)) {
1090				set_buffer_uptodate(bh);
1091			}
1092			continue;
1093		}
1094		if (buffer_new(bh))
1095			clear_buffer_new(bh);
1096		if (!buffer_mapped(bh)) {
1097			WARN_ON(bh->b_size != blocksize);
1098			err = get_block(inode, block, bh, 1);
1099			if (err)
1100				break;
1101			if (buffer_new(bh)) {
1102				if (PageUptodate(page)) {
1103					clear_buffer_new(bh);
 
 
 
 
 
 
 
 
 
 
 
 
 
1104					set_buffer_uptodate(bh);
1105					mark_buffer_dirty(bh);
1106					continue;
1107				}
1108				if (block_end > to || block_start < from)
1109					zero_user_segments(page, to, block_end,
1110							   block_start, from);
 
1111				continue;
1112			}
1113		}
1114		if (PageUptodate(page)) {
1115			set_buffer_uptodate(bh);
1116			continue;
1117		}
1118		if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1119		    !buffer_unwritten(bh) &&
1120		    (block_start < from || block_end > to)) {
1121			ext4_read_bh_lock(bh, 0, false);
1122			wait[nr_wait++] = bh;
1123		}
1124	}
1125	/*
1126	 * If we issued read requests, let them complete.
1127	 */
1128	for (i = 0; i < nr_wait; i++) {
1129		wait_on_buffer(wait[i]);
1130		if (!buffer_uptodate(wait[i]))
1131			err = -EIO;
1132	}
1133	if (unlikely(err)) {
1134		page_zero_new_buffers(page, from, to);
 
 
 
 
1135	} else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1136		for (i = 0; i < nr_wait; i++) {
1137			int err2;
1138
1139			err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1140								bh_offset(wait[i]));
1141			if (err2) {
1142				clear_buffer_uptodate(wait[i]);
1143				err = err2;
1144			}
1145		}
1146	}
1147
1148	return err;
1149}
1150#endif
1151
 
 
 
 
 
 
 
1152static int ext4_write_begin(struct file *file, struct address_space *mapping,
1153			    loff_t pos, unsigned len,
1154			    struct page **pagep, void **fsdata)
1155{
1156	struct inode *inode = mapping->host;
1157	int ret, needed_blocks;
1158	handle_t *handle;
1159	int retries = 0;
1160	struct page *page;
1161	pgoff_t index;
1162	unsigned from, to;
1163
1164	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1165		return -EIO;
1166
1167	trace_ext4_write_begin(inode, pos, len);
1168	/*
1169	 * Reserve one block more for addition to orphan list in case
1170	 * we allocate blocks but write fails for some reason
1171	 */
1172	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1173	index = pos >> PAGE_SHIFT;
1174	from = pos & (PAGE_SIZE - 1);
1175	to = from + len;
1176
1177	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1178		ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1179						    pagep);
1180		if (ret < 0)
1181			return ret;
1182		if (ret == 1)
1183			return 0;
1184	}
1185
1186	/*
1187	 * grab_cache_page_write_begin() can take a long time if the
1188	 * system is thrashing due to memory pressure, or if the page
1189	 * is being written back.  So grab it first before we start
1190	 * the transaction handle.  This also allows us to allocate
1191	 * the page (if needed) without using GFP_NOFS.
1192	 */
1193retry_grab:
1194	page = grab_cache_page_write_begin(mapping, index);
1195	if (!page)
1196		return -ENOMEM;
 
1197	/*
1198	 * The same as page allocation, we prealloc buffer heads before
1199	 * starting the handle.
1200	 */
1201	if (!page_has_buffers(page))
1202		create_empty_buffers(page, inode->i_sb->s_blocksize, 0);
1203
1204	unlock_page(page);
1205
1206retry_journal:
1207	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1208	if (IS_ERR(handle)) {
1209		put_page(page);
1210		return PTR_ERR(handle);
1211	}
1212
1213	lock_page(page);
1214	if (page->mapping != mapping) {
1215		/* The page got truncated from under us */
1216		unlock_page(page);
1217		put_page(page);
1218		ext4_journal_stop(handle);
1219		goto retry_grab;
1220	}
1221	/* In case writeback began while the page was unlocked */
1222	wait_for_stable_page(page);
1223
1224#ifdef CONFIG_FS_ENCRYPTION
1225	if (ext4_should_dioread_nolock(inode))
1226		ret = ext4_block_write_begin(page, pos, len,
1227					     ext4_get_block_unwritten);
1228	else
1229		ret = ext4_block_write_begin(page, pos, len,
1230					     ext4_get_block);
1231#else
1232	if (ext4_should_dioread_nolock(inode))
1233		ret = __block_write_begin(page, pos, len,
1234					  ext4_get_block_unwritten);
1235	else
1236		ret = __block_write_begin(page, pos, len, ext4_get_block);
1237#endif
1238	if (!ret && ext4_should_journal_data(inode)) {
1239		ret = ext4_walk_page_buffers(handle, inode,
1240					     page_buffers(page), from, to, NULL,
1241					     do_journal_get_write_access);
1242	}
1243
1244	if (ret) {
1245		bool extended = (pos + len > inode->i_size) &&
1246				!ext4_verity_in_progress(inode);
1247
1248		unlock_page(page);
1249		/*
1250		 * __block_write_begin may have instantiated a few blocks
1251		 * outside i_size.  Trim these off again. Don't need
1252		 * i_size_read because we hold i_rwsem.
1253		 *
1254		 * Add inode to orphan list in case we crash before
1255		 * truncate finishes
1256		 */
1257		if (extended && ext4_can_truncate(inode))
1258			ext4_orphan_add(handle, inode);
1259
1260		ext4_journal_stop(handle);
1261		if (extended) {
1262			ext4_truncate_failed_write(inode);
1263			/*
1264			 * If truncate failed early the inode might
1265			 * still be on the orphan list; we need to
1266			 * make sure the inode is removed from the
1267			 * orphan list in that case.
1268			 */
1269			if (inode->i_nlink)
1270				ext4_orphan_del(NULL, inode);
1271		}
1272
1273		if (ret == -ENOSPC &&
1274		    ext4_should_retry_alloc(inode->i_sb, &retries))
1275			goto retry_journal;
1276		put_page(page);
1277		return ret;
1278	}
1279	*pagep = page;
1280	return ret;
1281}
1282
1283/* For write_end() in data=journal mode */
1284static int write_end_fn(handle_t *handle, struct inode *inode,
1285			struct buffer_head *bh)
1286{
1287	int ret;
1288	if (!buffer_mapped(bh) || buffer_freed(bh))
1289		return 0;
1290	set_buffer_uptodate(bh);
1291	ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1292	clear_buffer_meta(bh);
1293	clear_buffer_prio(bh);
1294	return ret;
1295}
1296
1297/*
1298 * We need to pick up the new inode size which generic_commit_write gave us
1299 * `file' can be NULL - eg, when called from page_symlink().
1300 *
1301 * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1302 * buffers are managed internally.
1303 */
1304static int ext4_write_end(struct file *file,
1305			  struct address_space *mapping,
1306			  loff_t pos, unsigned len, unsigned copied,
1307			  struct page *page, void *fsdata)
1308{
1309	handle_t *handle = ext4_journal_current_handle();
1310	struct inode *inode = mapping->host;
1311	loff_t old_size = inode->i_size;
1312	int ret = 0, ret2;
1313	int i_size_changed = 0;
1314	bool verity = ext4_verity_in_progress(inode);
1315
1316	trace_ext4_write_end(inode, pos, len, copied);
1317
1318	if (ext4_has_inline_data(inode) &&
1319	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1320		return ext4_write_inline_data_end(inode, pos, len, copied, page);
 
1321
1322	copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1323	/*
1324	 * it's important to update i_size while still holding page lock:
1325	 * page writeout could otherwise come in and zero beyond i_size.
1326	 *
1327	 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1328	 * blocks are being written past EOF, so skip the i_size update.
1329	 */
1330	if (!verity)
1331		i_size_changed = ext4_update_inode_size(inode, pos + copied);
1332	unlock_page(page);
1333	put_page(page);
1334
1335	if (old_size < pos && !verity)
1336		pagecache_isize_extended(inode, old_size, pos);
 
 
1337	/*
1338	 * Don't mark the inode dirty under page lock. First, it unnecessarily
1339	 * makes the holding time of page lock longer. Second, it forces lock
1340	 * ordering of page lock and transaction start for journaling
1341	 * filesystems.
1342	 */
1343	if (i_size_changed)
1344		ret = ext4_mark_inode_dirty(handle, inode);
1345
1346	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1347		/* if we have allocated more blocks and copied
1348		 * less. We will have blocks allocated outside
1349		 * inode->i_size. So truncate them
1350		 */
1351		ext4_orphan_add(handle, inode);
1352
1353	ret2 = ext4_journal_stop(handle);
1354	if (!ret)
1355		ret = ret2;
1356
1357	if (pos + len > inode->i_size && !verity) {
1358		ext4_truncate_failed_write(inode);
1359		/*
1360		 * If truncate failed early the inode might still be
1361		 * on the orphan list; we need to make sure the inode
1362		 * is removed from the orphan list in that case.
1363		 */
1364		if (inode->i_nlink)
1365			ext4_orphan_del(NULL, inode);
1366	}
1367
1368	return ret ? ret : copied;
1369}
1370
1371/*
1372 * This is a private version of page_zero_new_buffers() which doesn't
1373 * set the buffer to be dirty, since in data=journalled mode we need
1374 * to call ext4_handle_dirty_metadata() instead.
1375 */
1376static void ext4_journalled_zero_new_buffers(handle_t *handle,
1377					    struct inode *inode,
1378					    struct page *page,
1379					    unsigned from, unsigned to)
1380{
1381	unsigned int block_start = 0, block_end;
1382	struct buffer_head *head, *bh;
1383
1384	bh = head = page_buffers(page);
1385	do {
1386		block_end = block_start + bh->b_size;
1387		if (buffer_new(bh)) {
1388			if (block_end > from && block_start < to) {
1389				if (!PageUptodate(page)) {
1390					unsigned start, size;
1391
1392					start = max(from, block_start);
1393					size = min(to, block_end) - start;
1394
1395					zero_user(page, start, size);
1396					write_end_fn(handle, inode, bh);
1397				}
1398				clear_buffer_new(bh);
 
1399			}
1400		}
1401		block_start = block_end;
1402		bh = bh->b_this_page;
1403	} while (bh != head);
1404}
1405
1406static int ext4_journalled_write_end(struct file *file,
1407				     struct address_space *mapping,
1408				     loff_t pos, unsigned len, unsigned copied,
1409				     struct page *page, void *fsdata)
1410{
1411	handle_t *handle = ext4_journal_current_handle();
1412	struct inode *inode = mapping->host;
1413	loff_t old_size = inode->i_size;
1414	int ret = 0, ret2;
1415	int partial = 0;
1416	unsigned from, to;
1417	int size_changed = 0;
1418	bool verity = ext4_verity_in_progress(inode);
1419
1420	trace_ext4_journalled_write_end(inode, pos, len, copied);
1421	from = pos & (PAGE_SIZE - 1);
1422	to = from + len;
1423
1424	BUG_ON(!ext4_handle_valid(handle));
1425
1426	if (ext4_has_inline_data(inode))
1427		return ext4_write_inline_data_end(inode, pos, len, copied, page);
 
1428
1429	if (unlikely(copied < len) && !PageUptodate(page)) {
1430		copied = 0;
1431		ext4_journalled_zero_new_buffers(handle, inode, page, from, to);
 
1432	} else {
1433		if (unlikely(copied < len))
1434			ext4_journalled_zero_new_buffers(handle, inode, page,
1435							 from + copied, to);
1436		ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
 
1437					     from, from + copied, &partial,
1438					     write_end_fn);
1439		if (!partial)
1440			SetPageUptodate(page);
1441	}
1442	if (!verity)
1443		size_changed = ext4_update_inode_size(inode, pos + copied);
1444	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1445	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1446	unlock_page(page);
1447	put_page(page);
1448
1449	if (old_size < pos && !verity)
1450		pagecache_isize_extended(inode, old_size, pos);
 
 
1451
1452	if (size_changed) {
1453		ret2 = ext4_mark_inode_dirty(handle, inode);
1454		if (!ret)
1455			ret = ret2;
1456	}
1457
1458	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1459		/* if we have allocated more blocks and copied
1460		 * less. We will have blocks allocated outside
1461		 * inode->i_size. So truncate them
1462		 */
1463		ext4_orphan_add(handle, inode);
1464
1465	ret2 = ext4_journal_stop(handle);
1466	if (!ret)
1467		ret = ret2;
1468	if (pos + len > inode->i_size && !verity) {
1469		ext4_truncate_failed_write(inode);
1470		/*
1471		 * If truncate failed early the inode might still be
1472		 * on the orphan list; we need to make sure the inode
1473		 * is removed from the orphan list in that case.
1474		 */
1475		if (inode->i_nlink)
1476			ext4_orphan_del(NULL, inode);
1477	}
1478
1479	return ret ? ret : copied;
1480}
1481
1482/*
1483 * Reserve space for a single cluster
1484 */
1485static int ext4_da_reserve_space(struct inode *inode)
1486{
1487	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1488	struct ext4_inode_info *ei = EXT4_I(inode);
1489	int ret;
1490
1491	/*
1492	 * We will charge metadata quota at writeout time; this saves
1493	 * us from metadata over-estimation, though we may go over by
1494	 * a small amount in the end.  Here we just reserve for data.
1495	 */
1496	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1497	if (ret)
1498		return ret;
1499
1500	spin_lock(&ei->i_block_reservation_lock);
1501	if (ext4_claim_free_clusters(sbi, 1, 0)) {
1502		spin_unlock(&ei->i_block_reservation_lock);
1503		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1504		return -ENOSPC;
1505	}
1506	ei->i_reserved_data_blocks++;
1507	trace_ext4_da_reserve_space(inode);
1508	spin_unlock(&ei->i_block_reservation_lock);
1509
1510	return 0;       /* success */
1511}
1512
1513void ext4_da_release_space(struct inode *inode, int to_free)
1514{
1515	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1516	struct ext4_inode_info *ei = EXT4_I(inode);
1517
1518	if (!to_free)
1519		return;		/* Nothing to release, exit */
1520
1521	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1522
1523	trace_ext4_da_release_space(inode, to_free);
1524	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1525		/*
1526		 * if there aren't enough reserved blocks, then the
1527		 * counter is messed up somewhere.  Since this
1528		 * function is called from invalidate page, it's
1529		 * harmless to return without any action.
1530		 */
1531		ext4_warning(inode->i_sb, "ext4_da_release_space: "
1532			 "ino %lu, to_free %d with only %d reserved "
1533			 "data blocks", inode->i_ino, to_free,
1534			 ei->i_reserved_data_blocks);
1535		WARN_ON(1);
1536		to_free = ei->i_reserved_data_blocks;
1537	}
1538	ei->i_reserved_data_blocks -= to_free;
1539
1540	/* update fs dirty data blocks counter */
1541	percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1542
1543	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1544
1545	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1546}
1547
1548/*
1549 * Delayed allocation stuff
1550 */
1551
1552struct mpage_da_data {
1553	/* These are input fields for ext4_do_writepages() */
1554	struct inode *inode;
1555	struct writeback_control *wbc;
1556	unsigned int can_map:1;	/* Can writepages call map blocks? */
1557
1558	/* These are internal state of ext4_do_writepages() */
1559	pgoff_t first_page;	/* The first page to write */
1560	pgoff_t next_page;	/* Current page to examine */
1561	pgoff_t last_page;	/* Last page to examine */
1562	/*
1563	 * Extent to map - this can be after first_page because that can be
1564	 * fully mapped. We somewhat abuse m_flags to store whether the extent
1565	 * is delalloc or unwritten.
1566	 */
1567	struct ext4_map_blocks map;
1568	struct ext4_io_submit io_submit;	/* IO submission data */
1569	unsigned int do_map:1;
1570	unsigned int scanned_until_end:1;
 
1571};
1572
1573static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1574				       bool invalidate)
1575{
1576	unsigned nr, i;
1577	pgoff_t index, end;
1578	struct folio_batch fbatch;
1579	struct inode *inode = mpd->inode;
1580	struct address_space *mapping = inode->i_mapping;
1581
1582	/* This is necessary when next_page == 0. */
1583	if (mpd->first_page >= mpd->next_page)
1584		return;
1585
1586	mpd->scanned_until_end = 0;
1587	index = mpd->first_page;
1588	end   = mpd->next_page - 1;
1589	if (invalidate) {
1590		ext4_lblk_t start, last;
1591		start = index << (PAGE_SHIFT - inode->i_blkbits);
1592		last = end << (PAGE_SHIFT - inode->i_blkbits);
1593
1594		/*
1595		 * avoid racing with extent status tree scans made by
1596		 * ext4_insert_delayed_block()
1597		 */
1598		down_write(&EXT4_I(inode)->i_data_sem);
1599		ext4_es_remove_extent(inode, start, last - start + 1);
1600		up_write(&EXT4_I(inode)->i_data_sem);
1601	}
1602
1603	folio_batch_init(&fbatch);
1604	while (index <= end) {
1605		nr = filemap_get_folios(mapping, &index, end, &fbatch);
1606		if (nr == 0)
1607			break;
1608		for (i = 0; i < nr; i++) {
1609			struct folio *folio = fbatch.folios[i];
1610
1611			if (folio->index < mpd->first_page)
1612				continue;
1613			if (folio->index + folio_nr_pages(folio) - 1 > end)
1614				continue;
1615			BUG_ON(!folio_test_locked(folio));
1616			BUG_ON(folio_test_writeback(folio));
1617			if (invalidate) {
1618				if (folio_mapped(folio))
1619					folio_clear_dirty_for_io(folio);
1620				block_invalidate_folio(folio, 0,
1621						folio_size(folio));
1622				folio_clear_uptodate(folio);
1623			}
1624			folio_unlock(folio);
1625		}
1626		folio_batch_release(&fbatch);
1627	}
1628}
1629
1630static void ext4_print_free_blocks(struct inode *inode)
1631{
1632	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1633	struct super_block *sb = inode->i_sb;
1634	struct ext4_inode_info *ei = EXT4_I(inode);
1635
1636	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1637	       EXT4_C2B(EXT4_SB(inode->i_sb),
1638			ext4_count_free_clusters(sb)));
1639	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1640	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1641	       (long long) EXT4_C2B(EXT4_SB(sb),
1642		percpu_counter_sum(&sbi->s_freeclusters_counter)));
1643	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1644	       (long long) EXT4_C2B(EXT4_SB(sb),
1645		percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1646	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1647	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1648		 ei->i_reserved_data_blocks);
1649	return;
1650}
1651
1652static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode,
1653				      struct buffer_head *bh)
 
 
 
 
 
 
 
1654{
1655	return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1656}
1657
1658/*
1659 * ext4_insert_delayed_block - adds a delayed block to the extents status
1660 *                             tree, incrementing the reserved cluster/block
1661 *                             count or making a pending reservation
1662 *                             where needed
1663 *
1664 * @inode - file containing the newly added block
1665 * @lblk - logical block to be added
 
1666 *
1667 * Returns 0 on success, negative error code on failure.
1668 */
1669static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
 
1670{
1671	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1672	int ret;
1673	bool allocated = false;
1674	bool reserved = false;
 
 
1675
1676	/*
1677	 * If the cluster containing lblk is shared with a delayed,
1678	 * written, or unwritten extent in a bigalloc file system, it's
1679	 * already been accounted for and does not need to be reserved.
1680	 * A pending reservation must be made for the cluster if it's
1681	 * shared with a written or unwritten extent and doesn't already
1682	 * have one.  Written and unwritten extents can be purged from the
1683	 * extents status tree if the system is under memory pressure, so
1684	 * it's necessary to examine the extent tree if a search of the
1685	 * extents status tree doesn't get a match.
1686	 */
1687	if (sbi->s_cluster_ratio == 1) {
1688		ret = ext4_da_reserve_space(inode);
1689		if (ret != 0)   /* ENOSPC */
1690			goto errout;
1691		reserved = true;
1692	} else {   /* bigalloc */
1693		if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1694			if (!ext4_es_scan_clu(inode,
1695					      &ext4_es_is_mapped, lblk)) {
1696				ret = ext4_clu_mapped(inode,
1697						      EXT4_B2C(sbi, lblk));
1698				if (ret < 0)
1699					goto errout;
1700				if (ret == 0) {
1701					ret = ext4_da_reserve_space(inode);
1702					if (ret != 0)   /* ENOSPC */
1703						goto errout;
1704					reserved = true;
1705				} else {
1706					allocated = true;
1707				}
1708			} else {
1709				allocated = true;
1710			}
1711		}
 
 
 
 
 
 
1712	}
1713
1714	ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1715	if (ret && reserved)
1716		ext4_da_release_space(inode, 1);
1717
1718errout:
1719	return ret;
1720}
1721
1722/*
1723 * This function is grabs code from the very beginning of
1724 * ext4_map_blocks, but assumes that the caller is from delayed write
1725 * time. This function looks up the requested blocks and sets the
1726 * buffer delay bit under the protection of i_data_sem.
1727 */
1728static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1729			      struct ext4_map_blocks *map,
1730			      struct buffer_head *bh)
 
1731{
1732	struct extent_status es;
1733	int retval;
1734	sector_t invalid_block = ~((sector_t) 0xffff);
1735#ifdef ES_AGGRESSIVE_TEST
1736	struct ext4_map_blocks orig_map;
1737
1738	memcpy(&orig_map, map, sizeof(*map));
1739#endif
1740
1741	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1742		invalid_block = ~0;
1743
1744	map->m_flags = 0;
1745	ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1746		  (unsigned long) map->m_lblk);
1747
1748	/* Lookup extent status tree firstly */
1749	if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1750		if (ext4_es_is_hole(&es)) {
1751			retval = 0;
1752			down_read(&EXT4_I(inode)->i_data_sem);
 
1753			goto add_delayed;
1754		}
1755
 
1756		/*
1757		 * Delayed extent could be allocated by fallocate.
1758		 * So we need to check it.
1759		 */
1760		if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1761			map_bh(bh, inode->i_sb, invalid_block);
1762			set_buffer_new(bh);
1763			set_buffer_delay(bh);
1764			return 0;
1765		}
1766
1767		map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1768		retval = es.es_len - (iblock - es.es_lblk);
1769		if (retval > map->m_len)
1770			retval = map->m_len;
1771		map->m_len = retval;
1772		if (ext4_es_is_written(&es))
1773			map->m_flags |= EXT4_MAP_MAPPED;
1774		else if (ext4_es_is_unwritten(&es))
1775			map->m_flags |= EXT4_MAP_UNWRITTEN;
1776		else
1777			BUG();
1778
1779#ifdef ES_AGGRESSIVE_TEST
1780		ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1781#endif
1782		return retval;
1783	}
1784
1785	/*
1786	 * Try to see if we can get the block without requesting a new
1787	 * file system block.
1788	 */
1789	down_read(&EXT4_I(inode)->i_data_sem);
1790	if (ext4_has_inline_data(inode))
1791		retval = 0;
1792	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1793		retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1794	else
1795		retval = ext4_ind_map_blocks(NULL, inode, map, 0);
 
 
 
1796
1797add_delayed:
1798	if (retval == 0) {
1799		int ret;
 
 
 
 
 
 
 
 
 
1800
1801		/*
1802		 * XXX: __block_prepare_write() unmaps passed block,
1803		 * is it OK?
1804		 */
1805
1806		ret = ext4_insert_delayed_block(inode, map->m_lblk);
1807		if (ret != 0) {
1808			retval = ret;
1809			goto out_unlock;
1810		}
1811
1812		map_bh(bh, inode->i_sb, invalid_block);
1813		set_buffer_new(bh);
1814		set_buffer_delay(bh);
1815	} else if (retval > 0) {
1816		int ret;
1817		unsigned int status;
1818
1819		if (unlikely(retval != map->m_len)) {
1820			ext4_warning(inode->i_sb,
1821				     "ES len assertion failed for inode "
1822				     "%lu: retval %d != map->m_len %d",
1823				     inode->i_ino, retval, map->m_len);
1824			WARN_ON(1);
1825		}
1826
1827		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1828				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1829		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1830					    map->m_pblk, status);
1831		if (ret != 0)
1832			retval = ret;
1833	}
1834
1835out_unlock:
1836	up_read((&EXT4_I(inode)->i_data_sem));
 
1837
1838	return retval;
1839}
1840
1841/*
1842 * This is a special get_block_t callback which is used by
1843 * ext4_da_write_begin().  It will either return mapped block or
1844 * reserve space for a single block.
1845 *
1846 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1847 * We also have b_blocknr = -1 and b_bdev initialized properly
1848 *
1849 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1850 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1851 * initialized properly.
1852 */
1853int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1854			   struct buffer_head *bh, int create)
1855{
1856	struct ext4_map_blocks map;
 
1857	int ret = 0;
1858
1859	BUG_ON(create == 0);
1860	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1861
 
 
 
1862	map.m_lblk = iblock;
1863	map.m_len = 1;
1864
1865	/*
1866	 * first, we need to know whether the block is allocated already
1867	 * preallocated blocks are unmapped but should treated
1868	 * the same as allocated blocks.
1869	 */
1870	ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1871	if (ret <= 0)
1872		return ret;
1873
 
 
 
 
 
 
 
1874	map_bh(bh, inode->i_sb, map.m_pblk);
1875	ext4_update_bh_state(bh, map.m_flags);
1876
1877	if (buffer_unwritten(bh)) {
1878		/* A delayed write to unwritten bh should be marked
1879		 * new and mapped.  Mapped ensures that we don't do
1880		 * get_block multiple times when we write to the same
1881		 * offset and new ensures that we do proper zero out
1882		 * for partial write.
1883		 */
1884		set_buffer_new(bh);
1885		set_buffer_mapped(bh);
1886	}
1887	return 0;
1888}
1889
1890static int __ext4_journalled_writepage(struct page *page,
1891				       unsigned int len)
1892{
1893	struct address_space *mapping = page->mapping;
1894	struct inode *inode = mapping->host;
1895	handle_t *handle = NULL;
1896	int ret = 0, err = 0;
1897	int inline_data = ext4_has_inline_data(inode);
1898	struct buffer_head *inode_bh = NULL;
1899	loff_t size;
1900
1901	ClearPageChecked(page);
1902
1903	if (inline_data) {
1904		BUG_ON(page->index != 0);
1905		BUG_ON(len > ext4_get_max_inline_size(inode));
1906		inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1907		if (inode_bh == NULL)
1908			goto out;
1909	}
1910	/*
1911	 * We need to release the page lock before we start the
1912	 * journal, so grab a reference so the page won't disappear
1913	 * out from under us.
1914	 */
1915	get_page(page);
1916	unlock_page(page);
1917
1918	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1919				    ext4_writepage_trans_blocks(inode));
1920	if (IS_ERR(handle)) {
1921		ret = PTR_ERR(handle);
1922		put_page(page);
1923		goto out_no_pagelock;
1924	}
1925	BUG_ON(!ext4_handle_valid(handle));
1926
1927	lock_page(page);
1928	put_page(page);
1929	size = i_size_read(inode);
1930	if (page->mapping != mapping || page_offset(page) > size) {
1931		/* The page got truncated from under us */
1932		ext4_journal_stop(handle);
1933		ret = 0;
1934		goto out;
1935	}
1936
1937	if (inline_data) {
1938		ret = ext4_mark_inode_dirty(handle, inode);
1939	} else {
1940		struct buffer_head *page_bufs = page_buffers(page);
1941
1942		if (page->index == size >> PAGE_SHIFT)
1943			len = size & ~PAGE_MASK;
1944		else
1945			len = PAGE_SIZE;
1946
1947		ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1948					     NULL, do_journal_get_write_access);
1949
1950		err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1951					     NULL, write_end_fn);
1952	}
1953	if (ret == 0)
1954		ret = err;
1955	err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1956	if (ret == 0)
1957		ret = err;
1958	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1959	err = ext4_journal_stop(handle);
1960	if (!ret)
1961		ret = err;
1962
1963	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1964out:
1965	unlock_page(page);
1966out_no_pagelock:
1967	brelse(inode_bh);
1968	return ret;
1969}
1970
1971/*
1972 * Note that we don't need to start a transaction unless we're journaling data
1973 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1974 * need to file the inode to the transaction's list in ordered mode because if
1975 * we are writing back data added by write(), the inode is already there and if
1976 * we are writing back data modified via mmap(), no one guarantees in which
1977 * transaction the data will hit the disk. In case we are journaling data, we
1978 * cannot start transaction directly because transaction start ranks above page
1979 * lock so we have to do some magic.
1980 *
1981 * This function can get called via...
1982 *   - ext4_writepages after taking page lock (have journal handle)
1983 *   - journal_submit_inode_data_buffers (no journal handle)
1984 *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1985 *   - grab_page_cache when doing write_begin (have journal handle)
1986 *
1987 * We don't do any block allocation in this function. If we have page with
1988 * multiple blocks we need to write those buffer_heads that are mapped. This
1989 * is important for mmaped based write. So if we do with blocksize 1K
1990 * truncate(f, 1024);
1991 * a = mmap(f, 0, 4096);
1992 * a[0] = 'a';
1993 * truncate(f, 4096);
1994 * we have in the page first buffer_head mapped via page_mkwrite call back
1995 * but other buffer_heads would be unmapped but dirty (dirty done via the
1996 * do_wp_page). So writepage should write the first block. If we modify
1997 * the mmap area beyond 1024 we will again get a page_fault and the
1998 * page_mkwrite callback will do the block allocation and mark the
1999 * buffer_heads mapped.
2000 *
2001 * We redirty the page if we have any buffer_heads that is either delay or
2002 * unwritten in the page.
2003 *
2004 * We can get recursively called as show below.
2005 *
2006 *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2007 *		ext4_writepage()
2008 *
2009 * But since we don't do any block allocation we should not deadlock.
2010 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2011 */
2012static int ext4_writepage(struct page *page,
2013			  struct writeback_control *wbc)
2014{
2015	struct folio *folio = page_folio(page);
2016	int ret = 0;
2017	loff_t size;
2018	unsigned int len;
2019	struct buffer_head *page_bufs = NULL;
2020	struct inode *inode = page->mapping->host;
2021	struct ext4_io_submit io_submit;
2022
2023	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2024		folio_invalidate(folio, 0, folio_size(folio));
2025		folio_unlock(folio);
2026		return -EIO;
2027	}
2028
2029	trace_ext4_writepage(page);
2030	size = i_size_read(inode);
2031	if (page->index == size >> PAGE_SHIFT &&
2032	    !ext4_verity_in_progress(inode))
2033		len = size & ~PAGE_MASK;
2034	else
2035		len = PAGE_SIZE;
2036
2037	/* Should never happen but for bugs in other kernel subsystems */
2038	if (!page_has_buffers(page)) {
2039		ext4_warning_inode(inode,
2040		   "page %lu does not have buffers attached", page->index);
2041		ClearPageDirty(page);
2042		unlock_page(page);
2043		return 0;
2044	}
2045
2046	page_bufs = page_buffers(page);
2047	/*
2048	 * We cannot do block allocation or other extent handling in this
2049	 * function. If there are buffers needing that, we have to redirty
2050	 * the page. But we may reach here when we do a journal commit via
2051	 * journal_submit_inode_data_buffers() and in that case we must write
2052	 * allocated buffers to achieve data=ordered mode guarantees.
2053	 *
2054	 * Also, if there is only one buffer per page (the fs block
2055	 * size == the page size), if one buffer needs block
2056	 * allocation or needs to modify the extent tree to clear the
2057	 * unwritten flag, we know that the page can't be written at
2058	 * all, so we might as well refuse the write immediately.
2059	 * Unfortunately if the block size != page size, we can't as
2060	 * easily detect this case using ext4_walk_page_buffers(), but
2061	 * for the extremely common case, this is an optimization that
2062	 * skips a useless round trip through ext4_bio_write_page().
2063	 */
2064	if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL,
2065				   ext4_bh_delay_or_unwritten)) {
2066		redirty_page_for_writepage(wbc, page);
2067		if ((current->flags & PF_MEMALLOC) ||
2068		    (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2069			/*
2070			 * For memory cleaning there's no point in writing only
2071			 * some buffers. So just bail out. Warn if we came here
2072			 * from direct reclaim.
2073			 */
2074			WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2075							== PF_MEMALLOC);
2076			unlock_page(page);
2077			return 0;
2078		}
2079	}
2080
2081	if (PageChecked(page) && ext4_should_journal_data(inode))
2082		/*
2083		 * It's mmapped pagecache.  Add buffers and journal it.  There
2084		 * doesn't seem much point in redirtying the page here.
2085		 */
2086		return __ext4_journalled_writepage(page, len);
2087
2088	ext4_io_submit_init(&io_submit, wbc);
2089	io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2090	if (!io_submit.io_end) {
2091		redirty_page_for_writepage(wbc, page);
2092		unlock_page(page);
2093		return -ENOMEM;
2094	}
2095	ret = ext4_bio_write_page(&io_submit, page, len);
2096	ext4_io_submit(&io_submit);
2097	/* Drop io_end reference we got from init */
2098	ext4_put_io_end_defer(io_submit.io_end);
2099	return ret;
2100}
2101
2102static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2103{
2104	int len;
2105	loff_t size;
2106	int err;
2107
2108	BUG_ON(page->index != mpd->first_page);
2109	clear_page_dirty_for_io(page);
2110	/*
2111	 * We have to be very careful here!  Nothing protects writeback path
2112	 * against i_size changes and the page can be writeably mapped into
2113	 * page tables. So an application can be growing i_size and writing
2114	 * data through mmap while writeback runs. clear_page_dirty_for_io()
2115	 * write-protects our page in page tables and the page cannot get
2116	 * written to again until we release page lock. So only after
2117	 * clear_page_dirty_for_io() we are safe to sample i_size for
2118	 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2119	 * on the barrier provided by TestClearPageDirty in
2120	 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2121	 * after page tables are updated.
2122	 */
2123	size = i_size_read(mpd->inode);
2124	if (page->index == size >> PAGE_SHIFT &&
 
2125	    !ext4_verity_in_progress(mpd->inode))
2126		len = size & ~PAGE_MASK;
2127	else
2128		len = PAGE_SIZE;
2129	err = ext4_bio_write_page(&mpd->io_submit, page, len);
2130	if (!err)
2131		mpd->wbc->nr_to_write--;
2132	mpd->first_page++;
2133
2134	return err;
2135}
2136
2137#define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2138
2139/*
2140 * mballoc gives us at most this number of blocks...
2141 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2142 * The rest of mballoc seems to handle chunks up to full group size.
2143 */
2144#define MAX_WRITEPAGES_EXTENT_LEN 2048
2145
2146/*
2147 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2148 *
2149 * @mpd - extent of blocks
2150 * @lblk - logical number of the block in the file
2151 * @bh - buffer head we want to add to the extent
2152 *
2153 * The function is used to collect contig. blocks in the same state. If the
2154 * buffer doesn't require mapping for writeback and we haven't started the
2155 * extent of buffers to map yet, the function returns 'true' immediately - the
2156 * caller can write the buffer right away. Otherwise the function returns true
2157 * if the block has been added to the extent, false if the block couldn't be
2158 * added.
2159 */
2160static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2161				   struct buffer_head *bh)
2162{
2163	struct ext4_map_blocks *map = &mpd->map;
2164
2165	/* Buffer that doesn't need mapping for writeback? */
2166	if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2167	    (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2168		/* So far no extent to map => we write the buffer right away */
2169		if (map->m_len == 0)
2170			return true;
2171		return false;
2172	}
2173
2174	/* First block in the extent? */
2175	if (map->m_len == 0) {
2176		/* We cannot map unless handle is started... */
2177		if (!mpd->do_map)
2178			return false;
2179		map->m_lblk = lblk;
2180		map->m_len = 1;
2181		map->m_flags = bh->b_state & BH_FLAGS;
2182		return true;
2183	}
2184
2185	/* Don't go larger than mballoc is willing to allocate */
2186	if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2187		return false;
2188
2189	/* Can we merge the block to our big extent? */
2190	if (lblk == map->m_lblk + map->m_len &&
2191	    (bh->b_state & BH_FLAGS) == map->m_flags) {
2192		map->m_len++;
2193		return true;
2194	}
2195	return false;
2196}
2197
2198/*
2199 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2200 *
2201 * @mpd - extent of blocks for mapping
2202 * @head - the first buffer in the page
2203 * @bh - buffer we should start processing from
2204 * @lblk - logical number of the block in the file corresponding to @bh
2205 *
2206 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2207 * the page for IO if all buffers in this page were mapped and there's no
2208 * accumulated extent of buffers to map or add buffers in the page to the
2209 * extent of buffers to map. The function returns 1 if the caller can continue
2210 * by processing the next page, 0 if it should stop adding buffers to the
2211 * extent to map because we cannot extend it anymore. It can also return value
2212 * < 0 in case of error during IO submission.
2213 */
2214static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2215				   struct buffer_head *head,
2216				   struct buffer_head *bh,
2217				   ext4_lblk_t lblk)
2218{
2219	struct inode *inode = mpd->inode;
2220	int err;
2221	ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2222							>> inode->i_blkbits;
2223
2224	if (ext4_verity_in_progress(inode))
2225		blocks = EXT_MAX_BLOCKS;
2226
2227	do {
2228		BUG_ON(buffer_locked(bh));
2229
2230		if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2231			/* Found extent to map? */
2232			if (mpd->map.m_len)
2233				return 0;
2234			/* Buffer needs mapping and handle is not started? */
2235			if (!mpd->do_map)
2236				return 0;
2237			/* Everything mapped so far and we hit EOF */
2238			break;
2239		}
2240	} while (lblk++, (bh = bh->b_this_page) != head);
2241	/* So far everything mapped? Submit the page for IO. */
2242	if (mpd->map.m_len == 0) {
2243		err = mpage_submit_page(mpd, head->b_page);
2244		if (err < 0)
2245			return err;
 
2246	}
2247	if (lblk >= blocks) {
2248		mpd->scanned_until_end = 1;
2249		return 0;
2250	}
2251	return 1;
2252}
2253
2254/*
2255 * mpage_process_page - update page buffers corresponding to changed extent and
2256 *		       may submit fully mapped page for IO
 
 
 
 
 
 
2257 *
2258 * @mpd		- description of extent to map, on return next extent to map
2259 * @m_lblk	- logical block mapping.
2260 * @m_pblk	- corresponding physical mapping.
2261 * @map_bh	- determines on return whether this page requires any further
2262 *		  mapping or not.
2263 * Scan given page buffers corresponding to changed extent and update buffer
2264 * state according to new extent state.
2265 * We map delalloc buffers to their physical location, clear unwritten bits.
2266 * If the given page is not fully mapped, we update @map to the next extent in
2267 * the given page that needs mapping & return @map_bh as true.
2268 */
2269static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2270			      ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2271			      bool *map_bh)
2272{
2273	struct buffer_head *head, *bh;
2274	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2275	ext4_lblk_t lblk = *m_lblk;
2276	ext4_fsblk_t pblock = *m_pblk;
2277	int err = 0;
2278	int blkbits = mpd->inode->i_blkbits;
2279	ssize_t io_end_size = 0;
2280	struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2281
2282	bh = head = page_buffers(page);
2283	do {
2284		if (lblk < mpd->map.m_lblk)
2285			continue;
2286		if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2287			/*
2288			 * Buffer after end of mapped extent.
2289			 * Find next buffer in the page to map.
2290			 */
2291			mpd->map.m_len = 0;
2292			mpd->map.m_flags = 0;
2293			io_end_vec->size += io_end_size;
2294
2295			err = mpage_process_page_bufs(mpd, head, bh, lblk);
2296			if (err > 0)
2297				err = 0;
2298			if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2299				io_end_vec = ext4_alloc_io_end_vec(io_end);
2300				if (IS_ERR(io_end_vec)) {
2301					err = PTR_ERR(io_end_vec);
2302					goto out;
2303				}
2304				io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2305			}
2306			*map_bh = true;
2307			goto out;
2308		}
2309		if (buffer_delay(bh)) {
2310			clear_buffer_delay(bh);
2311			bh->b_blocknr = pblock++;
2312		}
2313		clear_buffer_unwritten(bh);
2314		io_end_size += (1 << blkbits);
2315	} while (lblk++, (bh = bh->b_this_page) != head);
2316
2317	io_end_vec->size += io_end_size;
2318	*map_bh = false;
2319out:
2320	*m_lblk = lblk;
2321	*m_pblk = pblock;
2322	return err;
2323}
2324
2325/*
2326 * mpage_map_buffers - update buffers corresponding to changed extent and
2327 *		       submit fully mapped pages for IO
2328 *
2329 * @mpd - description of extent to map, on return next extent to map
2330 *
2331 * Scan buffers corresponding to changed extent (we expect corresponding pages
2332 * to be already locked) and update buffer state according to new extent state.
2333 * We map delalloc buffers to their physical location, clear unwritten bits,
2334 * and mark buffers as uninit when we perform writes to unwritten extents
2335 * and do extent conversion after IO is finished. If the last page is not fully
2336 * mapped, we update @map to the next extent in the last page that needs
2337 * mapping. Otherwise we submit the page for IO.
2338 */
2339static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2340{
2341	struct folio_batch fbatch;
2342	unsigned nr, i;
2343	struct inode *inode = mpd->inode;
2344	int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2345	pgoff_t start, end;
2346	ext4_lblk_t lblk;
2347	ext4_fsblk_t pblock;
2348	int err;
2349	bool map_bh = false;
2350
2351	start = mpd->map.m_lblk >> bpp_bits;
2352	end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2353	lblk = start << bpp_bits;
2354	pblock = mpd->map.m_pblk;
2355
2356	folio_batch_init(&fbatch);
2357	while (start <= end) {
2358		nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2359		if (nr == 0)
2360			break;
2361		for (i = 0; i < nr; i++) {
2362			struct page *page = &fbatch.folios[i]->page;
2363
2364			err = mpage_process_page(mpd, page, &lblk, &pblock,
2365						 &map_bh);
2366			/*
2367			 * If map_bh is true, means page may require further bh
2368			 * mapping, or maybe the page was submitted for IO.
2369			 * So we return to call further extent mapping.
2370			 */
2371			if (err < 0 || map_bh)
2372				goto out;
2373			/* Page fully mapped - let IO run! */
2374			err = mpage_submit_page(mpd, page);
2375			if (err < 0)
2376				goto out;
 
2377		}
2378		folio_batch_release(&fbatch);
2379	}
2380	/* Extent fully mapped and matches with page boundary. We are done. */
2381	mpd->map.m_len = 0;
2382	mpd->map.m_flags = 0;
2383	return 0;
2384out:
2385	folio_batch_release(&fbatch);
2386	return err;
2387}
2388
2389static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2390{
2391	struct inode *inode = mpd->inode;
2392	struct ext4_map_blocks *map = &mpd->map;
2393	int get_blocks_flags;
2394	int err, dioread_nolock;
2395
2396	trace_ext4_da_write_pages_extent(inode, map);
2397	/*
2398	 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2399	 * to convert an unwritten extent to be initialized (in the case
2400	 * where we have written into one or more preallocated blocks).  It is
2401	 * possible that we're going to need more metadata blocks than
2402	 * previously reserved. However we must not fail because we're in
2403	 * writeback and there is nothing we can do about it so it might result
2404	 * in data loss.  So use reserved blocks to allocate metadata if
2405	 * possible.
2406	 *
2407	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2408	 * the blocks in question are delalloc blocks.  This indicates
2409	 * that the blocks and quotas has already been checked when
2410	 * the data was copied into the page cache.
2411	 */
2412	get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2413			   EXT4_GET_BLOCKS_METADATA_NOFAIL |
2414			   EXT4_GET_BLOCKS_IO_SUBMIT;
2415	dioread_nolock = ext4_should_dioread_nolock(inode);
2416	if (dioread_nolock)
2417		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2418	if (map->m_flags & BIT(BH_Delay))
2419		get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2420
2421	err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2422	if (err < 0)
2423		return err;
2424	if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2425		if (!mpd->io_submit.io_end->handle &&
2426		    ext4_handle_valid(handle)) {
2427			mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2428			handle->h_rsv_handle = NULL;
2429		}
2430		ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2431	}
2432
2433	BUG_ON(map->m_len == 0);
2434	return 0;
2435}
2436
2437/*
2438 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2439 *				 mpd->len and submit pages underlying it for IO
2440 *
2441 * @handle - handle for journal operations
2442 * @mpd - extent to map
2443 * @give_up_on_write - we set this to true iff there is a fatal error and there
2444 *                     is no hope of writing the data. The caller should discard
2445 *                     dirty pages to avoid infinite loops.
2446 *
2447 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2448 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2449 * them to initialized or split the described range from larger unwritten
2450 * extent. Note that we need not map all the described range since allocation
2451 * can return less blocks or the range is covered by more unwritten extents. We
2452 * cannot map more because we are limited by reserved transaction credits. On
2453 * the other hand we always make sure that the last touched page is fully
2454 * mapped so that it can be written out (and thus forward progress is
2455 * guaranteed). After mapping we submit all mapped pages for IO.
2456 */
2457static int mpage_map_and_submit_extent(handle_t *handle,
2458				       struct mpage_da_data *mpd,
2459				       bool *give_up_on_write)
2460{
2461	struct inode *inode = mpd->inode;
2462	struct ext4_map_blocks *map = &mpd->map;
2463	int err;
2464	loff_t disksize;
2465	int progress = 0;
2466	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2467	struct ext4_io_end_vec *io_end_vec;
2468
2469	io_end_vec = ext4_alloc_io_end_vec(io_end);
2470	if (IS_ERR(io_end_vec))
2471		return PTR_ERR(io_end_vec);
2472	io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2473	do {
2474		err = mpage_map_one_extent(handle, mpd);
2475		if (err < 0) {
2476			struct super_block *sb = inode->i_sb;
2477
2478			if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2479			    ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2480				goto invalidate_dirty_pages;
2481			/*
2482			 * Let the uper layers retry transient errors.
2483			 * In the case of ENOSPC, if ext4_count_free_blocks()
2484			 * is non-zero, a commit should free up blocks.
2485			 */
2486			if ((err == -ENOMEM) ||
2487			    (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2488				if (progress)
2489					goto update_disksize;
2490				return err;
2491			}
2492			ext4_msg(sb, KERN_CRIT,
2493				 "Delayed block allocation failed for "
2494				 "inode %lu at logical offset %llu with"
2495				 " max blocks %u with error %d",
2496				 inode->i_ino,
2497				 (unsigned long long)map->m_lblk,
2498				 (unsigned)map->m_len, -err);
2499			ext4_msg(sb, KERN_CRIT,
2500				 "This should not happen!! Data will "
2501				 "be lost\n");
2502			if (err == -ENOSPC)
2503				ext4_print_free_blocks(inode);
2504		invalidate_dirty_pages:
2505			*give_up_on_write = true;
2506			return err;
2507		}
2508		progress = 1;
2509		/*
2510		 * Update buffer state, submit mapped pages, and get us new
2511		 * extent to map
2512		 */
2513		err = mpage_map_and_submit_buffers(mpd);
2514		if (err < 0)
2515			goto update_disksize;
2516	} while (map->m_len);
2517
2518update_disksize:
2519	/*
2520	 * Update on-disk size after IO is submitted.  Races with
2521	 * truncate are avoided by checking i_size under i_data_sem.
2522	 */
2523	disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2524	if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2525		int err2;
2526		loff_t i_size;
2527
2528		down_write(&EXT4_I(inode)->i_data_sem);
2529		i_size = i_size_read(inode);
2530		if (disksize > i_size)
2531			disksize = i_size;
2532		if (disksize > EXT4_I(inode)->i_disksize)
2533			EXT4_I(inode)->i_disksize = disksize;
2534		up_write(&EXT4_I(inode)->i_data_sem);
2535		err2 = ext4_mark_inode_dirty(handle, inode);
2536		if (err2) {
2537			ext4_error_err(inode->i_sb, -err2,
2538				       "Failed to mark inode %lu dirty",
2539				       inode->i_ino);
2540		}
2541		if (!err)
2542			err = err2;
2543	}
2544	return err;
2545}
2546
2547/*
2548 * Calculate the total number of credits to reserve for one writepages
2549 * iteration. This is called from ext4_writepages(). We map an extent of
2550 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2551 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2552 * bpp - 1 blocks in bpp different extents.
2553 */
2554static int ext4_da_writepages_trans_blocks(struct inode *inode)
2555{
2556	int bpp = ext4_journal_blocks_per_page(inode);
2557
2558	return ext4_meta_trans_blocks(inode,
2559				MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2560}
2561
2562/* Return true if the page needs to be written as part of transaction commit */
2563static bool ext4_page_nomap_can_writeout(struct page *page)
2564{
2565	struct buffer_head *bh, *head;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2566
2567	bh = head = page_buffers(page);
2568	do {
2569		if (buffer_dirty(bh) && buffer_mapped(bh) && !buffer_delay(bh))
2570			return true;
2571	} while ((bh = bh->b_this_page) != head);
2572	return false;
2573}
2574
2575/*
2576 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2577 * 				 needing mapping, submit mapped pages
2578 *
2579 * @mpd - where to look for pages
2580 *
2581 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2582 * IO immediately. If we cannot map blocks, we submit just already mapped
2583 * buffers in the page for IO and keep page dirty. When we can map blocks and
2584 * we find a page which isn't mapped we start accumulating extent of buffers
2585 * underlying these pages that needs mapping (formed by either delayed or
2586 * unwritten buffers). We also lock the pages containing these buffers. The
2587 * extent found is returned in @mpd structure (starting at mpd->lblk with
2588 * length mpd->len blocks).
2589 *
2590 * Note that this function can attach bios to one io_end structure which are
2591 * neither logically nor physically contiguous. Although it may seem as an
2592 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2593 * case as we need to track IO to all buffers underlying a page in one io_end.
2594 */
2595static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2596{
2597	struct address_space *mapping = mpd->inode->i_mapping;
2598	struct pagevec pvec;
2599	unsigned int nr_pages;
2600	long left = mpd->wbc->nr_to_write;
2601	pgoff_t index = mpd->first_page;
2602	pgoff_t end = mpd->last_page;
2603	xa_mark_t tag;
2604	int i, err = 0;
2605	int blkbits = mpd->inode->i_blkbits;
2606	ext4_lblk_t lblk;
2607	struct buffer_head *head;
 
 
2608
2609	if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2610		tag = PAGECACHE_TAG_TOWRITE;
2611	else
2612		tag = PAGECACHE_TAG_DIRTY;
2613
2614	pagevec_init(&pvec);
2615	mpd->map.m_len = 0;
2616	mpd->next_page = index;
 
 
 
 
 
 
 
2617	while (index <= end) {
2618		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2619				tag);
2620		if (nr_pages == 0)
2621			break;
2622
2623		for (i = 0; i < nr_pages; i++) {
2624			struct page *page = pvec.pages[i];
2625
2626			/*
2627			 * Accumulated enough dirty pages? This doesn't apply
2628			 * to WB_SYNC_ALL mode. For integrity sync we have to
2629			 * keep going because someone may be concurrently
2630			 * dirtying pages, and we might have synced a lot of
2631			 * newly appeared dirty pages, but have not synced all
2632			 * of the old dirty pages.
2633			 */
2634			if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
 
 
2635				goto out;
2636
2637			/* If we can't merge this page, we are done. */
2638			if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2639				goto out;
2640
2641			lock_page(page);
 
 
 
 
 
 
 
2642			/*
2643			 * If the page is no longer dirty, or its mapping no
2644			 * longer corresponds to inode we are writing (which
2645			 * means it has been truncated or invalidated), or the
2646			 * page is already under writeback and we are not doing
2647			 * a data integrity writeback, skip the page
2648			 */
2649			if (!PageDirty(page) ||
2650			    (PageWriteback(page) &&
2651			     (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2652			    unlikely(page->mapping != mapping)) {
2653				unlock_page(page);
2654				continue;
2655			}
2656
2657			wait_on_page_writeback(page);
2658			BUG_ON(PageWriteback(page));
2659
2660			/*
2661			 * Should never happen but for buggy code in
2662			 * other subsystems that call
2663			 * set_page_dirty() without properly warning
2664			 * the file system first.  See [1] for more
2665			 * information.
2666			 *
2667			 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2668			 */
2669			if (!page_has_buffers(page)) {
2670				ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2671				ClearPageDirty(page);
2672				unlock_page(page);
2673				continue;
2674			}
2675
2676			if (mpd->map.m_len == 0)
2677				mpd->first_page = page->index;
2678			mpd->next_page = page->index + 1;
2679			/*
2680			 * Writeout for transaction commit where we cannot
2681			 * modify metadata is simple. Just submit the page.
 
 
 
 
 
 
2682			 */
2683			if (!mpd->can_map) {
2684				if (ext4_page_nomap_can_writeout(page)) {
2685					err = mpage_submit_page(mpd, page);
 
 
 
 
 
2686					if (err < 0)
2687						goto out;
2688				} else {
2689					unlock_page(page);
2690					mpd->first_page++;
2691				}
 
2692			} else {
2693				/* Add all dirty buffers to mpd */
2694				lblk = ((ext4_lblk_t)page->index) <<
2695					(PAGE_SHIFT - blkbits);
2696				head = page_buffers(page);
2697				err = mpage_process_page_bufs(mpd, head, head,
2698							      lblk);
2699				if (err <= 0)
2700					goto out;
2701				err = 0;
2702			}
2703			left--;
2704		}
2705		pagevec_release(&pvec);
2706		cond_resched();
2707	}
2708	mpd->scanned_until_end = 1;
 
 
2709	return 0;
2710out:
2711	pagevec_release(&pvec);
 
 
2712	return err;
2713}
2714
2715static int ext4_writepage_cb(struct page *page, struct writeback_control *wbc,
2716			     void *data)
2717{
2718	return ext4_writepage(page, wbc);
2719}
2720
2721static int ext4_do_writepages(struct mpage_da_data *mpd)
2722{
2723	struct writeback_control *wbc = mpd->wbc;
2724	pgoff_t	writeback_index = 0;
2725	long nr_to_write = wbc->nr_to_write;
2726	int range_whole = 0;
2727	int cycled = 1;
2728	handle_t *handle = NULL;
2729	struct inode *inode = mpd->inode;
2730	struct address_space *mapping = inode->i_mapping;
2731	int needed_blocks, rsv_blocks = 0, ret = 0;
2732	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2733	struct blk_plug plug;
2734	bool give_up_on_write = false;
2735
2736	trace_ext4_writepages(inode, wbc);
2737
2738	/*
2739	 * No pages to write? This is mainly a kludge to avoid starting
2740	 * a transaction for special inodes like journal inode on last iput()
2741	 * because that could violate lock ordering on umount
2742	 */
2743	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2744		goto out_writepages;
2745
2746	if (ext4_should_journal_data(inode)) {
2747		blk_start_plug(&plug);
2748		ret = write_cache_pages(mapping, wbc, ext4_writepage_cb, NULL);
2749		blk_finish_plug(&plug);
2750		goto out_writepages;
2751	}
2752
2753	/*
2754	 * If the filesystem has aborted, it is read-only, so return
2755	 * right away instead of dumping stack traces later on that
2756	 * will obscure the real source of the problem.  We test
2757	 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2758	 * the latter could be true if the filesystem is mounted
2759	 * read-only, and in that case, ext4_writepages should
2760	 * *never* be called, so if that ever happens, we would want
2761	 * the stack trace.
2762	 */
2763	if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2764		     ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2765		ret = -EROFS;
2766		goto out_writepages;
2767	}
2768
2769	/*
2770	 * If we have inline data and arrive here, it means that
2771	 * we will soon create the block for the 1st page, so
2772	 * we'd better clear the inline data here.
2773	 */
2774	if (ext4_has_inline_data(inode)) {
2775		/* Just inode will be modified... */
2776		handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2777		if (IS_ERR(handle)) {
2778			ret = PTR_ERR(handle);
2779			goto out_writepages;
2780		}
2781		BUG_ON(ext4_test_inode_state(inode,
2782				EXT4_STATE_MAY_INLINE_DATA));
2783		ext4_destroy_inline_data(handle, inode);
2784		ext4_journal_stop(handle);
2785	}
2786
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2787	if (ext4_should_dioread_nolock(inode)) {
2788		/*
2789		 * We may need to convert up to one extent per block in
2790		 * the page and we may dirty the inode.
2791		 */
2792		rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2793						PAGE_SIZE >> inode->i_blkbits);
2794	}
2795
2796	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2797		range_whole = 1;
2798
2799	if (wbc->range_cyclic) {
2800		writeback_index = mapping->writeback_index;
2801		if (writeback_index)
2802			cycled = 0;
2803		mpd->first_page = writeback_index;
2804		mpd->last_page = -1;
2805	} else {
2806		mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2807		mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2808	}
2809
2810	ext4_io_submit_init(&mpd->io_submit, wbc);
2811retry:
2812	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2813		tag_pages_for_writeback(mapping, mpd->first_page,
2814					mpd->last_page);
2815	blk_start_plug(&plug);
2816
2817	/*
2818	 * First writeback pages that don't need mapping - we can avoid
2819	 * starting a transaction unnecessarily and also avoid being blocked
2820	 * in the block layer on device congestion while having transaction
2821	 * started.
2822	 */
2823	mpd->do_map = 0;
2824	mpd->scanned_until_end = 0;
2825	mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2826	if (!mpd->io_submit.io_end) {
2827		ret = -ENOMEM;
2828		goto unplug;
2829	}
2830	ret = mpage_prepare_extent_to_map(mpd);
2831	/* Unlock pages we didn't use */
2832	mpage_release_unused_pages(mpd, false);
2833	/* Submit prepared bio */
2834	ext4_io_submit(&mpd->io_submit);
2835	ext4_put_io_end_defer(mpd->io_submit.io_end);
2836	mpd->io_submit.io_end = NULL;
2837	if (ret < 0)
2838		goto unplug;
2839
2840	while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2841		/* For each extent of pages we use new io_end */
2842		mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2843		if (!mpd->io_submit.io_end) {
2844			ret = -ENOMEM;
2845			break;
2846		}
2847
2848		WARN_ON_ONCE(!mpd->can_map);
2849		/*
2850		 * We have two constraints: We find one extent to map and we
2851		 * must always write out whole page (makes a difference when
2852		 * blocksize < pagesize) so that we don't block on IO when we
2853		 * try to write out the rest of the page. Journalled mode is
2854		 * not supported by delalloc.
2855		 */
2856		BUG_ON(ext4_should_journal_data(inode));
2857		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2858
2859		/* start a new transaction */
2860		handle = ext4_journal_start_with_reserve(inode,
2861				EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2862		if (IS_ERR(handle)) {
2863			ret = PTR_ERR(handle);
2864			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2865			       "%ld pages, ino %lu; err %d", __func__,
2866				wbc->nr_to_write, inode->i_ino, ret);
2867			/* Release allocated io_end */
2868			ext4_put_io_end(mpd->io_submit.io_end);
2869			mpd->io_submit.io_end = NULL;
2870			break;
2871		}
2872		mpd->do_map = 1;
2873
2874		trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2875		ret = mpage_prepare_extent_to_map(mpd);
2876		if (!ret && mpd->map.m_len)
2877			ret = mpage_map_and_submit_extent(handle, mpd,
2878					&give_up_on_write);
2879		/*
2880		 * Caution: If the handle is synchronous,
2881		 * ext4_journal_stop() can wait for transaction commit
2882		 * to finish which may depend on writeback of pages to
2883		 * complete or on page lock to be released.  In that
2884		 * case, we have to wait until after we have
2885		 * submitted all the IO, released page locks we hold,
2886		 * and dropped io_end reference (for extent conversion
2887		 * to be able to complete) before stopping the handle.
2888		 */
2889		if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2890			ext4_journal_stop(handle);
2891			handle = NULL;
2892			mpd->do_map = 0;
2893		}
2894		/* Unlock pages we didn't use */
2895		mpage_release_unused_pages(mpd, give_up_on_write);
2896		/* Submit prepared bio */
2897		ext4_io_submit(&mpd->io_submit);
2898
2899		/*
2900		 * Drop our io_end reference we got from init. We have
2901		 * to be careful and use deferred io_end finishing if
2902		 * we are still holding the transaction as we can
2903		 * release the last reference to io_end which may end
2904		 * up doing unwritten extent conversion.
2905		 */
2906		if (handle) {
2907			ext4_put_io_end_defer(mpd->io_submit.io_end);
2908			ext4_journal_stop(handle);
2909		} else
2910			ext4_put_io_end(mpd->io_submit.io_end);
2911		mpd->io_submit.io_end = NULL;
2912
2913		if (ret == -ENOSPC && sbi->s_journal) {
2914			/*
2915			 * Commit the transaction which would
2916			 * free blocks released in the transaction
2917			 * and try again
2918			 */
2919			jbd2_journal_force_commit_nested(sbi->s_journal);
2920			ret = 0;
2921			continue;
2922		}
2923		/* Fatal error - ENOMEM, EIO... */
2924		if (ret)
2925			break;
2926	}
2927unplug:
2928	blk_finish_plug(&plug);
2929	if (!ret && !cycled && wbc->nr_to_write > 0) {
2930		cycled = 1;
2931		mpd->last_page = writeback_index - 1;
2932		mpd->first_page = 0;
2933		goto retry;
2934	}
2935
2936	/* Update index */
2937	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2938		/*
2939		 * Set the writeback_index so that range_cyclic
2940		 * mode will write it back later
2941		 */
2942		mapping->writeback_index = mpd->first_page;
2943
2944out_writepages:
2945	trace_ext4_writepages_result(inode, wbc, ret,
2946				     nr_to_write - wbc->nr_to_write);
2947	return ret;
2948}
2949
2950static int ext4_writepages(struct address_space *mapping,
2951			   struct writeback_control *wbc)
2952{
2953	struct super_block *sb = mapping->host->i_sb;
2954	struct mpage_da_data mpd = {
2955		.inode = mapping->host,
2956		.wbc = wbc,
2957		.can_map = 1,
2958	};
2959	int ret;
 
2960
2961	if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
2962		return -EIO;
2963
2964	percpu_down_read(&EXT4_SB(sb)->s_writepages_rwsem);
2965	ret = ext4_do_writepages(&mpd);
2966	percpu_up_read(&EXT4_SB(sb)->s_writepages_rwsem);
 
 
 
 
 
 
 
2967
2968	return ret;
2969}
2970
2971int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2972{
2973	struct writeback_control wbc = {
2974		.sync_mode = WB_SYNC_ALL,
2975		.nr_to_write = LONG_MAX,
2976		.range_start = jinode->i_dirty_start,
2977		.range_end = jinode->i_dirty_end,
2978	};
2979	struct mpage_da_data mpd = {
2980		.inode = jinode->i_vfs_inode,
2981		.wbc = &wbc,
2982		.can_map = 0,
2983	};
2984	return ext4_do_writepages(&mpd);
2985}
2986
2987static int ext4_dax_writepages(struct address_space *mapping,
2988			       struct writeback_control *wbc)
2989{
2990	int ret;
2991	long nr_to_write = wbc->nr_to_write;
2992	struct inode *inode = mapping->host;
2993	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2994
2995	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2996		return -EIO;
2997
2998	percpu_down_read(&sbi->s_writepages_rwsem);
2999	trace_ext4_writepages(inode, wbc);
3000
3001	ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
 
3002	trace_ext4_writepages_result(inode, wbc, ret,
3003				     nr_to_write - wbc->nr_to_write);
3004	percpu_up_read(&sbi->s_writepages_rwsem);
3005	return ret;
3006}
3007
3008static int ext4_nonda_switch(struct super_block *sb)
3009{
3010	s64 free_clusters, dirty_clusters;
3011	struct ext4_sb_info *sbi = EXT4_SB(sb);
3012
3013	/*
3014	 * switch to non delalloc mode if we are running low
3015	 * on free block. The free block accounting via percpu
3016	 * counters can get slightly wrong with percpu_counter_batch getting
3017	 * accumulated on each CPU without updating global counters
3018	 * Delalloc need an accurate free block accounting. So switch
3019	 * to non delalloc when we are near to error range.
3020	 */
3021	free_clusters =
3022		percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3023	dirty_clusters =
3024		percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3025	/*
3026	 * Start pushing delalloc when 1/2 of free blocks are dirty.
3027	 */
3028	if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3029		try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3030
3031	if (2 * free_clusters < 3 * dirty_clusters ||
3032	    free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3033		/*
3034		 * free block count is less than 150% of dirty blocks
3035		 * or free blocks is less than watermark
3036		 */
3037		return 1;
3038	}
3039	return 0;
3040}
3041
3042static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3043			       loff_t pos, unsigned len,
3044			       struct page **pagep, void **fsdata)
3045{
3046	int ret, retries = 0;
3047	struct page *page;
3048	pgoff_t index;
3049	struct inode *inode = mapping->host;
3050
3051	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3052		return -EIO;
3053
3054	index = pos >> PAGE_SHIFT;
3055
3056	if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
3057		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3058		return ext4_write_begin(file, mapping, pos,
3059					len, pagep, fsdata);
3060	}
3061	*fsdata = (void *)0;
3062	trace_ext4_da_write_begin(inode, pos, len);
3063
3064	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3065		ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
3066						      pagep, fsdata);
3067		if (ret < 0)
3068			return ret;
3069		if (ret == 1)
3070			return 0;
3071	}
3072
3073retry:
3074	page = grab_cache_page_write_begin(mapping, index);
3075	if (!page)
3076		return -ENOMEM;
3077
3078	/* In case writeback began while the page was unlocked */
3079	wait_for_stable_page(page);
3080
3081#ifdef CONFIG_FS_ENCRYPTION
3082	ret = ext4_block_write_begin(page, pos, len,
3083				     ext4_da_get_block_prep);
3084#else
3085	ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3086#endif
3087	if (ret < 0) {
3088		unlock_page(page);
3089		put_page(page);
3090		/*
3091		 * block_write_begin may have instantiated a few blocks
3092		 * outside i_size.  Trim these off again. Don't need
3093		 * i_size_read because we hold inode lock.
3094		 */
3095		if (pos + len > inode->i_size)
3096			ext4_truncate_failed_write(inode);
3097
3098		if (ret == -ENOSPC &&
3099		    ext4_should_retry_alloc(inode->i_sb, &retries))
3100			goto retry;
3101		return ret;
3102	}
3103
3104	*pagep = page;
3105	return ret;
3106}
3107
3108/*
3109 * Check if we should update i_disksize
3110 * when write to the end of file but not require block allocation
3111 */
3112static int ext4_da_should_update_i_disksize(struct page *page,
3113					    unsigned long offset)
3114{
3115	struct buffer_head *bh;
3116	struct inode *inode = page->mapping->host;
3117	unsigned int idx;
3118	int i;
3119
3120	bh = page_buffers(page);
3121	idx = offset >> inode->i_blkbits;
3122
3123	for (i = 0; i < idx; i++)
3124		bh = bh->b_this_page;
3125
3126	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3127		return 0;
3128	return 1;
3129}
3130
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3131static int ext4_da_write_end(struct file *file,
3132			     struct address_space *mapping,
3133			     loff_t pos, unsigned len, unsigned copied,
3134			     struct page *page, void *fsdata)
3135{
3136	struct inode *inode = mapping->host;
3137	loff_t new_i_size;
3138	unsigned long start, end;
3139	int write_mode = (int)(unsigned long)fsdata;
3140
3141	if (write_mode == FALL_BACK_TO_NONDELALLOC)
3142		return ext4_write_end(file, mapping, pos,
3143				      len, copied, page, fsdata);
3144
3145	trace_ext4_da_write_end(inode, pos, len, copied);
3146
3147	if (write_mode != CONVERT_INLINE_DATA &&
3148	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3149	    ext4_has_inline_data(inode))
3150		return ext4_write_inline_data_end(inode, pos, len, copied, page);
 
3151
3152	start = pos & (PAGE_SIZE - 1);
3153	end = start + copied - 1;
3154
3155	/*
3156	 * Since we are holding inode lock, we are sure i_disksize <=
3157	 * i_size. We also know that if i_disksize < i_size, there are
3158	 * delalloc writes pending in the range upto i_size. If the end of
3159	 * the current write is <= i_size, there's no need to touch
3160	 * i_disksize since writeback will push i_disksize upto i_size
3161	 * eventually. If the end of the current write is > i_size and
3162	 * inside an allocated block (ext4_da_should_update_i_disksize()
3163	 * check), we need to update i_disksize here as neither
3164	 * ext4_writepage() nor certain ext4_writepages() paths not
3165	 * allocating blocks update i_disksize.
3166	 *
3167	 * Note that we defer inode dirtying to generic_write_end() /
3168	 * ext4_da_write_inline_data_end().
3169	 */
3170	new_i_size = pos + copied;
3171	if (copied && new_i_size > inode->i_size &&
3172	    ext4_da_should_update_i_disksize(page, end))
3173		ext4_update_i_disksize(inode, new_i_size);
3174
3175	return generic_write_end(file, mapping, pos, len, copied, page, fsdata);
3176}
3177
3178/*
3179 * Force all delayed allocation blocks to be allocated for a given inode.
3180 */
3181int ext4_alloc_da_blocks(struct inode *inode)
3182{
3183	trace_ext4_alloc_da_blocks(inode);
3184
3185	if (!EXT4_I(inode)->i_reserved_data_blocks)
3186		return 0;
3187
3188	/*
3189	 * We do something simple for now.  The filemap_flush() will
3190	 * also start triggering a write of the data blocks, which is
3191	 * not strictly speaking necessary (and for users of
3192	 * laptop_mode, not even desirable).  However, to do otherwise
3193	 * would require replicating code paths in:
3194	 *
3195	 * ext4_writepages() ->
3196	 *    write_cache_pages() ---> (via passed in callback function)
3197	 *        __mpage_da_writepage() -->
3198	 *           mpage_add_bh_to_extent()
3199	 *           mpage_da_map_blocks()
3200	 *
3201	 * The problem is that write_cache_pages(), located in
3202	 * mm/page-writeback.c, marks pages clean in preparation for
3203	 * doing I/O, which is not desirable if we're not planning on
3204	 * doing I/O at all.
3205	 *
3206	 * We could call write_cache_pages(), and then redirty all of
3207	 * the pages by calling redirty_page_for_writepage() but that
3208	 * would be ugly in the extreme.  So instead we would need to
3209	 * replicate parts of the code in the above functions,
3210	 * simplifying them because we wouldn't actually intend to
3211	 * write out the pages, but rather only collect contiguous
3212	 * logical block extents, call the multi-block allocator, and
3213	 * then update the buffer heads with the block allocations.
3214	 *
3215	 * For now, though, we'll cheat by calling filemap_flush(),
3216	 * which will map the blocks, and start the I/O, but not
3217	 * actually wait for the I/O to complete.
3218	 */
3219	return filemap_flush(inode->i_mapping);
3220}
3221
3222/*
3223 * bmap() is special.  It gets used by applications such as lilo and by
3224 * the swapper to find the on-disk block of a specific piece of data.
3225 *
3226 * Naturally, this is dangerous if the block concerned is still in the
3227 * journal.  If somebody makes a swapfile on an ext4 data-journaling
3228 * filesystem and enables swap, then they may get a nasty shock when the
3229 * data getting swapped to that swapfile suddenly gets overwritten by
3230 * the original zero's written out previously to the journal and
3231 * awaiting writeback in the kernel's buffer cache.
3232 *
3233 * So, if we see any bmap calls here on a modified, data-journaled file,
3234 * take extra steps to flush any blocks which might be in the cache.
3235 */
3236static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3237{
3238	struct inode *inode = mapping->host;
3239	journal_t *journal;
3240	sector_t ret = 0;
3241	int err;
3242
3243	inode_lock_shared(inode);
3244	/*
3245	 * We can get here for an inline file via the FIBMAP ioctl
3246	 */
3247	if (ext4_has_inline_data(inode))
3248		goto out;
3249
3250	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3251			test_opt(inode->i_sb, DELALLOC)) {
 
3252		/*
3253		 * With delalloc we want to sync the file
3254		 * so that we can make sure we allocate
3255		 * blocks for file
3256		 */
3257		filemap_write_and_wait(mapping);
3258	}
3259
3260	if (EXT4_JOURNAL(inode) &&
3261	    ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3262		/*
3263		 * This is a REALLY heavyweight approach, but the use of
3264		 * bmap on dirty files is expected to be extremely rare:
3265		 * only if we run lilo or swapon on a freshly made file
3266		 * do we expect this to happen.
3267		 *
3268		 * (bmap requires CAP_SYS_RAWIO so this does not
3269		 * represent an unprivileged user DOS attack --- we'd be
3270		 * in trouble if mortal users could trigger this path at
3271		 * will.)
3272		 *
3273		 * NB. EXT4_STATE_JDATA is not set on files other than
3274		 * regular files.  If somebody wants to bmap a directory
3275		 * or symlink and gets confused because the buffer
3276		 * hasn't yet been flushed to disk, they deserve
3277		 * everything they get.
3278		 */
3279
3280		ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3281		journal = EXT4_JOURNAL(inode);
3282		jbd2_journal_lock_updates(journal);
3283		err = jbd2_journal_flush(journal, 0);
3284		jbd2_journal_unlock_updates(journal);
3285
3286		if (err)
3287			goto out;
3288	}
3289
3290	ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3291
3292out:
3293	inode_unlock_shared(inode);
3294	return ret;
3295}
3296
3297static int ext4_read_folio(struct file *file, struct folio *folio)
3298{
3299	struct page *page = &folio->page;
3300	int ret = -EAGAIN;
3301	struct inode *inode = page->mapping->host;
3302
3303	trace_ext4_readpage(page);
3304
3305	if (ext4_has_inline_data(inode))
3306		ret = ext4_readpage_inline(inode, page);
3307
3308	if (ret == -EAGAIN)
3309		return ext4_mpage_readpages(inode, NULL, page);
3310
3311	return ret;
3312}
3313
3314static void ext4_readahead(struct readahead_control *rac)
3315{
3316	struct inode *inode = rac->mapping->host;
3317
3318	/* If the file has inline data, no need to do readahead. */
3319	if (ext4_has_inline_data(inode))
3320		return;
3321
3322	ext4_mpage_readpages(inode, rac, NULL);
3323}
3324
3325static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3326				size_t length)
3327{
3328	trace_ext4_invalidate_folio(folio, offset, length);
3329
3330	/* No journalling happens on data buffers when this function is used */
3331	WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3332
3333	block_invalidate_folio(folio, offset, length);
3334}
3335
3336static int __ext4_journalled_invalidate_folio(struct folio *folio,
3337					    size_t offset, size_t length)
3338{
3339	journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3340
3341	trace_ext4_journalled_invalidate_folio(folio, offset, length);
3342
3343	/*
3344	 * If it's a full truncate we just forget about the pending dirtying
3345	 */
3346	if (offset == 0 && length == folio_size(folio))
3347		folio_clear_checked(folio);
3348
3349	return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3350}
3351
3352/* Wrapper for aops... */
3353static void ext4_journalled_invalidate_folio(struct folio *folio,
3354					   size_t offset,
3355					   size_t length)
3356{
3357	WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3358}
3359
3360static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3361{
3362	journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
 
3363
3364	trace_ext4_releasepage(&folio->page);
3365
3366	/* Page has dirty journalled data -> cannot release */
3367	if (folio_test_checked(folio))
3368		return false;
3369	if (journal)
3370		return jbd2_journal_try_to_free_buffers(journal, folio);
3371	else
3372		return try_to_free_buffers(folio);
3373}
3374
3375static bool ext4_inode_datasync_dirty(struct inode *inode)
3376{
3377	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3378
3379	if (journal) {
3380		if (jbd2_transaction_committed(journal,
3381			EXT4_I(inode)->i_datasync_tid))
3382			return false;
3383		if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3384			return !list_empty(&EXT4_I(inode)->i_fc_list);
3385		return true;
3386	}
3387
3388	/* Any metadata buffers to write? */
3389	if (!list_empty(&inode->i_mapping->private_list))
3390		return true;
3391	return inode->i_state & I_DIRTY_DATASYNC;
3392}
3393
3394static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3395			   struct ext4_map_blocks *map, loff_t offset,
3396			   loff_t length, unsigned int flags)
3397{
3398	u8 blkbits = inode->i_blkbits;
3399
3400	/*
3401	 * Writes that span EOF might trigger an I/O size update on completion,
3402	 * so consider them to be dirty for the purpose of O_DSYNC, even if
3403	 * there is no other metadata changes being made or are pending.
3404	 */
3405	iomap->flags = 0;
3406	if (ext4_inode_datasync_dirty(inode) ||
3407	    offset + length > i_size_read(inode))
3408		iomap->flags |= IOMAP_F_DIRTY;
3409
3410	if (map->m_flags & EXT4_MAP_NEW)
3411		iomap->flags |= IOMAP_F_NEW;
3412
3413	if (flags & IOMAP_DAX)
3414		iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3415	else
3416		iomap->bdev = inode->i_sb->s_bdev;
3417	iomap->offset = (u64) map->m_lblk << blkbits;
3418	iomap->length = (u64) map->m_len << blkbits;
3419
3420	if ((map->m_flags & EXT4_MAP_MAPPED) &&
3421	    !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3422		iomap->flags |= IOMAP_F_MERGED;
3423
3424	/*
3425	 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3426	 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3427	 * set. In order for any allocated unwritten extents to be converted
3428	 * into written extents correctly within the ->end_io() handler, we
3429	 * need to ensure that the iomap->type is set appropriately. Hence, the
3430	 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3431	 * been set first.
3432	 */
3433	if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3434		iomap->type = IOMAP_UNWRITTEN;
3435		iomap->addr = (u64) map->m_pblk << blkbits;
3436		if (flags & IOMAP_DAX)
3437			iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3438	} else if (map->m_flags & EXT4_MAP_MAPPED) {
3439		iomap->type = IOMAP_MAPPED;
3440		iomap->addr = (u64) map->m_pblk << blkbits;
3441		if (flags & IOMAP_DAX)
3442			iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
 
 
 
3443	} else {
3444		iomap->type = IOMAP_HOLE;
3445		iomap->addr = IOMAP_NULL_ADDR;
3446	}
3447}
3448
3449static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3450			    unsigned int flags)
3451{
3452	handle_t *handle;
3453	u8 blkbits = inode->i_blkbits;
3454	int ret, dio_credits, m_flags = 0, retries = 0;
3455
3456	/*
3457	 * Trim the mapping request to the maximum value that we can map at
3458	 * once for direct I/O.
3459	 */
3460	if (map->m_len > DIO_MAX_BLOCKS)
3461		map->m_len = DIO_MAX_BLOCKS;
3462	dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3463
3464retry:
3465	/*
3466	 * Either we allocate blocks and then don't get an unwritten extent, so
3467	 * in that case we have reserved enough credits. Or, the blocks are
3468	 * already allocated and unwritten. In that case, the extent conversion
3469	 * fits into the credits as well.
3470	 */
3471	handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3472	if (IS_ERR(handle))
3473		return PTR_ERR(handle);
3474
3475	/*
3476	 * DAX and direct I/O are the only two operations that are currently
3477	 * supported with IOMAP_WRITE.
3478	 */
3479	WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3480	if (flags & IOMAP_DAX)
3481		m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3482	/*
3483	 * We use i_size instead of i_disksize here because delalloc writeback
3484	 * can complete at any point during the I/O and subsequently push the
3485	 * i_disksize out to i_size. This could be beyond where direct I/O is
3486	 * happening and thus expose allocated blocks to direct I/O reads.
3487	 */
3488	else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3489		m_flags = EXT4_GET_BLOCKS_CREATE;
3490	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3491		m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3492
3493	ret = ext4_map_blocks(handle, inode, map, m_flags);
3494
3495	/*
3496	 * We cannot fill holes in indirect tree based inodes as that could
3497	 * expose stale data in the case of a crash. Use the magic error code
3498	 * to fallback to buffered I/O.
3499	 */
3500	if (!m_flags && !ret)
3501		ret = -ENOTBLK;
3502
3503	ext4_journal_stop(handle);
3504	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3505		goto retry;
3506
3507	return ret;
3508}
3509
3510
3511static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3512		unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3513{
3514	int ret;
3515	struct ext4_map_blocks map;
3516	u8 blkbits = inode->i_blkbits;
3517
3518	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3519		return -EINVAL;
3520
3521	if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3522		return -ERANGE;
3523
3524	/*
3525	 * Calculate the first and last logical blocks respectively.
3526	 */
3527	map.m_lblk = offset >> blkbits;
3528	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3529			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3530
3531	if (flags & IOMAP_WRITE) {
3532		/*
3533		 * We check here if the blocks are already allocated, then we
3534		 * don't need to start a journal txn and we can directly return
3535		 * the mapping information. This could boost performance
3536		 * especially in multi-threaded overwrite requests.
3537		 */
3538		if (offset + length <= i_size_read(inode)) {
3539			ret = ext4_map_blocks(NULL, inode, &map, 0);
3540			if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3541				goto out;
3542		}
3543		ret = ext4_iomap_alloc(inode, &map, flags);
3544	} else {
3545		ret = ext4_map_blocks(NULL, inode, &map, 0);
3546	}
3547
3548	if (ret < 0)
3549		return ret;
3550out:
3551	/*
3552	 * When inline encryption is enabled, sometimes I/O to an encrypted file
3553	 * has to be broken up to guarantee DUN contiguity.  Handle this by
3554	 * limiting the length of the mapping returned.
3555	 */
3556	map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3557
3558	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3559
3560	return 0;
3561}
3562
3563static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3564		loff_t length, unsigned flags, struct iomap *iomap,
3565		struct iomap *srcmap)
3566{
3567	int ret;
3568
3569	/*
3570	 * Even for writes we don't need to allocate blocks, so just pretend
3571	 * we are reading to save overhead of starting a transaction.
3572	 */
3573	flags &= ~IOMAP_WRITE;
3574	ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3575	WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3576	return ret;
3577}
3578
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3579static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3580			  ssize_t written, unsigned flags, struct iomap *iomap)
3581{
3582	/*
3583	 * Check to see whether an error occurred while writing out the data to
3584	 * the allocated blocks. If so, return the magic error code so that we
3585	 * fallback to buffered I/O and attempt to complete the remainder of
3586	 * the I/O. Any blocks that may have been allocated in preparation for
3587	 * the direct I/O will be reused during buffered I/O.
 
 
3588	 */
3589	if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3590		return -ENOTBLK;
3591
3592	return 0;
3593}
3594
3595const struct iomap_ops ext4_iomap_ops = {
3596	.iomap_begin		= ext4_iomap_begin,
3597	.iomap_end		= ext4_iomap_end,
3598};
3599
3600const struct iomap_ops ext4_iomap_overwrite_ops = {
3601	.iomap_begin		= ext4_iomap_overwrite_begin,
3602	.iomap_end		= ext4_iomap_end,
3603};
3604
3605static bool ext4_iomap_is_delalloc(struct inode *inode,
3606				   struct ext4_map_blocks *map)
3607{
3608	struct extent_status es;
3609	ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3610
3611	ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3612				  map->m_lblk, end, &es);
3613
3614	if (!es.es_len || es.es_lblk > end)
3615		return false;
3616
3617	if (es.es_lblk > map->m_lblk) {
3618		map->m_len = es.es_lblk - map->m_lblk;
3619		return false;
3620	}
3621
3622	offset = map->m_lblk - es.es_lblk;
3623	map->m_len = es.es_len - offset;
3624
3625	return true;
3626}
3627
3628static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3629				   loff_t length, unsigned int flags,
3630				   struct iomap *iomap, struct iomap *srcmap)
3631{
3632	int ret;
3633	bool delalloc = false;
3634	struct ext4_map_blocks map;
3635	u8 blkbits = inode->i_blkbits;
3636
3637	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3638		return -EINVAL;
3639
3640	if (ext4_has_inline_data(inode)) {
3641		ret = ext4_inline_data_iomap(inode, iomap);
3642		if (ret != -EAGAIN) {
3643			if (ret == 0 && offset >= iomap->length)
3644				ret = -ENOENT;
3645			return ret;
3646		}
3647	}
3648
3649	/*
3650	 * Calculate the first and last logical block respectively.
3651	 */
3652	map.m_lblk = offset >> blkbits;
3653	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3654			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3655
3656	/*
3657	 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3658	 * So handle it here itself instead of querying ext4_map_blocks().
3659	 * Since ext4_map_blocks() will warn about it and will return
3660	 * -EIO error.
3661	 */
3662	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3663		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3664
3665		if (offset >= sbi->s_bitmap_maxbytes) {
3666			map.m_flags = 0;
3667			goto set_iomap;
3668		}
3669	}
3670
3671	ret = ext4_map_blocks(NULL, inode, &map, 0);
3672	if (ret < 0)
3673		return ret;
3674	if (ret == 0)
3675		delalloc = ext4_iomap_is_delalloc(inode, &map);
3676
3677set_iomap:
3678	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3679	if (delalloc && iomap->type == IOMAP_HOLE)
3680		iomap->type = IOMAP_DELALLOC;
3681
3682	return 0;
3683}
3684
3685const struct iomap_ops ext4_iomap_report_ops = {
3686	.iomap_begin = ext4_iomap_begin_report,
3687};
3688
3689/*
3690 * Whenever the folio is being dirtied, corresponding buffers should already
3691 * be attached to the transaction (we take care of this in ext4_page_mkwrite()
3692 * and ext4_write_begin()). However we cannot move buffers to dirty transaction
3693 * lists here because ->dirty_folio is called under VFS locks and the folio
3694 * is not necessarily locked.
3695 *
3696 * We cannot just dirty the folio and leave attached buffers clean, because the
3697 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3698 * or jbddirty because all the journalling code will explode.
3699 *
3700 * So what we do is to mark the folio "pending dirty" and next time writepage
3701 * is called, propagate that into the buffers appropriately.
 
3702 */
3703static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3704		struct folio *folio)
3705{
3706	WARN_ON_ONCE(!folio_buffers(folio));
3707	folio_set_checked(folio);
 
3708	return filemap_dirty_folio(mapping, folio);
3709}
3710
3711static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3712{
3713	WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3714	WARN_ON_ONCE(!folio_buffers(folio));
3715	return block_dirty_folio(mapping, folio);
3716}
3717
3718static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3719				    struct file *file, sector_t *span)
3720{
3721	return iomap_swapfile_activate(sis, file, span,
3722				       &ext4_iomap_report_ops);
3723}
3724
3725static const struct address_space_operations ext4_aops = {
3726	.read_folio		= ext4_read_folio,
3727	.readahead		= ext4_readahead,
3728	.writepages		= ext4_writepages,
3729	.write_begin		= ext4_write_begin,
3730	.write_end		= ext4_write_end,
3731	.dirty_folio		= ext4_dirty_folio,
3732	.bmap			= ext4_bmap,
3733	.invalidate_folio	= ext4_invalidate_folio,
3734	.release_folio		= ext4_release_folio,
3735	.direct_IO		= noop_direct_IO,
3736	.migrate_folio		= buffer_migrate_folio,
3737	.is_partially_uptodate  = block_is_partially_uptodate,
3738	.error_remove_page	= generic_error_remove_page,
3739	.swap_activate		= ext4_iomap_swap_activate,
3740};
3741
3742static const struct address_space_operations ext4_journalled_aops = {
3743	.read_folio		= ext4_read_folio,
3744	.readahead		= ext4_readahead,
3745	.writepages		= ext4_writepages,
3746	.write_begin		= ext4_write_begin,
3747	.write_end		= ext4_journalled_write_end,
3748	.dirty_folio		= ext4_journalled_dirty_folio,
3749	.bmap			= ext4_bmap,
3750	.invalidate_folio	= ext4_journalled_invalidate_folio,
3751	.release_folio		= ext4_release_folio,
3752	.direct_IO		= noop_direct_IO,
3753	.migrate_folio		= buffer_migrate_folio_norefs,
3754	.is_partially_uptodate  = block_is_partially_uptodate,
3755	.error_remove_page	= generic_error_remove_page,
3756	.swap_activate		= ext4_iomap_swap_activate,
3757};
3758
3759static const struct address_space_operations ext4_da_aops = {
3760	.read_folio		= ext4_read_folio,
3761	.readahead		= ext4_readahead,
3762	.writepages		= ext4_writepages,
3763	.write_begin		= ext4_da_write_begin,
3764	.write_end		= ext4_da_write_end,
3765	.dirty_folio		= ext4_dirty_folio,
3766	.bmap			= ext4_bmap,
3767	.invalidate_folio	= ext4_invalidate_folio,
3768	.release_folio		= ext4_release_folio,
3769	.direct_IO		= noop_direct_IO,
3770	.migrate_folio		= buffer_migrate_folio,
3771	.is_partially_uptodate  = block_is_partially_uptodate,
3772	.error_remove_page	= generic_error_remove_page,
3773	.swap_activate		= ext4_iomap_swap_activate,
3774};
3775
3776static const struct address_space_operations ext4_dax_aops = {
3777	.writepages		= ext4_dax_writepages,
3778	.direct_IO		= noop_direct_IO,
3779	.dirty_folio		= noop_dirty_folio,
3780	.bmap			= ext4_bmap,
3781	.swap_activate		= ext4_iomap_swap_activate,
3782};
3783
3784void ext4_set_aops(struct inode *inode)
3785{
3786	switch (ext4_inode_journal_mode(inode)) {
3787	case EXT4_INODE_ORDERED_DATA_MODE:
3788	case EXT4_INODE_WRITEBACK_DATA_MODE:
3789		break;
3790	case EXT4_INODE_JOURNAL_DATA_MODE:
3791		inode->i_mapping->a_ops = &ext4_journalled_aops;
3792		return;
3793	default:
3794		BUG();
3795	}
3796	if (IS_DAX(inode))
3797		inode->i_mapping->a_ops = &ext4_dax_aops;
3798	else if (test_opt(inode->i_sb, DELALLOC))
3799		inode->i_mapping->a_ops = &ext4_da_aops;
3800	else
3801		inode->i_mapping->a_ops = &ext4_aops;
3802}
3803
 
 
 
 
 
 
3804static int __ext4_block_zero_page_range(handle_t *handle,
3805		struct address_space *mapping, loff_t from, loff_t length)
3806{
3807	ext4_fsblk_t index = from >> PAGE_SHIFT;
3808	unsigned offset = from & (PAGE_SIZE-1);
3809	unsigned blocksize, pos;
3810	ext4_lblk_t iblock;
3811	struct inode *inode = mapping->host;
3812	struct buffer_head *bh;
3813	struct page *page;
3814	int err = 0;
3815
3816	page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3817				   mapping_gfp_constraint(mapping, ~__GFP_FS));
3818	if (!page)
3819		return -ENOMEM;
 
3820
3821	blocksize = inode->i_sb->s_blocksize;
3822
3823	iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3824
3825	if (!page_has_buffers(page))
3826		create_empty_buffers(page, blocksize, 0);
 
3827
3828	/* Find the buffer that contains "offset" */
3829	bh = page_buffers(page);
3830	pos = blocksize;
3831	while (offset >= pos) {
3832		bh = bh->b_this_page;
3833		iblock++;
3834		pos += blocksize;
3835	}
3836	if (buffer_freed(bh)) {
3837		BUFFER_TRACE(bh, "freed: skip");
3838		goto unlock;
3839	}
3840	if (!buffer_mapped(bh)) {
3841		BUFFER_TRACE(bh, "unmapped");
3842		ext4_get_block(inode, iblock, bh, 0);
3843		/* unmapped? It's a hole - nothing to do */
3844		if (!buffer_mapped(bh)) {
3845			BUFFER_TRACE(bh, "still unmapped");
3846			goto unlock;
3847		}
3848	}
3849
3850	/* Ok, it's mapped. Make sure it's up-to-date */
3851	if (PageUptodate(page))
3852		set_buffer_uptodate(bh);
3853
3854	if (!buffer_uptodate(bh)) {
3855		err = ext4_read_bh_lock(bh, 0, true);
3856		if (err)
3857			goto unlock;
3858		if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3859			/* We expect the key to be set. */
3860			BUG_ON(!fscrypt_has_encryption_key(inode));
3861			err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
 
3862							       bh_offset(bh));
3863			if (err) {
3864				clear_buffer_uptodate(bh);
3865				goto unlock;
3866			}
3867		}
3868	}
3869	if (ext4_should_journal_data(inode)) {
3870		BUFFER_TRACE(bh, "get write access");
3871		err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3872						    EXT4_JTR_NONE);
3873		if (err)
3874			goto unlock;
3875	}
3876	zero_user(page, offset, length);
3877	BUFFER_TRACE(bh, "zeroed end of block");
3878
3879	if (ext4_should_journal_data(inode)) {
3880		err = ext4_handle_dirty_metadata(handle, inode, bh);
3881	} else {
3882		err = 0;
3883		mark_buffer_dirty(bh);
3884		if (ext4_should_order_data(inode))
3885			err = ext4_jbd2_inode_add_write(handle, inode, from,
3886					length);
3887	}
3888
3889unlock:
3890	unlock_page(page);
3891	put_page(page);
3892	return err;
3893}
3894
3895/*
3896 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3897 * starting from file offset 'from'.  The range to be zero'd must
3898 * be contained with in one block.  If the specified range exceeds
3899 * the end of the block it will be shortened to end of the block
3900 * that corresponds to 'from'
3901 */
3902static int ext4_block_zero_page_range(handle_t *handle,
3903		struct address_space *mapping, loff_t from, loff_t length)
3904{
3905	struct inode *inode = mapping->host;
3906	unsigned offset = from & (PAGE_SIZE-1);
3907	unsigned blocksize = inode->i_sb->s_blocksize;
3908	unsigned max = blocksize - (offset & (blocksize - 1));
3909
3910	/*
3911	 * correct length if it does not fall between
3912	 * 'from' and the end of the block
3913	 */
3914	if (length > max || length < 0)
3915		length = max;
3916
3917	if (IS_DAX(inode)) {
3918		return dax_zero_range(inode, from, length, NULL,
3919				      &ext4_iomap_ops);
3920	}
3921	return __ext4_block_zero_page_range(handle, mapping, from, length);
3922}
3923
3924/*
3925 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3926 * up to the end of the block which corresponds to `from'.
3927 * This required during truncate. We need to physically zero the tail end
3928 * of that block so it doesn't yield old data if the file is later grown.
3929 */
3930static int ext4_block_truncate_page(handle_t *handle,
3931		struct address_space *mapping, loff_t from)
3932{
3933	unsigned offset = from & (PAGE_SIZE-1);
3934	unsigned length;
3935	unsigned blocksize;
3936	struct inode *inode = mapping->host;
3937
3938	/* If we are processing an encrypted inode during orphan list handling */
3939	if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3940		return 0;
3941
3942	blocksize = inode->i_sb->s_blocksize;
3943	length = blocksize - (offset & (blocksize - 1));
3944
3945	return ext4_block_zero_page_range(handle, mapping, from, length);
3946}
3947
3948int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3949			     loff_t lstart, loff_t length)
3950{
3951	struct super_block *sb = inode->i_sb;
3952	struct address_space *mapping = inode->i_mapping;
3953	unsigned partial_start, partial_end;
3954	ext4_fsblk_t start, end;
3955	loff_t byte_end = (lstart + length - 1);
3956	int err = 0;
3957
3958	partial_start = lstart & (sb->s_blocksize - 1);
3959	partial_end = byte_end & (sb->s_blocksize - 1);
3960
3961	start = lstart >> sb->s_blocksize_bits;
3962	end = byte_end >> sb->s_blocksize_bits;
3963
3964	/* Handle partial zero within the single block */
3965	if (start == end &&
3966	    (partial_start || (partial_end != sb->s_blocksize - 1))) {
3967		err = ext4_block_zero_page_range(handle, mapping,
3968						 lstart, length);
3969		return err;
3970	}
3971	/* Handle partial zero out on the start of the range */
3972	if (partial_start) {
3973		err = ext4_block_zero_page_range(handle, mapping,
3974						 lstart, sb->s_blocksize);
3975		if (err)
3976			return err;
3977	}
3978	/* Handle partial zero out on the end of the range */
3979	if (partial_end != sb->s_blocksize - 1)
3980		err = ext4_block_zero_page_range(handle, mapping,
3981						 byte_end - partial_end,
3982						 partial_end + 1);
3983	return err;
3984}
3985
3986int ext4_can_truncate(struct inode *inode)
3987{
3988	if (S_ISREG(inode->i_mode))
3989		return 1;
3990	if (S_ISDIR(inode->i_mode))
3991		return 1;
3992	if (S_ISLNK(inode->i_mode))
3993		return !ext4_inode_is_fast_symlink(inode);
3994	return 0;
3995}
3996
3997/*
3998 * We have to make sure i_disksize gets properly updated before we truncate
3999 * page cache due to hole punching or zero range. Otherwise i_disksize update
4000 * can get lost as it may have been postponed to submission of writeback but
4001 * that will never happen after we truncate page cache.
4002 */
4003int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4004				      loff_t len)
4005{
4006	handle_t *handle;
4007	int ret;
4008
4009	loff_t size = i_size_read(inode);
4010
4011	WARN_ON(!inode_is_locked(inode));
4012	if (offset > size || offset + len < size)
4013		return 0;
4014
4015	if (EXT4_I(inode)->i_disksize >= size)
4016		return 0;
4017
4018	handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4019	if (IS_ERR(handle))
4020		return PTR_ERR(handle);
4021	ext4_update_i_disksize(inode, size);
4022	ret = ext4_mark_inode_dirty(handle, inode);
4023	ext4_journal_stop(handle);
4024
4025	return ret;
4026}
4027
4028static void ext4_wait_dax_page(struct inode *inode)
4029{
4030	filemap_invalidate_unlock(inode->i_mapping);
4031	schedule();
4032	filemap_invalidate_lock(inode->i_mapping);
4033}
4034
4035int ext4_break_layouts(struct inode *inode)
4036{
4037	struct page *page;
4038	int error;
4039
4040	if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
4041		return -EINVAL;
4042
4043	do {
4044		page = dax_layout_busy_page(inode->i_mapping);
4045		if (!page)
4046			return 0;
4047
4048		error = ___wait_var_event(&page->_refcount,
4049				atomic_read(&page->_refcount) == 1,
4050				TASK_INTERRUPTIBLE, 0, 0,
4051				ext4_wait_dax_page(inode));
4052	} while (error == 0);
4053
4054	return error;
4055}
4056
4057/*
4058 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4059 * associated with the given offset and length
4060 *
4061 * @inode:  File inode
4062 * @offset: The offset where the hole will begin
4063 * @len:    The length of the hole
4064 *
4065 * Returns: 0 on success or negative on failure
4066 */
4067
4068int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4069{
4070	struct inode *inode = file_inode(file);
4071	struct super_block *sb = inode->i_sb;
4072	ext4_lblk_t first_block, stop_block;
4073	struct address_space *mapping = inode->i_mapping;
4074	loff_t first_block_offset, last_block_offset, max_length;
4075	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4076	handle_t *handle;
4077	unsigned int credits;
4078	int ret = 0, ret2 = 0;
4079
4080	trace_ext4_punch_hole(inode, offset, length, 0);
4081
4082	/*
4083	 * Write out all dirty pages to avoid race conditions
4084	 * Then release them.
4085	 */
4086	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4087		ret = filemap_write_and_wait_range(mapping, offset,
4088						   offset + length - 1);
4089		if (ret)
4090			return ret;
4091	}
4092
4093	inode_lock(inode);
4094
4095	/* No need to punch hole beyond i_size */
4096	if (offset >= inode->i_size)
4097		goto out_mutex;
4098
4099	/*
4100	 * If the hole extends beyond i_size, set the hole
4101	 * to end after the page that contains i_size
4102	 */
4103	if (offset + length > inode->i_size) {
4104		length = inode->i_size +
4105		   PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4106		   offset;
4107	}
4108
4109	/*
4110	 * For punch hole the length + offset needs to be within one block
4111	 * before last range. Adjust the length if it goes beyond that limit.
4112	 */
4113	max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4114	if (offset + length > max_length)
4115		length = max_length - offset;
4116
4117	if (offset & (sb->s_blocksize - 1) ||
4118	    (offset + length) & (sb->s_blocksize - 1)) {
4119		/*
4120		 * Attach jinode to inode for jbd2 if we do any zeroing of
4121		 * partial block
4122		 */
4123		ret = ext4_inode_attach_jinode(inode);
4124		if (ret < 0)
4125			goto out_mutex;
4126
4127	}
4128
4129	/* Wait all existing dio workers, newcomers will block on i_rwsem */
4130	inode_dio_wait(inode);
4131
4132	ret = file_modified(file);
4133	if (ret)
4134		goto out_mutex;
4135
4136	/*
4137	 * Prevent page faults from reinstantiating pages we have released from
4138	 * page cache.
4139	 */
4140	filemap_invalidate_lock(mapping);
4141
4142	ret = ext4_break_layouts(inode);
4143	if (ret)
4144		goto out_dio;
4145
4146	first_block_offset = round_up(offset, sb->s_blocksize);
4147	last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4148
4149	/* Now release the pages and zero block aligned part of pages*/
4150	if (last_block_offset > first_block_offset) {
4151		ret = ext4_update_disksize_before_punch(inode, offset, length);
4152		if (ret)
4153			goto out_dio;
4154		truncate_pagecache_range(inode, first_block_offset,
4155					 last_block_offset);
4156	}
4157
4158	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4159		credits = ext4_writepage_trans_blocks(inode);
4160	else
4161		credits = ext4_blocks_for_truncate(inode);
4162	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4163	if (IS_ERR(handle)) {
4164		ret = PTR_ERR(handle);
4165		ext4_std_error(sb, ret);
4166		goto out_dio;
4167	}
4168
4169	ret = ext4_zero_partial_blocks(handle, inode, offset,
4170				       length);
4171	if (ret)
4172		goto out_stop;
4173
4174	first_block = (offset + sb->s_blocksize - 1) >>
4175		EXT4_BLOCK_SIZE_BITS(sb);
4176	stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4177
4178	/* If there are blocks to remove, do it */
4179	if (stop_block > first_block) {
 
4180
4181		down_write(&EXT4_I(inode)->i_data_sem);
4182		ext4_discard_preallocations(inode, 0);
4183
4184		ret = ext4_es_remove_extent(inode, first_block,
4185					    stop_block - first_block);
4186		if (ret) {
4187			up_write(&EXT4_I(inode)->i_data_sem);
4188			goto out_stop;
4189		}
4190
4191		if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4192			ret = ext4_ext_remove_space(inode, first_block,
4193						    stop_block - 1);
4194		else
4195			ret = ext4_ind_remove_space(handle, inode, first_block,
4196						    stop_block);
4197
 
 
4198		up_write(&EXT4_I(inode)->i_data_sem);
4199	}
4200	ext4_fc_track_range(handle, inode, first_block, stop_block);
4201	if (IS_SYNC(inode))
4202		ext4_handle_sync(handle);
4203
4204	inode->i_mtime = inode->i_ctime = current_time(inode);
4205	ret2 = ext4_mark_inode_dirty(handle, inode);
4206	if (unlikely(ret2))
4207		ret = ret2;
4208	if (ret >= 0)
4209		ext4_update_inode_fsync_trans(handle, inode, 1);
4210out_stop:
4211	ext4_journal_stop(handle);
4212out_dio:
4213	filemap_invalidate_unlock(mapping);
4214out_mutex:
4215	inode_unlock(inode);
4216	return ret;
4217}
4218
4219int ext4_inode_attach_jinode(struct inode *inode)
4220{
4221	struct ext4_inode_info *ei = EXT4_I(inode);
4222	struct jbd2_inode *jinode;
4223
4224	if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4225		return 0;
4226
4227	jinode = jbd2_alloc_inode(GFP_KERNEL);
4228	spin_lock(&inode->i_lock);
4229	if (!ei->jinode) {
4230		if (!jinode) {
4231			spin_unlock(&inode->i_lock);
4232			return -ENOMEM;
4233		}
4234		ei->jinode = jinode;
4235		jbd2_journal_init_jbd_inode(ei->jinode, inode);
4236		jinode = NULL;
4237	}
4238	spin_unlock(&inode->i_lock);
4239	if (unlikely(jinode != NULL))
4240		jbd2_free_inode(jinode);
4241	return 0;
4242}
4243
4244/*
4245 * ext4_truncate()
4246 *
4247 * We block out ext4_get_block() block instantiations across the entire
4248 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4249 * simultaneously on behalf of the same inode.
4250 *
4251 * As we work through the truncate and commit bits of it to the journal there
4252 * is one core, guiding principle: the file's tree must always be consistent on
4253 * disk.  We must be able to restart the truncate after a crash.
4254 *
4255 * The file's tree may be transiently inconsistent in memory (although it
4256 * probably isn't), but whenever we close off and commit a journal transaction,
4257 * the contents of (the filesystem + the journal) must be consistent and
4258 * restartable.  It's pretty simple, really: bottom up, right to left (although
4259 * left-to-right works OK too).
4260 *
4261 * Note that at recovery time, journal replay occurs *before* the restart of
4262 * truncate against the orphan inode list.
4263 *
4264 * The committed inode has the new, desired i_size (which is the same as
4265 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4266 * that this inode's truncate did not complete and it will again call
4267 * ext4_truncate() to have another go.  So there will be instantiated blocks
4268 * to the right of the truncation point in a crashed ext4 filesystem.  But
4269 * that's fine - as long as they are linked from the inode, the post-crash
4270 * ext4_truncate() run will find them and release them.
4271 */
4272int ext4_truncate(struct inode *inode)
4273{
4274	struct ext4_inode_info *ei = EXT4_I(inode);
4275	unsigned int credits;
4276	int err = 0, err2;
4277	handle_t *handle;
4278	struct address_space *mapping = inode->i_mapping;
4279
4280	/*
4281	 * There is a possibility that we're either freeing the inode
4282	 * or it's a completely new inode. In those cases we might not
4283	 * have i_rwsem locked because it's not necessary.
4284	 */
4285	if (!(inode->i_state & (I_NEW|I_FREEING)))
4286		WARN_ON(!inode_is_locked(inode));
4287	trace_ext4_truncate_enter(inode);
4288
4289	if (!ext4_can_truncate(inode))
4290		goto out_trace;
4291
4292	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4293		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4294
4295	if (ext4_has_inline_data(inode)) {
4296		int has_inline = 1;
4297
4298		err = ext4_inline_data_truncate(inode, &has_inline);
4299		if (err || has_inline)
4300			goto out_trace;
4301	}
4302
4303	/* If we zero-out tail of the page, we have to create jinode for jbd2 */
4304	if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4305		err = ext4_inode_attach_jinode(inode);
4306		if (err)
4307			goto out_trace;
4308	}
4309
4310	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4311		credits = ext4_writepage_trans_blocks(inode);
4312	else
4313		credits = ext4_blocks_for_truncate(inode);
4314
4315	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4316	if (IS_ERR(handle)) {
4317		err = PTR_ERR(handle);
4318		goto out_trace;
4319	}
4320
4321	if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4322		ext4_block_truncate_page(handle, mapping, inode->i_size);
4323
4324	/*
4325	 * We add the inode to the orphan list, so that if this
4326	 * truncate spans multiple transactions, and we crash, we will
4327	 * resume the truncate when the filesystem recovers.  It also
4328	 * marks the inode dirty, to catch the new size.
4329	 *
4330	 * Implication: the file must always be in a sane, consistent
4331	 * truncatable state while each transaction commits.
4332	 */
4333	err = ext4_orphan_add(handle, inode);
4334	if (err)
4335		goto out_stop;
4336
4337	down_write(&EXT4_I(inode)->i_data_sem);
4338
4339	ext4_discard_preallocations(inode, 0);
4340
4341	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4342		err = ext4_ext_truncate(handle, inode);
4343	else
4344		ext4_ind_truncate(handle, inode);
4345
4346	up_write(&ei->i_data_sem);
4347	if (err)
4348		goto out_stop;
4349
4350	if (IS_SYNC(inode))
4351		ext4_handle_sync(handle);
4352
4353out_stop:
4354	/*
4355	 * If this was a simple ftruncate() and the file will remain alive,
4356	 * then we need to clear up the orphan record which we created above.
4357	 * However, if this was a real unlink then we were called by
4358	 * ext4_evict_inode(), and we allow that function to clean up the
4359	 * orphan info for us.
4360	 */
4361	if (inode->i_nlink)
4362		ext4_orphan_del(handle, inode);
4363
4364	inode->i_mtime = inode->i_ctime = current_time(inode);
4365	err2 = ext4_mark_inode_dirty(handle, inode);
4366	if (unlikely(err2 && !err))
4367		err = err2;
4368	ext4_journal_stop(handle);
4369
4370out_trace:
4371	trace_ext4_truncate_exit(inode);
4372	return err;
4373}
4374
4375static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4376{
4377	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4378		return inode_peek_iversion_raw(inode);
4379	else
4380		return inode_peek_iversion(inode);
4381}
4382
4383static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4384				 struct ext4_inode_info *ei)
4385{
4386	struct inode *inode = &(ei->vfs_inode);
4387	u64 i_blocks = READ_ONCE(inode->i_blocks);
4388	struct super_block *sb = inode->i_sb;
4389
4390	if (i_blocks <= ~0U) {
4391		/*
4392		 * i_blocks can be represented in a 32 bit variable
4393		 * as multiple of 512 bytes
4394		 */
4395		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4396		raw_inode->i_blocks_high = 0;
4397		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4398		return 0;
4399	}
4400
4401	/*
4402	 * This should never happen since sb->s_maxbytes should not have
4403	 * allowed this, sb->s_maxbytes was set according to the huge_file
4404	 * feature in ext4_fill_super().
4405	 */
4406	if (!ext4_has_feature_huge_file(sb))
4407		return -EFSCORRUPTED;
4408
4409	if (i_blocks <= 0xffffffffffffULL) {
4410		/*
4411		 * i_blocks can be represented in a 48 bit variable
4412		 * as multiple of 512 bytes
4413		 */
4414		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4415		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4416		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4417	} else {
4418		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4419		/* i_block is stored in file system block size */
4420		i_blocks = i_blocks >> (inode->i_blkbits - 9);
4421		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4422		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4423	}
4424	return 0;
4425}
4426
4427static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4428{
4429	struct ext4_inode_info *ei = EXT4_I(inode);
4430	uid_t i_uid;
4431	gid_t i_gid;
4432	projid_t i_projid;
4433	int block;
4434	int err;
4435
4436	err = ext4_inode_blocks_set(raw_inode, ei);
4437
4438	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4439	i_uid = i_uid_read(inode);
4440	i_gid = i_gid_read(inode);
4441	i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4442	if (!(test_opt(inode->i_sb, NO_UID32))) {
4443		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4444		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4445		/*
4446		 * Fix up interoperability with old kernels. Otherwise,
4447		 * old inodes get re-used with the upper 16 bits of the
4448		 * uid/gid intact.
4449		 */
4450		if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4451			raw_inode->i_uid_high = 0;
4452			raw_inode->i_gid_high = 0;
4453		} else {
4454			raw_inode->i_uid_high =
4455				cpu_to_le16(high_16_bits(i_uid));
4456			raw_inode->i_gid_high =
4457				cpu_to_le16(high_16_bits(i_gid));
4458		}
4459	} else {
4460		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4461		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4462		raw_inode->i_uid_high = 0;
4463		raw_inode->i_gid_high = 0;
4464	}
4465	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4466
4467	EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4468	EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4469	EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4470	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4471
4472	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4473	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4474	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4475		raw_inode->i_file_acl_high =
4476			cpu_to_le16(ei->i_file_acl >> 32);
4477	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4478	ext4_isize_set(raw_inode, ei->i_disksize);
4479
4480	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4481	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4482		if (old_valid_dev(inode->i_rdev)) {
4483			raw_inode->i_block[0] =
4484				cpu_to_le32(old_encode_dev(inode->i_rdev));
4485			raw_inode->i_block[1] = 0;
4486		} else {
4487			raw_inode->i_block[0] = 0;
4488			raw_inode->i_block[1] =
4489				cpu_to_le32(new_encode_dev(inode->i_rdev));
4490			raw_inode->i_block[2] = 0;
4491		}
4492	} else if (!ext4_has_inline_data(inode)) {
4493		for (block = 0; block < EXT4_N_BLOCKS; block++)
4494			raw_inode->i_block[block] = ei->i_data[block];
4495	}
4496
4497	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4498		u64 ivers = ext4_inode_peek_iversion(inode);
4499
4500		raw_inode->i_disk_version = cpu_to_le32(ivers);
4501		if (ei->i_extra_isize) {
4502			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4503				raw_inode->i_version_hi =
4504					cpu_to_le32(ivers >> 32);
4505			raw_inode->i_extra_isize =
4506				cpu_to_le16(ei->i_extra_isize);
4507		}
4508	}
4509
4510	if (i_projid != EXT4_DEF_PROJID &&
4511	    !ext4_has_feature_project(inode->i_sb))
4512		err = err ?: -EFSCORRUPTED;
4513
4514	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4515	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4516		raw_inode->i_projid = cpu_to_le32(i_projid);
4517
4518	ext4_inode_csum_set(inode, raw_inode, ei);
4519	return err;
4520}
4521
4522/*
4523 * ext4_get_inode_loc returns with an extra refcount against the inode's
4524 * underlying buffer_head on success. If we pass 'inode' and it does not
4525 * have in-inode xattr, we have all inode data in memory that is needed
4526 * to recreate the on-disk version of this inode.
4527 */
4528static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4529				struct inode *inode, struct ext4_iloc *iloc,
4530				ext4_fsblk_t *ret_block)
4531{
4532	struct ext4_group_desc	*gdp;
4533	struct buffer_head	*bh;
4534	ext4_fsblk_t		block;
4535	struct blk_plug		plug;
4536	int			inodes_per_block, inode_offset;
4537
4538	iloc->bh = NULL;
4539	if (ino < EXT4_ROOT_INO ||
4540	    ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4541		return -EFSCORRUPTED;
4542
4543	iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4544	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4545	if (!gdp)
4546		return -EIO;
4547
4548	/*
4549	 * Figure out the offset within the block group inode table
4550	 */
4551	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4552	inode_offset = ((ino - 1) %
4553			EXT4_INODES_PER_GROUP(sb));
4554	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4555
4556	block = ext4_inode_table(sb, gdp);
4557	if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4558	    (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4559		ext4_error(sb, "Invalid inode table block %llu in "
4560			   "block_group %u", block, iloc->block_group);
4561		return -EFSCORRUPTED;
4562	}
4563	block += (inode_offset / inodes_per_block);
4564
4565	bh = sb_getblk(sb, block);
4566	if (unlikely(!bh))
4567		return -ENOMEM;
4568	if (ext4_buffer_uptodate(bh))
4569		goto has_buffer;
4570
4571	lock_buffer(bh);
4572	if (ext4_buffer_uptodate(bh)) {
4573		/* Someone brought it uptodate while we waited */
4574		unlock_buffer(bh);
4575		goto has_buffer;
4576	}
4577
4578	/*
4579	 * If we have all information of the inode in memory and this
4580	 * is the only valid inode in the block, we need not read the
4581	 * block.
4582	 */
4583	if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4584		struct buffer_head *bitmap_bh;
4585		int i, start;
4586
4587		start = inode_offset & ~(inodes_per_block - 1);
4588
4589		/* Is the inode bitmap in cache? */
4590		bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4591		if (unlikely(!bitmap_bh))
4592			goto make_io;
4593
4594		/*
4595		 * If the inode bitmap isn't in cache then the
4596		 * optimisation may end up performing two reads instead
4597		 * of one, so skip it.
4598		 */
4599		if (!buffer_uptodate(bitmap_bh)) {
4600			brelse(bitmap_bh);
4601			goto make_io;
4602		}
4603		for (i = start; i < start + inodes_per_block; i++) {
4604			if (i == inode_offset)
4605				continue;
4606			if (ext4_test_bit(i, bitmap_bh->b_data))
4607				break;
4608		}
4609		brelse(bitmap_bh);
4610		if (i == start + inodes_per_block) {
4611			struct ext4_inode *raw_inode =
4612				(struct ext4_inode *) (bh->b_data + iloc->offset);
4613
4614			/* all other inodes are free, so skip I/O */
4615			memset(bh->b_data, 0, bh->b_size);
4616			if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4617				ext4_fill_raw_inode(inode, raw_inode);
4618			set_buffer_uptodate(bh);
4619			unlock_buffer(bh);
4620			goto has_buffer;
4621		}
4622	}
4623
4624make_io:
4625	/*
4626	 * If we need to do any I/O, try to pre-readahead extra
4627	 * blocks from the inode table.
4628	 */
4629	blk_start_plug(&plug);
4630	if (EXT4_SB(sb)->s_inode_readahead_blks) {
4631		ext4_fsblk_t b, end, table;
4632		unsigned num;
4633		__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4634
4635		table = ext4_inode_table(sb, gdp);
4636		/* s_inode_readahead_blks is always a power of 2 */
4637		b = block & ~((ext4_fsblk_t) ra_blks - 1);
4638		if (table > b)
4639			b = table;
4640		end = b + ra_blks;
4641		num = EXT4_INODES_PER_GROUP(sb);
4642		if (ext4_has_group_desc_csum(sb))
4643			num -= ext4_itable_unused_count(sb, gdp);
4644		table += num / inodes_per_block;
4645		if (end > table)
4646			end = table;
4647		while (b <= end)
4648			ext4_sb_breadahead_unmovable(sb, b++);
4649	}
4650
4651	/*
4652	 * There are other valid inodes in the buffer, this inode
4653	 * has in-inode xattrs, or we don't have this inode in memory.
4654	 * Read the block from disk.
4655	 */
4656	trace_ext4_load_inode(sb, ino);
4657	ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
 
4658	blk_finish_plug(&plug);
4659	wait_on_buffer(bh);
4660	ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4661	if (!buffer_uptodate(bh)) {
4662		if (ret_block)
4663			*ret_block = block;
4664		brelse(bh);
4665		return -EIO;
4666	}
4667has_buffer:
4668	iloc->bh = bh;
4669	return 0;
4670}
4671
4672static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4673					struct ext4_iloc *iloc)
4674{
4675	ext4_fsblk_t err_blk = 0;
4676	int ret;
4677
4678	ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4679					&err_blk);
4680
4681	if (ret == -EIO)
4682		ext4_error_inode_block(inode, err_blk, EIO,
4683					"unable to read itable block");
4684
4685	return ret;
4686}
4687
4688int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4689{
4690	ext4_fsblk_t err_blk = 0;
4691	int ret;
4692
4693	ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4694					&err_blk);
4695
4696	if (ret == -EIO)
4697		ext4_error_inode_block(inode, err_blk, EIO,
4698					"unable to read itable block");
4699
4700	return ret;
4701}
4702
4703
4704int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4705			  struct ext4_iloc *iloc)
4706{
4707	return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4708}
4709
4710static bool ext4_should_enable_dax(struct inode *inode)
4711{
4712	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4713
4714	if (test_opt2(inode->i_sb, DAX_NEVER))
4715		return false;
4716	if (!S_ISREG(inode->i_mode))
4717		return false;
4718	if (ext4_should_journal_data(inode))
4719		return false;
4720	if (ext4_has_inline_data(inode))
4721		return false;
4722	if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4723		return false;
4724	if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4725		return false;
4726	if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4727		return false;
4728	if (test_opt(inode->i_sb, DAX_ALWAYS))
4729		return true;
4730
4731	return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4732}
4733
4734void ext4_set_inode_flags(struct inode *inode, bool init)
4735{
4736	unsigned int flags = EXT4_I(inode)->i_flags;
4737	unsigned int new_fl = 0;
4738
4739	WARN_ON_ONCE(IS_DAX(inode) && init);
4740
4741	if (flags & EXT4_SYNC_FL)
4742		new_fl |= S_SYNC;
4743	if (flags & EXT4_APPEND_FL)
4744		new_fl |= S_APPEND;
4745	if (flags & EXT4_IMMUTABLE_FL)
4746		new_fl |= S_IMMUTABLE;
4747	if (flags & EXT4_NOATIME_FL)
4748		new_fl |= S_NOATIME;
4749	if (flags & EXT4_DIRSYNC_FL)
4750		new_fl |= S_DIRSYNC;
4751
4752	/* Because of the way inode_set_flags() works we must preserve S_DAX
4753	 * here if already set. */
4754	new_fl |= (inode->i_flags & S_DAX);
4755	if (init && ext4_should_enable_dax(inode))
4756		new_fl |= S_DAX;
4757
4758	if (flags & EXT4_ENCRYPT_FL)
4759		new_fl |= S_ENCRYPTED;
4760	if (flags & EXT4_CASEFOLD_FL)
4761		new_fl |= S_CASEFOLD;
4762	if (flags & EXT4_VERITY_FL)
4763		new_fl |= S_VERITY;
4764	inode_set_flags(inode, new_fl,
4765			S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4766			S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4767}
4768
4769static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4770				  struct ext4_inode_info *ei)
4771{
4772	blkcnt_t i_blocks ;
4773	struct inode *inode = &(ei->vfs_inode);
4774	struct super_block *sb = inode->i_sb;
4775
4776	if (ext4_has_feature_huge_file(sb)) {
4777		/* we are using combined 48 bit field */
4778		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4779					le32_to_cpu(raw_inode->i_blocks_lo);
4780		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4781			/* i_blocks represent file system block size */
4782			return i_blocks  << (inode->i_blkbits - 9);
4783		} else {
4784			return i_blocks;
4785		}
4786	} else {
4787		return le32_to_cpu(raw_inode->i_blocks_lo);
4788	}
4789}
4790
4791static inline int ext4_iget_extra_inode(struct inode *inode,
4792					 struct ext4_inode *raw_inode,
4793					 struct ext4_inode_info *ei)
4794{
4795	__le32 *magic = (void *)raw_inode +
4796			EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4797
4798	if (EXT4_INODE_HAS_XATTR_SPACE(inode)  &&
4799	    *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
 
 
4800		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4801		return ext4_find_inline_data_nolock(inode);
 
 
 
4802	} else
4803		EXT4_I(inode)->i_inline_off = 0;
4804	return 0;
4805}
4806
4807int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4808{
4809	if (!ext4_has_feature_project(inode->i_sb))
4810		return -EOPNOTSUPP;
4811	*projid = EXT4_I(inode)->i_projid;
4812	return 0;
4813}
4814
4815/*
4816 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4817 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4818 * set.
4819 */
4820static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4821{
4822	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4823		inode_set_iversion_raw(inode, val);
4824	else
4825		inode_set_iversion_queried(inode, val);
4826}
4827
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4828struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4829			  ext4_iget_flags flags, const char *function,
4830			  unsigned int line)
4831{
4832	struct ext4_iloc iloc;
4833	struct ext4_inode *raw_inode;
4834	struct ext4_inode_info *ei;
4835	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4836	struct inode *inode;
 
4837	journal_t *journal = EXT4_SB(sb)->s_journal;
4838	long ret;
4839	loff_t size;
4840	int block;
4841	uid_t i_uid;
4842	gid_t i_gid;
4843	projid_t i_projid;
4844
4845	if ((!(flags & EXT4_IGET_SPECIAL) &&
4846	     ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4847	      ino == le32_to_cpu(es->s_usr_quota_inum) ||
4848	      ino == le32_to_cpu(es->s_grp_quota_inum) ||
4849	      ino == le32_to_cpu(es->s_prj_quota_inum) ||
4850	      ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4851	    (ino < EXT4_ROOT_INO) ||
4852	    (ino > le32_to_cpu(es->s_inodes_count))) {
4853		if (flags & EXT4_IGET_HANDLE)
4854			return ERR_PTR(-ESTALE);
4855		__ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4856			     "inode #%lu: comm %s: iget: illegal inode #",
4857			     ino, current->comm);
4858		return ERR_PTR(-EFSCORRUPTED);
4859	}
4860
4861	inode = iget_locked(sb, ino);
4862	if (!inode)
4863		return ERR_PTR(-ENOMEM);
4864	if (!(inode->i_state & I_NEW))
 
 
 
 
 
4865		return inode;
 
4866
4867	ei = EXT4_I(inode);
4868	iloc.bh = NULL;
4869
4870	ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4871	if (ret < 0)
4872		goto bad_inode;
4873	raw_inode = ext4_raw_inode(&iloc);
4874
4875	if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4876		ext4_error_inode(inode, function, line, 0,
4877				 "iget: root inode unallocated");
4878		ret = -EFSCORRUPTED;
4879		goto bad_inode;
4880	}
4881
4882	if ((flags & EXT4_IGET_HANDLE) &&
4883	    (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4884		ret = -ESTALE;
4885		goto bad_inode;
4886	}
4887
4888	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4889		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4890		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4891			EXT4_INODE_SIZE(inode->i_sb) ||
4892		    (ei->i_extra_isize & 3)) {
4893			ext4_error_inode(inode, function, line, 0,
4894					 "iget: bad extra_isize %u "
4895					 "(inode size %u)",
4896					 ei->i_extra_isize,
4897					 EXT4_INODE_SIZE(inode->i_sb));
4898			ret = -EFSCORRUPTED;
4899			goto bad_inode;
4900		}
4901	} else
4902		ei->i_extra_isize = 0;
4903
4904	/* Precompute checksum seed for inode metadata */
4905	if (ext4_has_metadata_csum(sb)) {
4906		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4907		__u32 csum;
4908		__le32 inum = cpu_to_le32(inode->i_ino);
4909		__le32 gen = raw_inode->i_generation;
4910		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4911				   sizeof(inum));
4912		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4913					      sizeof(gen));
4914	}
4915
4916	if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4917	    ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4918	     (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4919		ext4_error_inode_err(inode, function, line, 0,
4920				EFSBADCRC, "iget: checksum invalid");
4921		ret = -EFSBADCRC;
4922		goto bad_inode;
4923	}
4924
4925	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4926	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4927	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4928	if (ext4_has_feature_project(sb) &&
4929	    EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4930	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4931		i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4932	else
4933		i_projid = EXT4_DEF_PROJID;
4934
4935	if (!(test_opt(inode->i_sb, NO_UID32))) {
4936		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4937		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4938	}
4939	i_uid_write(inode, i_uid);
4940	i_gid_write(inode, i_gid);
4941	ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4942	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4943
4944	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
4945	ei->i_inline_off = 0;
4946	ei->i_dir_start_lookup = 0;
4947	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4948	/* We now have enough fields to check if the inode was active or not.
4949	 * This is needed because nfsd might try to access dead inodes
4950	 * the test is that same one that e2fsck uses
4951	 * NeilBrown 1999oct15
4952	 */
4953	if (inode->i_nlink == 0) {
4954		if ((inode->i_mode == 0 ||
4955		     !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4956		    ino != EXT4_BOOT_LOADER_INO) {
4957			/* this inode is deleted */
4958			ret = -ESTALE;
 
 
 
 
 
4959			goto bad_inode;
4960		}
4961		/* The only unlinked inodes we let through here have
4962		 * valid i_mode and are being read by the orphan
4963		 * recovery code: that's fine, we're about to complete
4964		 * the process of deleting those.
4965		 * OR it is the EXT4_BOOT_LOADER_INO which is
4966		 * not initialized on a new filesystem. */
4967	}
4968	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4969	ext4_set_inode_flags(inode, true);
4970	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4971	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4972	if (ext4_has_feature_64bit(sb))
4973		ei->i_file_acl |=
4974			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4975	inode->i_size = ext4_isize(sb, raw_inode);
4976	if ((size = i_size_read(inode)) < 0) {
4977		ext4_error_inode(inode, function, line, 0,
4978				 "iget: bad i_size value: %lld", size);
4979		ret = -EFSCORRUPTED;
4980		goto bad_inode;
4981	}
4982	/*
4983	 * If dir_index is not enabled but there's dir with INDEX flag set,
4984	 * we'd normally treat htree data as empty space. But with metadata
4985	 * checksumming that corrupts checksums so forbid that.
4986	 */
4987	if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4988	    ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4989		ext4_error_inode(inode, function, line, 0,
4990			 "iget: Dir with htree data on filesystem without dir_index feature.");
4991		ret = -EFSCORRUPTED;
4992		goto bad_inode;
4993	}
4994	ei->i_disksize = inode->i_size;
4995#ifdef CONFIG_QUOTA
4996	ei->i_reserved_quota = 0;
4997#endif
4998	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4999	ei->i_block_group = iloc.block_group;
5000	ei->i_last_alloc_group = ~0;
5001	/*
5002	 * NOTE! The in-memory inode i_data array is in little-endian order
5003	 * even on big-endian machines: we do NOT byteswap the block numbers!
5004	 */
5005	for (block = 0; block < EXT4_N_BLOCKS; block++)
5006		ei->i_data[block] = raw_inode->i_block[block];
5007	INIT_LIST_HEAD(&ei->i_orphan);
5008	ext4_fc_init_inode(&ei->vfs_inode);
5009
5010	/*
5011	 * Set transaction id's of transactions that have to be committed
5012	 * to finish f[data]sync. We set them to currently running transaction
5013	 * as we cannot be sure that the inode or some of its metadata isn't
5014	 * part of the transaction - the inode could have been reclaimed and
5015	 * now it is reread from disk.
5016	 */
5017	if (journal) {
5018		transaction_t *transaction;
5019		tid_t tid;
5020
5021		read_lock(&journal->j_state_lock);
5022		if (journal->j_running_transaction)
5023			transaction = journal->j_running_transaction;
5024		else
5025			transaction = journal->j_committing_transaction;
5026		if (transaction)
5027			tid = transaction->t_tid;
5028		else
5029			tid = journal->j_commit_sequence;
5030		read_unlock(&journal->j_state_lock);
5031		ei->i_sync_tid = tid;
5032		ei->i_datasync_tid = tid;
5033	}
5034
5035	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5036		if (ei->i_extra_isize == 0) {
5037			/* The extra space is currently unused. Use it. */
5038			BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5039			ei->i_extra_isize = sizeof(struct ext4_inode) -
5040					    EXT4_GOOD_OLD_INODE_SIZE;
5041		} else {
5042			ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5043			if (ret)
5044				goto bad_inode;
5045		}
5046	}
5047
5048	EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5049	EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5050	EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5051	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5052
5053	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5054		u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5055
5056		if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5057			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5058				ivers |=
5059		    (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5060		}
5061		ext4_inode_set_iversion_queried(inode, ivers);
5062	}
5063
5064	ret = 0;
5065	if (ei->i_file_acl &&
5066	    !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
5067		ext4_error_inode(inode, function, line, 0,
5068				 "iget: bad extended attribute block %llu",
5069				 ei->i_file_acl);
5070		ret = -EFSCORRUPTED;
5071		goto bad_inode;
5072	} else if (!ext4_has_inline_data(inode)) {
5073		/* validate the block references in the inode */
5074		if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
5075			(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5076			(S_ISLNK(inode->i_mode) &&
5077			!ext4_inode_is_fast_symlink(inode)))) {
5078			if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5079				ret = ext4_ext_check_inode(inode);
5080			else
5081				ret = ext4_ind_check_inode(inode);
5082		}
5083	}
5084	if (ret)
5085		goto bad_inode;
5086
5087	if (S_ISREG(inode->i_mode)) {
5088		inode->i_op = &ext4_file_inode_operations;
5089		inode->i_fop = &ext4_file_operations;
5090		ext4_set_aops(inode);
5091	} else if (S_ISDIR(inode->i_mode)) {
5092		inode->i_op = &ext4_dir_inode_operations;
5093		inode->i_fop = &ext4_dir_operations;
5094	} else if (S_ISLNK(inode->i_mode)) {
5095		/* VFS does not allow setting these so must be corruption */
5096		if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5097			ext4_error_inode(inode, function, line, 0,
5098					 "iget: immutable or append flags "
5099					 "not allowed on symlinks");
5100			ret = -EFSCORRUPTED;
5101			goto bad_inode;
5102		}
5103		if (IS_ENCRYPTED(inode)) {
5104			inode->i_op = &ext4_encrypted_symlink_inode_operations;
5105		} else if (ext4_inode_is_fast_symlink(inode)) {
5106			inode->i_link = (char *)ei->i_data;
5107			inode->i_op = &ext4_fast_symlink_inode_operations;
5108			nd_terminate_link(ei->i_data, inode->i_size,
5109				sizeof(ei->i_data) - 1);
5110		} else {
5111			inode->i_op = &ext4_symlink_inode_operations;
5112		}
5113	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5114	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5115		inode->i_op = &ext4_special_inode_operations;
5116		if (raw_inode->i_block[0])
5117			init_special_inode(inode, inode->i_mode,
5118			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5119		else
5120			init_special_inode(inode, inode->i_mode,
5121			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5122	} else if (ino == EXT4_BOOT_LOADER_INO) {
5123		make_bad_inode(inode);
5124	} else {
5125		ret = -EFSCORRUPTED;
5126		ext4_error_inode(inode, function, line, 0,
5127				 "iget: bogus i_mode (%o)", inode->i_mode);
5128		goto bad_inode;
5129	}
5130	if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
5131		ext4_error_inode(inode, function, line, 0,
5132				 "casefold flag without casefold feature");
5133	if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) {
5134		ext4_error_inode(inode, function, line, 0,
5135				 "bad inode without EXT4_IGET_BAD flag");
5136		ret = -EUCLEAN;
 
 
5137		goto bad_inode;
5138	}
5139
5140	brelse(iloc.bh);
5141	unlock_new_inode(inode);
5142	return inode;
5143
5144bad_inode:
5145	brelse(iloc.bh);
5146	iget_failed(inode);
5147	return ERR_PTR(ret);
5148}
5149
5150static void __ext4_update_other_inode_time(struct super_block *sb,
5151					   unsigned long orig_ino,
5152					   unsigned long ino,
5153					   struct ext4_inode *raw_inode)
5154{
5155	struct inode *inode;
5156
5157	inode = find_inode_by_ino_rcu(sb, ino);
5158	if (!inode)
5159		return;
5160
5161	if (!inode_is_dirtytime_only(inode))
5162		return;
5163
5164	spin_lock(&inode->i_lock);
5165	if (inode_is_dirtytime_only(inode)) {
5166		struct ext4_inode_info	*ei = EXT4_I(inode);
5167
5168		inode->i_state &= ~I_DIRTY_TIME;
5169		spin_unlock(&inode->i_lock);
5170
5171		spin_lock(&ei->i_raw_lock);
5172		EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5173		EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5174		EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5175		ext4_inode_csum_set(inode, raw_inode, ei);
5176		spin_unlock(&ei->i_raw_lock);
5177		trace_ext4_other_inode_update_time(inode, orig_ino);
5178		return;
5179	}
5180	spin_unlock(&inode->i_lock);
5181}
5182
5183/*
5184 * Opportunistically update the other time fields for other inodes in
5185 * the same inode table block.
5186 */
5187static void ext4_update_other_inodes_time(struct super_block *sb,
5188					  unsigned long orig_ino, char *buf)
5189{
5190	unsigned long ino;
5191	int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5192	int inode_size = EXT4_INODE_SIZE(sb);
5193
5194	/*
5195	 * Calculate the first inode in the inode table block.  Inode
5196	 * numbers are one-based.  That is, the first inode in a block
5197	 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5198	 */
5199	ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5200	rcu_read_lock();
5201	for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5202		if (ino == orig_ino)
5203			continue;
5204		__ext4_update_other_inode_time(sb, orig_ino, ino,
5205					       (struct ext4_inode *)buf);
5206	}
5207	rcu_read_unlock();
5208}
5209
5210/*
5211 * Post the struct inode info into an on-disk inode location in the
5212 * buffer-cache.  This gobbles the caller's reference to the
5213 * buffer_head in the inode location struct.
5214 *
5215 * The caller must have write access to iloc->bh.
5216 */
5217static int ext4_do_update_inode(handle_t *handle,
5218				struct inode *inode,
5219				struct ext4_iloc *iloc)
5220{
5221	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5222	struct ext4_inode_info *ei = EXT4_I(inode);
5223	struct buffer_head *bh = iloc->bh;
5224	struct super_block *sb = inode->i_sb;
5225	int err;
5226	int need_datasync = 0, set_large_file = 0;
5227
5228	spin_lock(&ei->i_raw_lock);
5229
5230	/*
5231	 * For fields not tracked in the in-memory inode, initialise them
5232	 * to zero for new inodes.
5233	 */
5234	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5235		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5236
5237	if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5238		need_datasync = 1;
5239	if (ei->i_disksize > 0x7fffffffULL) {
5240		if (!ext4_has_feature_large_file(sb) ||
5241		    EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5242			set_large_file = 1;
5243	}
5244
5245	err = ext4_fill_raw_inode(inode, raw_inode);
5246	spin_unlock(&ei->i_raw_lock);
5247	if (err) {
5248		EXT4_ERROR_INODE(inode, "corrupted inode contents");
5249		goto out_brelse;
5250	}
5251
5252	if (inode->i_sb->s_flags & SB_LAZYTIME)
5253		ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5254					      bh->b_data);
5255
5256	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5257	err = ext4_handle_dirty_metadata(handle, NULL, bh);
5258	if (err)
5259		goto out_error;
5260	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5261	if (set_large_file) {
5262		BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5263		err = ext4_journal_get_write_access(handle, sb,
5264						    EXT4_SB(sb)->s_sbh,
5265						    EXT4_JTR_NONE);
5266		if (err)
5267			goto out_error;
5268		lock_buffer(EXT4_SB(sb)->s_sbh);
5269		ext4_set_feature_large_file(sb);
5270		ext4_superblock_csum_set(sb);
5271		unlock_buffer(EXT4_SB(sb)->s_sbh);
5272		ext4_handle_sync(handle);
5273		err = ext4_handle_dirty_metadata(handle, NULL,
5274						 EXT4_SB(sb)->s_sbh);
5275	}
5276	ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5277out_error:
5278	ext4_std_error(inode->i_sb, err);
5279out_brelse:
5280	brelse(bh);
5281	return err;
5282}
5283
5284/*
5285 * ext4_write_inode()
5286 *
5287 * We are called from a few places:
5288 *
5289 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5290 *   Here, there will be no transaction running. We wait for any running
5291 *   transaction to commit.
5292 *
5293 * - Within flush work (sys_sync(), kupdate and such).
5294 *   We wait on commit, if told to.
5295 *
5296 * - Within iput_final() -> write_inode_now()
5297 *   We wait on commit, if told to.
5298 *
5299 * In all cases it is actually safe for us to return without doing anything,
5300 * because the inode has been copied into a raw inode buffer in
5301 * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
5302 * writeback.
5303 *
5304 * Note that we are absolutely dependent upon all inode dirtiers doing the
5305 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5306 * which we are interested.
5307 *
5308 * It would be a bug for them to not do this.  The code:
5309 *
5310 *	mark_inode_dirty(inode)
5311 *	stuff();
5312 *	inode->i_size = expr;
5313 *
5314 * is in error because write_inode() could occur while `stuff()' is running,
5315 * and the new i_size will be lost.  Plus the inode will no longer be on the
5316 * superblock's dirty inode list.
5317 */
5318int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5319{
5320	int err;
5321
5322	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5323	    sb_rdonly(inode->i_sb))
5324		return 0;
5325
5326	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5327		return -EIO;
5328
5329	if (EXT4_SB(inode->i_sb)->s_journal) {
5330		if (ext4_journal_current_handle()) {
5331			ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5332			dump_stack();
5333			return -EIO;
5334		}
5335
5336		/*
5337		 * No need to force transaction in WB_SYNC_NONE mode. Also
5338		 * ext4_sync_fs() will force the commit after everything is
5339		 * written.
5340		 */
5341		if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5342			return 0;
5343
5344		err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5345						EXT4_I(inode)->i_sync_tid);
5346	} else {
5347		struct ext4_iloc iloc;
5348
5349		err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5350		if (err)
5351			return err;
5352		/*
5353		 * sync(2) will flush the whole buffer cache. No need to do
5354		 * it here separately for each inode.
5355		 */
5356		if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5357			sync_dirty_buffer(iloc.bh);
5358		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5359			ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5360					       "IO error syncing inode");
5361			err = -EIO;
5362		}
5363		brelse(iloc.bh);
5364	}
5365	return err;
5366}
5367
5368/*
5369 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5370 * buffers that are attached to a folio straddling i_size and are undergoing
5371 * commit. In that case we have to wait for commit to finish and try again.
5372 */
5373static void ext4_wait_for_tail_page_commit(struct inode *inode)
5374{
5375	unsigned offset;
5376	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5377	tid_t commit_tid = 0;
5378	int ret;
 
5379
5380	offset = inode->i_size & (PAGE_SIZE - 1);
5381	/*
5382	 * If the folio is fully truncated, we don't need to wait for any commit
5383	 * (and we even should not as __ext4_journalled_invalidate_folio() may
5384	 * strip all buffers from the folio but keep the folio dirty which can then
5385	 * confuse e.g. concurrent ext4_writepage() seeing dirty folio without
5386	 * buffers). Also we don't need to wait for any commit if all buffers in
5387	 * the folio remain valid. This is most beneficial for the common case of
5388	 * blocksize == PAGESIZE.
5389	 */
5390	if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5391		return;
5392	while (1) {
5393		struct folio *folio = filemap_lock_folio(inode->i_mapping,
5394				      inode->i_size >> PAGE_SHIFT);
5395		if (!folio)
5396			return;
5397		ret = __ext4_journalled_invalidate_folio(folio, offset,
5398						folio_size(folio) - offset);
5399		folio_unlock(folio);
5400		folio_put(folio);
5401		if (ret != -EBUSY)
5402			return;
5403		commit_tid = 0;
5404		read_lock(&journal->j_state_lock);
5405		if (journal->j_committing_transaction)
5406			commit_tid = journal->j_committing_transaction->t_tid;
 
 
5407		read_unlock(&journal->j_state_lock);
5408		if (commit_tid)
5409			jbd2_log_wait_commit(journal, commit_tid);
5410	}
5411}
5412
5413/*
5414 * ext4_setattr()
5415 *
5416 * Called from notify_change.
5417 *
5418 * We want to trap VFS attempts to truncate the file as soon as
5419 * possible.  In particular, we want to make sure that when the VFS
5420 * shrinks i_size, we put the inode on the orphan list and modify
5421 * i_disksize immediately, so that during the subsequent flushing of
5422 * dirty pages and freeing of disk blocks, we can guarantee that any
5423 * commit will leave the blocks being flushed in an unused state on
5424 * disk.  (On recovery, the inode will get truncated and the blocks will
5425 * be freed, so we have a strong guarantee that no future commit will
5426 * leave these blocks visible to the user.)
5427 *
5428 * Another thing we have to assure is that if we are in ordered mode
5429 * and inode is still attached to the committing transaction, we must
5430 * we start writeout of all the dirty pages which are being truncated.
5431 * This way we are sure that all the data written in the previous
5432 * transaction are already on disk (truncate waits for pages under
5433 * writeback).
5434 *
5435 * Called with inode->i_rwsem down.
5436 */
5437int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5438		 struct iattr *attr)
5439{
5440	struct inode *inode = d_inode(dentry);
5441	int error, rc = 0;
5442	int orphan = 0;
5443	const unsigned int ia_valid = attr->ia_valid;
5444	bool inc_ivers = true;
5445
5446	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5447		return -EIO;
5448
5449	if (unlikely(IS_IMMUTABLE(inode)))
5450		return -EPERM;
5451
5452	if (unlikely(IS_APPEND(inode) &&
5453		     (ia_valid & (ATTR_MODE | ATTR_UID |
5454				  ATTR_GID | ATTR_TIMES_SET))))
5455		return -EPERM;
5456
5457	error = setattr_prepare(mnt_userns, dentry, attr);
5458	if (error)
5459		return error;
5460
5461	error = fscrypt_prepare_setattr(dentry, attr);
5462	if (error)
5463		return error;
5464
5465	error = fsverity_prepare_setattr(dentry, attr);
5466	if (error)
5467		return error;
5468
5469	if (is_quota_modification(mnt_userns, inode, attr)) {
5470		error = dquot_initialize(inode);
5471		if (error)
5472			return error;
5473	}
5474
5475	if (i_uid_needs_update(mnt_userns, attr, inode) ||
5476	    i_gid_needs_update(mnt_userns, attr, inode)) {
5477		handle_t *handle;
5478
5479		/* (user+group)*(old+new) structure, inode write (sb,
5480		 * inode block, ? - but truncate inode update has it) */
5481		handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5482			(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5483			 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5484		if (IS_ERR(handle)) {
5485			error = PTR_ERR(handle);
5486			goto err_out;
5487		}
5488
5489		/* dquot_transfer() calls back ext4_get_inode_usage() which
5490		 * counts xattr inode references.
5491		 */
5492		down_read(&EXT4_I(inode)->xattr_sem);
5493		error = dquot_transfer(mnt_userns, inode, attr);
5494		up_read(&EXT4_I(inode)->xattr_sem);
5495
5496		if (error) {
5497			ext4_journal_stop(handle);
5498			return error;
5499		}
5500		/* Update corresponding info in inode so that everything is in
5501		 * one transaction */
5502		i_uid_update(mnt_userns, attr, inode);
5503		i_gid_update(mnt_userns, attr, inode);
5504		error = ext4_mark_inode_dirty(handle, inode);
5505		ext4_journal_stop(handle);
5506		if (unlikely(error)) {
5507			return error;
5508		}
5509	}
5510
5511	if (attr->ia_valid & ATTR_SIZE) {
5512		handle_t *handle;
5513		loff_t oldsize = inode->i_size;
5514		loff_t old_disksize;
5515		int shrink = (attr->ia_size < inode->i_size);
5516
5517		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5518			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5519
5520			if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5521				return -EFBIG;
5522			}
5523		}
5524		if (!S_ISREG(inode->i_mode)) {
5525			return -EINVAL;
5526		}
5527
5528		if (attr->ia_size == inode->i_size)
5529			inc_ivers = false;
5530
5531		if (shrink) {
5532			if (ext4_should_order_data(inode)) {
5533				error = ext4_begin_ordered_truncate(inode,
5534							    attr->ia_size);
5535				if (error)
5536					goto err_out;
5537			}
5538			/*
5539			 * Blocks are going to be removed from the inode. Wait
5540			 * for dio in flight.
5541			 */
5542			inode_dio_wait(inode);
5543		}
5544
5545		filemap_invalidate_lock(inode->i_mapping);
5546
5547		rc = ext4_break_layouts(inode);
5548		if (rc) {
5549			filemap_invalidate_unlock(inode->i_mapping);
5550			goto err_out;
5551		}
5552
5553		if (attr->ia_size != inode->i_size) {
 
 
 
 
 
 
 
 
5554			handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5555			if (IS_ERR(handle)) {
5556				error = PTR_ERR(handle);
5557				goto out_mmap_sem;
5558			}
5559			if (ext4_handle_valid(handle) && shrink) {
5560				error = ext4_orphan_add(handle, inode);
5561				orphan = 1;
5562			}
5563			/*
5564			 * Update c/mtime on truncate up, ext4_truncate() will
5565			 * update c/mtime in shrink case below
 
5566			 */
5567			if (!shrink) {
5568				inode->i_mtime = current_time(inode);
5569				inode->i_ctime = inode->i_mtime;
 
 
 
5570			}
5571
5572			if (shrink)
5573				ext4_fc_track_range(handle, inode,
5574					(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5575					inode->i_sb->s_blocksize_bits,
5576					EXT_MAX_BLOCKS - 1);
5577			else
5578				ext4_fc_track_range(
5579					handle, inode,
5580					(oldsize > 0 ? oldsize - 1 : oldsize) >>
5581					inode->i_sb->s_blocksize_bits,
5582					(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5583					inode->i_sb->s_blocksize_bits);
5584
5585			down_write(&EXT4_I(inode)->i_data_sem);
5586			old_disksize = EXT4_I(inode)->i_disksize;
5587			EXT4_I(inode)->i_disksize = attr->ia_size;
5588			rc = ext4_mark_inode_dirty(handle, inode);
5589			if (!error)
5590				error = rc;
5591			/*
5592			 * We have to update i_size under i_data_sem together
5593			 * with i_disksize to avoid races with writeback code
5594			 * running ext4_wb_update_i_disksize().
5595			 */
5596			if (!error)
5597				i_size_write(inode, attr->ia_size);
5598			else
5599				EXT4_I(inode)->i_disksize = old_disksize;
5600			up_write(&EXT4_I(inode)->i_data_sem);
5601			ext4_journal_stop(handle);
5602			if (error)
5603				goto out_mmap_sem;
5604			if (!shrink) {
5605				pagecache_isize_extended(inode, oldsize,
5606							 inode->i_size);
5607			} else if (ext4_should_journal_data(inode)) {
5608				ext4_wait_for_tail_page_commit(inode);
5609			}
5610		}
5611
5612		/*
5613		 * Truncate pagecache after we've waited for commit
5614		 * in data=journal mode to make pages freeable.
5615		 */
5616		truncate_pagecache(inode, inode->i_size);
5617		/*
5618		 * Call ext4_truncate() even if i_size didn't change to
5619		 * truncate possible preallocated blocks.
5620		 */
5621		if (attr->ia_size <= oldsize) {
5622			rc = ext4_truncate(inode);
5623			if (rc)
5624				error = rc;
5625		}
5626out_mmap_sem:
5627		filemap_invalidate_unlock(inode->i_mapping);
5628	}
5629
5630	if (!error) {
5631		if (inc_ivers)
5632			inode_inc_iversion(inode);
5633		setattr_copy(mnt_userns, inode, attr);
5634		mark_inode_dirty(inode);
5635	}
5636
5637	/*
5638	 * If the call to ext4_truncate failed to get a transaction handle at
5639	 * all, we need to clean up the in-core orphan list manually.
5640	 */
5641	if (orphan && inode->i_nlink)
5642		ext4_orphan_del(NULL, inode);
5643
5644	if (!error && (ia_valid & ATTR_MODE))
5645		rc = posix_acl_chmod(mnt_userns, dentry, inode->i_mode);
5646
5647err_out:
5648	if  (error)
5649		ext4_std_error(inode->i_sb, error);
5650	if (!error)
5651		error = rc;
5652	return error;
5653}
5654
5655u32 ext4_dio_alignment(struct inode *inode)
5656{
5657	if (fsverity_active(inode))
5658		return 0;
5659	if (ext4_should_journal_data(inode))
5660		return 0;
5661	if (ext4_has_inline_data(inode))
5662		return 0;
5663	if (IS_ENCRYPTED(inode)) {
5664		if (!fscrypt_dio_supported(inode))
5665			return 0;
5666		return i_blocksize(inode);
5667	}
5668	return 1; /* use the iomap defaults */
5669}
5670
5671int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5672		 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5673{
5674	struct inode *inode = d_inode(path->dentry);
5675	struct ext4_inode *raw_inode;
5676	struct ext4_inode_info *ei = EXT4_I(inode);
5677	unsigned int flags;
5678
5679	if ((request_mask & STATX_BTIME) &&
5680	    EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5681		stat->result_mask |= STATX_BTIME;
5682		stat->btime.tv_sec = ei->i_crtime.tv_sec;
5683		stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5684	}
5685
5686	/*
5687	 * Return the DIO alignment restrictions if requested.  We only return
5688	 * this information when requested, since on encrypted files it might
5689	 * take a fair bit of work to get if the file wasn't opened recently.
5690	 */
5691	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5692		u32 dio_align = ext4_dio_alignment(inode);
5693
5694		stat->result_mask |= STATX_DIOALIGN;
5695		if (dio_align == 1) {
5696			struct block_device *bdev = inode->i_sb->s_bdev;
5697
5698			/* iomap defaults */
5699			stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5700			stat->dio_offset_align = bdev_logical_block_size(bdev);
5701		} else {
5702			stat->dio_mem_align = dio_align;
5703			stat->dio_offset_align = dio_align;
5704		}
5705	}
5706
 
 
 
 
 
 
 
 
 
 
 
 
5707	flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5708	if (flags & EXT4_APPEND_FL)
5709		stat->attributes |= STATX_ATTR_APPEND;
5710	if (flags & EXT4_COMPR_FL)
5711		stat->attributes |= STATX_ATTR_COMPRESSED;
5712	if (flags & EXT4_ENCRYPT_FL)
5713		stat->attributes |= STATX_ATTR_ENCRYPTED;
5714	if (flags & EXT4_IMMUTABLE_FL)
5715		stat->attributes |= STATX_ATTR_IMMUTABLE;
5716	if (flags & EXT4_NODUMP_FL)
5717		stat->attributes |= STATX_ATTR_NODUMP;
5718	if (flags & EXT4_VERITY_FL)
5719		stat->attributes |= STATX_ATTR_VERITY;
5720
5721	stat->attributes_mask |= (STATX_ATTR_APPEND |
5722				  STATX_ATTR_COMPRESSED |
5723				  STATX_ATTR_ENCRYPTED |
5724				  STATX_ATTR_IMMUTABLE |
5725				  STATX_ATTR_NODUMP |
5726				  STATX_ATTR_VERITY);
5727
5728	generic_fillattr(mnt_userns, inode, stat);
5729	return 0;
5730}
5731
5732int ext4_file_getattr(struct user_namespace *mnt_userns,
5733		      const struct path *path, struct kstat *stat,
5734		      u32 request_mask, unsigned int query_flags)
5735{
5736	struct inode *inode = d_inode(path->dentry);
5737	u64 delalloc_blocks;
5738
5739	ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5740
5741	/*
5742	 * If there is inline data in the inode, the inode will normally not
5743	 * have data blocks allocated (it may have an external xattr block).
5744	 * Report at least one sector for such files, so tools like tar, rsync,
5745	 * others don't incorrectly think the file is completely sparse.
5746	 */
5747	if (unlikely(ext4_has_inline_data(inode)))
5748		stat->blocks += (stat->size + 511) >> 9;
5749
5750	/*
5751	 * We can't update i_blocks if the block allocation is delayed
5752	 * otherwise in the case of system crash before the real block
5753	 * allocation is done, we will have i_blocks inconsistent with
5754	 * on-disk file blocks.
5755	 * We always keep i_blocks updated together with real
5756	 * allocation. But to not confuse with user, stat
5757	 * will return the blocks that include the delayed allocation
5758	 * blocks for this file.
5759	 */
5760	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5761				   EXT4_I(inode)->i_reserved_data_blocks);
5762	stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5763	return 0;
5764}
5765
5766static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5767				   int pextents)
5768{
5769	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5770		return ext4_ind_trans_blocks(inode, lblocks);
5771	return ext4_ext_index_trans_blocks(inode, pextents);
5772}
5773
5774/*
5775 * Account for index blocks, block groups bitmaps and block group
5776 * descriptor blocks if modify datablocks and index blocks
5777 * worse case, the indexs blocks spread over different block groups
5778 *
5779 * If datablocks are discontiguous, they are possible to spread over
5780 * different block groups too. If they are contiguous, with flexbg,
5781 * they could still across block group boundary.
5782 *
5783 * Also account for superblock, inode, quota and xattr blocks
5784 */
5785static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5786				  int pextents)
5787{
5788	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5789	int gdpblocks;
5790	int idxblocks;
5791	int ret = 0;
5792
5793	/*
5794	 * How many index blocks need to touch to map @lblocks logical blocks
5795	 * to @pextents physical extents?
5796	 */
5797	idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5798
5799	ret = idxblocks;
5800
5801	/*
5802	 * Now let's see how many group bitmaps and group descriptors need
5803	 * to account
5804	 */
5805	groups = idxblocks + pextents;
5806	gdpblocks = groups;
5807	if (groups > ngroups)
5808		groups = ngroups;
5809	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5810		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5811
5812	/* bitmaps and block group descriptor blocks */
5813	ret += groups + gdpblocks;
5814
5815	/* Blocks for super block, inode, quota and xattr blocks */
5816	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5817
5818	return ret;
5819}
5820
5821/*
5822 * Calculate the total number of credits to reserve to fit
5823 * the modification of a single pages into a single transaction,
5824 * which may include multiple chunks of block allocations.
5825 *
5826 * This could be called via ext4_write_begin()
5827 *
5828 * We need to consider the worse case, when
5829 * one new block per extent.
5830 */
5831int ext4_writepage_trans_blocks(struct inode *inode)
5832{
5833	int bpp = ext4_journal_blocks_per_page(inode);
5834	int ret;
5835
5836	ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5837
5838	/* Account for data blocks for journalled mode */
5839	if (ext4_should_journal_data(inode))
5840		ret += bpp;
5841	return ret;
5842}
5843
5844/*
5845 * Calculate the journal credits for a chunk of data modification.
5846 *
5847 * This is called from DIO, fallocate or whoever calling
5848 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5849 *
5850 * journal buffers for data blocks are not included here, as DIO
5851 * and fallocate do no need to journal data buffers.
5852 */
5853int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5854{
5855	return ext4_meta_trans_blocks(inode, nrblocks, 1);
5856}
5857
5858/*
5859 * The caller must have previously called ext4_reserve_inode_write().
5860 * Give this, we know that the caller already has write access to iloc->bh.
5861 */
5862int ext4_mark_iloc_dirty(handle_t *handle,
5863			 struct inode *inode, struct ext4_iloc *iloc)
5864{
5865	int err = 0;
5866
5867	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5868		put_bh(iloc->bh);
5869		return -EIO;
5870	}
5871	ext4_fc_track_inode(handle, inode);
5872
5873	/* the do_update_inode consumes one bh->b_count */
5874	get_bh(iloc->bh);
5875
5876	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5877	err = ext4_do_update_inode(handle, inode, iloc);
5878	put_bh(iloc->bh);
5879	return err;
5880}
5881
5882/*
5883 * On success, We end up with an outstanding reference count against
5884 * iloc->bh.  This _must_ be cleaned up later.
5885 */
5886
5887int
5888ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5889			 struct ext4_iloc *iloc)
5890{
5891	int err;
5892
5893	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5894		return -EIO;
5895
5896	err = ext4_get_inode_loc(inode, iloc);
5897	if (!err) {
5898		BUFFER_TRACE(iloc->bh, "get_write_access");
5899		err = ext4_journal_get_write_access(handle, inode->i_sb,
5900						    iloc->bh, EXT4_JTR_NONE);
5901		if (err) {
5902			brelse(iloc->bh);
5903			iloc->bh = NULL;
5904		}
5905	}
5906	ext4_std_error(inode->i_sb, err);
5907	return err;
5908}
5909
5910static int __ext4_expand_extra_isize(struct inode *inode,
5911				     unsigned int new_extra_isize,
5912				     struct ext4_iloc *iloc,
5913				     handle_t *handle, int *no_expand)
5914{
5915	struct ext4_inode *raw_inode;
5916	struct ext4_xattr_ibody_header *header;
5917	unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5918	struct ext4_inode_info *ei = EXT4_I(inode);
5919	int error;
5920
5921	/* this was checked at iget time, but double check for good measure */
5922	if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5923	    (ei->i_extra_isize & 3)) {
5924		EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5925				 ei->i_extra_isize,
5926				 EXT4_INODE_SIZE(inode->i_sb));
5927		return -EFSCORRUPTED;
5928	}
5929	if ((new_extra_isize < ei->i_extra_isize) ||
5930	    (new_extra_isize < 4) ||
5931	    (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5932		return -EINVAL;	/* Should never happen */
5933
5934	raw_inode = ext4_raw_inode(iloc);
5935
5936	header = IHDR(inode, raw_inode);
5937
5938	/* No extended attributes present */
5939	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5940	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5941		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5942		       EXT4_I(inode)->i_extra_isize, 0,
5943		       new_extra_isize - EXT4_I(inode)->i_extra_isize);
5944		EXT4_I(inode)->i_extra_isize = new_extra_isize;
5945		return 0;
5946	}
5947
5948	/*
5949	 * We may need to allocate external xattr block so we need quotas
5950	 * initialized. Here we can be called with various locks held so we
5951	 * cannot affort to initialize quotas ourselves. So just bail.
5952	 */
5953	if (dquot_initialize_needed(inode))
5954		return -EAGAIN;
5955
5956	/* try to expand with EAs present */
5957	error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5958					   raw_inode, handle);
5959	if (error) {
5960		/*
5961		 * Inode size expansion failed; don't try again
5962		 */
5963		*no_expand = 1;
5964	}
5965
5966	return error;
5967}
5968
5969/*
5970 * Expand an inode by new_extra_isize bytes.
5971 * Returns 0 on success or negative error number on failure.
5972 */
5973static int ext4_try_to_expand_extra_isize(struct inode *inode,
5974					  unsigned int new_extra_isize,
5975					  struct ext4_iloc iloc,
5976					  handle_t *handle)
5977{
5978	int no_expand;
5979	int error;
5980
5981	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5982		return -EOVERFLOW;
5983
5984	/*
5985	 * In nojournal mode, we can immediately attempt to expand
5986	 * the inode.  When journaled, we first need to obtain extra
5987	 * buffer credits since we may write into the EA block
5988	 * with this same handle. If journal_extend fails, then it will
5989	 * only result in a minor loss of functionality for that inode.
5990	 * If this is felt to be critical, then e2fsck should be run to
5991	 * force a large enough s_min_extra_isize.
5992	 */
5993	if (ext4_journal_extend(handle,
5994				EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5995		return -ENOSPC;
5996
5997	if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5998		return -EBUSY;
5999
6000	error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
6001					  handle, &no_expand);
6002	ext4_write_unlock_xattr(inode, &no_expand);
6003
6004	return error;
6005}
6006
6007int ext4_expand_extra_isize(struct inode *inode,
6008			    unsigned int new_extra_isize,
6009			    struct ext4_iloc *iloc)
6010{
6011	handle_t *handle;
6012	int no_expand;
6013	int error, rc;
6014
6015	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
6016		brelse(iloc->bh);
6017		return -EOVERFLOW;
6018	}
6019
6020	handle = ext4_journal_start(inode, EXT4_HT_INODE,
6021				    EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6022	if (IS_ERR(handle)) {
6023		error = PTR_ERR(handle);
6024		brelse(iloc->bh);
6025		return error;
6026	}
6027
6028	ext4_write_lock_xattr(inode, &no_expand);
6029
6030	BUFFER_TRACE(iloc->bh, "get_write_access");
6031	error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
6032					      EXT4_JTR_NONE);
6033	if (error) {
6034		brelse(iloc->bh);
6035		goto out_unlock;
6036	}
6037
6038	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6039					  handle, &no_expand);
6040
6041	rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6042	if (!error)
6043		error = rc;
6044
6045out_unlock:
6046	ext4_write_unlock_xattr(inode, &no_expand);
6047	ext4_journal_stop(handle);
6048	return error;
6049}
6050
6051/*
6052 * What we do here is to mark the in-core inode as clean with respect to inode
6053 * dirtiness (it may still be data-dirty).
6054 * This means that the in-core inode may be reaped by prune_icache
6055 * without having to perform any I/O.  This is a very good thing,
6056 * because *any* task may call prune_icache - even ones which
6057 * have a transaction open against a different journal.
6058 *
6059 * Is this cheating?  Not really.  Sure, we haven't written the
6060 * inode out, but prune_icache isn't a user-visible syncing function.
6061 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6062 * we start and wait on commits.
6063 */
6064int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
6065				const char *func, unsigned int line)
6066{
6067	struct ext4_iloc iloc;
6068	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6069	int err;
6070
6071	might_sleep();
6072	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6073	err = ext4_reserve_inode_write(handle, inode, &iloc);
6074	if (err)
6075		goto out;
6076
6077	if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6078		ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6079					       iloc, handle);
6080
6081	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
6082out:
6083	if (unlikely(err))
6084		ext4_error_inode_err(inode, func, line, 0, err,
6085					"mark_inode_dirty error");
6086	return err;
6087}
6088
6089/*
6090 * ext4_dirty_inode() is called from __mark_inode_dirty()
6091 *
6092 * We're really interested in the case where a file is being extended.
6093 * i_size has been changed by generic_commit_write() and we thus need
6094 * to include the updated inode in the current transaction.
6095 *
6096 * Also, dquot_alloc_block() will always dirty the inode when blocks
6097 * are allocated to the file.
6098 *
6099 * If the inode is marked synchronous, we don't honour that here - doing
6100 * so would cause a commit on atime updates, which we don't bother doing.
6101 * We handle synchronous inodes at the highest possible level.
6102 */
6103void ext4_dirty_inode(struct inode *inode, int flags)
6104{
6105	handle_t *handle;
6106
6107	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6108	if (IS_ERR(handle))
6109		return;
6110	ext4_mark_inode_dirty(handle, inode);
6111	ext4_journal_stop(handle);
6112}
6113
6114int ext4_change_inode_journal_flag(struct inode *inode, int val)
6115{
6116	journal_t *journal;
6117	handle_t *handle;
6118	int err;
6119	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6120
6121	/*
6122	 * We have to be very careful here: changing a data block's
6123	 * journaling status dynamically is dangerous.  If we write a
6124	 * data block to the journal, change the status and then delete
6125	 * that block, we risk forgetting to revoke the old log record
6126	 * from the journal and so a subsequent replay can corrupt data.
6127	 * So, first we make sure that the journal is empty and that
6128	 * nobody is changing anything.
6129	 */
6130
6131	journal = EXT4_JOURNAL(inode);
6132	if (!journal)
6133		return 0;
6134	if (is_journal_aborted(journal))
6135		return -EROFS;
6136
6137	/* Wait for all existing dio workers */
6138	inode_dio_wait(inode);
6139
6140	/*
6141	 * Before flushing the journal and switching inode's aops, we have
6142	 * to flush all dirty data the inode has. There can be outstanding
6143	 * delayed allocations, there can be unwritten extents created by
6144	 * fallocate or buffered writes in dioread_nolock mode covered by
6145	 * dirty data which can be converted only after flushing the dirty
6146	 * data (and journalled aops don't know how to handle these cases).
6147	 */
6148	if (val) {
6149		filemap_invalidate_lock(inode->i_mapping);
6150		err = filemap_write_and_wait(inode->i_mapping);
6151		if (err < 0) {
6152			filemap_invalidate_unlock(inode->i_mapping);
6153			return err;
6154		}
6155	}
6156
6157	percpu_down_write(&sbi->s_writepages_rwsem);
6158	jbd2_journal_lock_updates(journal);
6159
6160	/*
6161	 * OK, there are no updates running now, and all cached data is
6162	 * synced to disk.  We are now in a completely consistent state
6163	 * which doesn't have anything in the journal, and we know that
6164	 * no filesystem updates are running, so it is safe to modify
6165	 * the inode's in-core data-journaling state flag now.
6166	 */
6167
6168	if (val)
6169		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6170	else {
6171		err = jbd2_journal_flush(journal, 0);
6172		if (err < 0) {
6173			jbd2_journal_unlock_updates(journal);
6174			percpu_up_write(&sbi->s_writepages_rwsem);
6175			return err;
6176		}
6177		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6178	}
6179	ext4_set_aops(inode);
6180
6181	jbd2_journal_unlock_updates(journal);
6182	percpu_up_write(&sbi->s_writepages_rwsem);
6183
6184	if (val)
6185		filemap_invalidate_unlock(inode->i_mapping);
6186
6187	/* Finally we can mark the inode as dirty. */
6188
6189	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6190	if (IS_ERR(handle))
6191		return PTR_ERR(handle);
6192
6193	ext4_fc_mark_ineligible(inode->i_sb,
6194		EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6195	err = ext4_mark_inode_dirty(handle, inode);
6196	ext4_handle_sync(handle);
6197	ext4_journal_stop(handle);
6198	ext4_std_error(inode->i_sb, err);
6199
6200	return err;
6201}
6202
6203static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6204			    struct buffer_head *bh)
6205{
6206	return !buffer_mapped(bh);
6207}
6208
6209vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6210{
6211	struct vm_area_struct *vma = vmf->vma;
6212	struct page *page = vmf->page;
6213	loff_t size;
6214	unsigned long len;
6215	int err;
6216	vm_fault_t ret;
6217	struct file *file = vma->vm_file;
6218	struct inode *inode = file_inode(file);
6219	struct address_space *mapping = inode->i_mapping;
6220	handle_t *handle;
6221	get_block_t *get_block;
6222	int retries = 0;
6223
6224	if (unlikely(IS_IMMUTABLE(inode)))
6225		return VM_FAULT_SIGBUS;
6226
6227	sb_start_pagefault(inode->i_sb);
6228	file_update_time(vma->vm_file);
6229
6230	filemap_invalidate_lock_shared(mapping);
6231
6232	err = ext4_convert_inline_data(inode);
6233	if (err)
6234		goto out_ret;
6235
6236	/*
6237	 * On data journalling we skip straight to the transaction handle:
6238	 * there's no delalloc; page truncated will be checked later; the
6239	 * early return w/ all buffers mapped (calculates size/len) can't
6240	 * be used; and there's no dioread_nolock, so only ext4_get_block.
6241	 */
6242	if (ext4_should_journal_data(inode))
6243		goto retry_alloc;
6244
6245	/* Delalloc case is easy... */
6246	if (test_opt(inode->i_sb, DELALLOC) &&
6247	    !ext4_nonda_switch(inode->i_sb)) {
6248		do {
6249			err = block_page_mkwrite(vma, vmf,
6250						   ext4_da_get_block_prep);
6251		} while (err == -ENOSPC &&
6252		       ext4_should_retry_alloc(inode->i_sb, &retries));
6253		goto out_ret;
6254	}
6255
6256	lock_page(page);
6257	size = i_size_read(inode);
6258	/* Page got truncated from under us? */
6259	if (page->mapping != mapping || page_offset(page) > size) {
6260		unlock_page(page);
6261		ret = VM_FAULT_NOPAGE;
6262		goto out;
6263	}
6264
6265	if (page->index == size >> PAGE_SHIFT)
6266		len = size & ~PAGE_MASK;
6267	else
6268		len = PAGE_SIZE;
6269	/*
6270	 * Return if we have all the buffers mapped. This avoids the need to do
6271	 * journal_start/journal_stop which can block and take a long time
6272	 *
6273	 * This cannot be done for data journalling, as we have to add the
6274	 * inode to the transaction's list to writeprotect pages on commit.
6275	 */
6276	if (page_has_buffers(page)) {
6277		if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page),
6278					    0, len, NULL,
6279					    ext4_bh_unmapped)) {
6280			/* Wait so that we don't change page under IO */
6281			wait_for_stable_page(page);
6282			ret = VM_FAULT_LOCKED;
6283			goto out;
6284		}
6285	}
6286	unlock_page(page);
6287	/* OK, we need to fill the hole... */
6288	if (ext4_should_dioread_nolock(inode))
6289		get_block = ext4_get_block_unwritten;
6290	else
6291		get_block = ext4_get_block;
6292retry_alloc:
6293	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6294				    ext4_writepage_trans_blocks(inode));
6295	if (IS_ERR(handle)) {
6296		ret = VM_FAULT_SIGBUS;
6297		goto out;
6298	}
6299	/*
6300	 * Data journalling can't use block_page_mkwrite() because it
6301	 * will set_buffer_dirty() before do_journal_get_write_access()
6302	 * thus might hit warning messages for dirty metadata buffers.
6303	 */
6304	if (!ext4_should_journal_data(inode)) {
6305		err = block_page_mkwrite(vma, vmf, get_block);
6306	} else {
6307		lock_page(page);
6308		size = i_size_read(inode);
6309		/* Page got truncated from under us? */
6310		if (page->mapping != mapping || page_offset(page) > size) {
6311			ret = VM_FAULT_NOPAGE;
6312			goto out_error;
6313		}
6314
6315		if (page->index == size >> PAGE_SHIFT)
6316			len = size & ~PAGE_MASK;
6317		else
6318			len = PAGE_SIZE;
6319
6320		err = __block_write_begin(page, 0, len, ext4_get_block);
 
6321		if (!err) {
6322			ret = VM_FAULT_SIGBUS;
6323			if (ext4_walk_page_buffers(handle, inode,
6324					page_buffers(page), 0, len, NULL,
6325					do_journal_get_write_access))
6326				goto out_error;
6327			if (ext4_walk_page_buffers(handle, inode,
6328					page_buffers(page), 0, len, NULL,
6329					write_end_fn))
6330				goto out_error;
6331			if (ext4_jbd2_inode_add_write(handle, inode,
6332						      page_offset(page), len))
6333				goto out_error;
6334			ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6335		} else {
6336			unlock_page(page);
6337		}
6338	}
6339	ext4_journal_stop(handle);
6340	if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6341		goto retry_alloc;
6342out_ret:
6343	ret = block_page_mkwrite_return(err);
6344out:
6345	filemap_invalidate_unlock_shared(mapping);
6346	sb_end_pagefault(inode->i_sb);
6347	return ret;
6348out_error:
6349	unlock_page(page);
6350	ext4_journal_stop(handle);
6351	goto out;
6352}