Linux Audio

Check our new training course

Loading...
v3.5.6
 
   1/*
   2 *  linux/fs/ext4/inode.c
   3 *
   4 * Copyright (C) 1992, 1993, 1994, 1995
   5 * Remy Card (card@masi.ibp.fr)
   6 * Laboratoire MASI - Institut Blaise Pascal
   7 * Universite Pierre et Marie Curie (Paris VI)
   8 *
   9 *  from
  10 *
  11 *  linux/fs/minix/inode.c
  12 *
  13 *  Copyright (C) 1991, 1992  Linus Torvalds
  14 *
  15 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  16 *	(jj@sunsite.ms.mff.cuni.cz)
  17 *
  18 *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  19 */
  20
  21#include <linux/fs.h>
 
  22#include <linux/time.h>
  23#include <linux/jbd2.h>
  24#include <linux/highuid.h>
  25#include <linux/pagemap.h>
 
  26#include <linux/quotaops.h>
  27#include <linux/string.h>
  28#include <linux/buffer_head.h>
  29#include <linux/writeback.h>
  30#include <linux/pagevec.h>
  31#include <linux/mpage.h>
  32#include <linux/namei.h>
  33#include <linux/uio.h>
  34#include <linux/bio.h>
  35#include <linux/workqueue.h>
  36#include <linux/kernel.h>
  37#include <linux/printk.h>
  38#include <linux/slab.h>
  39#include <linux/ratelimit.h>
 
 
  40
  41#include "ext4_jbd2.h"
  42#include "xattr.h"
  43#include "acl.h"
  44#include "truncate.h"
  45
  46#include <trace/events/ext4.h>
  47
  48#define MPAGE_DA_EXTENT_TAIL 0x01
  49
  50static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
  51			      struct ext4_inode_info *ei)
  52{
  53	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  54	__u16 csum_lo;
  55	__u16 csum_hi = 0;
  56	__u32 csum;
 
 
 
 
 
 
 
 
 
  57
  58	csum_lo = raw->i_checksum_lo;
  59	raw->i_checksum_lo = 0;
  60	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  61	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
  62		csum_hi = raw->i_checksum_hi;
  63		raw->i_checksum_hi = 0;
 
 
 
 
 
 
  64	}
  65
  66	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
  67			   EXT4_INODE_SIZE(inode->i_sb));
  68
  69	raw->i_checksum_lo = csum_lo;
  70	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  71	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
  72		raw->i_checksum_hi = csum_hi;
  73
  74	return csum;
  75}
  76
  77static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
  78				  struct ext4_inode_info *ei)
  79{
  80	__u32 provided, calculated;
  81
  82	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  83	    cpu_to_le32(EXT4_OS_LINUX) ||
  84	    !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
  85		EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
  86		return 1;
  87
  88	provided = le16_to_cpu(raw->i_checksum_lo);
  89	calculated = ext4_inode_csum(inode, raw, ei);
  90	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  91	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
  92		provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
  93	else
  94		calculated &= 0xFFFF;
  95
  96	return provided == calculated;
  97}
  98
  99static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
 100				struct ext4_inode_info *ei)
 101{
 102	__u32 csum;
 103
 104	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
 105	    cpu_to_le32(EXT4_OS_LINUX) ||
 106	    !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 107		EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 108		return;
 109
 110	csum = ext4_inode_csum(inode, raw, ei);
 111	raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
 112	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
 113	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
 114		raw->i_checksum_hi = cpu_to_le16(csum >> 16);
 115}
 116
 117static inline int ext4_begin_ordered_truncate(struct inode *inode,
 118					      loff_t new_size)
 119{
 120	trace_ext4_begin_ordered_truncate(inode, new_size);
 121	/*
 122	 * If jinode is zero, then we never opened the file for
 123	 * writing, so there's no need to call
 124	 * jbd2_journal_begin_ordered_truncate() since there's no
 125	 * outstanding writes we need to flush.
 126	 */
 127	if (!EXT4_I(inode)->jinode)
 128		return 0;
 129	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
 130						   EXT4_I(inode)->jinode,
 131						   new_size);
 132}
 133
 134static void ext4_invalidatepage(struct page *page, unsigned long offset);
 135static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
 136				   struct buffer_head *bh_result, int create);
 137static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
 138static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
 139static int __ext4_journalled_writepage(struct page *page, unsigned int len);
 140static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
 141static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
 142		struct inode *inode, struct page *page, loff_t from,
 143		loff_t length, int flags);
 144
 145/*
 146 * Test whether an inode is a fast symlink.
 
 147 */
 148static int ext4_inode_is_fast_symlink(struct inode *inode)
 149{
 150	int ea_blocks = EXT4_I(inode)->i_file_acl ?
 151		(inode->i_sb->s_blocksize >> 9) : 0;
 152
 153	return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
 154}
 155
 156/*
 157 * Restart the transaction associated with *handle.  This does a commit,
 158 * so before we call here everything must be consistently dirtied against
 159 * this transaction.
 160 */
 161int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
 162				 int nblocks)
 163{
 164	int ret;
 165
 166	/*
 167	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
 168	 * moment, get_block can be called only for blocks inside i_size since
 169	 * page cache has been already dropped and writes are blocked by
 170	 * i_mutex. So we can safely drop the i_data_sem here.
 171	 */
 172	BUG_ON(EXT4_JOURNAL(inode) == NULL);
 173	jbd_debug(2, "restarting handle %p\n", handle);
 174	up_write(&EXT4_I(inode)->i_data_sem);
 175	ret = ext4_journal_restart(handle, nblocks);
 176	down_write(&EXT4_I(inode)->i_data_sem);
 177	ext4_discard_preallocations(inode);
 178
 179	return ret;
 
 
 
 180}
 181
 182/*
 183 * Called at the last iput() if i_nlink is zero.
 184 */
 185void ext4_evict_inode(struct inode *inode)
 186{
 187	handle_t *handle;
 188	int err;
 
 
 
 
 
 
 
 
 189
 190	trace_ext4_evict_inode(inode);
 191
 192	ext4_ioend_wait(inode);
 193
 194	if (inode->i_nlink) {
 195		/*
 196		 * When journalling data dirty buffers are tracked only in the
 197		 * journal. So although mm thinks everything is clean and
 198		 * ready for reaping the inode might still have some pages to
 199		 * write in the running transaction or waiting to be
 200		 * checkpointed. Thus calling jbd2_journal_invalidatepage()
 201		 * (via truncate_inode_pages()) to discard these buffers can
 202		 * cause data loss. Also even if we did not discard these
 203		 * buffers, we would have no way to find them after the inode
 204		 * is reaped and thus user could see stale data if he tries to
 205		 * read them before the transaction is checkpointed. So be
 206		 * careful and force everything to disk here... We use
 207		 * ei->i_datasync_tid to store the newest transaction
 208		 * containing inode's data.
 209		 *
 210		 * Note that directories do not have this problem because they
 211		 * don't use page cache.
 212		 */
 213		if (ext4_should_journal_data(inode) &&
 214		    (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
 215			journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 216			tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
 217
 218			jbd2_log_start_commit(journal, commit_tid);
 219			jbd2_log_wait_commit(journal, commit_tid);
 220			filemap_write_and_wait(&inode->i_data);
 221		}
 222		truncate_inode_pages(&inode->i_data, 0);
 223		goto no_delete;
 224	}
 225
 226	if (!is_bad_inode(inode))
 227		dquot_initialize(inode);
 
 228
 229	if (ext4_should_order_data(inode))
 230		ext4_begin_ordered_truncate(inode, 0);
 231	truncate_inode_pages(&inode->i_data, 0);
 232
 233	if (is_bad_inode(inode))
 234		goto no_delete;
 
 
 
 
 
 
 
 235
 236	handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 237	if (IS_ERR(handle)) {
 238		ext4_std_error(inode->i_sb, PTR_ERR(handle));
 239		/*
 240		 * If we're going to skip the normal cleanup, we still need to
 241		 * make sure that the in-core orphan linked list is properly
 242		 * cleaned up.
 243		 */
 244		ext4_orphan_del(NULL, inode);
 
 
 245		goto no_delete;
 246	}
 247
 248	if (IS_SYNC(inode))
 249		ext4_handle_sync(handle);
 
 
 
 
 
 
 
 
 
 
 250	inode->i_size = 0;
 251	err = ext4_mark_inode_dirty(handle, inode);
 252	if (err) {
 253		ext4_warning(inode->i_sb,
 254			     "couldn't mark inode dirty (err %d)", err);
 255		goto stop_handle;
 256	}
 257	if (inode->i_blocks)
 258		ext4_truncate(inode);
 259
 260	/*
 261	 * ext4_ext_truncate() doesn't reserve any slop when it
 262	 * restarts journal transactions; therefore there may not be
 263	 * enough credits left in the handle to remove the inode from
 264	 * the orphan list and set the dtime field.
 265	 */
 266	if (!ext4_handle_has_enough_credits(handle, 3)) {
 267		err = ext4_journal_extend(handle, 3);
 268		if (err > 0)
 269			err = ext4_journal_restart(handle, 3);
 270		if (err != 0) {
 271			ext4_warning(inode->i_sb,
 272				     "couldn't extend journal (err %d)", err);
 273		stop_handle:
 274			ext4_journal_stop(handle);
 275			ext4_orphan_del(NULL, inode);
 276			goto no_delete;
 277		}
 278	}
 279
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 280	/*
 281	 * Kill off the orphan record which ext4_truncate created.
 282	 * AKPM: I think this can be inside the above `if'.
 283	 * Note that ext4_orphan_del() has to be able to cope with the
 284	 * deletion of a non-existent orphan - this is because we don't
 285	 * know if ext4_truncate() actually created an orphan record.
 286	 * (Well, we could do this if we need to, but heck - it works)
 287	 */
 288	ext4_orphan_del(handle, inode);
 289	EXT4_I(inode)->i_dtime	= get_seconds();
 290
 291	/*
 292	 * One subtle ordering requirement: if anything has gone wrong
 293	 * (transaction abort, IO errors, whatever), then we can still
 294	 * do these next steps (the fs will already have been marked as
 295	 * having errors), but we can't free the inode if the mark_dirty
 296	 * fails.
 297	 */
 298	if (ext4_mark_inode_dirty(handle, inode))
 299		/* If that failed, just do the required in-core inode clear. */
 300		ext4_clear_inode(inode);
 301	else
 302		ext4_free_inode(handle, inode);
 303	ext4_journal_stop(handle);
 
 
 
 304	return;
 305no_delete:
 
 
 
 
 
 
 
 
 306	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
 307}
 308
 309#ifdef CONFIG_QUOTA
 310qsize_t *ext4_get_reserved_space(struct inode *inode)
 311{
 312	return &EXT4_I(inode)->i_reserved_quota;
 313}
 314#endif
 315
 316/*
 317 * Calculate the number of metadata blocks need to reserve
 318 * to allocate a block located at @lblock
 319 */
 320static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 321{
 322	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 323		return ext4_ext_calc_metadata_amount(inode, lblock);
 324
 325	return ext4_ind_calc_metadata_amount(inode, lblock);
 326}
 327
 328/*
 329 * Called with i_data_sem down, which is important since we can call
 330 * ext4_discard_preallocations() from here.
 331 */
 332void ext4_da_update_reserve_space(struct inode *inode,
 333					int used, int quota_claim)
 334{
 335	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 336	struct ext4_inode_info *ei = EXT4_I(inode);
 337
 338	spin_lock(&ei->i_block_reservation_lock);
 339	trace_ext4_da_update_reserve_space(inode, used, quota_claim);
 340	if (unlikely(used > ei->i_reserved_data_blocks)) {
 341		ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
 342			 "with only %d reserved data blocks",
 343			 __func__, inode->i_ino, used,
 344			 ei->i_reserved_data_blocks);
 345		WARN_ON(1);
 346		used = ei->i_reserved_data_blocks;
 347	}
 348
 349	if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
 350		ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
 351			 "with only %d reserved metadata blocks\n", __func__,
 352			 inode->i_ino, ei->i_allocated_meta_blocks,
 353			 ei->i_reserved_meta_blocks);
 354		WARN_ON(1);
 355		ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
 356	}
 357
 358	/* Update per-inode reservations */
 359	ei->i_reserved_data_blocks -= used;
 360	ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
 361	percpu_counter_sub(&sbi->s_dirtyclusters_counter,
 362			   used + ei->i_allocated_meta_blocks);
 363	ei->i_allocated_meta_blocks = 0;
 364
 365	if (ei->i_reserved_data_blocks == 0) {
 366		/*
 367		 * We can release all of the reserved metadata blocks
 368		 * only when we have written all of the delayed
 369		 * allocation blocks.
 370		 */
 371		percpu_counter_sub(&sbi->s_dirtyclusters_counter,
 372				   ei->i_reserved_meta_blocks);
 373		ei->i_reserved_meta_blocks = 0;
 374		ei->i_da_metadata_calc_len = 0;
 375	}
 376	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 377
 378	/* Update quota subsystem for data blocks */
 379	if (quota_claim)
 380		dquot_claim_block(inode, EXT4_C2B(sbi, used));
 381	else {
 382		/*
 383		 * We did fallocate with an offset that is already delayed
 384		 * allocated. So on delayed allocated writeback we should
 385		 * not re-claim the quota for fallocated blocks.
 386		 */
 387		dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
 388	}
 389
 390	/*
 391	 * If we have done all the pending block allocations and if
 392	 * there aren't any writers on the inode, we can discard the
 393	 * inode's preallocations.
 394	 */
 395	if ((ei->i_reserved_data_blocks == 0) &&
 396	    (atomic_read(&inode->i_writecount) == 0))
 397		ext4_discard_preallocations(inode);
 398}
 399
 400static int __check_block_validity(struct inode *inode, const char *func,
 401				unsigned int line,
 402				struct ext4_map_blocks *map)
 403{
 404	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
 405				   map->m_len)) {
 
 
 
 406		ext4_error_inode(inode, func, line, map->m_pblk,
 407				 "lblock %lu mapped to illegal pblock "
 408				 "(length %d)", (unsigned long) map->m_lblk,
 409				 map->m_len);
 410		return -EIO;
 411	}
 412	return 0;
 413}
 414
 415#define check_block_validity(inode, map)	\
 416	__check_block_validity((inode), __func__, __LINE__, (map))
 417
 418/*
 419 * Return the number of contiguous dirty pages in a given inode
 420 * starting at page frame idx.
 421 */
 422static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
 423				    unsigned int max_pages)
 424{
 425	struct address_space *mapping = inode->i_mapping;
 426	pgoff_t	index;
 427	struct pagevec pvec;
 428	pgoff_t num = 0;
 429	int i, nr_pages, done = 0;
 430
 431	if (max_pages == 0)
 432		return 0;
 433	pagevec_init(&pvec, 0);
 434	while (!done) {
 435		index = idx;
 436		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 437					      PAGECACHE_TAG_DIRTY,
 438					      (pgoff_t)PAGEVEC_SIZE);
 439		if (nr_pages == 0)
 440			break;
 441		for (i = 0; i < nr_pages; i++) {
 442			struct page *page = pvec.pages[i];
 443			struct buffer_head *bh, *head;
 444
 445			lock_page(page);
 446			if (unlikely(page->mapping != mapping) ||
 447			    !PageDirty(page) ||
 448			    PageWriteback(page) ||
 449			    page->index != idx) {
 450				done = 1;
 451				unlock_page(page);
 452				break;
 453			}
 454			if (page_has_buffers(page)) {
 455				bh = head = page_buffers(page);
 456				do {
 457					if (!buffer_delay(bh) &&
 458					    !buffer_unwritten(bh))
 459						done = 1;
 460					bh = bh->b_this_page;
 461				} while (!done && (bh != head));
 462			}
 463			unlock_page(page);
 464			if (done)
 465				break;
 466			idx++;
 467			num++;
 468			if (num >= max_pages) {
 469				done = 1;
 470				break;
 471			}
 472		}
 473		pagevec_release(&pvec);
 474	}
 475	return num;
 476}
 477
 478/*
 479 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
 480 */
 481static void set_buffers_da_mapped(struct inode *inode,
 482				   struct ext4_map_blocks *map)
 483{
 484	struct address_space *mapping = inode->i_mapping;
 485	struct pagevec pvec;
 486	int i, nr_pages;
 487	pgoff_t index, end;
 488
 489	index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
 490	end = (map->m_lblk + map->m_len - 1) >>
 491		(PAGE_CACHE_SHIFT - inode->i_blkbits);
 492
 493	pagevec_init(&pvec, 0);
 494	while (index <= end) {
 495		nr_pages = pagevec_lookup(&pvec, mapping, index,
 496					  min(end - index + 1,
 497					      (pgoff_t)PAGEVEC_SIZE));
 498		if (nr_pages == 0)
 499			break;
 500		for (i = 0; i < nr_pages; i++) {
 501			struct page *page = pvec.pages[i];
 502			struct buffer_head *bh, *head;
 503
 504			if (unlikely(page->mapping != mapping) ||
 505			    !PageDirty(page))
 506				break;
 
 
 
 
 
 507
 508			if (page_has_buffers(page)) {
 509				bh = head = page_buffers(page);
 510				do {
 511					set_buffer_da_mapped(bh);
 512					bh = bh->b_this_page;
 513				} while (bh != head);
 514			}
 515			index++;
 516		}
 517		pagevec_release(&pvec);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 518	}
 519}
 
 520
 521/*
 522 * The ext4_map_blocks() function tries to look up the requested blocks,
 523 * and returns if the blocks are already mapped.
 524 *
 525 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 526 * and store the allocated blocks in the result buffer head and mark it
 527 * mapped.
 528 *
 529 * If file type is extents based, it will call ext4_ext_map_blocks(),
 530 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 531 * based files
 532 *
 533 * On success, it returns the number of blocks being mapped or allocate.
 534 * if create==0 and the blocks are pre-allocated and uninitialized block,
 535 * the result buffer head is unmapped. If the create ==1, it will make sure
 536 * the buffer head is mapped.
 537 *
 538 * It returns 0 if plain look up failed (blocks have not been allocated), in
 539 * that case, buffer head is unmapped
 
 540 *
 541 * It returns the error in case of allocation failure.
 542 */
 543int ext4_map_blocks(handle_t *handle, struct inode *inode,
 544		    struct ext4_map_blocks *map, int flags)
 545{
 
 546	int retval;
 
 
 
 
 
 
 547
 548	map->m_flags = 0;
 549	ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
 550		  "logical block %lu\n", inode->i_ino, flags, map->m_len,
 551		  (unsigned long) map->m_lblk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 552	/*
 553	 * Try to see if we can get the block without requesting a new
 554	 * file system block.
 555	 */
 556	down_read((&EXT4_I(inode)->i_data_sem));
 557	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 558		retval = ext4_ext_map_blocks(handle, inode, map, flags &
 559					     EXT4_GET_BLOCKS_KEEP_SIZE);
 560	} else {
 561		retval = ext4_ind_map_blocks(handle, inode, map, flags &
 562					     EXT4_GET_BLOCKS_KEEP_SIZE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 563	}
 564	up_read((&EXT4_I(inode)->i_data_sem));
 565
 
 566	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 567		int ret = check_block_validity(inode, map);
 568		if (ret != 0)
 569			return ret;
 570	}
 571
 572	/* If it is only a block(s) look up */
 573	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 574		return retval;
 575
 576	/*
 577	 * Returns if the blocks have already allocated
 578	 *
 579	 * Note that if blocks have been preallocated
 580	 * ext4_ext_get_block() returns the create = 0
 581	 * with buffer head unmapped.
 582	 */
 583	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 584		return retval;
 
 
 
 
 
 
 585
 586	/*
 587	 * When we call get_blocks without the create flag, the
 588	 * BH_Unwritten flag could have gotten set if the blocks
 589	 * requested were part of a uninitialized extent.  We need to
 590	 * clear this flag now that we are committed to convert all or
 591	 * part of the uninitialized extent to be an initialized
 592	 * extent.  This is because we need to avoid the combination
 593	 * of BH_Unwritten and BH_Mapped flags being simultaneously
 594	 * set on the buffer_head.
 595	 */
 596	map->m_flags &= ~EXT4_MAP_UNWRITTEN;
 597
 598	/*
 599	 * New blocks allocate and/or writing to uninitialized extent
 600	 * will possibly result in updating i_data, so we take
 601	 * the write lock of i_data_sem, and call get_blocks()
 602	 * with create == 1 flag.
 603	 */
 604	down_write((&EXT4_I(inode)->i_data_sem));
 605
 606	/*
 607	 * if the caller is from delayed allocation writeout path
 608	 * we have already reserved fs blocks for allocation
 609	 * let the underlying get_block() function know to
 610	 * avoid double accounting
 611	 */
 612	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 613		ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
 614	/*
 615	 * We need to check for EXT4 here because migrate
 616	 * could have changed the inode type in between
 617	 */
 618	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 619		retval = ext4_ext_map_blocks(handle, inode, map, flags);
 620	} else {
 621		retval = ext4_ind_map_blocks(handle, inode, map, flags);
 622
 623		if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
 624			/*
 625			 * We allocated new blocks which will result in
 626			 * i_data's format changing.  Force the migrate
 627			 * to fail by clearing migrate flags
 628			 */
 629			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 630		}
 
 
 
 
 
 
 
 
 
 
 
 
 631
 632		/*
 633		 * Update reserved blocks/metadata blocks after successful
 634		 * block allocation which had been deferred till now. We don't
 635		 * support fallocate for non extent files. So we can update
 636		 * reserve space here.
 637		 */
 638		if ((retval > 0) &&
 639			(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
 640			ext4_da_update_reserve_space(inode, retval, 1);
 641	}
 642	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
 643		ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
 644
 645		/* If we have successfully mapped the delayed allocated blocks,
 646		 * set the BH_Da_Mapped bit on them. Its important to do this
 647		 * under the protection of i_data_sem.
 
 
 
 
 
 648		 */
 649		if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 650			set_buffers_da_mapped(inode, map);
 
 
 
 
 
 
 
 
 
 
 
 
 651	}
 652
 
 653	up_write((&EXT4_I(inode)->i_data_sem));
 654	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 655		int ret = check_block_validity(inode, map);
 656		if (ret != 0)
 657			return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 658	}
 
 
 
 
 
 
 659	return retval;
 660}
 661
 662/* Maximum number of blocks we map for direct IO at once. */
 663#define DIO_MAX_BLOCKS 4096
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 664
 665static int _ext4_get_block(struct inode *inode, sector_t iblock,
 666			   struct buffer_head *bh, int flags)
 667{
 668	handle_t *handle = ext4_journal_current_handle();
 669	struct ext4_map_blocks map;
 670	int ret = 0, started = 0;
 671	int dio_credits;
 
 
 672
 673	map.m_lblk = iblock;
 674	map.m_len = bh->b_size >> inode->i_blkbits;
 675
 676	if (flags && !handle) {
 677		/* Direct IO write... */
 678		if (map.m_len > DIO_MAX_BLOCKS)
 679			map.m_len = DIO_MAX_BLOCKS;
 680		dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
 681		handle = ext4_journal_start(inode, dio_credits);
 682		if (IS_ERR(handle)) {
 683			ret = PTR_ERR(handle);
 684			return ret;
 685		}
 686		started = 1;
 687	}
 688
 689	ret = ext4_map_blocks(handle, inode, &map, flags);
 690	if (ret > 0) {
 691		map_bh(bh, inode->i_sb, map.m_pblk);
 692		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
 693		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 694		ret = 0;
 
 
 
 695	}
 696	if (started)
 697		ext4_journal_stop(handle);
 698	return ret;
 699}
 700
 701int ext4_get_block(struct inode *inode, sector_t iblock,
 702		   struct buffer_head *bh, int create)
 703{
 704	return _ext4_get_block(inode, iblock, bh,
 705			       create ? EXT4_GET_BLOCKS_CREATE : 0);
 706}
 707
 708/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 709 * `handle' can be NULL if create is zero
 710 */
 711struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 712				ext4_lblk_t block, int create, int *errp)
 713{
 714	struct ext4_map_blocks map;
 715	struct buffer_head *bh;
 716	int fatal = 0, err;
 
 
 717
 718	J_ASSERT(handle != NULL || create == 0);
 
 
 719
 720	map.m_lblk = block;
 721	map.m_len = 1;
 722	err = ext4_map_blocks(handle, inode, &map,
 723			      create ? EXT4_GET_BLOCKS_CREATE : 0);
 724
 
 
 725	if (err < 0)
 726		*errp = err;
 727	if (err <= 0)
 728		return NULL;
 729	*errp = 0;
 730
 731	bh = sb_getblk(inode->i_sb, map.m_pblk);
 732	if (!bh) {
 733		*errp = -EIO;
 734		return NULL;
 735	}
 736	if (map.m_flags & EXT4_MAP_NEW) {
 737		J_ASSERT(create != 0);
 738		J_ASSERT(handle != NULL);
 
 739
 740		/*
 741		 * Now that we do not always journal data, we should
 742		 * keep in mind whether this should always journal the
 743		 * new buffer as metadata.  For now, regular file
 744		 * writes use ext4_get_block instead, so it's not a
 745		 * problem.
 746		 */
 747		lock_buffer(bh);
 748		BUFFER_TRACE(bh, "call get_create_access");
 749		fatal = ext4_journal_get_create_access(handle, bh);
 750		if (!fatal && !buffer_uptodate(bh)) {
 
 
 
 
 
 751			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 752			set_buffer_uptodate(bh);
 753		}
 754		unlock_buffer(bh);
 755		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 756		err = ext4_handle_dirty_metadata(handle, inode, bh);
 757		if (!fatal)
 758			fatal = err;
 759	} else {
 760		BUFFER_TRACE(bh, "not a new buffer");
 761	}
 762	if (fatal) {
 763		*errp = fatal;
 764		brelse(bh);
 765		bh = NULL;
 766	}
 767	return bh;
 
 
 
 768}
 769
 770struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 771			       ext4_lblk_t block, int create, int *err)
 772{
 773	struct buffer_head *bh;
 
 774
 775	bh = ext4_getblk(handle, inode, block, create, err);
 776	if (!bh)
 777		return bh;
 778	if (buffer_uptodate(bh))
 779		return bh;
 780	ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
 781	wait_on_buffer(bh);
 782	if (buffer_uptodate(bh))
 783		return bh;
 784	put_bh(bh);
 785	*err = -EIO;
 786	return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 787}
 788
 789static int walk_page_buffers(handle_t *handle,
 790			     struct buffer_head *head,
 791			     unsigned from,
 792			     unsigned to,
 793			     int *partial,
 794			     int (*fn)(handle_t *handle,
 795				       struct buffer_head *bh))
 796{
 797	struct buffer_head *bh;
 798	unsigned block_start, block_end;
 799	unsigned blocksize = head->b_size;
 800	int err, ret = 0;
 801	struct buffer_head *next;
 802
 803	for (bh = head, block_start = 0;
 804	     ret == 0 && (bh != head || !block_start);
 805	     block_start = block_end, bh = next) {
 806		next = bh->b_this_page;
 807		block_end = block_start + blocksize;
 808		if (block_end <= from || block_start >= to) {
 809			if (partial && !buffer_uptodate(bh))
 810				*partial = 1;
 811			continue;
 812		}
 813		err = (*fn)(handle, bh);
 814		if (!ret)
 815			ret = err;
 816	}
 817	return ret;
 818}
 819
 820/*
 821 * To preserve ordering, it is essential that the hole instantiation and
 822 * the data write be encapsulated in a single transaction.  We cannot
 823 * close off a transaction and start a new one between the ext4_get_block()
 824 * and the commit_write().  So doing the jbd2_journal_start at the start of
 825 * prepare_write() is the right place.
 826 *
 827 * Also, this function can nest inside ext4_writepage() ->
 828 * block_write_full_page(). In that case, we *know* that ext4_writepage()
 829 * has generated enough buffer credits to do the whole page.  So we won't
 830 * block on the journal in that case, which is good, because the caller may
 831 * be PF_MEMALLOC.
 832 *
 833 * By accident, ext4 can be reentered when a transaction is open via
 834 * quota file writes.  If we were to commit the transaction while thus
 835 * reentered, there can be a deadlock - we would be holding a quota
 836 * lock, and the commit would never complete if another thread had a
 837 * transaction open and was blocking on the quota lock - a ranking
 838 * violation.
 839 *
 840 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
 841 * will _not_ run commit under these circumstances because handle->h_ref
 842 * is elevated.  We'll still have enough credits for the tiny quotafile
 843 * write.
 844 */
 845static int do_journal_get_write_access(handle_t *handle,
 846				       struct buffer_head *bh)
 
 
 
 
 
 
 847{
 848	int dirty = buffer_dirty(bh);
 849	int ret;
 850
 851	if (!buffer_mapped(bh) || buffer_freed(bh))
 852		return 0;
 853	/*
 854	 * __block_write_begin() could have dirtied some buffers. Clean
 855	 * the dirty bit as jbd2_journal_get_write_access() could complain
 856	 * otherwise about fs integrity issues. Setting of the dirty bit
 857	 * by __block_write_begin() isn't a real problem here as we clear
 858	 * the bit before releasing a page lock and thus writeback cannot
 859	 * ever write the buffer.
 860	 */
 861	if (dirty)
 862		clear_buffer_dirty(bh);
 863	ret = ext4_journal_get_write_access(handle, bh);
 
 
 864	if (!ret && dirty)
 865		ret = ext4_handle_dirty_metadata(handle, NULL, bh);
 866	return ret;
 867}
 868
 869static int ext4_get_block_write(struct inode *inode, sector_t iblock,
 870		   struct buffer_head *bh_result, int create);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 871static int ext4_write_begin(struct file *file, struct address_space *mapping,
 872			    loff_t pos, unsigned len, unsigned flags,
 873			    struct page **pagep, void **fsdata)
 874{
 875	struct inode *inode = mapping->host;
 876	int ret, needed_blocks;
 877	handle_t *handle;
 878	int retries = 0;
 879	struct page *page;
 880	pgoff_t index;
 881	unsigned from, to;
 882
 883	trace_ext4_write_begin(inode, pos, len, flags);
 
 
 
 884	/*
 885	 * Reserve one block more for addition to orphan list in case
 886	 * we allocate blocks but write fails for some reason
 887	 */
 888	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
 889	index = pos >> PAGE_CACHE_SHIFT;
 890	from = pos & (PAGE_CACHE_SIZE - 1);
 891	to = from + len;
 892
 893retry:
 894	handle = ext4_journal_start(inode, needed_blocks);
 895	if (IS_ERR(handle)) {
 896		ret = PTR_ERR(handle);
 897		goto out;
 
 
 898	}
 899
 900	/* We cannot recurse into the filesystem as the transaction is already
 901	 * started */
 902	flags |= AOP_FLAG_NOFS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 903
 904	page = grab_cache_page_write_begin(mapping, index, flags);
 905	if (!page) {
 
 
 
 906		ext4_journal_stop(handle);
 907		ret = -ENOMEM;
 908		goto out;
 909	}
 910	*pagep = page;
 
 911
 
 912	if (ext4_should_dioread_nolock(inode))
 913		ret = __block_write_begin(page, pos, len, ext4_get_block_write);
 
 914	else
 915		ret = __block_write_begin(page, pos, len, ext4_get_block);
 916
 
 
 
 
 
 
 917	if (!ret && ext4_should_journal_data(inode)) {
 918		ret = walk_page_buffers(handle, page_buffers(page),
 919				from, to, NULL, do_journal_get_write_access);
 
 920	}
 921
 922	if (ret) {
 923		unlock_page(page);
 924		page_cache_release(page);
 
 
 925		/*
 926		 * __block_write_begin may have instantiated a few blocks
 927		 * outside i_size.  Trim these off again. Don't need
 928		 * i_size_read because we hold i_mutex.
 929		 *
 930		 * Add inode to orphan list in case we crash before
 931		 * truncate finishes
 932		 */
 933		if (pos + len > inode->i_size && ext4_can_truncate(inode))
 934			ext4_orphan_add(handle, inode);
 935
 936		ext4_journal_stop(handle);
 937		if (pos + len > inode->i_size) {
 938			ext4_truncate_failed_write(inode);
 939			/*
 940			 * If truncate failed early the inode might
 941			 * still be on the orphan list; we need to
 942			 * make sure the inode is removed from the
 943			 * orphan list in that case.
 944			 */
 945			if (inode->i_nlink)
 946				ext4_orphan_del(NULL, inode);
 947		}
 948	}
 949
 950	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 951		goto retry;
 952out:
 
 
 
 
 953	return ret;
 954}
 955
 956/* For write_end() in data=journal mode */
 957static int write_end_fn(handle_t *handle, struct buffer_head *bh)
 
 958{
 
 959	if (!buffer_mapped(bh) || buffer_freed(bh))
 960		return 0;
 961	set_buffer_uptodate(bh);
 962	return ext4_handle_dirty_metadata(handle, NULL, bh);
 963}
 964
 965static int ext4_generic_write_end(struct file *file,
 966				  struct address_space *mapping,
 967				  loff_t pos, unsigned len, unsigned copied,
 968				  struct page *page, void *fsdata)
 969{
 970	int i_size_changed = 0;
 971	struct inode *inode = mapping->host;
 972	handle_t *handle = ext4_journal_current_handle();
 973
 974	copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 975
 976	/*
 977	 * No need to use i_size_read() here, the i_size
 978	 * cannot change under us because we hold i_mutex.
 979	 *
 980	 * But it's important to update i_size while still holding page lock:
 981	 * page writeout could otherwise come in and zero beyond i_size.
 982	 */
 983	if (pos + copied > inode->i_size) {
 984		i_size_write(inode, pos + copied);
 985		i_size_changed = 1;
 986	}
 987
 988	if (pos + copied >  EXT4_I(inode)->i_disksize) {
 989		/* We need to mark inode dirty even if
 990		 * new_i_size is less that inode->i_size
 991		 * bu greater than i_disksize.(hint delalloc)
 992		 */
 993		ext4_update_i_disksize(inode, (pos + copied));
 994		i_size_changed = 1;
 995	}
 996	unlock_page(page);
 997	page_cache_release(page);
 998
 999	/*
1000	 * Don't mark the inode dirty under page lock. First, it unnecessarily
1001	 * makes the holding time of page lock longer. Second, it forces lock
1002	 * ordering of page lock and transaction start for journaling
1003	 * filesystems.
1004	 */
1005	if (i_size_changed)
1006		ext4_mark_inode_dirty(handle, inode);
1007
1008	return copied;
1009}
1010
1011/*
1012 * We need to pick up the new inode size which generic_commit_write gave us
1013 * `file' can be NULL - eg, when called from page_symlink().
1014 *
1015 * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1016 * buffers are managed internally.
1017 */
1018static int ext4_ordered_write_end(struct file *file,
1019				  struct address_space *mapping,
1020				  loff_t pos, unsigned len, unsigned copied,
1021				  struct page *page, void *fsdata)
1022{
 
1023	handle_t *handle = ext4_journal_current_handle();
1024	struct inode *inode = mapping->host;
 
1025	int ret = 0, ret2;
 
 
1026
1027	trace_ext4_ordered_write_end(inode, pos, len, copied);
1028	ret = ext4_jbd2_file_inode(handle, inode);
1029
1030	if (ret == 0) {
1031		ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1032							page, fsdata);
1033		copied = ret2;
1034		if (pos + len > inode->i_size && ext4_can_truncate(inode))
1035			/* if we have allocated more blocks and copied
1036			 * less. We will have blocks allocated outside
1037			 * inode->i_size. So truncate them
1038			 */
1039			ext4_orphan_add(handle, inode);
1040		if (ret2 < 0)
1041			ret = ret2;
1042	} else {
1043		unlock_page(page);
1044		page_cache_release(page);
1045	}
1046
1047	ret2 = ext4_journal_stop(handle);
1048	if (!ret)
1049		ret = ret2;
1050
1051	if (pos + len > inode->i_size) {
1052		ext4_truncate_failed_write(inode);
1053		/*
1054		 * If truncate failed early the inode might still be
1055		 * on the orphan list; we need to make sure the inode
1056		 * is removed from the orphan list in that case.
1057		 */
1058		if (inode->i_nlink)
1059			ext4_orphan_del(NULL, inode);
1060	}
1061
1062
1063	return ret ? ret : copied;
1064}
 
 
1065
1066static int ext4_writeback_write_end(struct file *file,
1067				    struct address_space *mapping,
1068				    loff_t pos, unsigned len, unsigned copied,
1069				    struct page *page, void *fsdata)
1070{
1071	handle_t *handle = ext4_journal_current_handle();
1072	struct inode *inode = mapping->host;
1073	int ret = 0, ret2;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1074
1075	trace_ext4_writeback_write_end(inode, pos, len, copied);
1076	ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1077							page, fsdata);
1078	copied = ret2;
1079	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1080		/* if we have allocated more blocks and copied
1081		 * less. We will have blocks allocated outside
1082		 * inode->i_size. So truncate them
1083		 */
1084		ext4_orphan_add(handle, inode);
1085
1086	if (ret2 < 0)
1087		ret = ret2;
1088
1089	ret2 = ext4_journal_stop(handle);
1090	if (!ret)
1091		ret = ret2;
1092
1093	if (pos + len > inode->i_size) {
1094		ext4_truncate_failed_write(inode);
1095		/*
1096		 * If truncate failed early the inode might still be
1097		 * on the orphan list; we need to make sure the inode
1098		 * is removed from the orphan list in that case.
1099		 */
1100		if (inode->i_nlink)
1101			ext4_orphan_del(NULL, inode);
1102	}
1103
1104	return ret ? ret : copied;
1105}
1106
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1107static int ext4_journalled_write_end(struct file *file,
1108				     struct address_space *mapping,
1109				     loff_t pos, unsigned len, unsigned copied,
1110				     struct page *page, void *fsdata)
1111{
 
1112	handle_t *handle = ext4_journal_current_handle();
1113	struct inode *inode = mapping->host;
 
1114	int ret = 0, ret2;
1115	int partial = 0;
1116	unsigned from, to;
1117	loff_t new_i_size;
 
1118
1119	trace_ext4_journalled_write_end(inode, pos, len, copied);
1120	from = pos & (PAGE_CACHE_SIZE - 1);
1121	to = from + len;
1122
1123	BUG_ON(!ext4_handle_valid(handle));
1124
1125	if (copied < len) {
1126		if (!PageUptodate(page))
1127			copied = 0;
1128		page_zero_new_buffers(page, from+copied, to);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1129	}
1130
1131	ret = walk_page_buffers(handle, page_buffers(page), from,
1132				to, &partial, write_end_fn);
1133	if (!partial)
1134		SetPageUptodate(page);
1135	new_i_size = pos + copied;
1136	if (new_i_size > inode->i_size)
1137		i_size_write(inode, pos+copied);
1138	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1139	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1140	if (new_i_size > EXT4_I(inode)->i_disksize) {
1141		ext4_update_i_disksize(inode, new_i_size);
 
 
 
 
 
1142		ret2 = ext4_mark_inode_dirty(handle, inode);
1143		if (!ret)
1144			ret = ret2;
1145	}
1146
1147	unlock_page(page);
1148	page_cache_release(page);
1149	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1150		/* if we have allocated more blocks and copied
1151		 * less. We will have blocks allocated outside
1152		 * inode->i_size. So truncate them
1153		 */
1154		ext4_orphan_add(handle, inode);
1155
1156	ret2 = ext4_journal_stop(handle);
1157	if (!ret)
1158		ret = ret2;
1159	if (pos + len > inode->i_size) {
1160		ext4_truncate_failed_write(inode);
1161		/*
1162		 * If truncate failed early the inode might still be
1163		 * on the orphan list; we need to make sure the inode
1164		 * is removed from the orphan list in that case.
1165		 */
1166		if (inode->i_nlink)
1167			ext4_orphan_del(NULL, inode);
1168	}
1169
1170	return ret ? ret : copied;
1171}
1172
1173/*
1174 * Reserve a single cluster located at lblock
1175 */
1176static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1177{
1178	int retries = 0;
1179	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1180	struct ext4_inode_info *ei = EXT4_I(inode);
1181	unsigned int md_needed;
1182	int ret;
1183	ext4_lblk_t save_last_lblock;
1184	int save_len;
1185
1186	/*
1187	 * We will charge metadata quota at writeout time; this saves
1188	 * us from metadata over-estimation, though we may go over by
1189	 * a small amount in the end.  Here we just reserve for data.
1190	 */
1191	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1192	if (ret)
1193		return ret;
1194
1195	/*
1196	 * recalculate the amount of metadata blocks to reserve
1197	 * in order to allocate nrblocks
1198	 * worse case is one extent per block
1199	 */
1200repeat:
1201	spin_lock(&ei->i_block_reservation_lock);
1202	/*
1203	 * ext4_calc_metadata_amount() has side effects, which we have
1204	 * to be prepared undo if we fail to claim space.
1205	 */
1206	save_len = ei->i_da_metadata_calc_len;
1207	save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1208	md_needed = EXT4_NUM_B2C(sbi,
1209				 ext4_calc_metadata_amount(inode, lblock));
1210	trace_ext4_da_reserve_space(inode, md_needed);
1211
1212	/*
1213	 * We do still charge estimated metadata to the sb though;
1214	 * we cannot afford to run out of free blocks.
1215	 */
1216	if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1217		ei->i_da_metadata_calc_len = save_len;
1218		ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1219		spin_unlock(&ei->i_block_reservation_lock);
1220		if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1221			yield();
1222			goto repeat;
1223		}
1224		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1225		return -ENOSPC;
1226	}
1227	ei->i_reserved_data_blocks++;
1228	ei->i_reserved_meta_blocks += md_needed;
1229	spin_unlock(&ei->i_block_reservation_lock);
1230
1231	return 0;       /* success */
1232}
1233
1234static void ext4_da_release_space(struct inode *inode, int to_free)
1235{
1236	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1237	struct ext4_inode_info *ei = EXT4_I(inode);
1238
1239	if (!to_free)
1240		return;		/* Nothing to release, exit */
1241
1242	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1243
1244	trace_ext4_da_release_space(inode, to_free);
1245	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1246		/*
1247		 * if there aren't enough reserved blocks, then the
1248		 * counter is messed up somewhere.  Since this
1249		 * function is called from invalidate page, it's
1250		 * harmless to return without any action.
1251		 */
1252		ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1253			 "ino %lu, to_free %d with only %d reserved "
1254			 "data blocks", inode->i_ino, to_free,
1255			 ei->i_reserved_data_blocks);
1256		WARN_ON(1);
1257		to_free = ei->i_reserved_data_blocks;
1258	}
1259	ei->i_reserved_data_blocks -= to_free;
1260
1261	if (ei->i_reserved_data_blocks == 0) {
1262		/*
1263		 * We can release all of the reserved metadata blocks
1264		 * only when we have written all of the delayed
1265		 * allocation blocks.
1266		 * Note that in case of bigalloc, i_reserved_meta_blocks,
1267		 * i_reserved_data_blocks, etc. refer to number of clusters.
1268		 */
1269		percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1270				   ei->i_reserved_meta_blocks);
1271		ei->i_reserved_meta_blocks = 0;
1272		ei->i_da_metadata_calc_len = 0;
1273	}
1274
1275	/* update fs dirty data blocks counter */
1276	percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1277
1278	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1279
1280	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1281}
1282
1283static void ext4_da_page_release_reservation(struct page *page,
1284					     unsigned long offset)
1285{
1286	int to_release = 0;
1287	struct buffer_head *head, *bh;
1288	unsigned int curr_off = 0;
1289	struct inode *inode = page->mapping->host;
1290	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1291	int num_clusters;
1292
1293	head = page_buffers(page);
1294	bh = head;
1295	do {
1296		unsigned int next_off = curr_off + bh->b_size;
1297
1298		if ((offset <= curr_off) && (buffer_delay(bh))) {
1299			to_release++;
1300			clear_buffer_delay(bh);
1301			clear_buffer_da_mapped(bh);
1302		}
1303		curr_off = next_off;
1304	} while ((bh = bh->b_this_page) != head);
1305
1306	/* If we have released all the blocks belonging to a cluster, then we
1307	 * need to release the reserved space for that cluster. */
1308	num_clusters = EXT4_NUM_B2C(sbi, to_release);
1309	while (num_clusters > 0) {
1310		ext4_fsblk_t lblk;
1311		lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1312			((num_clusters - 1) << sbi->s_cluster_bits);
1313		if (sbi->s_cluster_ratio == 1 ||
1314		    !ext4_find_delalloc_cluster(inode, lblk, 1))
1315			ext4_da_release_space(inode, 1);
1316
1317		num_clusters--;
1318	}
1319}
1320
1321/*
1322 * Delayed allocation stuff
1323 */
1324
1325/*
1326 * mpage_da_submit_io - walks through extent of pages and try to write
1327 * them with writepage() call back
1328 *
1329 * @mpd->inode: inode
1330 * @mpd->first_page: first page of the extent
1331 * @mpd->next_page: page after the last page of the extent
1332 *
1333 * By the time mpage_da_submit_io() is called we expect all blocks
1334 * to be allocated. this may be wrong if allocation failed.
1335 *
1336 * As pages are already locked by write_cache_pages(), we can't use it
1337 */
1338static int mpage_da_submit_io(struct mpage_da_data *mpd,
1339			      struct ext4_map_blocks *map)
1340{
1341	struct pagevec pvec;
1342	unsigned long index, end;
1343	int ret = 0, err, nr_pages, i;
1344	struct inode *inode = mpd->inode;
1345	struct address_space *mapping = inode->i_mapping;
1346	loff_t size = i_size_read(inode);
1347	unsigned int len, block_start;
1348	struct buffer_head *bh, *page_bufs = NULL;
1349	int journal_data = ext4_should_journal_data(inode);
1350	sector_t pblock = 0, cur_logical = 0;
1351	struct ext4_io_submit io_submit;
1352
1353	BUG_ON(mpd->next_page <= mpd->first_page);
1354	memset(&io_submit, 0, sizeof(io_submit));
1355	/*
1356	 * We need to start from the first_page to the next_page - 1
1357	 * to make sure we also write the mapped dirty buffer_heads.
1358	 * If we look at mpd->b_blocknr we would only be looking
1359	 * at the currently mapped buffer_heads.
1360	 */
1361	index = mpd->first_page;
1362	end = mpd->next_page - 1;
1363
1364	pagevec_init(&pvec, 0);
1365	while (index <= end) {
1366		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1367		if (nr_pages == 0)
1368			break;
1369		for (i = 0; i < nr_pages; i++) {
1370			int commit_write = 0, skip_page = 0;
1371			struct page *page = pvec.pages[i];
1372
1373			index = page->index;
1374			if (index > end)
1375				break;
1376
1377			if (index == size >> PAGE_CACHE_SHIFT)
1378				len = size & ~PAGE_CACHE_MASK;
1379			else
1380				len = PAGE_CACHE_SIZE;
1381			if (map) {
1382				cur_logical = index << (PAGE_CACHE_SHIFT -
1383							inode->i_blkbits);
1384				pblock = map->m_pblk + (cur_logical -
1385							map->m_lblk);
1386			}
1387			index++;
1388
1389			BUG_ON(!PageLocked(page));
1390			BUG_ON(PageWriteback(page));
1391
1392			/*
1393			 * If the page does not have buffers (for
1394			 * whatever reason), try to create them using
1395			 * __block_write_begin.  If this fails,
1396			 * skip the page and move on.
1397			 */
1398			if (!page_has_buffers(page)) {
1399				if (__block_write_begin(page, 0, len,
1400						noalloc_get_block_write)) {
1401				skip_page:
1402					unlock_page(page);
1403					continue;
1404				}
1405				commit_write = 1;
1406			}
1407
1408			bh = page_bufs = page_buffers(page);
1409			block_start = 0;
1410			do {
1411				if (!bh)
1412					goto skip_page;
1413				if (map && (cur_logical >= map->m_lblk) &&
1414				    (cur_logical <= (map->m_lblk +
1415						     (map->m_len - 1)))) {
1416					if (buffer_delay(bh)) {
1417						clear_buffer_delay(bh);
1418						bh->b_blocknr = pblock;
1419					}
1420					if (buffer_da_mapped(bh))
1421						clear_buffer_da_mapped(bh);
1422					if (buffer_unwritten(bh) ||
1423					    buffer_mapped(bh))
1424						BUG_ON(bh->b_blocknr != pblock);
1425					if (map->m_flags & EXT4_MAP_UNINIT)
1426						set_buffer_uninit(bh);
1427					clear_buffer_unwritten(bh);
1428				}
1429
1430				/*
1431				 * skip page if block allocation undone and
1432				 * block is dirty
1433				 */
1434				if (ext4_bh_delay_or_unwritten(NULL, bh))
1435					skip_page = 1;
1436				bh = bh->b_this_page;
1437				block_start += bh->b_size;
1438				cur_logical++;
1439				pblock++;
1440			} while (bh != page_bufs);
1441
1442			if (skip_page)
1443				goto skip_page;
1444
1445			if (commit_write)
1446				/* mark the buffer_heads as dirty & uptodate */
1447				block_commit_write(page, 0, len);
1448
1449			clear_page_dirty_for_io(page);
1450			/*
1451			 * Delalloc doesn't support data journalling,
1452			 * but eventually maybe we'll lift this
1453			 * restriction.
1454			 */
1455			if (unlikely(journal_data && PageChecked(page)))
1456				err = __ext4_journalled_writepage(page, len);
1457			else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1458				err = ext4_bio_write_page(&io_submit, page,
1459							  len, mpd->wbc);
1460			else if (buffer_uninit(page_bufs)) {
1461				ext4_set_bh_endio(page_bufs, inode);
1462				err = block_write_full_page_endio(page,
1463					noalloc_get_block_write,
1464					mpd->wbc, ext4_end_io_buffer_write);
1465			} else
1466				err = block_write_full_page(page,
1467					noalloc_get_block_write, mpd->wbc);
1468
1469			if (!err)
1470				mpd->pages_written++;
1471			/*
1472			 * In error case, we have to continue because
1473			 * remaining pages are still locked
1474			 */
1475			if (ret == 0)
1476				ret = err;
1477		}
1478		pagevec_release(&pvec);
1479	}
1480	ext4_io_submit(&io_submit);
1481	return ret;
1482}
 
1483
1484static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
 
1485{
1486	int nr_pages, i;
1487	pgoff_t index, end;
1488	struct pagevec pvec;
1489	struct inode *inode = mpd->inode;
1490	struct address_space *mapping = inode->i_mapping;
1491
 
 
 
 
 
1492	index = mpd->first_page;
1493	end   = mpd->next_page - 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1494	while (index <= end) {
1495		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1496		if (nr_pages == 0)
1497			break;
1498		for (i = 0; i < nr_pages; i++) {
1499			struct page *page = pvec.pages[i];
1500			if (page->index > end)
1501				break;
1502			BUG_ON(!PageLocked(page));
1503			BUG_ON(PageWriteback(page));
1504			block_invalidatepage(page, 0);
1505			ClearPageUptodate(page);
1506			unlock_page(page);
 
 
 
 
 
 
 
 
1507		}
1508		index = pvec.pages[nr_pages - 1]->index + 1;
1509		pagevec_release(&pvec);
1510	}
1511	return;
1512}
1513
1514static void ext4_print_free_blocks(struct inode *inode)
1515{
1516	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1517	struct super_block *sb = inode->i_sb;
 
1518
1519	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1520	       EXT4_C2B(EXT4_SB(inode->i_sb),
1521			ext4_count_free_clusters(inode->i_sb)));
1522	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1523	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1524	       (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1525		percpu_counter_sum(&sbi->s_freeclusters_counter)));
1526	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1527	       (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1528		percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1529	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1530	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1531		 EXT4_I(inode)->i_reserved_data_blocks);
1532	ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1533	       EXT4_I(inode)->i_reserved_meta_blocks);
1534	return;
1535}
1536
1537/*
1538 * mpage_da_map_and_submit - go through given space, map them
1539 *       if necessary, and then submit them for I/O
 
 
1540 *
1541 * @mpd - bh describing space
1542 *
1543 * The function skips space we know is already mapped to disk blocks.
1544 *
 
1545 */
1546static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1547{
1548	int err, blks, get_blocks_flags;
1549	struct ext4_map_blocks map, *mapp = NULL;
1550	sector_t next = mpd->b_blocknr;
1551	unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1552	loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1553	handle_t *handle = NULL;
1554
1555	/*
1556	 * If the blocks are mapped already, or we couldn't accumulate
1557	 * any blocks, then proceed immediately to the submission stage.
1558	 */
1559	if ((mpd->b_size == 0) ||
1560	    ((mpd->b_state  & (1 << BH_Mapped)) &&
1561	     !(mpd->b_state & (1 << BH_Delay)) &&
1562	     !(mpd->b_state & (1 << BH_Unwritten))))
1563		goto submit_io;
1564
1565	handle = ext4_journal_current_handle();
1566	BUG_ON(!handle);
1567
1568	/*
1569	 * Call ext4_map_blocks() to allocate any delayed allocation
1570	 * blocks, or to convert an uninitialized extent to be
1571	 * initialized (in the case where we have written into
1572	 * one or more preallocated blocks).
1573	 *
1574	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1575	 * indicate that we are on the delayed allocation path.  This
1576	 * affects functions in many different parts of the allocation
1577	 * call path.  This flag exists primarily because we don't
1578	 * want to change *many* call functions, so ext4_map_blocks()
1579	 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1580	 * inode's allocation semaphore is taken.
1581	 *
1582	 * If the blocks in questions were delalloc blocks, set
1583	 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1584	 * variables are updated after the blocks have been allocated.
1585	 */
1586	map.m_lblk = next;
1587	map.m_len = max_blocks;
1588	get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1589	if (ext4_should_dioread_nolock(mpd->inode))
1590		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1591	if (mpd->b_state & (1 << BH_Delay))
1592		get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1593
1594	blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1595	if (blks < 0) {
1596		struct super_block *sb = mpd->inode->i_sb;
1597
1598		err = blks;
1599		/*
1600		 * If get block returns EAGAIN or ENOSPC and there
1601		 * appears to be free blocks we will just let
1602		 * mpage_da_submit_io() unlock all of the pages.
1603		 */
1604		if (err == -EAGAIN)
1605			goto submit_io;
1606
1607		if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1608			mpd->retval = err;
1609			goto submit_io;
1610		}
1611
1612		/*
1613		 * get block failure will cause us to loop in
1614		 * writepages, because a_ops->writepage won't be able
1615		 * to make progress. The page will be redirtied by
1616		 * writepage and writepages will again try to write
1617		 * the same.
1618		 */
1619		if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1620			ext4_msg(sb, KERN_CRIT,
1621				 "delayed block allocation failed for inode %lu "
1622				 "at logical offset %llu with max blocks %zd "
1623				 "with error %d", mpd->inode->i_ino,
1624				 (unsigned long long) next,
1625				 mpd->b_size >> mpd->inode->i_blkbits, err);
1626			ext4_msg(sb, KERN_CRIT,
1627				"This should not happen!! Data will be lost\n");
1628			if (err == -ENOSPC)
1629				ext4_print_free_blocks(mpd->inode);
1630		}
1631		/* invalidate all the pages */
1632		ext4_da_block_invalidatepages(mpd);
1633
1634		/* Mark this page range as having been completed */
1635		mpd->io_done = 1;
1636		return;
1637	}
1638	BUG_ON(blks == 0);
1639
1640	mapp = &map;
1641	if (map.m_flags & EXT4_MAP_NEW) {
1642		struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1643		int i;
1644
1645		for (i = 0; i < map.m_len; i++)
1646			unmap_underlying_metadata(bdev, map.m_pblk + i);
1647
1648		if (ext4_should_order_data(mpd->inode)) {
1649			err = ext4_jbd2_file_inode(handle, mpd->inode);
1650			if (err) {
1651				/* Only if the journal is aborted */
1652				mpd->retval = err;
1653				goto submit_io;
1654			}
1655		}
1656	}
1657
1658	/*
1659	 * Update on-disk size along with block allocation.
1660	 */
1661	disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1662	if (disksize > i_size_read(mpd->inode))
1663		disksize = i_size_read(mpd->inode);
1664	if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1665		ext4_update_i_disksize(mpd->inode, disksize);
1666		err = ext4_mark_inode_dirty(handle, mpd->inode);
1667		if (err)
1668			ext4_error(mpd->inode->i_sb,
1669				   "Failed to mark inode %lu dirty",
1670				   mpd->inode->i_ino);
1671	}
1672
1673submit_io:
1674	mpage_da_submit_io(mpd, mapp);
1675	mpd->io_done = 1;
1676}
1677
1678#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1679		(1 << BH_Delay) | (1 << BH_Unwritten))
1680
1681/*
1682 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1683 *
1684 * @mpd->lbh - extent of blocks
1685 * @logical - logical number of the block in the file
1686 * @bh - bh of the block (used to access block's state)
1687 *
1688 * the function is used to collect contig. blocks in same state
1689 */
1690static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1691				   sector_t logical, size_t b_size,
1692				   unsigned long b_state)
1693{
1694	sector_t next;
1695	int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1696
1697	/*
1698	 * XXX Don't go larger than mballoc is willing to allocate
1699	 * This is a stopgap solution.  We eventually need to fold
1700	 * mpage_da_submit_io() into this function and then call
1701	 * ext4_map_blocks() multiple times in a loop
1702	 */
1703	if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1704		goto flush_it;
1705
1706	/* check if thereserved journal credits might overflow */
1707	if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1708		if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1709			/*
1710			 * With non-extent format we are limited by the journal
1711			 * credit available.  Total credit needed to insert
1712			 * nrblocks contiguous blocks is dependent on the
1713			 * nrblocks.  So limit nrblocks.
1714			 */
1715			goto flush_it;
1716		} else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1717				EXT4_MAX_TRANS_DATA) {
1718			/*
1719			 * Adding the new buffer_head would make it cross the
1720			 * allowed limit for which we have journal credit
1721			 * reserved. So limit the new bh->b_size
1722			 */
1723			b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1724						mpd->inode->i_blkbits;
1725			/* we will do mpage_da_submit_io in the next loop */
1726		}
1727	}
1728	/*
1729	 * First block in the extent
1730	 */
1731	if (mpd->b_size == 0) {
1732		mpd->b_blocknr = logical;
1733		mpd->b_size = b_size;
1734		mpd->b_state = b_state & BH_FLAGS;
1735		return;
1736	}
1737
1738	next = mpd->b_blocknr + nrblocks;
1739	/*
1740	 * Can we merge the block to our big extent?
1741	 */
1742	if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1743		mpd->b_size += b_size;
1744		return;
1745	}
1746
1747flush_it:
1748	/*
1749	 * We couldn't merge the block to our extent, so we
1750	 * need to flush current  extent and start new one
1751	 */
1752	mpage_da_map_and_submit(mpd);
1753	return;
1754}
1755
1756static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1757{
1758	return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1759}
1760
1761/*
1762 * This function is grabs code from the very beginning of
1763 * ext4_map_blocks, but assumes that the caller is from delayed write
1764 * time. This function looks up the requested blocks and sets the
1765 * buffer delay bit under the protection of i_data_sem.
1766 */
1767static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1768			      struct ext4_map_blocks *map,
1769			      struct buffer_head *bh)
1770{
 
1771	int retval;
1772	sector_t invalid_block = ~((sector_t) 0xffff);
 
 
 
 
 
1773
1774	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1775		invalid_block = ~0;
1776
1777	map->m_flags = 0;
1778	ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1779		  "logical block %lu\n", inode->i_ino, map->m_len,
1780		  (unsigned long) map->m_lblk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1781	/*
1782	 * Try to see if we can get the block without requesting a new
1783	 * file system block.
1784	 */
1785	down_read((&EXT4_I(inode)->i_data_sem));
1786	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 
 
1787		retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1788	else
1789		retval = ext4_ind_map_blocks(NULL, inode, map, 0);
 
 
 
 
 
 
1790
1791	if (retval == 0) {
1792		/*
1793		 * XXX: __block_prepare_write() unmaps passed block,
1794		 * is it OK?
1795		 */
1796		/* If the block was allocated from previously allocated cluster,
1797		 * then we dont need to reserve it again. */
1798		if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1799			retval = ext4_da_reserve_space(inode, iblock);
1800			if (retval)
1801				/* not enough space to reserve */
1802				goto out_unlock;
1803		}
1804
1805		/* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1806		 * and it should not appear on the bh->b_state.
1807		 */
1808		map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1809
1810		map_bh(bh, inode->i_sb, invalid_block);
1811		set_buffer_new(bh);
1812		set_buffer_delay(bh);
1813	}
 
1814
1815out_unlock:
1816	up_read((&EXT4_I(inode)->i_data_sem));
 
 
 
 
1817
 
 
 
1818	return retval;
1819}
1820
1821/*
1822 * This is a special get_blocks_t callback which is used by
1823 * ext4_da_write_begin().  It will either return mapped block or
1824 * reserve space for a single block.
1825 *
1826 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1827 * We also have b_blocknr = -1 and b_bdev initialized properly
1828 *
1829 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1830 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1831 * initialized properly.
1832 */
1833static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1834				  struct buffer_head *bh, int create)
1835{
1836	struct ext4_map_blocks map;
1837	int ret = 0;
1838
1839	BUG_ON(create == 0);
1840	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1841
1842	map.m_lblk = iblock;
1843	map.m_len = 1;
1844
1845	/*
1846	 * first, we need to know whether the block is allocated already
1847	 * preallocated blocks are unmapped but should treated
1848	 * the same as allocated blocks.
1849	 */
1850	ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1851	if (ret <= 0)
1852		return ret;
1853
1854	map_bh(bh, inode->i_sb, map.m_pblk);
1855	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1856
1857	if (buffer_unwritten(bh)) {
1858		/* A delayed write to unwritten bh should be marked
1859		 * new and mapped.  Mapped ensures that we don't do
1860		 * get_block multiple times when we write to the same
1861		 * offset and new ensures that we do proper zero out
1862		 * for partial write.
1863		 */
1864		set_buffer_new(bh);
1865		set_buffer_mapped(bh);
1866	}
1867	return 0;
1868}
1869
1870/*
1871 * This function is used as a standard get_block_t calback function
1872 * when there is no desire to allocate any blocks.  It is used as a
1873 * callback function for block_write_begin() and block_write_full_page().
1874 * These functions should only try to map a single block at a time.
1875 *
1876 * Since this function doesn't do block allocations even if the caller
1877 * requests it by passing in create=1, it is critically important that
1878 * any caller checks to make sure that any buffer heads are returned
1879 * by this function are either all already mapped or marked for
1880 * delayed allocation before calling  block_write_full_page().  Otherwise,
1881 * b_blocknr could be left unitialized, and the page write functions will
1882 * be taken by surprise.
1883 */
1884static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1885				   struct buffer_head *bh_result, int create)
1886{
1887	BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1888	return _ext4_get_block(inode, iblock, bh_result, 0);
1889}
1890
1891static int bget_one(handle_t *handle, struct buffer_head *bh)
1892{
1893	get_bh(bh);
1894	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1895}
1896
1897static int bput_one(handle_t *handle, struct buffer_head *bh)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1898{
1899	put_bh(bh);
1900	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1901}
1902
1903static int __ext4_journalled_writepage(struct page *page,
1904				       unsigned int len)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1905{
1906	struct address_space *mapping = page->mapping;
1907	struct inode *inode = mapping->host;
1908	struct buffer_head *page_bufs;
1909	handle_t *handle = NULL;
1910	int ret = 0;
1911	int err;
 
 
1912
1913	ClearPageChecked(page);
1914	page_bufs = page_buffers(page);
1915	BUG_ON(!page_bufs);
1916	walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1917	/* As soon as we unlock the page, it can go away, but we have
1918	 * references to buffers so we are safe */
1919	unlock_page(page);
1920
1921	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1922	if (IS_ERR(handle)) {
1923		ret = PTR_ERR(handle);
1924		goto out;
1925	}
1926
1927	BUG_ON(!ext4_handle_valid(handle));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1928
1929	ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1930				do_journal_get_write_access);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1931
1932	err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1933				write_end_fn);
1934	if (ret == 0)
1935		ret = err;
1936	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1937	err = ext4_journal_stop(handle);
1938	if (!ret)
1939		ret = err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1940
1941	walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1942	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1943out:
1944	return ret;
 
 
1945}
1946
1947static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1948static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1949
1950/*
1951 * Note that we don't need to start a transaction unless we're journaling data
1952 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1953 * need to file the inode to the transaction's list in ordered mode because if
1954 * we are writing back data added by write(), the inode is already there and if
1955 * we are writing back data modified via mmap(), no one guarantees in which
1956 * transaction the data will hit the disk. In case we are journaling data, we
1957 * cannot start transaction directly because transaction start ranks above page
1958 * lock so we have to do some magic.
1959 *
1960 * This function can get called via...
1961 *   - ext4_da_writepages after taking page lock (have journal handle)
1962 *   - journal_submit_inode_data_buffers (no journal handle)
1963 *   - shrink_page_list via pdflush (no journal handle)
1964 *   - grab_page_cache when doing write_begin (have journal handle)
1965 *
1966 * We don't do any block allocation in this function. If we have page with
1967 * multiple blocks we need to write those buffer_heads that are mapped. This
1968 * is important for mmaped based write. So if we do with blocksize 1K
1969 * truncate(f, 1024);
1970 * a = mmap(f, 0, 4096);
1971 * a[0] = 'a';
1972 * truncate(f, 4096);
1973 * we have in the page first buffer_head mapped via page_mkwrite call back
1974 * but other buffer_heads would be unmapped but dirty (dirty done via the
1975 * do_wp_page). So writepage should write the first block. If we modify
1976 * the mmap area beyond 1024 we will again get a page_fault and the
1977 * page_mkwrite callback will do the block allocation and mark the
1978 * buffer_heads mapped.
1979 *
1980 * We redirty the page if we have any buffer_heads that is either delay or
1981 * unwritten in the page.
1982 *
1983 * We can get recursively called as show below.
1984 *
1985 *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1986 *		ext4_writepage()
1987 *
1988 * But since we don't do any block allocation we should not deadlock.
1989 * Page also have the dirty flag cleared so we don't get recurive page_lock.
 
 
 
 
 
 
 
1990 */
1991static int ext4_writepage(struct page *page,
1992			  struct writeback_control *wbc)
1993{
1994	int ret = 0, commit_write = 0;
1995	loff_t size;
1996	unsigned int len;
1997	struct buffer_head *page_bufs = NULL;
1998	struct inode *inode = page->mapping->host;
 
 
 
 
1999
2000	trace_ext4_writepage(page);
2001	size = i_size_read(inode);
2002	if (page->index == size >> PAGE_CACHE_SHIFT)
2003		len = size & ~PAGE_CACHE_MASK;
2004	else
2005		len = PAGE_CACHE_SIZE;
 
 
 
 
 
 
2006
2007	/*
2008	 * If the page does not have buffers (for whatever reason),
2009	 * try to create them using __block_write_begin.  If this
2010	 * fails, redirty the page and move on.
2011	 */
2012	if (!page_has_buffers(page)) {
2013		if (__block_write_begin(page, 0, len,
2014					noalloc_get_block_write)) {
2015		redirty_page:
2016			redirty_page_for_writepage(wbc, page);
2017			unlock_page(page);
2018			return 0;
 
 
2019		}
2020		commit_write = 1;
2021	}
2022	page_bufs = page_buffers(page);
2023	if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2024			      ext4_bh_delay_or_unwritten)) {
2025		/*
2026		 * We don't want to do block allocation, so redirty
2027		 * the page and return.  We may reach here when we do
2028		 * a journal commit via journal_submit_inode_data_buffers.
2029		 * We can also reach here via shrink_page_list but it
2030		 * should never be for direct reclaim so warn if that
2031		 * happens
2032		 */
2033		WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2034								PF_MEMALLOC);
2035		goto redirty_page;
2036	}
2037	if (commit_write)
2038		/* now mark the buffer_heads as dirty and uptodate */
2039		block_commit_write(page, 0, len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2040
2041	if (PageChecked(page) && ext4_should_journal_data(inode))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2042		/*
2043		 * It's mmapped pagecache.  Add buffers and journal it.  There
2044		 * doesn't seem much point in redirtying the page here.
2045		 */
2046		return __ext4_journalled_writepage(page, len);
2047
2048	if (buffer_uninit(page_bufs)) {
2049		ext4_set_bh_endio(page_bufs, inode);
2050		ret = block_write_full_page_endio(page, noalloc_get_block_write,
2051					    wbc, ext4_end_io_buffer_write);
2052	} else
2053		ret = block_write_full_page(page, noalloc_get_block_write,
2054					    wbc);
2055
2056	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2057}
2058
2059/*
2060 * This is called via ext4_da_writepages() to
2061 * calculate the total number of credits to reserve to fit
2062 * a single extent allocation into a single transaction,
2063 * ext4_da_writpeages() will loop calling this before
2064 * the block allocation.
2065 */
2066
2067static int ext4_da_writepages_trans_blocks(struct inode *inode)
2068{
2069	int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2070
2071	/*
2072	 * With non-extent format the journal credit needed to
2073	 * insert nrblocks contiguous block is dependent on
2074	 * number of contiguous block. So we will limit
2075	 * number of contiguous block to a sane value
2076	 */
2077	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2078	    (max_blocks > EXT4_MAX_TRANS_DATA))
2079		max_blocks = EXT4_MAX_TRANS_DATA;
2080
2081	return ext4_chunk_trans_blocks(inode, max_blocks);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2082}
2083
2084/*
2085 * write_cache_pages_da - walk the list of dirty pages of the given
2086 * address space and accumulate pages that need writing, and call
2087 * mpage_da_map_and_submit to map a single contiguous memory region
2088 * and then write them.
2089 */
2090static int write_cache_pages_da(struct address_space *mapping,
2091				struct writeback_control *wbc,
2092				struct mpage_da_data *mpd,
2093				pgoff_t *done_index)
2094{
2095	struct buffer_head	*bh, *head;
2096	struct inode		*inode = mapping->host;
2097	struct pagevec		pvec;
2098	unsigned int		nr_pages;
2099	sector_t		logical;
2100	pgoff_t			index, end;
2101	long			nr_to_write = wbc->nr_to_write;
2102	int			i, tag, ret = 0;
2103
2104	memset(mpd, 0, sizeof(struct mpage_da_data));
2105	mpd->wbc = wbc;
2106	mpd->inode = inode;
2107	pagevec_init(&pvec, 0);
2108	index = wbc->range_start >> PAGE_CACHE_SHIFT;
2109	end = wbc->range_end >> PAGE_CACHE_SHIFT;
 
 
 
 
 
 
 
 
2110
2111	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2112		tag = PAGECACHE_TAG_TOWRITE;
2113	else
2114		tag = PAGECACHE_TAG_DIRTY;
2115
2116	*done_index = index;
 
 
 
 
 
 
 
 
2117	while (index <= end) {
2118		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2119			      min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2120		if (nr_pages == 0)
2121			return 0;
2122
2123		for (i = 0; i < nr_pages; i++) {
2124			struct page *page = pvec.pages[i];
2125
2126			/*
2127			 * At this point, the page may be truncated or
2128			 * invalidated (changing page->mapping to NULL), or
2129			 * even swizzled back from swapper_space to tmpfs file
2130			 * mapping. However, page->index will not change
2131			 * because we have a reference on the page.
 
2132			 */
2133			if (page->index > end)
 
 
2134				goto out;
2135
2136			*done_index = page->index + 1;
 
 
2137
 
 
 
 
 
 
 
 
2138			/*
2139			 * If we can't merge this page, and we have
2140			 * accumulated an contiguous region, write it
 
 
 
2141			 */
2142			if ((mpd->next_page != page->index) &&
2143			    (mpd->next_page != mpd->first_page)) {
2144				mpage_da_map_and_submit(mpd);
2145				goto ret_extent_tail;
 
 
2146			}
2147
2148			lock_page(page);
 
2149
2150			/*
2151			 * If the page is no longer dirty, or its
2152			 * mapping no longer corresponds to inode we
2153			 * are writing (which means it has been
2154			 * truncated or invalidated), or the page is
2155			 * already under writeback and we are not
2156			 * doing a data integrity writeback, skip the page
 
2157			 */
2158			if (!PageDirty(page) ||
2159			    (PageWriteback(page) &&
2160			     (wbc->sync_mode == WB_SYNC_NONE)) ||
2161			    unlikely(page->mapping != mapping)) {
2162				unlock_page(page);
2163				continue;
2164			}
2165
2166			wait_on_page_writeback(page);
2167			BUG_ON(PageWriteback(page));
2168
2169			if (mpd->next_page != page->index)
2170				mpd->first_page = page->index;
2171			mpd->next_page = page->index + 1;
2172			logical = (sector_t) page->index <<
2173				(PAGE_CACHE_SHIFT - inode->i_blkbits);
2174
2175			if (!page_has_buffers(page)) {
2176				mpage_add_bh_to_extent(mpd, logical,
2177						       PAGE_CACHE_SIZE,
2178						       (1 << BH_Dirty) | (1 << BH_Uptodate));
2179				if (mpd->io_done)
2180					goto ret_extent_tail;
 
 
 
 
 
 
 
 
 
 
 
2181			} else {
2182				/*
2183				 * Page with regular buffer heads,
2184				 * just add all dirty ones
2185				 */
2186				head = page_buffers(page);
2187				bh = head;
2188				do {
2189					BUG_ON(buffer_locked(bh));
2190					/*
2191					 * We need to try to allocate
2192					 * unmapped blocks in the same page.
2193					 * Otherwise we won't make progress
2194					 * with the page in ext4_writepage
2195					 */
2196					if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2197						mpage_add_bh_to_extent(mpd, logical,
2198								       bh->b_size,
2199								       bh->b_state);
2200						if (mpd->io_done)
2201							goto ret_extent_tail;
2202					} else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2203						/*
2204						 * mapped dirty buffer. We need
2205						 * to update the b_state
2206						 * because we look at b_state
2207						 * in mpage_da_map_blocks.  We
2208						 * don't update b_size because
2209						 * if we find an unmapped
2210						 * buffer_head later we need to
2211						 * use the b_state flag of that
2212						 * buffer_head.
2213						 */
2214						if (mpd->b_size == 0)
2215							mpd->b_state = bh->b_state & BH_FLAGS;
2216					}
2217					logical++;
2218				} while ((bh = bh->b_this_page) != head);
2219			}
2220
2221			if (nr_to_write > 0) {
2222				nr_to_write--;
2223				if (nr_to_write == 0 &&
2224				    wbc->sync_mode == WB_SYNC_NONE)
2225					/*
2226					 * We stop writing back only if we are
2227					 * not doing integrity sync. In case of
2228					 * integrity sync we have to keep going
2229					 * because someone may be concurrently
2230					 * dirtying pages, and we might have
2231					 * synced a lot of newly appeared dirty
2232					 * pages, but have not synced all of the
2233					 * old dirty pages.
2234					 */
2235					goto out;
 
2236			}
2237		}
2238		pagevec_release(&pvec);
2239		cond_resched();
2240	}
 
 
 
2241	return 0;
2242ret_extent_tail:
2243	ret = MPAGE_DA_EXTENT_TAIL;
2244out:
2245	pagevec_release(&pvec);
2246	cond_resched();
2247	return ret;
 
2248}
2249
2250
2251static int ext4_da_writepages(struct address_space *mapping,
2252			      struct writeback_control *wbc)
2253{
2254	pgoff_t	index;
 
 
2255	int range_whole = 0;
 
2256	handle_t *handle = NULL;
2257	struct mpage_da_data mpd;
2258	struct inode *inode = mapping->host;
2259	int pages_written = 0;
2260	unsigned int max_pages;
2261	int range_cyclic, cycled = 1, io_done = 0;
2262	int needed_blocks, ret = 0;
2263	long desired_nr_to_write, nr_to_writebump = 0;
2264	loff_t range_start = wbc->range_start;
2265	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2266	pgoff_t done_index = 0;
2267	pgoff_t end;
2268	struct blk_plug plug;
 
2269
2270	trace_ext4_da_writepages(inode, wbc);
2271
2272	/*
2273	 * No pages to write? This is mainly a kludge to avoid starting
2274	 * a transaction for special inodes like journal inode on last iput()
2275	 * because that could violate lock ordering on umount
2276	 */
2277	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2278		return 0;
2279
2280	/*
2281	 * If the filesystem has aborted, it is read-only, so return
2282	 * right away instead of dumping stack traces later on that
2283	 * will obscure the real source of the problem.  We test
2284	 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2285	 * the latter could be true if the filesystem is mounted
2286	 * read-only, and in that case, ext4_da_writepages should
2287	 * *never* be called, so if that ever happens, we would want
2288	 * the stack trace.
2289	 */
2290	if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2291		return -EROFS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2292
2293	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2294		range_whole = 1;
2295
2296	range_cyclic = wbc->range_cyclic;
2297	if (wbc->range_cyclic) {
2298		index = mapping->writeback_index;
2299		if (index)
2300			cycled = 0;
2301		wbc->range_start = index << PAGE_CACHE_SHIFT;
2302		wbc->range_end  = LLONG_MAX;
2303		wbc->range_cyclic = 0;
2304		end = -1;
2305	} else {
2306		index = wbc->range_start >> PAGE_CACHE_SHIFT;
2307		end = wbc->range_end >> PAGE_CACHE_SHIFT;
2308	}
2309
2310	/*
2311	 * This works around two forms of stupidity.  The first is in
2312	 * the writeback code, which caps the maximum number of pages
2313	 * written to be 1024 pages.  This is wrong on multiple
2314	 * levels; different architectues have a different page size,
2315	 * which changes the maximum amount of data which gets
2316	 * written.  Secondly, 4 megabytes is way too small.  XFS
2317	 * forces this value to be 16 megabytes by multiplying
2318	 * nr_to_write parameter by four, and then relies on its
2319	 * allocator to allocate larger extents to make them
2320	 * contiguous.  Unfortunately this brings us to the second
2321	 * stupidity, which is that ext4's mballoc code only allocates
2322	 * at most 2048 blocks.  So we force contiguous writes up to
2323	 * the number of dirty blocks in the inode, or
2324	 * sbi->max_writeback_mb_bump whichever is smaller.
2325	 */
2326	max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2327	if (!range_cyclic && range_whole) {
2328		if (wbc->nr_to_write == LONG_MAX)
2329			desired_nr_to_write = wbc->nr_to_write;
2330		else
2331			desired_nr_to_write = wbc->nr_to_write * 8;
2332	} else
2333		desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2334							   max_pages);
2335	if (desired_nr_to_write > max_pages)
2336		desired_nr_to_write = max_pages;
2337
2338	if (wbc->nr_to_write < desired_nr_to_write) {
2339		nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2340		wbc->nr_to_write = desired_nr_to_write;
2341	}
2342
 
2343retry:
2344	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2345		tag_pages_for_writeback(mapping, index, end);
2346
2347	blk_start_plug(&plug);
2348	while (!ret && wbc->nr_to_write > 0) {
2349
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2350		/*
2351		 * we  insert one extent at a time. So we need
2352		 * credit needed for single extent allocation.
2353		 * journalled mode is currently not supported
2354		 * by delalloc
 
2355		 */
2356		BUG_ON(ext4_should_journal_data(inode));
2357		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2358
2359		/* start a new transaction*/
2360		handle = ext4_journal_start(inode, needed_blocks);
 
2361		if (IS_ERR(handle)) {
2362			ret = PTR_ERR(handle);
2363			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2364			       "%ld pages, ino %lu; err %d", __func__,
2365				wbc->nr_to_write, inode->i_ino, ret);
2366			blk_finish_plug(&plug);
2367			goto out_writepages;
 
 
2368		}
 
2369
 
 
 
 
 
2370		/*
2371		 * Now call write_cache_pages_da() to find the next
2372		 * contiguous region of logical blocks that need
2373		 * blocks to be allocated by ext4 and submit them.
2374		 */
2375		ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2376		/*
2377		 * If we have a contiguous extent of pages and we
2378		 * haven't done the I/O yet, map the blocks and submit
2379		 * them for I/O.
2380		 */
2381		if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2382			mpage_da_map_and_submit(&mpd);
2383			ret = MPAGE_DA_EXTENT_TAIL;
 
2384		}
2385		trace_ext4_da_write_pages(inode, &mpd);
2386		wbc->nr_to_write -= mpd.pages_written;
 
 
2387
2388		ext4_journal_stop(handle);
 
 
 
 
 
 
 
 
 
 
 
 
2389
2390		if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2391			/* commit the transaction which would
 
2392			 * free blocks released in the transaction
2393			 * and try again
2394			 */
2395			jbd2_journal_force_commit_nested(sbi->s_journal);
2396			ret = 0;
2397		} else if (ret == MPAGE_DA_EXTENT_TAIL) {
2398			/*
2399			 * Got one extent now try with rest of the pages.
2400			 * If mpd.retval is set -EIO, journal is aborted.
2401			 * So we don't need to write any more.
2402			 */
2403			pages_written += mpd.pages_written;
2404			ret = mpd.retval;
2405			io_done = 1;
2406		} else if (wbc->nr_to_write)
2407			/*
2408			 * There is no more writeout needed
2409			 * or we requested for a noblocking writeout
2410			 * and we found the device congested
2411			 */
2412			break;
2413	}
 
2414	blk_finish_plug(&plug);
2415	if (!io_done && !cycled) {
2416		cycled = 1;
2417		index = 0;
2418		wbc->range_start = index << PAGE_CACHE_SHIFT;
2419		wbc->range_end  = mapping->writeback_index - 1;
2420		goto retry;
2421	}
2422
2423	/* Update index */
2424	wbc->range_cyclic = range_cyclic;
2425	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2426		/*
2427		 * set the writeback_index so that range_cyclic
2428		 * mode will write it back later
2429		 */
2430		mapping->writeback_index = done_index;
2431
2432out_writepages:
2433	wbc->nr_to_write -= nr_to_writebump;
2434	wbc->range_start = range_start;
2435	trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2436	return ret;
2437}
2438
2439#define FALL_BACK_TO_NONDELALLOC 1
2440static int ext4_nonda_switch(struct super_block *sb)
2441{
2442	s64 free_blocks, dirty_blocks;
2443	struct ext4_sb_info *sbi = EXT4_SB(sb);
2444
2445	/*
2446	 * switch to non delalloc mode if we are running low
2447	 * on free block. The free block accounting via percpu
2448	 * counters can get slightly wrong with percpu_counter_batch getting
2449	 * accumulated on each CPU without updating global counters
2450	 * Delalloc need an accurate free block accounting. So switch
2451	 * to non delalloc when we are near to error range.
2452	 */
2453	free_blocks  = EXT4_C2B(sbi,
2454		percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2455	dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2456	if (2 * free_blocks < 3 * dirty_blocks ||
2457		free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
 
 
 
 
 
 
 
2458		/*
2459		 * free block count is less than 150% of dirty blocks
2460		 * or free blocks is less than watermark
2461		 */
2462		return 1;
2463	}
2464	/*
2465	 * Even if we don't switch but are nearing capacity,
2466	 * start pushing delalloc when 1/2 of free blocks are dirty.
2467	 */
2468	if (free_blocks < 2 * dirty_blocks)
2469		writeback_inodes_sb_if_idle(sb, WB_REASON_FS_FREE_SPACE);
2470
2471	return 0;
2472}
2473
2474static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2475			       loff_t pos, unsigned len, unsigned flags,
2476			       struct page **pagep, void **fsdata)
2477{
2478	int ret, retries = 0;
2479	struct page *page;
2480	pgoff_t index;
2481	struct inode *inode = mapping->host;
2482	handle_t *handle;
2483
2484	index = pos >> PAGE_CACHE_SHIFT;
 
 
 
2485
2486	if (ext4_nonda_switch(inode->i_sb)) {
2487		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2488		return ext4_write_begin(file, mapping, pos,
2489					len, flags, pagep, fsdata);
2490	}
2491	*fsdata = (void *)0;
2492	trace_ext4_da_write_begin(inode, pos, len, flags);
2493retry:
2494	/*
2495	 * With delayed allocation, we don't log the i_disksize update
2496	 * if there is delayed block allocation. But we still need
2497	 * to journalling the i_disksize update if writes to the end
2498	 * of file which has an already mapped buffer.
2499	 */
2500	handle = ext4_journal_start(inode, 1);
2501	if (IS_ERR(handle)) {
2502		ret = PTR_ERR(handle);
2503		goto out;
2504	}
2505	/* We cannot recurse into the filesystem as the transaction is already
2506	 * started */
2507	flags |= AOP_FLAG_NOFS;
2508
2509	page = grab_cache_page_write_begin(mapping, index, flags);
2510	if (!page) {
2511		ext4_journal_stop(handle);
2512		ret = -ENOMEM;
2513		goto out;
 
 
2514	}
2515	*pagep = page;
2516
2517	ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
 
 
 
 
 
 
 
 
 
 
2518	if (ret < 0) {
2519		unlock_page(page);
2520		ext4_journal_stop(handle);
2521		page_cache_release(page);
2522		/*
2523		 * block_write_begin may have instantiated a few blocks
2524		 * outside i_size.  Trim these off again. Don't need
2525		 * i_size_read because we hold i_mutex.
2526		 */
2527		if (pos + len > inode->i_size)
2528			ext4_truncate_failed_write(inode);
 
 
 
 
 
2529	}
2530
2531	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2532		goto retry;
2533out:
2534	return ret;
2535}
2536
2537/*
2538 * Check if we should update i_disksize
2539 * when write to the end of file but not require block allocation
2540 */
2541static int ext4_da_should_update_i_disksize(struct page *page,
2542					    unsigned long offset)
2543{
2544	struct buffer_head *bh;
2545	struct inode *inode = page->mapping->host;
2546	unsigned int idx;
2547	int i;
2548
2549	bh = page_buffers(page);
2550	idx = offset >> inode->i_blkbits;
2551
2552	for (i = 0; i < idx; i++)
2553		bh = bh->b_this_page;
2554
2555	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2556		return 0;
2557	return 1;
2558}
2559
2560static int ext4_da_write_end(struct file *file,
2561			     struct address_space *mapping,
2562			     loff_t pos, unsigned len, unsigned copied,
2563			     struct page *page, void *fsdata)
2564{
2565	struct inode *inode = mapping->host;
2566	int ret = 0, ret2;
2567	handle_t *handle = ext4_journal_current_handle();
2568	loff_t new_i_size;
2569	unsigned long start, end;
2570	int write_mode = (int)(unsigned long)fsdata;
2571
2572	if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2573		switch (ext4_inode_journal_mode(inode)) {
2574		case EXT4_INODE_ORDERED_DATA_MODE:
2575			return ext4_ordered_write_end(file, mapping, pos,
2576					len, copied, page, fsdata);
2577		case EXT4_INODE_WRITEBACK_DATA_MODE:
2578			return ext4_writeback_write_end(file, mapping, pos,
2579					len, copied, page, fsdata);
2580		default:
2581			BUG();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2582		}
2583	}
2584
2585	trace_ext4_da_write_end(inode, pos, len, copied);
2586	start = pos & (PAGE_CACHE_SIZE - 1);
2587	end = start + copied - 1;
2588
2589	/*
2590	 * generic_write_end() will run mark_inode_dirty() if i_size
2591	 * changes.  So let's piggyback the i_disksize mark_inode_dirty
2592	 * into that.
2593	 */
2594
2595	new_i_size = pos + copied;
2596	if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2597		if (ext4_da_should_update_i_disksize(page, end)) {
2598			down_write(&EXT4_I(inode)->i_data_sem);
2599			if (new_i_size > EXT4_I(inode)->i_disksize) {
2600				/*
2601				 * Updating i_disksize when extending file
2602				 * without needing block allocation
2603				 */
2604				if (ext4_should_order_data(inode))
2605					ret = ext4_jbd2_file_inode(handle,
2606								   inode);
2607
2608				EXT4_I(inode)->i_disksize = new_i_size;
2609			}
2610			up_write(&EXT4_I(inode)->i_data_sem);
2611			/* We need to mark inode dirty even if
2612			 * new_i_size is less that inode->i_size
2613			 * bu greater than i_disksize.(hint delalloc)
2614			 */
2615			ext4_mark_inode_dirty(handle, inode);
2616		}
2617	}
2618	ret2 = generic_write_end(file, mapping, pos, len, copied,
2619							page, fsdata);
2620	copied = ret2;
2621	if (ret2 < 0)
2622		ret = ret2;
2623	ret2 = ext4_journal_stop(handle);
2624	if (!ret)
2625		ret = ret2;
2626
2627	return ret ? ret : copied;
2628}
2629
2630static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
 
 
 
2631{
2632	/*
2633	 * Drop reserved blocks
2634	 */
2635	BUG_ON(!PageLocked(page));
2636	if (!page_has_buffers(page))
2637		goto out;
2638
2639	ext4_da_page_release_reservation(page, offset);
 
 
2640
2641out:
2642	ext4_invalidatepage(page, offset);
2643
2644	return;
 
 
 
 
 
 
 
 
 
2645}
2646
2647/*
2648 * Force all delayed allocation blocks to be allocated for a given inode.
2649 */
2650int ext4_alloc_da_blocks(struct inode *inode)
2651{
2652	trace_ext4_alloc_da_blocks(inode);
2653
2654	if (!EXT4_I(inode)->i_reserved_data_blocks &&
2655	    !EXT4_I(inode)->i_reserved_meta_blocks)
2656		return 0;
2657
2658	/*
2659	 * We do something simple for now.  The filemap_flush() will
2660	 * also start triggering a write of the data blocks, which is
2661	 * not strictly speaking necessary (and for users of
2662	 * laptop_mode, not even desirable).  However, to do otherwise
2663	 * would require replicating code paths in:
2664	 *
2665	 * ext4_da_writepages() ->
2666	 *    write_cache_pages() ---> (via passed in callback function)
2667	 *        __mpage_da_writepage() -->
2668	 *           mpage_add_bh_to_extent()
2669	 *           mpage_da_map_blocks()
2670	 *
2671	 * The problem is that write_cache_pages(), located in
2672	 * mm/page-writeback.c, marks pages clean in preparation for
2673	 * doing I/O, which is not desirable if we're not planning on
2674	 * doing I/O at all.
2675	 *
2676	 * We could call write_cache_pages(), and then redirty all of
2677	 * the pages by calling redirty_page_for_writepage() but that
2678	 * would be ugly in the extreme.  So instead we would need to
2679	 * replicate parts of the code in the above functions,
2680	 * simplifying them because we wouldn't actually intend to
2681	 * write out the pages, but rather only collect contiguous
2682	 * logical block extents, call the multi-block allocator, and
2683	 * then update the buffer heads with the block allocations.
2684	 *
2685	 * For now, though, we'll cheat by calling filemap_flush(),
2686	 * which will map the blocks, and start the I/O, but not
2687	 * actually wait for the I/O to complete.
2688	 */
2689	return filemap_flush(inode->i_mapping);
2690}
2691
2692/*
2693 * bmap() is special.  It gets used by applications such as lilo and by
2694 * the swapper to find the on-disk block of a specific piece of data.
2695 *
2696 * Naturally, this is dangerous if the block concerned is still in the
2697 * journal.  If somebody makes a swapfile on an ext4 data-journaling
2698 * filesystem and enables swap, then they may get a nasty shock when the
2699 * data getting swapped to that swapfile suddenly gets overwritten by
2700 * the original zero's written out previously to the journal and
2701 * awaiting writeback in the kernel's buffer cache.
2702 *
2703 * So, if we see any bmap calls here on a modified, data-journaled file,
2704 * take extra steps to flush any blocks which might be in the cache.
2705 */
2706static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2707{
2708	struct inode *inode = mapping->host;
2709	journal_t *journal;
2710	int err;
 
 
 
 
 
 
2711
2712	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2713			test_opt(inode->i_sb, DELALLOC)) {
 
2714		/*
2715		 * With delalloc we want to sync the file
2716		 * so that we can make sure we allocate
2717		 * blocks for file
2718		 */
2719		filemap_write_and_wait(mapping);
2720	}
2721
2722	if (EXT4_JOURNAL(inode) &&
2723	    ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2724		/*
2725		 * This is a REALLY heavyweight approach, but the use of
2726		 * bmap on dirty files is expected to be extremely rare:
2727		 * only if we run lilo or swapon on a freshly made file
2728		 * do we expect this to happen.
2729		 *
2730		 * (bmap requires CAP_SYS_RAWIO so this does not
2731		 * represent an unprivileged user DOS attack --- we'd be
2732		 * in trouble if mortal users could trigger this path at
2733		 * will.)
2734		 *
2735		 * NB. EXT4_STATE_JDATA is not set on files other than
2736		 * regular files.  If somebody wants to bmap a directory
2737		 * or symlink and gets confused because the buffer
2738		 * hasn't yet been flushed to disk, they deserve
2739		 * everything they get.
2740		 */
2741
2742		ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2743		journal = EXT4_JOURNAL(inode);
2744		jbd2_journal_lock_updates(journal);
2745		err = jbd2_journal_flush(journal);
2746		jbd2_journal_unlock_updates(journal);
2747
2748		if (err)
2749			return 0;
2750	}
2751
2752	return generic_block_bmap(mapping, block, ext4_get_block);
2753}
2754
2755static int ext4_readpage(struct file *file, struct page *page)
2756{
2757	trace_ext4_readpage(page);
2758	return mpage_readpage(page, ext4_get_block);
 
 
 
 
 
 
 
 
 
 
2759}
2760
2761static int
2762ext4_readpages(struct file *file, struct address_space *mapping,
2763		struct list_head *pages, unsigned nr_pages)
2764{
2765	return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
 
 
 
 
 
 
2766}
2767
2768static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
 
2769{
2770	struct buffer_head *head, *bh;
2771	unsigned int curr_off = 0;
2772
2773	if (!page_has_buffers(page))
2774		return;
2775	head = bh = page_buffers(page);
2776	do {
2777		if (offset <= curr_off && test_clear_buffer_uninit(bh)
2778					&& bh->b_private) {
2779			ext4_free_io_end(bh->b_private);
2780			bh->b_private = NULL;
2781			bh->b_end_io = NULL;
2782		}
2783		curr_off = curr_off + bh->b_size;
2784		bh = bh->b_this_page;
2785	} while (bh != head);
2786}
2787
2788static void ext4_invalidatepage(struct page *page, unsigned long offset)
 
2789{
2790	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2791
2792	trace_ext4_invalidatepage(page, offset);
2793
2794	/*
2795	 * free any io_end structure allocated for buffers to be discarded
2796	 */
2797	if (ext4_should_dioread_nolock(page->mapping->host))
2798		ext4_invalidatepage_free_endio(page, offset);
2799	/*
2800	 * If it's a full truncate we just forget about the pending dirtying
2801	 */
2802	if (offset == 0)
2803		ClearPageChecked(page);
2804
2805	if (journal)
2806		jbd2_journal_invalidatepage(journal, page, offset);
2807	else
2808		block_invalidatepage(page, offset);
2809}
2810
2811static int ext4_releasepage(struct page *page, gfp_t wait)
 
 
 
2812{
2813	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
 
 
 
 
 
 
2814
2815	trace_ext4_releasepage(page);
2816
2817	WARN_ON(PageChecked(page));
2818	if (!page_has_buffers(page))
2819		return 0;
2820	if (journal)
2821		return jbd2_journal_try_to_free_buffers(journal, page, wait);
2822	else
2823		return try_to_free_buffers(page);
2824}
2825
2826/*
2827 * ext4_get_block used when preparing for a DIO write or buffer write.
2828 * We allocate an uinitialized extent if blocks haven't been allocated.
2829 * The extent will be converted to initialized after the IO is complete.
2830 */
2831static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2832		   struct buffer_head *bh_result, int create)
2833{
2834	ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2835		   inode->i_ino, create);
2836	return _ext4_get_block(inode, iblock, bh_result,
2837			       EXT4_GET_BLOCKS_IO_CREATE_EXT);
2838}
2839
2840static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2841			    ssize_t size, void *private, int ret,
2842			    bool is_async)
2843{
2844	struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2845        ext4_io_end_t *io_end = iocb->private;
2846	struct workqueue_struct *wq;
2847	unsigned long flags;
2848	struct ext4_inode_info *ei;
2849
2850	/* if not async direct IO or dio with 0 bytes write, just return */
2851	if (!io_end || !size)
2852		goto out;
 
 
2853
2854	ext_debug("ext4_end_io_dio(): io_end 0x%p "
2855		  "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2856 		  iocb->private, io_end->inode->i_ino, iocb, offset,
2857		  size);
2858
2859	iocb->private = NULL;
2860
2861	/* if not aio dio with unwritten extents, just free io and return */
2862	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2863		ext4_free_io_end(io_end);
2864out:
2865		if (is_async)
2866			aio_complete(iocb, ret, 0);
2867		inode_dio_done(inode);
2868		return;
2869	}
2870
2871	io_end->offset = offset;
2872	io_end->size = size;
2873	if (is_async) {
2874		io_end->iocb = iocb;
2875		io_end->result = ret;
2876	}
2877	wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
 
 
2878
2879	/* Add the io_end to per-inode completed aio dio list*/
2880	ei = EXT4_I(io_end->inode);
2881	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2882	list_add_tail(&io_end->list, &ei->i_completed_io_list);
2883	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2884
2885	/* queue the work to convert unwritten extents to written */
2886	queue_work(wq, &io_end->work);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2887}
2888
2889static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
 
2890{
2891	ext4_io_end_t *io_end = bh->b_private;
2892	struct workqueue_struct *wq;
2893	struct inode *inode;
2894	unsigned long flags;
2895
2896	if (!test_clear_buffer_uninit(bh) || !io_end)
2897		goto out;
2898
2899	if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2900		ext4_msg(io_end->inode->i_sb, KERN_INFO,
2901			 "sb umounted, discard end_io request for inode %lu",
2902			 io_end->inode->i_ino);
2903		ext4_free_io_end(io_end);
2904		goto out;
2905	}
2906
 
2907	/*
2908	 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2909	 * but being more careful is always safe for the future change.
 
 
2910	 */
2911	inode = io_end->inode;
2912	ext4_set_io_unwritten_flag(inode, io_end);
 
2913
2914	/* Add the io_end to per-inode completed io list*/
2915	spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2916	list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2917	spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2918
2919	wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2920	/* queue the work to convert unwritten extents to written */
2921	queue_work(wq, &io_end->work);
2922out:
2923	bh->b_private = NULL;
2924	bh->b_end_io = NULL;
2925	clear_buffer_uninit(bh);
2926	end_buffer_async_write(bh, uptodate);
2927}
 
 
 
2928
2929static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2930{
2931	ext4_io_end_t *io_end;
2932	struct page *page = bh->b_page;
2933	loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2934	size_t size = bh->b_size;
2935
2936retry:
2937	io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2938	if (!io_end) {
2939		pr_warn_ratelimited("%s: allocation fail\n", __func__);
2940		schedule();
2941		goto retry;
2942	}
2943	io_end->offset = offset;
2944	io_end->size = size;
2945	/*
2946	 * We need to hold a reference to the page to make sure it
2947	 * doesn't get evicted before ext4_end_io_work() has a chance
2948	 * to convert the extent from written to unwritten.
2949	 */
2950	io_end->page = page;
2951	get_page(io_end->page);
2952
2953	bh->b_private = io_end;
2954	bh->b_end_io = ext4_end_io_buffer_write;
2955	return 0;
 
 
2956}
2957
2958/*
2959 * For ext4 extent files, ext4 will do direct-io write to holes,
2960 * preallocated extents, and those write extend the file, no need to
2961 * fall back to buffered IO.
2962 *
2963 * For holes, we fallocate those blocks, mark them as uninitialized
2964 * If those blocks were preallocated, we mark sure they are splited, but
2965 * still keep the range to write as uninitialized.
2966 *
2967 * The unwrritten extents will be converted to written when DIO is completed.
2968 * For async direct IO, since the IO may still pending when return, we
2969 * set up an end_io call back function, which will do the conversion
2970 * when async direct IO completed.
2971 *
2972 * If the O_DIRECT write will extend the file then add this inode to the
2973 * orphan list.  So recovery will truncate it back to the original size
2974 * if the machine crashes during the write.
2975 *
2976 */
2977static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2978			      const struct iovec *iov, loff_t offset,
2979			      unsigned long nr_segs)
2980{
2981	struct file *file = iocb->ki_filp;
2982	struct inode *inode = file->f_mapping->host;
2983	ssize_t ret;
2984	size_t count = iov_length(iov, nr_segs);
2985
2986	loff_t final_size = offset + count;
2987	if (rw == WRITE && final_size <= inode->i_size) {
2988		/*
2989 		 * We could direct write to holes and fallocate.
2990		 *
2991 		 * Allocated blocks to fill the hole are marked as uninitialized
2992 		 * to prevent parallel buffered read to expose the stale data
2993 		 * before DIO complete the data IO.
2994		 *
2995 		 * As to previously fallocated extents, ext4 get_block
2996 		 * will just simply mark the buffer mapped but still
2997 		 * keep the extents uninitialized.
2998 		 *
2999		 * for non AIO case, we will convert those unwritten extents
3000		 * to written after return back from blockdev_direct_IO.
3001		 *
3002		 * for async DIO, the conversion needs to be defered when
3003		 * the IO is completed. The ext4 end_io callback function
3004		 * will be called to take care of the conversion work.
3005		 * Here for async case, we allocate an io_end structure to
3006		 * hook to the iocb.
3007 		 */
3008		iocb->private = NULL;
3009		EXT4_I(inode)->cur_aio_dio = NULL;
3010		if (!is_sync_kiocb(iocb)) {
3011			ext4_io_end_t *io_end =
3012				ext4_init_io_end(inode, GFP_NOFS);
3013			if (!io_end)
3014				return -ENOMEM;
3015			io_end->flag |= EXT4_IO_END_DIRECT;
3016			iocb->private = io_end;
3017			/*
3018			 * we save the io structure for current async
3019			 * direct IO, so that later ext4_map_blocks()
3020			 * could flag the io structure whether there
3021			 * is a unwritten extents needs to be converted
3022			 * when IO is completed.
3023			 */
3024			EXT4_I(inode)->cur_aio_dio = iocb->private;
3025		}
3026
3027		ret = __blockdev_direct_IO(rw, iocb, inode,
3028					 inode->i_sb->s_bdev, iov,
3029					 offset, nr_segs,
3030					 ext4_get_block_write,
3031					 ext4_end_io_dio,
3032					 NULL,
3033					 DIO_LOCKING);
3034		if (iocb->private)
3035			EXT4_I(inode)->cur_aio_dio = NULL;
 
 
 
 
 
3036		/*
3037		 * The io_end structure takes a reference to the inode,
3038		 * that structure needs to be destroyed and the
3039		 * reference to the inode need to be dropped, when IO is
3040		 * complete, even with 0 byte write, or failed.
3041		 *
3042		 * In the successful AIO DIO case, the io_end structure will be
3043		 * desctroyed and the reference to the inode will be dropped
3044		 * after the end_io call back function is called.
3045		 *
3046		 * In the case there is 0 byte write, or error case, since
3047		 * VFS direct IO won't invoke the end_io call back function,
3048		 * we need to free the end_io structure here.
3049		 */
3050		if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3051			ext4_free_io_end(iocb->private);
3052			iocb->private = NULL;
3053		} else if (ret > 0 && ext4_test_inode_state(inode,
3054						EXT4_STATE_DIO_UNWRITTEN)) {
3055			int err;
3056			/*
3057			 * for non AIO case, since the IO is already
3058			 * completed, we could do the conversion right here
3059			 */
3060			err = ext4_convert_unwritten_extents(inode,
3061							     offset, ret);
3062			if (err < 0)
3063				ret = err;
3064			ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3065		}
3066		return ret;
 
 
3067	}
3068
3069	/* for write the the end of file case, we fall back to old way */
3070	return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
 
 
 
 
 
 
 
 
 
 
 
3071}
3072
3073static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3074			      const struct iovec *iov, loff_t offset,
3075			      unsigned long nr_segs)
3076{
3077	struct file *file = iocb->ki_filp;
3078	struct inode *inode = file->f_mapping->host;
3079	ssize_t ret;
3080
3081	/*
3082	 * If we are doing data journalling we don't support O_DIRECT
 
3083	 */
3084	if (ext4_should_journal_data(inode))
3085		return 0;
3086
3087	trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3088	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3089		ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3090	else
3091		ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3092	trace_ext4_direct_IO_exit(inode, offset,
3093				iov_length(iov, nr_segs), rw, ret);
3094	return ret;
3095}
3096
3097/*
3098 * Pages can be marked dirty completely asynchronously from ext4's journalling
3099 * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
3100 * much here because ->set_page_dirty is called under VFS locks.  The page is
3101 * not necessarily locked.
3102 *
3103 * We cannot just dirty the page and leave attached buffers clean, because the
3104 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3105 * or jbddirty because all the journalling code will explode.
3106 *
3107 * So what we do is to mark the page "pending dirty" and next time writepage
3108 * is called, propagate that into the buffers appropriately.
3109 */
3110static int ext4_journalled_set_page_dirty(struct page *page)
3111{
3112	SetPageChecked(page);
3113	return __set_page_dirty_nobuffers(page);
 
 
 
 
 
 
 
 
 
3114}
3115
3116static const struct address_space_operations ext4_ordered_aops = {
3117	.readpage		= ext4_readpage,
3118	.readpages		= ext4_readpages,
3119	.writepage		= ext4_writepage,
3120	.write_begin		= ext4_write_begin,
3121	.write_end		= ext4_ordered_write_end,
3122	.bmap			= ext4_bmap,
3123	.invalidatepage		= ext4_invalidatepage,
3124	.releasepage		= ext4_releasepage,
3125	.direct_IO		= ext4_direct_IO,
3126	.migratepage		= buffer_migrate_page,
3127	.is_partially_uptodate  = block_is_partially_uptodate,
3128	.error_remove_page	= generic_error_remove_page,
3129};
3130
3131static const struct address_space_operations ext4_writeback_aops = {
3132	.readpage		= ext4_readpage,
3133	.readpages		= ext4_readpages,
3134	.writepage		= ext4_writepage,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3135	.write_begin		= ext4_write_begin,
3136	.write_end		= ext4_writeback_write_end,
 
3137	.bmap			= ext4_bmap,
3138	.invalidatepage		= ext4_invalidatepage,
3139	.releasepage		= ext4_releasepage,
3140	.direct_IO		= ext4_direct_IO,
3141	.migratepage		= buffer_migrate_page,
3142	.is_partially_uptodate  = block_is_partially_uptodate,
3143	.error_remove_page	= generic_error_remove_page,
 
3144};
3145
3146static const struct address_space_operations ext4_journalled_aops = {
3147	.readpage		= ext4_readpage,
3148	.readpages		= ext4_readpages,
3149	.writepage		= ext4_writepage,
3150	.write_begin		= ext4_write_begin,
3151	.write_end		= ext4_journalled_write_end,
3152	.set_page_dirty		= ext4_journalled_set_page_dirty,
3153	.bmap			= ext4_bmap,
3154	.invalidatepage		= ext4_invalidatepage,
3155	.releasepage		= ext4_releasepage,
3156	.direct_IO		= ext4_direct_IO,
 
3157	.is_partially_uptodate  = block_is_partially_uptodate,
3158	.error_remove_page	= generic_error_remove_page,
 
3159};
3160
3161static const struct address_space_operations ext4_da_aops = {
3162	.readpage		= ext4_readpage,
3163	.readpages		= ext4_readpages,
3164	.writepage		= ext4_writepage,
3165	.writepages		= ext4_da_writepages,
3166	.write_begin		= ext4_da_write_begin,
3167	.write_end		= ext4_da_write_end,
 
3168	.bmap			= ext4_bmap,
3169	.invalidatepage		= ext4_da_invalidatepage,
3170	.releasepage		= ext4_releasepage,
3171	.direct_IO		= ext4_direct_IO,
3172	.migratepage		= buffer_migrate_page,
3173	.is_partially_uptodate  = block_is_partially_uptodate,
3174	.error_remove_page	= generic_error_remove_page,
 
 
 
 
 
 
 
 
 
3175};
3176
3177void ext4_set_aops(struct inode *inode)
3178{
3179	switch (ext4_inode_journal_mode(inode)) {
3180	case EXT4_INODE_ORDERED_DATA_MODE:
3181		if (test_opt(inode->i_sb, DELALLOC))
3182			inode->i_mapping->a_ops = &ext4_da_aops;
3183		else
3184			inode->i_mapping->a_ops = &ext4_ordered_aops;
3185		break;
3186	case EXT4_INODE_WRITEBACK_DATA_MODE:
3187		if (test_opt(inode->i_sb, DELALLOC))
3188			inode->i_mapping->a_ops = &ext4_da_aops;
3189		else
3190			inode->i_mapping->a_ops = &ext4_writeback_aops;
3191		break;
3192	case EXT4_INODE_JOURNAL_DATA_MODE:
3193		inode->i_mapping->a_ops = &ext4_journalled_aops;
3194		break;
3195	default:
3196		BUG();
3197	}
 
 
 
 
 
 
3198}
3199
3200
3201/*
3202 * ext4_discard_partial_page_buffers()
3203 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3204 * This function finds and locks the page containing the offset
3205 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3206 * Calling functions that already have the page locked should call
3207 * ext4_discard_partial_page_buffers_no_lock directly.
3208 */
3209int ext4_discard_partial_page_buffers(handle_t *handle,
3210		struct address_space *mapping, loff_t from,
3211		loff_t length, int flags)
3212{
3213	struct inode *inode = mapping->host;
3214	struct page *page;
3215	int err = 0;
3216
3217	page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3218				   mapping_gfp_mask(mapping) & ~__GFP_FS);
3219	if (!page)
3220		return -ENOMEM;
3221
3222	err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3223		from, length, flags);
3224
3225	unlock_page(page);
3226	page_cache_release(page);
3227	return err;
3228}
3229
3230/*
3231 * ext4_discard_partial_page_buffers_no_lock()
3232 * Zeros a page range of length 'length' starting from offset 'from'.
3233 * Buffer heads that correspond to the block aligned regions of the
3234 * zeroed range will be unmapped.  Unblock aligned regions
3235 * will have the corresponding buffer head mapped if needed so that
3236 * that region of the page can be updated with the partial zero out.
3237 *
3238 * This function assumes that the page has already been  locked.  The
3239 * The range to be discarded must be contained with in the given page.
3240 * If the specified range exceeds the end of the page it will be shortened
3241 * to the end of the page that corresponds to 'from'.  This function is
3242 * appropriate for updating a page and it buffer heads to be unmapped and
3243 * zeroed for blocks that have been either released, or are going to be
3244 * released.
3245 *
3246 * handle: The journal handle
3247 * inode:  The files inode
3248 * page:   A locked page that contains the offset "from"
3249 * from:   The starting byte offset (from the begining of the file)
3250 *         to begin discarding
3251 * len:    The length of bytes to discard
3252 * flags:  Optional flags that may be used:
3253 *
3254 *         EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3255 *         Only zero the regions of the page whose buffer heads
3256 *         have already been unmapped.  This flag is appropriate
3257 *         for updateing the contents of a page whose blocks may
3258 *         have already been released, and we only want to zero
3259 *         out the regions that correspond to those released blocks.
3260 *
3261 * Returns zero on sucess or negative on failure.
3262 */
3263static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3264		struct inode *inode, struct page *page, loff_t from,
3265		loff_t length, int flags)
3266{
3267	ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3268	unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3269	unsigned int blocksize, max, pos;
3270	ext4_lblk_t iblock;
 
3271	struct buffer_head *bh;
 
3272	int err = 0;
3273
3274	blocksize = inode->i_sb->s_blocksize;
3275	max = PAGE_CACHE_SIZE - offset;
3276
3277	if (index != page->index)
3278		return -EINVAL;
3279
3280	/*
3281	 * correct length if it does not fall between
3282	 * 'from' and the end of the page
3283	 */
3284	if (length > max || length < 0)
3285		length = max;
3286
3287	iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3288
3289	if (!page_has_buffers(page))
3290		create_empty_buffers(page, blocksize, 0);
 
3291
3292	/* Find the buffer that contains "offset" */
3293	bh = page_buffers(page);
3294	pos = blocksize;
3295	while (offset >= pos) {
3296		bh = bh->b_this_page;
3297		iblock++;
3298		pos += blocksize;
3299	}
 
 
 
 
 
 
 
 
 
 
 
 
 
3300
3301	pos = offset;
3302	while (pos < offset + length) {
3303		unsigned int end_of_block, range_to_discard;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3304
 
 
 
3305		err = 0;
 
 
 
 
 
3306
3307		/* The length of space left to zero and unmap */
3308		range_to_discard = offset + length - pos;
 
 
 
3309
3310		/* The length of space until the end of the block */
3311		end_of_block = blocksize - (pos & (blocksize-1));
 
 
 
 
 
 
 
 
 
 
 
 
3312
3313		/*
3314		 * Do not unmap or zero past end of block
3315		 * for this buffer head
3316		 */
3317		if (range_to_discard > end_of_block)
3318			range_to_discard = end_of_block;
3319
 
 
 
 
 
 
3320
3321		/*
3322		 * Skip this buffer head if we are only zeroing unampped
3323		 * regions of the page
3324		 */
3325		if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3326			buffer_mapped(bh))
3327				goto next;
3328
3329		/* If the range is block aligned, unmap */
3330		if (range_to_discard == blocksize) {
3331			clear_buffer_dirty(bh);
3332			bh->b_bdev = NULL;
3333			clear_buffer_mapped(bh);
3334			clear_buffer_req(bh);
3335			clear_buffer_new(bh);
3336			clear_buffer_delay(bh);
3337			clear_buffer_unwritten(bh);
3338			clear_buffer_uptodate(bh);
3339			zero_user(page, pos, range_to_discard);
3340			BUFFER_TRACE(bh, "Buffer discarded");
3341			goto next;
3342		}
3343
3344		/*
3345		 * If this block is not completely contained in the range
3346		 * to be discarded, then it is not going to be released. Because
3347		 * we need to keep this block, we need to make sure this part
3348		 * of the page is uptodate before we modify it by writeing
3349		 * partial zeros on it.
3350		 */
3351		if (!buffer_mapped(bh)) {
3352			/*
3353			 * Buffer head must be mapped before we can read
3354			 * from the block
3355			 */
3356			BUFFER_TRACE(bh, "unmapped");
3357			ext4_get_block(inode, iblock, bh, 0);
3358			/* unmapped? It's a hole - nothing to do */
3359			if (!buffer_mapped(bh)) {
3360				BUFFER_TRACE(bh, "still unmapped");
3361				goto next;
3362			}
3363		}
3364
3365		/* Ok, it's mapped. Make sure it's up-to-date */
3366		if (PageUptodate(page))
3367			set_buffer_uptodate(bh);
3368
3369		if (!buffer_uptodate(bh)) {
3370			err = -EIO;
3371			ll_rw_block(READ, 1, &bh);
3372			wait_on_buffer(bh);
3373			/* Uhhuh. Read error. Complain and punt.*/
3374			if (!buffer_uptodate(bh))
3375				goto next;
3376		}
3377
3378		if (ext4_should_journal_data(inode)) {
3379			BUFFER_TRACE(bh, "get write access");
3380			err = ext4_journal_get_write_access(handle, bh);
3381			if (err)
3382				goto next;
3383		}
 
 
 
3384
3385		zero_user(page, pos, range_to_discard);
 
3386
3387		err = 0;
3388		if (ext4_should_journal_data(inode)) {
3389			err = ext4_handle_dirty_metadata(handle, inode, bh);
3390		} else
3391			mark_buffer_dirty(bh);
3392
3393		BUFFER_TRACE(bh, "Partial buffer zeroed");
3394next:
3395		bh = bh->b_this_page;
3396		iblock++;
3397		pos += range_to_discard;
 
 
 
 
 
 
 
 
3398	}
3399
 
 
 
 
3400	return err;
3401}
3402
3403int ext4_can_truncate(struct inode *inode)
3404{
3405	if (S_ISREG(inode->i_mode))
3406		return 1;
3407	if (S_ISDIR(inode->i_mode))
3408		return 1;
3409	if (S_ISLNK(inode->i_mode))
3410		return !ext4_inode_is_fast_symlink(inode);
3411	return 0;
3412}
3413
3414/*
3415 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3416 * associated with the given offset and length
3417 *
3418 * @inode:  File inode
3419 * @offset: The offset where the hole will begin
3420 * @len:    The length of the hole
3421 *
3422 * Returns: 0 on sucess or negative on failure
3423 */
3424
3425int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3426{
3427	struct inode *inode = file->f_path.dentry->d_inode;
3428	if (!S_ISREG(inode->i_mode))
3429		return -EOPNOTSUPP;
 
 
 
 
 
 
3430
3431	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3432		/* TODO: Add support for non extent hole punching */
3433		return -EOPNOTSUPP;
 
 
 
 
 
 
 
 
3434	}
3435
3436	if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3437		/* TODO: Add support for bigalloc file systems */
3438		return -EOPNOTSUPP;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3439	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3440
3441	return ext4_ext_punch_hole(file, offset, length);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3442}
3443
3444/*
3445 * ext4_truncate()
3446 *
3447 * We block out ext4_get_block() block instantiations across the entire
3448 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3449 * simultaneously on behalf of the same inode.
3450 *
3451 * As we work through the truncate and commit bits of it to the journal there
3452 * is one core, guiding principle: the file's tree must always be consistent on
3453 * disk.  We must be able to restart the truncate after a crash.
3454 *
3455 * The file's tree may be transiently inconsistent in memory (although it
3456 * probably isn't), but whenever we close off and commit a journal transaction,
3457 * the contents of (the filesystem + the journal) must be consistent and
3458 * restartable.  It's pretty simple, really: bottom up, right to left (although
3459 * left-to-right works OK too).
3460 *
3461 * Note that at recovery time, journal replay occurs *before* the restart of
3462 * truncate against the orphan inode list.
3463 *
3464 * The committed inode has the new, desired i_size (which is the same as
3465 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
3466 * that this inode's truncate did not complete and it will again call
3467 * ext4_truncate() to have another go.  So there will be instantiated blocks
3468 * to the right of the truncation point in a crashed ext4 filesystem.  But
3469 * that's fine - as long as they are linked from the inode, the post-crash
3470 * ext4_truncate() run will find them and release them.
3471 */
3472void ext4_truncate(struct inode *inode)
3473{
 
 
 
 
 
 
 
 
 
 
 
 
 
3474	trace_ext4_truncate_enter(inode);
3475
3476	if (!ext4_can_truncate(inode))
3477		return;
3478
3479	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3480
3481	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3482		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3483
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3484	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3485		ext4_ext_truncate(inode);
3486	else
3487		ext4_ind_truncate(inode);
 
 
 
 
 
 
 
 
 
3488
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3489	trace_ext4_truncate_exit(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3490}
3491
3492/*
3493 * ext4_get_inode_loc returns with an extra refcount against the inode's
3494 * underlying buffer_head on success. If 'in_mem' is true, we have all
3495 * data in memory that is needed to recreate the on-disk version of this
3496 * inode.
3497 */
3498static int __ext4_get_inode_loc(struct inode *inode,
3499				struct ext4_iloc *iloc, int in_mem)
 
3500{
3501	struct ext4_group_desc	*gdp;
3502	struct buffer_head	*bh;
3503	struct super_block	*sb = inode->i_sb;
3504	ext4_fsblk_t		block;
 
3505	int			inodes_per_block, inode_offset;
3506
3507	iloc->bh = NULL;
3508	if (!ext4_valid_inum(sb, inode->i_ino))
3509		return -EIO;
 
3510
3511	iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3512	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3513	if (!gdp)
3514		return -EIO;
3515
3516	/*
3517	 * Figure out the offset within the block group inode table
3518	 */
3519	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3520	inode_offset = ((inode->i_ino - 1) %
3521			EXT4_INODES_PER_GROUP(sb));
3522	block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3523	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3524
 
 
 
 
 
 
 
 
 
3525	bh = sb_getblk(sb, block);
3526	if (!bh) {
3527		EXT4_ERROR_INODE_BLOCK(inode, block,
3528				       "unable to read itable block");
3529		return -EIO;
 
 
 
 
 
 
3530	}
3531	if (!buffer_uptodate(bh)) {
3532		lock_buffer(bh);
3533
3534		/*
3535		 * If the buffer has the write error flag, we have failed
3536		 * to write out another inode in the same block.  In this
3537		 * case, we don't have to read the block because we may
3538		 * read the old inode data successfully.
3539		 */
3540		if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3541			set_buffer_uptodate(bh);
3542
3543		if (buffer_uptodate(bh)) {
3544			/* someone brought it uptodate while we waited */
3545			unlock_buffer(bh);
3546			goto has_buffer;
3547		}
 
3548
3549		/*
3550		 * If we have all information of the inode in memory and this
3551		 * is the only valid inode in the block, we need not read the
3552		 * block.
3553		 */
3554		if (in_mem) {
3555			struct buffer_head *bitmap_bh;
3556			int i, start;
3557
3558			start = inode_offset & ~(inodes_per_block - 1);
3559
3560			/* Is the inode bitmap in cache? */
3561			bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3562			if (!bitmap_bh)
3563				goto make_io;
3564
3565			/*
3566			 * If the inode bitmap isn't in cache then the
3567			 * optimisation may end up performing two reads instead
3568			 * of one, so skip it.
3569			 */
3570			if (!buffer_uptodate(bitmap_bh)) {
3571				brelse(bitmap_bh);
3572				goto make_io;
3573			}
3574			for (i = start; i < start + inodes_per_block; i++) {
3575				if (i == inode_offset)
3576					continue;
3577				if (ext4_test_bit(i, bitmap_bh->b_data))
3578					break;
3579			}
3580			brelse(bitmap_bh);
3581			if (i == start + inodes_per_block) {
3582				/* all other inodes are free, so skip I/O */
3583				memset(bh->b_data, 0, bh->b_size);
3584				set_buffer_uptodate(bh);
3585				unlock_buffer(bh);
3586				goto has_buffer;
3587			}
 
 
 
 
 
 
 
 
 
 
 
 
 
3588		}
 
3589
3590make_io:
3591		/*
3592		 * If we need to do any I/O, try to pre-readahead extra
3593		 * blocks from the inode table.
3594		 */
3595		if (EXT4_SB(sb)->s_inode_readahead_blks) {
3596			ext4_fsblk_t b, end, table;
3597			unsigned num;
3598
3599			table = ext4_inode_table(sb, gdp);
3600			/* s_inode_readahead_blks is always a power of 2 */
3601			b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3602			if (table > b)
3603				b = table;
3604			end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3605			num = EXT4_INODES_PER_GROUP(sb);
3606			if (ext4_has_group_desc_csum(sb))
3607				num -= ext4_itable_unused_count(sb, gdp);
3608			table += num / inodes_per_block;
3609			if (end > table)
3610				end = table;
3611			while (b <= end)
3612				sb_breadahead(sb, b++);
3613		}
 
 
3614
3615		/*
3616		 * There are other valid inodes in the buffer, this inode
3617		 * has in-inode xattrs, or we don't have this inode in memory.
3618		 * Read the block from disk.
3619		 */
3620		trace_ext4_load_inode(inode);
3621		get_bh(bh);
3622		bh->b_end_io = end_buffer_read_sync;
3623		submit_bh(READ | REQ_META | REQ_PRIO, bh);
3624		wait_on_buffer(bh);
3625		if (!buffer_uptodate(bh)) {
3626			EXT4_ERROR_INODE_BLOCK(inode, block,
3627					       "unable to read itable block");
3628			brelse(bh);
3629			return -EIO;
3630		}
3631	}
3632has_buffer:
3633	iloc->bh = bh;
3634	return 0;
3635}
3636
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3637int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3638{
3639	/* We have all inode data except xattrs in memory here. */
3640	return __ext4_get_inode_loc(inode, iloc,
3641		!ext4_test_inode_state(inode, EXT4_STATE_XATTR));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3642}
3643
3644void ext4_set_inode_flags(struct inode *inode)
3645{
3646	unsigned int flags = EXT4_I(inode)->i_flags;
 
 
 
3647
3648	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3649	if (flags & EXT4_SYNC_FL)
3650		inode->i_flags |= S_SYNC;
3651	if (flags & EXT4_APPEND_FL)
3652		inode->i_flags |= S_APPEND;
3653	if (flags & EXT4_IMMUTABLE_FL)
3654		inode->i_flags |= S_IMMUTABLE;
3655	if (flags & EXT4_NOATIME_FL)
3656		inode->i_flags |= S_NOATIME;
3657	if (flags & EXT4_DIRSYNC_FL)
3658		inode->i_flags |= S_DIRSYNC;
3659}
3660
3661/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3662void ext4_get_inode_flags(struct ext4_inode_info *ei)
3663{
3664	unsigned int vfs_fl;
3665	unsigned long old_fl, new_fl;
3666
3667	do {
3668		vfs_fl = ei->vfs_inode.i_flags;
3669		old_fl = ei->i_flags;
3670		new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3671				EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3672				EXT4_DIRSYNC_FL);
3673		if (vfs_fl & S_SYNC)
3674			new_fl |= EXT4_SYNC_FL;
3675		if (vfs_fl & S_APPEND)
3676			new_fl |= EXT4_APPEND_FL;
3677		if (vfs_fl & S_IMMUTABLE)
3678			new_fl |= EXT4_IMMUTABLE_FL;
3679		if (vfs_fl & S_NOATIME)
3680			new_fl |= EXT4_NOATIME_FL;
3681		if (vfs_fl & S_DIRSYNC)
3682			new_fl |= EXT4_DIRSYNC_FL;
3683	} while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3684}
3685
3686static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3687				  struct ext4_inode_info *ei)
3688{
3689	blkcnt_t i_blocks ;
3690	struct inode *inode = &(ei->vfs_inode);
3691	struct super_block *sb = inode->i_sb;
3692
3693	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3694				EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3695		/* we are using combined 48 bit field */
3696		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3697					le32_to_cpu(raw_inode->i_blocks_lo);
3698		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3699			/* i_blocks represent file system block size */
3700			return i_blocks  << (inode->i_blkbits - 9);
3701		} else {
3702			return i_blocks;
3703		}
3704	} else {
3705		return le32_to_cpu(raw_inode->i_blocks_lo);
3706	}
3707}
3708
3709struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3710{
3711	struct ext4_iloc iloc;
3712	struct ext4_inode *raw_inode;
3713	struct ext4_inode_info *ei;
 
3714	struct inode *inode;
 
3715	journal_t *journal = EXT4_SB(sb)->s_journal;
3716	long ret;
 
3717	int block;
3718	uid_t i_uid;
3719	gid_t i_gid;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3720
3721	inode = iget_locked(sb, ino);
3722	if (!inode)
3723		return ERR_PTR(-ENOMEM);
3724	if (!(inode->i_state & I_NEW))
 
 
 
 
 
3725		return inode;
 
3726
3727	ei = EXT4_I(inode);
3728	iloc.bh = NULL;
3729
3730	ret = __ext4_get_inode_loc(inode, &iloc, 0);
3731	if (ret < 0)
3732		goto bad_inode;
3733	raw_inode = ext4_raw_inode(&iloc);
3734
 
 
 
 
 
 
3735	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3736		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3737		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3738		    EXT4_INODE_SIZE(inode->i_sb)) {
3739			EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3740				EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3741				EXT4_INODE_SIZE(inode->i_sb));
3742			ret = -EIO;
 
 
 
3743			goto bad_inode;
3744		}
3745	} else
3746		ei->i_extra_isize = 0;
3747
3748	/* Precompute checksum seed for inode metadata */
3749	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3750			EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3751		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3752		__u32 csum;
3753		__le32 inum = cpu_to_le32(inode->i_ino);
3754		__le32 gen = raw_inode->i_generation;
3755		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3756				   sizeof(inum));
3757		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3758					      sizeof(gen));
3759	}
3760
3761	if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3762		EXT4_ERROR_INODE(inode, "checksum invalid");
3763		ret = -EIO;
 
 
 
3764		goto bad_inode;
3765	}
3766
3767	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3768	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3769	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
 
 
 
 
 
 
 
3770	if (!(test_opt(inode->i_sb, NO_UID32))) {
3771		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3772		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3773	}
3774	i_uid_write(inode, i_uid);
3775	i_gid_write(inode, i_gid);
 
3776	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3777
3778	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
 
3779	ei->i_dir_start_lookup = 0;
3780	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3781	/* We now have enough fields to check if the inode was active or not.
3782	 * This is needed because nfsd might try to access dead inodes
3783	 * the test is that same one that e2fsck uses
3784	 * NeilBrown 1999oct15
3785	 */
3786	if (inode->i_nlink == 0) {
3787		if (inode->i_mode == 0 ||
3788		    !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3789			/* this inode is deleted */
3790			ret = -ESTALE;
 
 
 
 
 
 
3791			goto bad_inode;
3792		}
3793		/* The only unlinked inodes we let through here have
3794		 * valid i_mode and are being read by the orphan
3795		 * recovery code: that's fine, we're about to complete
3796		 * the process of deleting those. */
 
 
3797	}
3798	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
 
3799	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3800	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3801	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3802		ei->i_file_acl |=
3803			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3804	inode->i_size = ext4_isize(raw_inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3805	ei->i_disksize = inode->i_size;
3806#ifdef CONFIG_QUOTA
3807	ei->i_reserved_quota = 0;
3808#endif
3809	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3810	ei->i_block_group = iloc.block_group;
3811	ei->i_last_alloc_group = ~0;
3812	/*
3813	 * NOTE! The in-memory inode i_data array is in little-endian order
3814	 * even on big-endian machines: we do NOT byteswap the block numbers!
3815	 */
3816	for (block = 0; block < EXT4_N_BLOCKS; block++)
3817		ei->i_data[block] = raw_inode->i_block[block];
3818	INIT_LIST_HEAD(&ei->i_orphan);
 
3819
3820	/*
3821	 * Set transaction id's of transactions that have to be committed
3822	 * to finish f[data]sync. We set them to currently running transaction
3823	 * as we cannot be sure that the inode or some of its metadata isn't
3824	 * part of the transaction - the inode could have been reclaimed and
3825	 * now it is reread from disk.
3826	 */
3827	if (journal) {
3828		transaction_t *transaction;
3829		tid_t tid;
3830
3831		read_lock(&journal->j_state_lock);
3832		if (journal->j_running_transaction)
3833			transaction = journal->j_running_transaction;
3834		else
3835			transaction = journal->j_committing_transaction;
3836		if (transaction)
3837			tid = transaction->t_tid;
3838		else
3839			tid = journal->j_commit_sequence;
3840		read_unlock(&journal->j_state_lock);
3841		ei->i_sync_tid = tid;
3842		ei->i_datasync_tid = tid;
3843	}
3844
3845	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3846		if (ei->i_extra_isize == 0) {
3847			/* The extra space is currently unused. Use it. */
 
3848			ei->i_extra_isize = sizeof(struct ext4_inode) -
3849					    EXT4_GOOD_OLD_INODE_SIZE;
3850		} else {
3851			__le32 *magic = (void *)raw_inode +
3852					EXT4_GOOD_OLD_INODE_SIZE +
3853					ei->i_extra_isize;
3854			if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3855				ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3856		}
3857	}
3858
3859	EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3860	EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3861	EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3862	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3863
3864	inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3865	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3866		if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3867			inode->i_version |=
3868			(__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
 
 
 
 
3869	}
3870
3871	ret = 0;
3872	if (ei->i_file_acl &&
3873	    !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3874		EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
 
3875				 ei->i_file_acl);
3876		ret = -EIO;
3877		goto bad_inode;
3878	} else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3879		if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3880		    (S_ISLNK(inode->i_mode) &&
3881		     !ext4_inode_is_fast_symlink(inode)))
3882			/* Validate extent which is part of inode */
3883			ret = ext4_ext_check_inode(inode);
3884	} else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3885		   (S_ISLNK(inode->i_mode) &&
3886		    !ext4_inode_is_fast_symlink(inode))) {
3887		/* Validate block references which are part of inode */
3888		ret = ext4_ind_check_inode(inode);
3889	}
3890	if (ret)
3891		goto bad_inode;
3892
3893	if (S_ISREG(inode->i_mode)) {
3894		inode->i_op = &ext4_file_inode_operations;
3895		inode->i_fop = &ext4_file_operations;
3896		ext4_set_aops(inode);
3897	} else if (S_ISDIR(inode->i_mode)) {
3898		inode->i_op = &ext4_dir_inode_operations;
3899		inode->i_fop = &ext4_dir_operations;
3900	} else if (S_ISLNK(inode->i_mode)) {
3901		if (ext4_inode_is_fast_symlink(inode)) {
 
 
 
 
 
 
 
 
 
 
 
3902			inode->i_op = &ext4_fast_symlink_inode_operations;
3903			nd_terminate_link(ei->i_data, inode->i_size,
3904				sizeof(ei->i_data) - 1);
3905		} else {
3906			inode->i_op = &ext4_symlink_inode_operations;
3907			ext4_set_aops(inode);
3908		}
3909	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3910	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3911		inode->i_op = &ext4_special_inode_operations;
3912		if (raw_inode->i_block[0])
3913			init_special_inode(inode, inode->i_mode,
3914			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3915		else
3916			init_special_inode(inode, inode->i_mode,
3917			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
 
 
3918	} else {
3919		ret = -EIO;
3920		EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
 
3921		goto bad_inode;
3922	}
 
 
 
 
 
 
 
 
 
 
 
 
3923	brelse(iloc.bh);
3924	ext4_set_inode_flags(inode);
3925	unlock_new_inode(inode);
3926	return inode;
3927
3928bad_inode:
3929	brelse(iloc.bh);
3930	iget_failed(inode);
3931	return ERR_PTR(ret);
3932}
3933
3934static int ext4_inode_blocks_set(handle_t *handle,
3935				struct ext4_inode *raw_inode,
3936				struct ext4_inode_info *ei)
 
3937{
3938	struct inode *inode = &(ei->vfs_inode);
3939	u64 i_blocks = inode->i_blocks;
3940	struct super_block *sb = inode->i_sb;
3941
3942	if (i_blocks <= ~0U) {
3943		/*
3944		 * i_blocks can be represnted in a 32 bit variable
3945		 * as multiple of 512 bytes
3946		 */
3947		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3948		raw_inode->i_blocks_high = 0;
3949		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3950		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
3951	}
3952	if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3953		return -EFBIG;
3954
3955	if (i_blocks <= 0xffffffffffffULL) {
3956		/*
3957		 * i_blocks can be represented in a 48 bit variable
3958		 * as multiple of 512 bytes
3959		 */
3960		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3961		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3962		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3963	} else {
3964		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3965		/* i_block is stored in file system block size */
3966		i_blocks = i_blocks >> (inode->i_blkbits - 9);
3967		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3968		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
 
 
 
 
 
 
 
 
 
3969	}
3970	return 0;
3971}
3972
3973/*
3974 * Post the struct inode info into an on-disk inode location in the
3975 * buffer-cache.  This gobbles the caller's reference to the
3976 * buffer_head in the inode location struct.
3977 *
3978 * The caller must have write access to iloc->bh.
3979 */
3980static int ext4_do_update_inode(handle_t *handle,
3981				struct inode *inode,
3982				struct ext4_iloc *iloc)
3983{
3984	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3985	struct ext4_inode_info *ei = EXT4_I(inode);
3986	struct buffer_head *bh = iloc->bh;
3987	int err = 0, rc, block;
3988	uid_t i_uid;
3989	gid_t i_gid;
 
 
3990
3991	/* For fields not not tracking in the in-memory inode,
3992	 * initialise them to zero for new inodes. */
 
 
3993	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
3994		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3995
3996	ext4_get_inode_flags(ei);
3997	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3998	i_uid = i_uid_read(inode);
3999	i_gid = i_gid_read(inode);
4000	if (!(test_opt(inode->i_sb, NO_UID32))) {
4001		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4002		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4003/*
4004 * Fix up interoperability with old kernels. Otherwise, old inodes get
4005 * re-used with the upper 16 bits of the uid/gid intact
4006 */
4007		if (!ei->i_dtime) {
4008			raw_inode->i_uid_high =
4009				cpu_to_le16(high_16_bits(i_uid));
4010			raw_inode->i_gid_high =
4011				cpu_to_le16(high_16_bits(i_gid));
4012		} else {
4013			raw_inode->i_uid_high = 0;
4014			raw_inode->i_gid_high = 0;
4015		}
4016	} else {
4017		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4018		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4019		raw_inode->i_uid_high = 0;
4020		raw_inode->i_gid_high = 0;
4021	}
4022	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4023
4024	EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4025	EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4026	EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4027	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4028
4029	if (ext4_inode_blocks_set(handle, raw_inode, ei))
4030		goto out_brelse;
4031	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4032	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4033	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4034	    cpu_to_le32(EXT4_OS_HURD))
4035		raw_inode->i_file_acl_high =
4036			cpu_to_le16(ei->i_file_acl >> 32);
4037	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4038	ext4_isize_set(raw_inode, ei->i_disksize);
4039	if (ei->i_disksize > 0x7fffffffULL) {
4040		struct super_block *sb = inode->i_sb;
4041		if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4042				EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4043				EXT4_SB(sb)->s_es->s_rev_level ==
4044				cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4045			/* If this is the first large file
4046			 * created, add a flag to the superblock.
4047			 */
4048			err = ext4_journal_get_write_access(handle,
4049					EXT4_SB(sb)->s_sbh);
4050			if (err)
4051				goto out_brelse;
4052			ext4_update_dynamic_rev(sb);
4053			EXT4_SET_RO_COMPAT_FEATURE(sb,
4054					EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4055			ext4_handle_sync(handle);
4056			err = ext4_handle_dirty_super_now(handle, sb);
4057		}
4058	}
4059	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4060	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4061		if (old_valid_dev(inode->i_rdev)) {
4062			raw_inode->i_block[0] =
4063				cpu_to_le32(old_encode_dev(inode->i_rdev));
4064			raw_inode->i_block[1] = 0;
4065		} else {
4066			raw_inode->i_block[0] = 0;
4067			raw_inode->i_block[1] =
4068				cpu_to_le32(new_encode_dev(inode->i_rdev));
4069			raw_inode->i_block[2] = 0;
4070		}
4071	} else
4072		for (block = 0; block < EXT4_N_BLOCKS; block++)
4073			raw_inode->i_block[block] = ei->i_data[block];
4074
4075	raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4076	if (ei->i_extra_isize) {
4077		if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4078			raw_inode->i_version_hi =
4079			cpu_to_le32(inode->i_version >> 32);
4080		raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4081	}
4082
4083	ext4_inode_csum_set(inode, raw_inode, ei);
 
 
4084
4085	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4086	rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4087	if (!err)
4088		err = rc;
4089	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4090
4091	ext4_update_inode_fsync_trans(handle, inode, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4092out_brelse:
4093	brelse(bh);
4094	ext4_std_error(inode->i_sb, err);
4095	return err;
4096}
4097
4098/*
4099 * ext4_write_inode()
4100 *
4101 * We are called from a few places:
4102 *
4103 * - Within generic_file_write() for O_SYNC files.
4104 *   Here, there will be no transaction running. We wait for any running
4105 *   trasnaction to commit.
4106 *
4107 * - Within sys_sync(), kupdate and such.
4108 *   We wait on commit, if tol to.
4109 *
4110 * - Within prune_icache() (PF_MEMALLOC == true)
4111 *   Here we simply return.  We can't afford to block kswapd on the
4112 *   journal commit.
4113 *
4114 * In all cases it is actually safe for us to return without doing anything,
4115 * because the inode has been copied into a raw inode buffer in
4116 * ext4_mark_inode_dirty().  This is a correctness thing for O_SYNC and for
4117 * knfsd.
4118 *
4119 * Note that we are absolutely dependent upon all inode dirtiers doing the
4120 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4121 * which we are interested.
4122 *
4123 * It would be a bug for them to not do this.  The code:
4124 *
4125 *	mark_inode_dirty(inode)
4126 *	stuff();
4127 *	inode->i_size = expr;
4128 *
4129 * is in error because a kswapd-driven write_inode() could occur while
4130 * `stuff()' is running, and the new i_size will be lost.  Plus the inode
4131 * will no longer be on the superblock's dirty inode list.
4132 */
4133int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4134{
4135	int err;
4136
4137	if (current->flags & PF_MEMALLOC)
4138		return 0;
4139
 
 
 
4140	if (EXT4_SB(inode->i_sb)->s_journal) {
4141		if (ext4_journal_current_handle()) {
4142			jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4143			dump_stack();
4144			return -EIO;
4145		}
4146
4147		if (wbc->sync_mode != WB_SYNC_ALL)
 
 
 
 
 
4148			return 0;
4149
4150		err = ext4_force_commit(inode->i_sb);
 
4151	} else {
4152		struct ext4_iloc iloc;
4153
4154		err = __ext4_get_inode_loc(inode, &iloc, 0);
4155		if (err)
4156			return err;
4157		if (wbc->sync_mode == WB_SYNC_ALL)
 
 
 
 
4158			sync_dirty_buffer(iloc.bh);
4159		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4160			EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4161					 "IO error syncing inode");
4162			err = -EIO;
4163		}
4164		brelse(iloc.bh);
4165	}
4166	return err;
4167}
4168
4169/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4170 * ext4_setattr()
4171 *
4172 * Called from notify_change.
4173 *
4174 * We want to trap VFS attempts to truncate the file as soon as
4175 * possible.  In particular, we want to make sure that when the VFS
4176 * shrinks i_size, we put the inode on the orphan list and modify
4177 * i_disksize immediately, so that during the subsequent flushing of
4178 * dirty pages and freeing of disk blocks, we can guarantee that any
4179 * commit will leave the blocks being flushed in an unused state on
4180 * disk.  (On recovery, the inode will get truncated and the blocks will
4181 * be freed, so we have a strong guarantee that no future commit will
4182 * leave these blocks visible to the user.)
4183 *
4184 * Another thing we have to assure is that if we are in ordered mode
4185 * and inode is still attached to the committing transaction, we must
4186 * we start writeout of all the dirty pages which are being truncated.
4187 * This way we are sure that all the data written in the previous
4188 * transaction are already on disk (truncate waits for pages under
4189 * writeback).
4190 *
4191 * Called with inode->i_mutex down.
4192 */
4193int ext4_setattr(struct dentry *dentry, struct iattr *attr)
 
4194{
4195	struct inode *inode = dentry->d_inode;
4196	int error, rc = 0;
4197	int orphan = 0;
4198	const unsigned int ia_valid = attr->ia_valid;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4199
4200	error = inode_change_ok(inode, attr);
4201	if (error)
4202		return error;
4203
4204	if (is_quota_modification(inode, attr))
4205		dquot_initialize(inode);
4206	if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4207	    (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
 
 
 
 
4208		handle_t *handle;
4209
4210		/* (user+group)*(old+new) structure, inode write (sb,
4211		 * inode block, ? - but truncate inode update has it) */
4212		handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4213					EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
 
4214		if (IS_ERR(handle)) {
4215			error = PTR_ERR(handle);
4216			goto err_out;
4217		}
4218		error = dquot_transfer(inode, attr);
 
 
 
 
 
 
 
4219		if (error) {
4220			ext4_journal_stop(handle);
4221			return error;
4222		}
4223		/* Update corresponding info in inode so that everything is in
4224		 * one transaction */
4225		if (attr->ia_valid & ATTR_UID)
4226			inode->i_uid = attr->ia_uid;
4227		if (attr->ia_valid & ATTR_GID)
4228			inode->i_gid = attr->ia_gid;
4229		error = ext4_mark_inode_dirty(handle, inode);
4230		ext4_journal_stop(handle);
 
 
 
4231	}
4232
4233	if (attr->ia_valid & ATTR_SIZE) {
4234		inode_dio_wait(inode);
 
 
 
4235
4236		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4237			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4238
4239			if (attr->ia_size > sbi->s_bitmap_maxbytes)
4240				return -EFBIG;
 
4241		}
4242	}
4243
4244	if (S_ISREG(inode->i_mode) &&
4245	    attr->ia_valid & ATTR_SIZE &&
4246	    (attr->ia_size < inode->i_size)) {
4247		handle_t *handle;
4248
4249		handle = ext4_journal_start(inode, 3);
4250		if (IS_ERR(handle)) {
4251			error = PTR_ERR(handle);
4252			goto err_out;
4253		}
4254		if (ext4_handle_valid(handle)) {
4255			error = ext4_orphan_add(handle, inode);
4256			orphan = 1;
4257		}
4258		EXT4_I(inode)->i_disksize = attr->ia_size;
4259		rc = ext4_mark_inode_dirty(handle, inode);
4260		if (!error)
4261			error = rc;
4262		ext4_journal_stop(handle);
4263
4264		if (ext4_should_order_data(inode)) {
4265			error = ext4_begin_ordered_truncate(inode,
 
 
 
 
4266							    attr->ia_size);
4267			if (error) {
4268				/* Do as much error cleanup as possible */
4269				handle = ext4_journal_start(inode, 3);
4270				if (IS_ERR(handle)) {
4271					ext4_orphan_del(NULL, inode);
4272					goto err_out;
4273				}
4274				ext4_orphan_del(handle, inode);
4275				orphan = 0;
4276				ext4_journal_stop(handle);
4277				goto err_out;
4278			}
 
 
 
 
 
4279		}
4280	}
4281
4282	if (attr->ia_valid & ATTR_SIZE) {
4283		if (attr->ia_size != i_size_read(inode))
4284			truncate_setsize(inode, attr->ia_size);
4285		ext4_truncate(inode);
4286	}
4287
4288	if (!rc) {
4289		setattr_copy(inode, attr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4290		mark_inode_dirty(inode);
4291	}
4292
4293	/*
4294	 * If the call to ext4_truncate failed to get a transaction handle at
4295	 * all, we need to clean up the in-core orphan list manually.
4296	 */
4297	if (orphan && inode->i_nlink)
4298		ext4_orphan_del(NULL, inode);
4299
4300	if (!rc && (ia_valid & ATTR_MODE))
4301		rc = ext4_acl_chmod(inode);
4302
4303err_out:
4304	ext4_std_error(inode->i_sb, error);
 
4305	if (!error)
4306		error = rc;
4307	return error;
4308}
4309
4310int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4311		 struct kstat *stat)
4312{
4313	struct inode *inode;
4314	unsigned long delalloc_blocks;
 
 
 
 
 
 
 
 
 
 
 
4315
4316	inode = dentry->d_inode;
4317	generic_fillattr(inode, stat);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4318
4319	/*
4320	 * We can't update i_blocks if the block allocation is delayed
4321	 * otherwise in the case of system crash before the real block
4322	 * allocation is done, we will have i_blocks inconsistent with
4323	 * on-disk file blocks.
4324	 * We always keep i_blocks updated together with real
4325	 * allocation. But to not confuse with user, stat
4326	 * will return the blocks that include the delayed allocation
4327	 * blocks for this file.
4328	 */
4329	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4330				EXT4_I(inode)->i_reserved_data_blocks);
4331
4332	stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4333	return 0;
4334}
4335
4336static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 
4337{
4338	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4339		return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4340	return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4341}
4342
4343/*
4344 * Account for index blocks, block groups bitmaps and block group
4345 * descriptor blocks if modify datablocks and index blocks
4346 * worse case, the indexs blocks spread over different block groups
4347 *
4348 * If datablocks are discontiguous, they are possible to spread over
4349 * different block groups too. If they are contiuguous, with flexbg,
4350 * they could still across block group boundary.
4351 *
4352 * Also account for superblock, inode, quota and xattr blocks
4353 */
4354static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 
4355{
4356	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4357	int gdpblocks;
4358	int idxblocks;
4359	int ret = 0;
4360
4361	/*
4362	 * How many index blocks need to touch to modify nrblocks?
4363	 * The "Chunk" flag indicating whether the nrblocks is
4364	 * physically contiguous on disk
4365	 *
4366	 * For Direct IO and fallocate, they calls get_block to allocate
4367	 * one single extent at a time, so they could set the "Chunk" flag
4368	 */
4369	idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4370
4371	ret = idxblocks;
4372
4373	/*
4374	 * Now let's see how many group bitmaps and group descriptors need
4375	 * to account
4376	 */
4377	groups = idxblocks;
4378	if (chunk)
4379		groups += 1;
4380	else
4381		groups += nrblocks;
4382
4383	gdpblocks = groups;
4384	if (groups > ngroups)
4385		groups = ngroups;
4386	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4387		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4388
4389	/* bitmaps and block group descriptor blocks */
4390	ret += groups + gdpblocks;
4391
4392	/* Blocks for super block, inode, quota and xattr blocks */
4393	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4394
4395	return ret;
4396}
4397
4398/*
4399 * Calculate the total number of credits to reserve to fit
4400 * the modification of a single pages into a single transaction,
4401 * which may include multiple chunks of block allocations.
4402 *
4403 * This could be called via ext4_write_begin()
4404 *
4405 * We need to consider the worse case, when
4406 * one new block per extent.
4407 */
4408int ext4_writepage_trans_blocks(struct inode *inode)
4409{
4410	int bpp = ext4_journal_blocks_per_page(inode);
4411	int ret;
4412
4413	ret = ext4_meta_trans_blocks(inode, bpp, 0);
4414
4415	/* Account for data blocks for journalled mode */
4416	if (ext4_should_journal_data(inode))
4417		ret += bpp;
4418	return ret;
4419}
4420
4421/*
4422 * Calculate the journal credits for a chunk of data modification.
4423 *
4424 * This is called from DIO, fallocate or whoever calling
4425 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4426 *
4427 * journal buffers for data blocks are not included here, as DIO
4428 * and fallocate do no need to journal data buffers.
4429 */
4430int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4431{
4432	return ext4_meta_trans_blocks(inode, nrblocks, 1);
4433}
4434
4435/*
4436 * The caller must have previously called ext4_reserve_inode_write().
4437 * Give this, we know that the caller already has write access to iloc->bh.
4438 */
4439int ext4_mark_iloc_dirty(handle_t *handle,
4440			 struct inode *inode, struct ext4_iloc *iloc)
4441{
4442	int err = 0;
4443
4444	if (IS_I_VERSION(inode))
4445		inode_inc_iversion(inode);
 
 
 
4446
4447	/* the do_update_inode consumes one bh->b_count */
4448	get_bh(iloc->bh);
4449
4450	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4451	err = ext4_do_update_inode(handle, inode, iloc);
4452	put_bh(iloc->bh);
4453	return err;
4454}
4455
4456/*
4457 * On success, We end up with an outstanding reference count against
4458 * iloc->bh.  This _must_ be cleaned up later.
4459 */
4460
4461int
4462ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4463			 struct ext4_iloc *iloc)
4464{
4465	int err;
4466
 
 
 
4467	err = ext4_get_inode_loc(inode, iloc);
4468	if (!err) {
4469		BUFFER_TRACE(iloc->bh, "get_write_access");
4470		err = ext4_journal_get_write_access(handle, iloc->bh);
 
4471		if (err) {
4472			brelse(iloc->bh);
4473			iloc->bh = NULL;
4474		}
4475	}
4476	ext4_std_error(inode->i_sb, err);
4477	return err;
4478}
4479
4480/*
4481 * Expand an inode by new_extra_isize bytes.
4482 * Returns 0 on success or negative error number on failure.
4483 */
4484static int ext4_expand_extra_isize(struct inode *inode,
4485				   unsigned int new_extra_isize,
4486				   struct ext4_iloc iloc,
4487				   handle_t *handle)
4488{
4489	struct ext4_inode *raw_inode;
4490	struct ext4_xattr_ibody_header *header;
 
 
 
4491
4492	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4493		return 0;
 
 
 
 
 
 
 
 
 
 
4494
4495	raw_inode = ext4_raw_inode(&iloc);
4496
4497	header = IHDR(inode, raw_inode);
4498
4499	/* No extended attributes present */
4500	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4501	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4502		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4503			new_extra_isize);
 
4504		EXT4_I(inode)->i_extra_isize = new_extra_isize;
4505		return 0;
4506	}
4507
 
 
 
 
 
 
 
 
4508	/* try to expand with EAs present */
4509	return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4510					  raw_inode, handle);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4511}
4512
4513/*
4514 * What we do here is to mark the in-core inode as clean with respect to inode
4515 * dirtiness (it may still be data-dirty).
4516 * This means that the in-core inode may be reaped by prune_icache
4517 * without having to perform any I/O.  This is a very good thing,
4518 * because *any* task may call prune_icache - even ones which
4519 * have a transaction open against a different journal.
4520 *
4521 * Is this cheating?  Not really.  Sure, we haven't written the
4522 * inode out, but prune_icache isn't a user-visible syncing function.
4523 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4524 * we start and wait on commits.
4525 *
4526 * Is this efficient/effective?  Well, we're being nice to the system
4527 * by cleaning up our inodes proactively so they can be reaped
4528 * without I/O.  But we are potentially leaving up to five seconds'
4529 * worth of inodes floating about which prune_icache wants us to
4530 * write out.  One way to fix that would be to get prune_icache()
4531 * to do a write_super() to free up some memory.  It has the desired
4532 * effect.
4533 */
4534int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
 
4535{
4536	struct ext4_iloc iloc;
4537	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4538	static unsigned int mnt_count;
4539	int err, ret;
4540
4541	might_sleep();
4542	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4543	err = ext4_reserve_inode_write(handle, inode, &iloc);
4544	if (ext4_handle_valid(handle) &&
4545	    EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4546	    !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4547		/*
4548		 * We need extra buffer credits since we may write into EA block
4549		 * with this same handle. If journal_extend fails, then it will
4550		 * only result in a minor loss of functionality for that inode.
4551		 * If this is felt to be critical, then e2fsck should be run to
4552		 * force a large enough s_min_extra_isize.
4553		 */
4554		if ((jbd2_journal_extend(handle,
4555			     EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4556			ret = ext4_expand_extra_isize(inode,
4557						      sbi->s_want_extra_isize,
4558						      iloc, handle);
4559			if (ret) {
4560				ext4_set_inode_state(inode,
4561						     EXT4_STATE_NO_EXPAND);
4562				if (mnt_count !=
4563					le16_to_cpu(sbi->s_es->s_mnt_count)) {
4564					ext4_warning(inode->i_sb,
4565					"Unable to expand inode %lu. Delete"
4566					" some EAs or run e2fsck.",
4567					inode->i_ino);
4568					mnt_count =
4569					  le16_to_cpu(sbi->s_es->s_mnt_count);
4570				}
4571			}
4572		}
4573	}
4574	if (!err)
4575		err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4576	return err;
4577}
4578
4579/*
4580 * ext4_dirty_inode() is called from __mark_inode_dirty()
4581 *
4582 * We're really interested in the case where a file is being extended.
4583 * i_size has been changed by generic_commit_write() and we thus need
4584 * to include the updated inode in the current transaction.
4585 *
4586 * Also, dquot_alloc_block() will always dirty the inode when blocks
4587 * are allocated to the file.
4588 *
4589 * If the inode is marked synchronous, we don't honour that here - doing
4590 * so would cause a commit on atime updates, which we don't bother doing.
4591 * We handle synchronous inodes at the highest possible level.
4592 */
4593void ext4_dirty_inode(struct inode *inode, int flags)
4594{
4595	handle_t *handle;
4596
4597	handle = ext4_journal_start(inode, 2);
4598	if (IS_ERR(handle))
4599		goto out;
4600
4601	ext4_mark_inode_dirty(handle, inode);
4602
4603	ext4_journal_stop(handle);
4604out:
4605	return;
4606}
4607
4608#if 0
4609/*
4610 * Bind an inode's backing buffer_head into this transaction, to prevent
4611 * it from being flushed to disk early.  Unlike
4612 * ext4_reserve_inode_write, this leaves behind no bh reference and
4613 * returns no iloc structure, so the caller needs to repeat the iloc
4614 * lookup to mark the inode dirty later.
4615 */
4616static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4617{
4618	struct ext4_iloc iloc;
4619
4620	int err = 0;
4621	if (handle) {
4622		err = ext4_get_inode_loc(inode, &iloc);
4623		if (!err) {
4624			BUFFER_TRACE(iloc.bh, "get_write_access");
4625			err = jbd2_journal_get_write_access(handle, iloc.bh);
4626			if (!err)
4627				err = ext4_handle_dirty_metadata(handle,
4628								 NULL,
4629								 iloc.bh);
4630			brelse(iloc.bh);
4631		}
4632	}
4633	ext4_std_error(inode->i_sb, err);
4634	return err;
4635}
4636#endif
4637
4638int ext4_change_inode_journal_flag(struct inode *inode, int val)
4639{
4640	journal_t *journal;
4641	handle_t *handle;
4642	int err;
 
4643
4644	/*
4645	 * We have to be very careful here: changing a data block's
4646	 * journaling status dynamically is dangerous.  If we write a
4647	 * data block to the journal, change the status and then delete
4648	 * that block, we risk forgetting to revoke the old log record
4649	 * from the journal and so a subsequent replay can corrupt data.
4650	 * So, first we make sure that the journal is empty and that
4651	 * nobody is changing anything.
4652	 */
4653
4654	journal = EXT4_JOURNAL(inode);
4655	if (!journal)
4656		return 0;
4657	if (is_journal_aborted(journal))
4658		return -EROFS;
4659	/* We have to allocate physical blocks for delalloc blocks
4660	 * before flushing journal. otherwise delalloc blocks can not
4661	 * be allocated any more. even more truncate on delalloc blocks
4662	 * could trigger BUG by flushing delalloc blocks in journal.
4663	 * There is no delalloc block in non-journal data mode.
4664	 */
4665	if (val && test_opt(inode->i_sb, DELALLOC)) {
4666		err = ext4_alloc_da_blocks(inode);
4667		if (err < 0)
 
 
 
 
 
 
 
 
4668			return err;
 
4669	}
4670
 
4671	jbd2_journal_lock_updates(journal);
4672
4673	/*
4674	 * OK, there are no updates running now, and all cached data is
4675	 * synced to disk.  We are now in a completely consistent state
4676	 * which doesn't have anything in the journal, and we know that
4677	 * no filesystem updates are running, so it is safe to modify
4678	 * the inode's in-core data-journaling state flag now.
4679	 */
4680
4681	if (val)
4682		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4683	else {
4684		jbd2_journal_flush(journal);
 
 
 
 
 
4685		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4686	}
4687	ext4_set_aops(inode);
4688
4689	jbd2_journal_unlock_updates(journal);
 
 
 
 
4690
4691	/* Finally we can mark the inode as dirty. */
4692
4693	handle = ext4_journal_start(inode, 1);
4694	if (IS_ERR(handle))
4695		return PTR_ERR(handle);
4696
 
 
4697	err = ext4_mark_inode_dirty(handle, inode);
4698	ext4_handle_sync(handle);
4699	ext4_journal_stop(handle);
4700	ext4_std_error(inode->i_sb, err);
4701
4702	return err;
4703}
4704
4705static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
 
4706{
4707	return !buffer_mapped(bh);
4708}
4709
4710int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4711{
4712	struct page *page = vmf->page;
 
4713	loff_t size;
4714	unsigned long len;
4715	int ret;
 
4716	struct file *file = vma->vm_file;
4717	struct inode *inode = file->f_path.dentry->d_inode;
4718	struct address_space *mapping = inode->i_mapping;
4719	handle_t *handle;
4720	get_block_t *get_block;
4721	int retries = 0;
4722
 
 
 
 
 
 
 
 
 
 
 
 
4723	/*
4724	 * This check is racy but catches the common case. We rely on
4725	 * __block_page_mkwrite() to do a reliable check.
 
 
4726	 */
4727	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 
 
4728	/* Delalloc case is easy... */
4729	if (test_opt(inode->i_sb, DELALLOC) &&
4730	    !ext4_should_journal_data(inode) &&
4731	    !ext4_nonda_switch(inode->i_sb)) {
4732		do {
4733			ret = __block_page_mkwrite(vma, vmf,
4734						   ext4_da_get_block_prep);
4735		} while (ret == -ENOSPC &&
4736		       ext4_should_retry_alloc(inode->i_sb, &retries));
4737		goto out_ret;
4738	}
4739
4740	lock_page(page);
4741	size = i_size_read(inode);
4742	/* Page got truncated from under us? */
4743	if (page->mapping != mapping || page_offset(page) > size) {
4744		unlock_page(page);
4745		ret = VM_FAULT_NOPAGE;
4746		goto out;
4747	}
4748
4749	if (page->index == size >> PAGE_CACHE_SHIFT)
4750		len = size & ~PAGE_CACHE_MASK;
4751	else
4752		len = PAGE_CACHE_SIZE;
4753	/*
4754	 * Return if we have all the buffers mapped. This avoids the need to do
4755	 * journal_start/journal_stop which can block and take a long time
 
 
 
4756	 */
4757	if (page_has_buffers(page)) {
4758		if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4759					ext4_bh_unmapped)) {
 
4760			/* Wait so that we don't change page under IO */
4761			wait_on_page_writeback(page);
4762			ret = VM_FAULT_LOCKED;
4763			goto out;
4764		}
4765	}
4766	unlock_page(page);
4767	/* OK, we need to fill the hole... */
4768	if (ext4_should_dioread_nolock(inode))
4769		get_block = ext4_get_block_write;
4770	else
4771		get_block = ext4_get_block;
4772retry_alloc:
4773	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
 
4774	if (IS_ERR(handle)) {
4775		ret = VM_FAULT_SIGBUS;
4776		goto out;
4777	}
4778	ret = __block_page_mkwrite(vma, vmf, get_block);
4779	if (!ret && ext4_should_journal_data(inode)) {
4780		if (walk_page_buffers(handle, page_buffers(page), 0,
4781			  PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4782			unlock_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4783			ret = VM_FAULT_SIGBUS;
4784			ext4_journal_stop(handle);
4785			goto out;
 
 
4786		}
4787		ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4788	}
4789	ext4_journal_stop(handle);
4790	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4791		goto retry_alloc;
4792out_ret:
4793	ret = block_page_mkwrite_return(ret);
4794out:
 
 
4795	return ret;
 
 
 
 
4796}
v6.9.4
   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_meta_trans_blocks(struct inode *inode, int lblocks,
 140				  int pextents);
 
 
 
 
 
 
 
 
 141
 142/*
 143 * Test whether an inode is a fast symlink.
 144 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
 145 */
 146int ext4_inode_is_fast_symlink(struct inode *inode)
 147{
 148	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 149		int ea_blocks = EXT4_I(inode)->i_file_acl ?
 150				EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
 
 
 
 
 
 
 
 
 
 
 
 
 151
 152		if (ext4_has_inline_data(inode))
 153			return 0;
 
 
 
 
 
 
 
 
 
 
 154
 155		return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
 156	}
 157	return S_ISLNK(inode->i_mode) && inode->i_size &&
 158	       (inode->i_size < EXT4_N_BLOCKS * 4);
 159}
 160
 161/*
 162 * Called at the last iput() if i_nlink is zero.
 163 */
 164void ext4_evict_inode(struct inode *inode)
 165{
 166	handle_t *handle;
 167	int err;
 168	/*
 169	 * Credits for final inode cleanup and freeing:
 170	 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
 171	 * (xattr block freeing), bitmap, group descriptor (inode freeing)
 172	 */
 173	int extra_credits = 6;
 174	struct ext4_xattr_inode_array *ea_inode_array = NULL;
 175	bool freeze_protected = false;
 176
 177	trace_ext4_evict_inode(inode);
 178
 179	if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
 180		ext4_evict_ea_inode(inode);
 181	if (inode->i_nlink) {
 182		truncate_inode_pages_final(&inode->i_data);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 183
 
 
 
 
 
 184		goto no_delete;
 185	}
 186
 187	if (is_bad_inode(inode))
 188		goto no_delete;
 189	dquot_initialize(inode);
 190
 191	if (ext4_should_order_data(inode))
 192		ext4_begin_ordered_truncate(inode, 0);
 193	truncate_inode_pages_final(&inode->i_data);
 194
 195	/*
 196	 * For inodes with journalled data, transaction commit could have
 197	 * dirtied the inode. And for inodes with dioread_nolock, unwritten
 198	 * extents converting worker could merge extents and also have dirtied
 199	 * the inode. Flush worker is ignoring it because of I_FREEING flag but
 200	 * we still need to remove the inode from the writeback lists.
 201	 */
 202	if (!list_empty_careful(&inode->i_io_list))
 203		inode_io_list_del(inode);
 204
 205	/*
 206	 * Protect us against freezing - iput() caller didn't have to have any
 207	 * protection against it. When we are in a running transaction though,
 208	 * we are already protected against freezing and we cannot grab further
 209	 * protection due to lock ordering constraints.
 210	 */
 211	if (!ext4_journal_current_handle()) {
 212		sb_start_intwrite(inode->i_sb);
 213		freeze_protected = true;
 214	}
 215
 216	if (!IS_NOQUOTA(inode))
 217		extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
 218
 219	/*
 220	 * Block bitmap, group descriptor, and inode are accounted in both
 221	 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
 222	 */
 223	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
 224			 ext4_blocks_for_truncate(inode) + extra_credits - 3);
 225	if (IS_ERR(handle)) {
 226		ext4_std_error(inode->i_sb, PTR_ERR(handle));
 227		/*
 228		 * If we're going to skip the normal cleanup, we still need to
 229		 * make sure that the in-core orphan linked list is properly
 230		 * cleaned up.
 231		 */
 232		ext4_orphan_del(NULL, inode);
 233		if (freeze_protected)
 234			sb_end_intwrite(inode->i_sb);
 235		goto no_delete;
 236	}
 237
 238	if (IS_SYNC(inode))
 239		ext4_handle_sync(handle);
 240
 241	/*
 242	 * Set inode->i_size to 0 before calling ext4_truncate(). We need
 243	 * special handling of symlinks here because i_size is used to
 244	 * determine whether ext4_inode_info->i_data contains symlink data or
 245	 * block mappings. Setting i_size to 0 will remove its fast symlink
 246	 * status. Erase i_data so that it becomes a valid empty block map.
 247	 */
 248	if (ext4_inode_is_fast_symlink(inode))
 249		memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
 250	inode->i_size = 0;
 251	err = ext4_mark_inode_dirty(handle, inode);
 252	if (err) {
 253		ext4_warning(inode->i_sb,
 254			     "couldn't mark inode dirty (err %d)", err);
 255		goto stop_handle;
 256	}
 257	if (inode->i_blocks) {
 258		err = ext4_truncate(inode);
 259		if (err) {
 260			ext4_error_err(inode->i_sb, -err,
 261				       "couldn't truncate inode %lu (err %d)",
 262				       inode->i_ino, err);
 263			goto stop_handle;
 
 
 
 
 
 
 
 
 
 
 
 
 
 264		}
 265	}
 266
 267	/* Remove xattr references. */
 268	err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
 269				      extra_credits);
 270	if (err) {
 271		ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
 272stop_handle:
 273		ext4_journal_stop(handle);
 274		ext4_orphan_del(NULL, inode);
 275		if (freeze_protected)
 276			sb_end_intwrite(inode->i_sb);
 277		ext4_xattr_inode_array_free(ea_inode_array);
 278		goto no_delete;
 279	}
 280
 281	/*
 282	 * Kill off the orphan record which ext4_truncate created.
 283	 * AKPM: I think this can be inside the above `if'.
 284	 * Note that ext4_orphan_del() has to be able to cope with the
 285	 * deletion of a non-existent orphan - this is because we don't
 286	 * know if ext4_truncate() actually created an orphan record.
 287	 * (Well, we could do this if we need to, but heck - it works)
 288	 */
 289	ext4_orphan_del(handle, inode);
 290	EXT4_I(inode)->i_dtime	= (__u32)ktime_get_real_seconds();
 291
 292	/*
 293	 * One subtle ordering requirement: if anything has gone wrong
 294	 * (transaction abort, IO errors, whatever), then we can still
 295	 * do these next steps (the fs will already have been marked as
 296	 * having errors), but we can't free the inode if the mark_dirty
 297	 * fails.
 298	 */
 299	if (ext4_mark_inode_dirty(handle, inode))
 300		/* If that failed, just do the required in-core inode clear. */
 301		ext4_clear_inode(inode);
 302	else
 303		ext4_free_inode(handle, inode);
 304	ext4_journal_stop(handle);
 305	if (freeze_protected)
 306		sb_end_intwrite(inode->i_sb);
 307	ext4_xattr_inode_array_free(ea_inode_array);
 308	return;
 309no_delete:
 310	/*
 311	 * Check out some where else accidentally dirty the evicting inode,
 312	 * which may probably cause inode use-after-free issues later.
 313	 */
 314	WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
 315
 316	if (!list_empty(&EXT4_I(inode)->i_fc_list))
 317		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
 318	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
 319}
 320
 321#ifdef CONFIG_QUOTA
 322qsize_t *ext4_get_reserved_space(struct inode *inode)
 323{
 324	return &EXT4_I(inode)->i_reserved_quota;
 325}
 326#endif
 327
 328/*
 
 
 
 
 
 
 
 
 
 
 
 
 329 * Called with i_data_sem down, which is important since we can call
 330 * ext4_discard_preallocations() from here.
 331 */
 332void ext4_da_update_reserve_space(struct inode *inode,
 333					int used, int quota_claim)
 334{
 335	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 336	struct ext4_inode_info *ei = EXT4_I(inode);
 337
 338	spin_lock(&ei->i_block_reservation_lock);
 339	trace_ext4_da_update_reserve_space(inode, used, quota_claim);
 340	if (unlikely(used > ei->i_reserved_data_blocks)) {
 341		ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
 342			 "with only %d reserved data blocks",
 343			 __func__, inode->i_ino, used,
 344			 ei->i_reserved_data_blocks);
 345		WARN_ON(1);
 346		used = ei->i_reserved_data_blocks;
 347	}
 348
 
 
 
 
 
 
 
 
 
 349	/* Update per-inode reservations */
 350	ei->i_reserved_data_blocks -= used;
 351	percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
 
 
 
 352
 353	spin_unlock(&ei->i_block_reservation_lock);
 
 
 
 
 
 
 
 
 
 
 
 354
 355	/* Update quota subsystem for data blocks */
 356	if (quota_claim)
 357		dquot_claim_block(inode, EXT4_C2B(sbi, used));
 358	else {
 359		/*
 360		 * We did fallocate with an offset that is already delayed
 361		 * allocated. So on delayed allocated writeback we should
 362		 * not re-claim the quota for fallocated blocks.
 363		 */
 364		dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
 365	}
 366
 367	/*
 368	 * If we have done all the pending block allocations and if
 369	 * there aren't any writers on the inode, we can discard the
 370	 * inode's preallocations.
 371	 */
 372	if ((ei->i_reserved_data_blocks == 0) &&
 373	    !inode_is_open_for_write(inode))
 374		ext4_discard_preallocations(inode);
 375}
 376
 377static int __check_block_validity(struct inode *inode, const char *func,
 378				unsigned int line,
 379				struct ext4_map_blocks *map)
 380{
 381	if (ext4_has_feature_journal(inode->i_sb) &&
 382	    (inode->i_ino ==
 383	     le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
 384		return 0;
 385	if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
 386		ext4_error_inode(inode, func, line, map->m_pblk,
 387				 "lblock %lu mapped to illegal pblock %llu "
 388				 "(length %d)", (unsigned long) map->m_lblk,
 389				 map->m_pblk, map->m_len);
 390		return -EFSCORRUPTED;
 391	}
 392	return 0;
 393}
 394
 395int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
 396		       ext4_lblk_t len)
 
 
 
 
 
 
 
 397{
 398	int ret;
 
 
 
 
 399
 400	if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
 401		return fscrypt_zeroout_range(inode, lblk, pblk, len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 402
 403	ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
 404	if (ret > 0)
 405		ret = 0;
 
 
 
 
 
 
 
 406
 407	return ret;
 408}
 
 409
 410#define check_block_validity(inode, map)	\
 411	__check_block_validity((inode), __func__, __LINE__, (map))
 
 
 
 
 
 
 
 
 412
 413#ifdef ES_AGGRESSIVE_TEST
 414static void ext4_map_blocks_es_recheck(handle_t *handle,
 415				       struct inode *inode,
 416				       struct ext4_map_blocks *es_map,
 417				       struct ext4_map_blocks *map,
 418				       int flags)
 419{
 420	int retval;
 421
 422	map->m_flags = 0;
 423	/*
 424	 * There is a race window that the result is not the same.
 425	 * e.g. xfstests #223 when dioread_nolock enables.  The reason
 426	 * is that we lookup a block mapping in extent status tree with
 427	 * out taking i_data_sem.  So at the time the unwritten extent
 428	 * could be converted.
 429	 */
 430	down_read(&EXT4_I(inode)->i_data_sem);
 431	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 432		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 433	} else {
 434		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 435	}
 436	up_read((&EXT4_I(inode)->i_data_sem));
 437
 438	/*
 439	 * We don't check m_len because extent will be collpased in status
 440	 * tree.  So the m_len might not equal.
 441	 */
 442	if (es_map->m_lblk != map->m_lblk ||
 443	    es_map->m_flags != map->m_flags ||
 444	    es_map->m_pblk != map->m_pblk) {
 445		printk("ES cache assertion failed for inode: %lu "
 446		       "es_cached ex [%d/%d/%llu/%x] != "
 447		       "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
 448		       inode->i_ino, es_map->m_lblk, es_map->m_len,
 449		       es_map->m_pblk, es_map->m_flags, map->m_lblk,
 450		       map->m_len, map->m_pblk, map->m_flags,
 451		       retval, flags);
 452	}
 453}
 454#endif /* ES_AGGRESSIVE_TEST */
 455
 456/*
 457 * The ext4_map_blocks() function tries to look up the requested blocks,
 458 * and returns if the blocks are already mapped.
 459 *
 460 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 461 * and store the allocated blocks in the result buffer head and mark it
 462 * mapped.
 463 *
 464 * If file type is extents based, it will call ext4_ext_map_blocks(),
 465 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 466 * based files
 467 *
 468 * On success, it returns the number of blocks being mapped or allocated.
 469 * If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are
 470 * pre-allocated and unwritten, the resulting @map is marked as unwritten.
 471 * If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped.
 472 *
 473 * It returns 0 if plain look up failed (blocks have not been allocated), in
 474 * that case, @map is returned as unmapped but we still do fill map->m_len to
 475 * indicate the length of a hole starting at map->m_lblk.
 476 *
 477 * It returns the error in case of allocation failure.
 478 */
 479int ext4_map_blocks(handle_t *handle, struct inode *inode,
 480		    struct ext4_map_blocks *map, int flags)
 481{
 482	struct extent_status es;
 483	int retval;
 484	int ret = 0;
 485#ifdef ES_AGGRESSIVE_TEST
 486	struct ext4_map_blocks orig_map;
 487
 488	memcpy(&orig_map, map, sizeof(*map));
 489#endif
 490
 491	map->m_flags = 0;
 492	ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
 493		  flags, map->m_len, (unsigned long) map->m_lblk);
 494
 495	/*
 496	 * ext4_map_blocks returns an int, and m_len is an unsigned int
 497	 */
 498	if (unlikely(map->m_len > INT_MAX))
 499		map->m_len = INT_MAX;
 500
 501	/* We can handle the block number less than EXT_MAX_BLOCKS */
 502	if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
 503		return -EFSCORRUPTED;
 504
 505	/* Lookup extent status tree firstly */
 506	if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
 507	    ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
 508		if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
 509			map->m_pblk = ext4_es_pblock(&es) +
 510					map->m_lblk - es.es_lblk;
 511			map->m_flags |= ext4_es_is_written(&es) ?
 512					EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
 513			retval = es.es_len - (map->m_lblk - es.es_lblk);
 514			if (retval > map->m_len)
 515				retval = map->m_len;
 516			map->m_len = retval;
 517		} else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
 518			map->m_pblk = 0;
 519			map->m_flags |= ext4_es_is_delayed(&es) ?
 520					EXT4_MAP_DELAYED : 0;
 521			retval = es.es_len - (map->m_lblk - es.es_lblk);
 522			if (retval > map->m_len)
 523				retval = map->m_len;
 524			map->m_len = retval;
 525			retval = 0;
 526		} else {
 527			BUG();
 528		}
 529
 530		if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
 531			return retval;
 532#ifdef ES_AGGRESSIVE_TEST
 533		ext4_map_blocks_es_recheck(handle, inode, map,
 534					   &orig_map, flags);
 535#endif
 536		goto found;
 537	}
 538	/*
 539	 * In the query cache no-wait mode, nothing we can do more if we
 540	 * cannot find extent in the cache.
 541	 */
 542	if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
 543		return 0;
 544
 545	/*
 546	 * Try to see if we can get the block without requesting a new
 547	 * file system block.
 548	 */
 549	down_read(&EXT4_I(inode)->i_data_sem);
 550	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 551		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 
 552	} else {
 553		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 554	}
 555	if (retval > 0) {
 556		unsigned int status;
 557
 558		if (unlikely(retval != map->m_len)) {
 559			ext4_warning(inode->i_sb,
 560				     "ES len assertion failed for inode "
 561				     "%lu: retval %d != map->m_len %d",
 562				     inode->i_ino, retval, map->m_len);
 563			WARN_ON(1);
 564		}
 565
 566		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 567				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 568		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
 569		    !(status & EXTENT_STATUS_WRITTEN) &&
 570		    ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
 571				       map->m_lblk + map->m_len - 1))
 572			status |= EXTENT_STATUS_DELAYED;
 573		ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
 574				      map->m_pblk, status);
 575	}
 576	up_read((&EXT4_I(inode)->i_data_sem));
 577
 578found:
 579	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 580		ret = check_block_validity(inode, map);
 581		if (ret != 0)
 582			return ret;
 583	}
 584
 585	/* If it is only a block(s) look up */
 586	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 587		return retval;
 588
 589	/*
 590	 * Returns if the blocks have already allocated
 591	 *
 592	 * Note that if blocks have been preallocated
 593	 * ext4_ext_map_blocks() returns with buffer head unmapped
 
 594	 */
 595	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 596		/*
 597		 * If we need to convert extent to unwritten
 598		 * we continue and do the actual work in
 599		 * ext4_ext_map_blocks()
 600		 */
 601		if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
 602			return retval;
 603
 604	/*
 605	 * Here we clear m_flags because after allocating an new extent,
 606	 * it will be set again.
 
 
 
 
 
 
 607	 */
 608	map->m_flags &= ~EXT4_MAP_FLAGS;
 609
 610	/*
 611	 * New blocks allocate and/or writing to unwritten extent
 612	 * will possibly result in updating i_data, so we take
 613	 * the write lock of i_data_sem, and call get_block()
 614	 * with create == 1 flag.
 615	 */
 616	down_write(&EXT4_I(inode)->i_data_sem);
 617
 618	/*
 
 
 
 
 
 
 
 
 619	 * We need to check for EXT4 here because migrate
 620	 * could have changed the inode type in between
 621	 */
 622	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 623		retval = ext4_ext_map_blocks(handle, inode, map, flags);
 624	} else {
 625		retval = ext4_ind_map_blocks(handle, inode, map, flags);
 626
 627		if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
 628			/*
 629			 * We allocated new blocks which will result in
 630			 * i_data's format changing.  Force the migrate
 631			 * to fail by clearing migrate flags
 632			 */
 633			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 634		}
 635	}
 636
 637	if (retval > 0) {
 638		unsigned int status;
 639
 640		if (unlikely(retval != map->m_len)) {
 641			ext4_warning(inode->i_sb,
 642				     "ES len assertion failed for inode "
 643				     "%lu: retval %d != map->m_len %d",
 644				     inode->i_ino, retval, map->m_len);
 645			WARN_ON(1);
 646		}
 647
 648		/*
 649		 * We have to zeroout blocks before inserting them into extent
 650		 * status tree. Otherwise someone could look them up there and
 651		 * use them before they are really zeroed. We also have to
 652		 * unmap metadata before zeroing as otherwise writeback can
 653		 * overwrite zeros with stale data from block device.
 654		 */
 655		if (flags & EXT4_GET_BLOCKS_ZERO &&
 656		    map->m_flags & EXT4_MAP_MAPPED &&
 657		    map->m_flags & EXT4_MAP_NEW) {
 658			ret = ext4_issue_zeroout(inode, map->m_lblk,
 659						 map->m_pblk, map->m_len);
 660			if (ret) {
 661				retval = ret;
 662				goto out_sem;
 663			}
 664		}
 665
 666		/*
 667		 * If the extent has been zeroed out, we don't need to update
 668		 * extent status tree.
 669		 */
 670		if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
 671		    ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
 672			if (ext4_es_is_written(&es))
 673				goto out_sem;
 674		}
 675		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 676				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 677		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
 678		    !(status & EXTENT_STATUS_WRITTEN) &&
 679		    ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
 680				       map->m_lblk + map->m_len - 1))
 681			status |= EXTENT_STATUS_DELAYED;
 682		ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
 683				      map->m_pblk, status);
 684	}
 685
 686out_sem:
 687	up_write((&EXT4_I(inode)->i_data_sem));
 688	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 689		ret = check_block_validity(inode, map);
 690		if (ret != 0)
 691			return ret;
 692
 693		/*
 694		 * Inodes with freshly allocated blocks where contents will be
 695		 * visible after transaction commit must be on transaction's
 696		 * ordered data list.
 697		 */
 698		if (map->m_flags & EXT4_MAP_NEW &&
 699		    !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
 700		    !(flags & EXT4_GET_BLOCKS_ZERO) &&
 701		    !ext4_is_quota_file(inode) &&
 702		    ext4_should_order_data(inode)) {
 703			loff_t start_byte =
 704				(loff_t)map->m_lblk << inode->i_blkbits;
 705			loff_t length = (loff_t)map->m_len << inode->i_blkbits;
 706
 707			if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
 708				ret = ext4_jbd2_inode_add_wait(handle, inode,
 709						start_byte, length);
 710			else
 711				ret = ext4_jbd2_inode_add_write(handle, inode,
 712						start_byte, length);
 713			if (ret)
 714				return ret;
 715		}
 716	}
 717	if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
 718				map->m_flags & EXT4_MAP_MAPPED))
 719		ext4_fc_track_range(handle, inode, map->m_lblk,
 720					map->m_lblk + map->m_len - 1);
 721	if (retval < 0)
 722		ext_debug(inode, "failed with err %d\n", retval);
 723	return retval;
 724}
 725
 726/*
 727 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
 728 * we have to be careful as someone else may be manipulating b_state as well.
 729 */
 730static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
 731{
 732	unsigned long old_state;
 733	unsigned long new_state;
 734
 735	flags &= EXT4_MAP_FLAGS;
 736
 737	/* Dummy buffer_head? Set non-atomically. */
 738	if (!bh->b_page) {
 739		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
 740		return;
 741	}
 742	/*
 743	 * Someone else may be modifying b_state. Be careful! This is ugly but
 744	 * once we get rid of using bh as a container for mapping information
 745	 * to pass to / from get_block functions, this can go away.
 746	 */
 747	old_state = READ_ONCE(bh->b_state);
 748	do {
 749		new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
 750	} while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
 751}
 752
 753static int _ext4_get_block(struct inode *inode, sector_t iblock,
 754			   struct buffer_head *bh, int flags)
 755{
 
 756	struct ext4_map_blocks map;
 757	int ret = 0;
 758
 759	if (ext4_has_inline_data(inode))
 760		return -ERANGE;
 761
 762	map.m_lblk = iblock;
 763	map.m_len = bh->b_size >> inode->i_blkbits;
 764
 765	ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
 766			      flags);
 
 
 
 
 
 
 
 
 
 
 
 
 767	if (ret > 0) {
 768		map_bh(bh, inode->i_sb, map.m_pblk);
 769		ext4_update_bh_state(bh, map.m_flags);
 770		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 771		ret = 0;
 772	} else if (ret == 0) {
 773		/* hole case, need to fill in bh->b_size */
 774		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 775	}
 
 
 776	return ret;
 777}
 778
 779int ext4_get_block(struct inode *inode, sector_t iblock,
 780		   struct buffer_head *bh, int create)
 781{
 782	return _ext4_get_block(inode, iblock, bh,
 783			       create ? EXT4_GET_BLOCKS_CREATE : 0);
 784}
 785
 786/*
 787 * Get block function used when preparing for buffered write if we require
 788 * creating an unwritten extent if blocks haven't been allocated.  The extent
 789 * will be converted to written after the IO is complete.
 790 */
 791int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
 792			     struct buffer_head *bh_result, int create)
 793{
 794	int ret = 0;
 795
 796	ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
 797		   inode->i_ino, create);
 798	ret = _ext4_get_block(inode, iblock, bh_result,
 799			       EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
 800
 801	/*
 802	 * If the buffer is marked unwritten, mark it as new to make sure it is
 803	 * zeroed out correctly in case of partial writes. Otherwise, there is
 804	 * a chance of stale data getting exposed.
 805	 */
 806	if (ret == 0 && buffer_unwritten(bh_result))
 807		set_buffer_new(bh_result);
 808
 809	return ret;
 810}
 811
 812/* Maximum number of blocks we map for direct IO at once. */
 813#define DIO_MAX_BLOCKS 4096
 814
 815/*
 816 * `handle' can be NULL if create is zero
 817 */
 818struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 819				ext4_lblk_t block, int map_flags)
 820{
 821	struct ext4_map_blocks map;
 822	struct buffer_head *bh;
 823	int create = map_flags & EXT4_GET_BLOCKS_CREATE;
 824	bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
 825	int err;
 826
 827	ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 828		    || handle != NULL || create == 0);
 829	ASSERT(create == 0 || !nowait);
 830
 831	map.m_lblk = block;
 832	map.m_len = 1;
 833	err = ext4_map_blocks(handle, inode, &map, map_flags);
 
 834
 835	if (err == 0)
 836		return create ? ERR_PTR(-ENOSPC) : NULL;
 837	if (err < 0)
 838		return ERR_PTR(err);
 839
 840	if (nowait)
 841		return sb_find_get_block(inode->i_sb, map.m_pblk);
 842
 843	bh = sb_getblk(inode->i_sb, map.m_pblk);
 844	if (unlikely(!bh))
 845		return ERR_PTR(-ENOMEM);
 
 
 846	if (map.m_flags & EXT4_MAP_NEW) {
 847		ASSERT(create != 0);
 848		ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 849			    || (handle != NULL));
 850
 851		/*
 852		 * Now that we do not always journal data, we should
 853		 * keep in mind whether this should always journal the
 854		 * new buffer as metadata.  For now, regular file
 855		 * writes use ext4_get_block instead, so it's not a
 856		 * problem.
 857		 */
 858		lock_buffer(bh);
 859		BUFFER_TRACE(bh, "call get_create_access");
 860		err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
 861						     EXT4_JTR_NONE);
 862		if (unlikely(err)) {
 863			unlock_buffer(bh);
 864			goto errout;
 865		}
 866		if (!buffer_uptodate(bh)) {
 867			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 868			set_buffer_uptodate(bh);
 869		}
 870		unlock_buffer(bh);
 871		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 872		err = ext4_handle_dirty_metadata(handle, inode, bh);
 873		if (unlikely(err))
 874			goto errout;
 875	} else
 876		BUFFER_TRACE(bh, "not a new buffer");
 
 
 
 
 
 
 877	return bh;
 878errout:
 879	brelse(bh);
 880	return ERR_PTR(err);
 881}
 882
 883struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 884			       ext4_lblk_t block, int map_flags)
 885{
 886	struct buffer_head *bh;
 887	int ret;
 888
 889	bh = ext4_getblk(handle, inode, block, map_flags);
 890	if (IS_ERR(bh))
 891		return bh;
 892	if (!bh || ext4_buffer_uptodate(bh))
 893		return bh;
 894
 895	ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
 896	if (ret) {
 897		put_bh(bh);
 898		return ERR_PTR(ret);
 899	}
 900	return bh;
 901}
 902
 903/* Read a contiguous batch of blocks. */
 904int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
 905		     bool wait, struct buffer_head **bhs)
 906{
 907	int i, err;
 908
 909	for (i = 0; i < bh_count; i++) {
 910		bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
 911		if (IS_ERR(bhs[i])) {
 912			err = PTR_ERR(bhs[i]);
 913			bh_count = i;
 914			goto out_brelse;
 915		}
 916	}
 917
 918	for (i = 0; i < bh_count; i++)
 919		/* Note that NULL bhs[i] is valid because of holes. */
 920		if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
 921			ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
 922
 923	if (!wait)
 924		return 0;
 925
 926	for (i = 0; i < bh_count; i++)
 927		if (bhs[i])
 928			wait_on_buffer(bhs[i]);
 929
 930	for (i = 0; i < bh_count; i++) {
 931		if (bhs[i] && !buffer_uptodate(bhs[i])) {
 932			err = -EIO;
 933			goto out_brelse;
 934		}
 935	}
 936	return 0;
 937
 938out_brelse:
 939	for (i = 0; i < bh_count; i++) {
 940		brelse(bhs[i]);
 941		bhs[i] = NULL;
 942	}
 943	return err;
 944}
 945
 946int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
 947			   struct buffer_head *head,
 948			   unsigned from,
 949			   unsigned to,
 950			   int *partial,
 951			   int (*fn)(handle_t *handle, struct inode *inode,
 952				     struct buffer_head *bh))
 953{
 954	struct buffer_head *bh;
 955	unsigned block_start, block_end;
 956	unsigned blocksize = head->b_size;
 957	int err, ret = 0;
 958	struct buffer_head *next;
 959
 960	for (bh = head, block_start = 0;
 961	     ret == 0 && (bh != head || !block_start);
 962	     block_start = block_end, bh = next) {
 963		next = bh->b_this_page;
 964		block_end = block_start + blocksize;
 965		if (block_end <= from || block_start >= to) {
 966			if (partial && !buffer_uptodate(bh))
 967				*partial = 1;
 968			continue;
 969		}
 970		err = (*fn)(handle, inode, bh);
 971		if (!ret)
 972			ret = err;
 973	}
 974	return ret;
 975}
 976
 977/*
 978 * Helper for handling dirtying of journalled data. We also mark the folio as
 979 * dirty so that writeback code knows about this page (and inode) contains
 980 * dirty data. ext4_writepages() then commits appropriate transaction to
 981 * make data stable.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 982 */
 983static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
 984{
 985	folio_mark_dirty(bh->b_folio);
 986	return ext4_handle_dirty_metadata(handle, NULL, bh);
 987}
 988
 989int do_journal_get_write_access(handle_t *handle, struct inode *inode,
 990				struct buffer_head *bh)
 991{
 992	int dirty = buffer_dirty(bh);
 993	int ret;
 994
 995	if (!buffer_mapped(bh) || buffer_freed(bh))
 996		return 0;
 997	/*
 998	 * __block_write_begin() could have dirtied some buffers. Clean
 999	 * the dirty bit as jbd2_journal_get_write_access() could complain
1000	 * otherwise about fs integrity issues. Setting of the dirty bit
1001	 * by __block_write_begin() isn't a real problem here as we clear
1002	 * the bit before releasing a page lock and thus writeback cannot
1003	 * ever write the buffer.
1004	 */
1005	if (dirty)
1006		clear_buffer_dirty(bh);
1007	BUFFER_TRACE(bh, "get write access");
1008	ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1009					    EXT4_JTR_NONE);
1010	if (!ret && dirty)
1011		ret = ext4_dirty_journalled_data(handle, bh);
1012	return ret;
1013}
1014
1015#ifdef CONFIG_FS_ENCRYPTION
1016static int ext4_block_write_begin(struct folio *folio, loff_t pos, unsigned len,
1017				  get_block_t *get_block)
1018{
1019	unsigned from = pos & (PAGE_SIZE - 1);
1020	unsigned to = from + len;
1021	struct inode *inode = folio->mapping->host;
1022	unsigned block_start, block_end;
1023	sector_t block;
1024	int err = 0;
1025	unsigned blocksize = inode->i_sb->s_blocksize;
1026	unsigned bbits;
1027	struct buffer_head *bh, *head, *wait[2];
1028	int nr_wait = 0;
1029	int i;
1030
1031	BUG_ON(!folio_test_locked(folio));
1032	BUG_ON(from > PAGE_SIZE);
1033	BUG_ON(to > PAGE_SIZE);
1034	BUG_ON(from > to);
1035
1036	head = folio_buffers(folio);
1037	if (!head)
1038		head = create_empty_buffers(folio, blocksize, 0);
1039	bbits = ilog2(blocksize);
1040	block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1041
1042	for (bh = head, block_start = 0; bh != head || !block_start;
1043	    block++, block_start = block_end, bh = bh->b_this_page) {
1044		block_end = block_start + blocksize;
1045		if (block_end <= from || block_start >= to) {
1046			if (folio_test_uptodate(folio)) {
1047				set_buffer_uptodate(bh);
1048			}
1049			continue;
1050		}
1051		if (buffer_new(bh))
1052			clear_buffer_new(bh);
1053		if (!buffer_mapped(bh)) {
1054			WARN_ON(bh->b_size != blocksize);
1055			err = get_block(inode, block, bh, 1);
1056			if (err)
1057				break;
1058			if (buffer_new(bh)) {
1059				if (folio_test_uptodate(folio)) {
1060					clear_buffer_new(bh);
1061					set_buffer_uptodate(bh);
1062					mark_buffer_dirty(bh);
1063					continue;
1064				}
1065				if (block_end > to || block_start < from)
1066					folio_zero_segments(folio, to,
1067							    block_end,
1068							    block_start, from);
1069				continue;
1070			}
1071		}
1072		if (folio_test_uptodate(folio)) {
1073			set_buffer_uptodate(bh);
1074			continue;
1075		}
1076		if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1077		    !buffer_unwritten(bh) &&
1078		    (block_start < from || block_end > to)) {
1079			ext4_read_bh_lock(bh, 0, false);
1080			wait[nr_wait++] = bh;
1081		}
1082	}
1083	/*
1084	 * If we issued read requests, let them complete.
1085	 */
1086	for (i = 0; i < nr_wait; i++) {
1087		wait_on_buffer(wait[i]);
1088		if (!buffer_uptodate(wait[i]))
1089			err = -EIO;
1090	}
1091	if (unlikely(err)) {
1092		folio_zero_new_buffers(folio, from, to);
1093	} else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1094		for (i = 0; i < nr_wait; i++) {
1095			int err2;
1096
1097			err2 = fscrypt_decrypt_pagecache_blocks(folio,
1098						blocksize, bh_offset(wait[i]));
1099			if (err2) {
1100				clear_buffer_uptodate(wait[i]);
1101				err = err2;
1102			}
1103		}
1104	}
1105
1106	return err;
1107}
1108#endif
1109
1110/*
1111 * To preserve ordering, it is essential that the hole instantiation and
1112 * the data write be encapsulated in a single transaction.  We cannot
1113 * close off a transaction and start a new one between the ext4_get_block()
1114 * and the ext4_write_end().  So doing the jbd2_journal_start at the start of
1115 * ext4_write_begin() is the right place.
1116 */
1117static int ext4_write_begin(struct file *file, struct address_space *mapping,
1118			    loff_t pos, unsigned len,
1119			    struct page **pagep, void **fsdata)
1120{
1121	struct inode *inode = mapping->host;
1122	int ret, needed_blocks;
1123	handle_t *handle;
1124	int retries = 0;
1125	struct folio *folio;
1126	pgoff_t index;
1127	unsigned from, to;
1128
1129	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
1130		return -EIO;
1131
1132	trace_ext4_write_begin(inode, pos, len);
1133	/*
1134	 * Reserve one block more for addition to orphan list in case
1135	 * we allocate blocks but write fails for some reason
1136	 */
1137	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1138	index = pos >> PAGE_SHIFT;
1139	from = pos & (PAGE_SIZE - 1);
1140	to = from + len;
1141
1142	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1143		ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1144						    pagep);
1145		if (ret < 0)
1146			return ret;
1147		if (ret == 1)
1148			return 0;
1149	}
1150
1151	/*
1152	 * __filemap_get_folio() can take a long time if the
1153	 * system is thrashing due to memory pressure, or if the folio
1154	 * is being written back.  So grab it first before we start
1155	 * the transaction handle.  This also allows us to allocate
1156	 * the folio (if needed) without using GFP_NOFS.
1157	 */
1158retry_grab:
1159	folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
1160					mapping_gfp_mask(mapping));
1161	if (IS_ERR(folio))
1162		return PTR_ERR(folio);
1163	/*
1164	 * The same as page allocation, we prealloc buffer heads before
1165	 * starting the handle.
1166	 */
1167	if (!folio_buffers(folio))
1168		create_empty_buffers(folio, inode->i_sb->s_blocksize, 0);
1169
1170	folio_unlock(folio);
1171
1172retry_journal:
1173	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1174	if (IS_ERR(handle)) {
1175		folio_put(folio);
1176		return PTR_ERR(handle);
1177	}
1178
1179	folio_lock(folio);
1180	if (folio->mapping != mapping) {
1181		/* The folio got truncated from under us */
1182		folio_unlock(folio);
1183		folio_put(folio);
1184		ext4_journal_stop(handle);
1185		goto retry_grab;
 
1186	}
1187	/* In case writeback began while the folio was unlocked */
1188	folio_wait_stable(folio);
1189
1190#ifdef CONFIG_FS_ENCRYPTION
1191	if (ext4_should_dioread_nolock(inode))
1192		ret = ext4_block_write_begin(folio, pos, len,
1193					     ext4_get_block_unwritten);
1194	else
1195		ret = ext4_block_write_begin(folio, pos, len, ext4_get_block);
1196#else
1197	if (ext4_should_dioread_nolock(inode))
1198		ret = __block_write_begin(&folio->page, pos, len,
1199					  ext4_get_block_unwritten);
1200	else
1201		ret = __block_write_begin(&folio->page, pos, len, ext4_get_block);
1202#endif
1203	if (!ret && ext4_should_journal_data(inode)) {
1204		ret = ext4_walk_page_buffers(handle, inode,
1205					     folio_buffers(folio), from, to,
1206					     NULL, do_journal_get_write_access);
1207	}
1208
1209	if (ret) {
1210		bool extended = (pos + len > inode->i_size) &&
1211				!ext4_verity_in_progress(inode);
1212
1213		folio_unlock(folio);
1214		/*
1215		 * __block_write_begin may have instantiated a few blocks
1216		 * outside i_size.  Trim these off again. Don't need
1217		 * i_size_read because we hold i_rwsem.
1218		 *
1219		 * Add inode to orphan list in case we crash before
1220		 * truncate finishes
1221		 */
1222		if (extended && ext4_can_truncate(inode))
1223			ext4_orphan_add(handle, inode);
1224
1225		ext4_journal_stop(handle);
1226		if (extended) {
1227			ext4_truncate_failed_write(inode);
1228			/*
1229			 * If truncate failed early the inode might
1230			 * still be on the orphan list; we need to
1231			 * make sure the inode is removed from the
1232			 * orphan list in that case.
1233			 */
1234			if (inode->i_nlink)
1235				ext4_orphan_del(NULL, inode);
1236		}
 
1237
1238		if (ret == -ENOSPC &&
1239		    ext4_should_retry_alloc(inode->i_sb, &retries))
1240			goto retry_journal;
1241		folio_put(folio);
1242		return ret;
1243	}
1244	*pagep = &folio->page;
1245	return ret;
1246}
1247
1248/* For write_end() in data=journal mode */
1249static int write_end_fn(handle_t *handle, struct inode *inode,
1250			struct buffer_head *bh)
1251{
1252	int ret;
1253	if (!buffer_mapped(bh) || buffer_freed(bh))
1254		return 0;
1255	set_buffer_uptodate(bh);
1256	ret = ext4_dirty_journalled_data(handle, bh);
1257	clear_buffer_meta(bh);
1258	clear_buffer_prio(bh);
1259	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1260}
1261
1262/*
1263 * We need to pick up the new inode size which generic_commit_write gave us
1264 * `file' can be NULL - eg, when called from page_symlink().
1265 *
1266 * ext4 never places buffers on inode->i_mapping->i_private_list.  metadata
1267 * buffers are managed internally.
1268 */
1269static int ext4_write_end(struct file *file,
1270			  struct address_space *mapping,
1271			  loff_t pos, unsigned len, unsigned copied,
1272			  struct page *page, void *fsdata)
1273{
1274	struct folio *folio = page_folio(page);
1275	handle_t *handle = ext4_journal_current_handle();
1276	struct inode *inode = mapping->host;
1277	loff_t old_size = inode->i_size;
1278	int ret = 0, ret2;
1279	int i_size_changed = 0;
1280	bool verity = ext4_verity_in_progress(inode);
1281
1282	trace_ext4_write_end(inode, pos, len, copied);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1283
1284	if (ext4_has_inline_data(inode) &&
1285	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1286		return ext4_write_inline_data_end(inode, pos, len, copied,
1287						  folio);
1288
1289	copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1290	/*
1291	 * it's important to update i_size while still holding folio lock:
1292	 * page writeout could otherwise come in and zero beyond i_size.
1293	 *
1294	 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1295	 * blocks are being written past EOF, so skip the i_size update.
1296	 */
1297	if (!verity)
1298		i_size_changed = ext4_update_inode_size(inode, pos + copied);
1299	folio_unlock(folio);
1300	folio_put(folio);
1301
1302	if (old_size < pos && !verity)
1303		pagecache_isize_extended(inode, old_size, pos);
1304	/*
1305	 * Don't mark the inode dirty under folio lock. First, it unnecessarily
1306	 * makes the holding time of folio lock longer. Second, it forces lock
1307	 * ordering of folio lock and transaction start for journaling
1308	 * filesystems.
1309	 */
1310	if (i_size_changed)
1311		ret = ext4_mark_inode_dirty(handle, inode);
1312
1313	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
 
 
 
 
1314		/* if we have allocated more blocks and copied
1315		 * less. We will have blocks allocated outside
1316		 * inode->i_size. So truncate them
1317		 */
1318		ext4_orphan_add(handle, inode);
1319
 
 
 
1320	ret2 = ext4_journal_stop(handle);
1321	if (!ret)
1322		ret = ret2;
1323
1324	if (pos + len > inode->i_size && !verity) {
1325		ext4_truncate_failed_write(inode);
1326		/*
1327		 * If truncate failed early the inode might still be
1328		 * on the orphan list; we need to make sure the inode
1329		 * is removed from the orphan list in that case.
1330		 */
1331		if (inode->i_nlink)
1332			ext4_orphan_del(NULL, inode);
1333	}
1334
1335	return ret ? ret : copied;
1336}
1337
1338/*
1339 * This is a private version of folio_zero_new_buffers() which doesn't
1340 * set the buffer to be dirty, since in data=journalled mode we need
1341 * to call ext4_dirty_journalled_data() instead.
1342 */
1343static void ext4_journalled_zero_new_buffers(handle_t *handle,
1344					    struct inode *inode,
1345					    struct folio *folio,
1346					    unsigned from, unsigned to)
1347{
1348	unsigned int block_start = 0, block_end;
1349	struct buffer_head *head, *bh;
1350
1351	bh = head = folio_buffers(folio);
1352	do {
1353		block_end = block_start + bh->b_size;
1354		if (buffer_new(bh)) {
1355			if (block_end > from && block_start < to) {
1356				if (!folio_test_uptodate(folio)) {
1357					unsigned start, size;
1358
1359					start = max(from, block_start);
1360					size = min(to, block_end) - start;
1361
1362					folio_zero_range(folio, start, size);
1363					write_end_fn(handle, inode, bh);
1364				}
1365				clear_buffer_new(bh);
1366			}
1367		}
1368		block_start = block_end;
1369		bh = bh->b_this_page;
1370	} while (bh != head);
1371}
1372
1373static int ext4_journalled_write_end(struct file *file,
1374				     struct address_space *mapping,
1375				     loff_t pos, unsigned len, unsigned copied,
1376				     struct page *page, void *fsdata)
1377{
1378	struct folio *folio = page_folio(page);
1379	handle_t *handle = ext4_journal_current_handle();
1380	struct inode *inode = mapping->host;
1381	loff_t old_size = inode->i_size;
1382	int ret = 0, ret2;
1383	int partial = 0;
1384	unsigned from, to;
1385	int size_changed = 0;
1386	bool verity = ext4_verity_in_progress(inode);
1387
1388	trace_ext4_journalled_write_end(inode, pos, len, copied);
1389	from = pos & (PAGE_SIZE - 1);
1390	to = from + len;
1391
1392	BUG_ON(!ext4_handle_valid(handle));
1393
1394	if (ext4_has_inline_data(inode))
1395		return ext4_write_inline_data_end(inode, pos, len, copied,
1396						  folio);
1397
1398	if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1399		copied = 0;
1400		ext4_journalled_zero_new_buffers(handle, inode, folio,
1401						 from, to);
1402	} else {
1403		if (unlikely(copied < len))
1404			ext4_journalled_zero_new_buffers(handle, inode, folio,
1405							 from + copied, to);
1406		ret = ext4_walk_page_buffers(handle, inode,
1407					     folio_buffers(folio),
1408					     from, from + copied, &partial,
1409					     write_end_fn);
1410		if (!partial)
1411			folio_mark_uptodate(folio);
1412	}
1413	if (!verity)
1414		size_changed = ext4_update_inode_size(inode, pos + copied);
 
 
 
 
 
 
 
1415	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1416	folio_unlock(folio);
1417	folio_put(folio);
1418
1419	if (old_size < pos && !verity)
1420		pagecache_isize_extended(inode, old_size, pos);
1421
1422	if (size_changed) {
1423		ret2 = ext4_mark_inode_dirty(handle, inode);
1424		if (!ret)
1425			ret = ret2;
1426	}
1427
1428	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
 
 
1429		/* if we have allocated more blocks and copied
1430		 * less. We will have blocks allocated outside
1431		 * inode->i_size. So truncate them
1432		 */
1433		ext4_orphan_add(handle, inode);
1434
1435	ret2 = ext4_journal_stop(handle);
1436	if (!ret)
1437		ret = ret2;
1438	if (pos + len > inode->i_size && !verity) {
1439		ext4_truncate_failed_write(inode);
1440		/*
1441		 * If truncate failed early the inode might still be
1442		 * on the orphan list; we need to make sure the inode
1443		 * is removed from the orphan list in that case.
1444		 */
1445		if (inode->i_nlink)
1446			ext4_orphan_del(NULL, inode);
1447	}
1448
1449	return ret ? ret : copied;
1450}
1451
1452/*
1453 * Reserve space for a single cluster
1454 */
1455static int ext4_da_reserve_space(struct inode *inode)
1456{
 
1457	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1458	struct ext4_inode_info *ei = EXT4_I(inode);
 
1459	int ret;
 
 
1460
1461	/*
1462	 * We will charge metadata quota at writeout time; this saves
1463	 * us from metadata over-estimation, though we may go over by
1464	 * a small amount in the end.  Here we just reserve for data.
1465	 */
1466	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1467	if (ret)
1468		return ret;
1469
 
 
 
 
 
 
1470	spin_lock(&ei->i_block_reservation_lock);
1471	if (ext4_claim_free_clusters(sbi, 1, 0)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1472		spin_unlock(&ei->i_block_reservation_lock);
 
 
 
 
1473		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1474		return -ENOSPC;
1475	}
1476	ei->i_reserved_data_blocks++;
1477	trace_ext4_da_reserve_space(inode);
1478	spin_unlock(&ei->i_block_reservation_lock);
1479
1480	return 0;       /* success */
1481}
1482
1483void ext4_da_release_space(struct inode *inode, int to_free)
1484{
1485	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1486	struct ext4_inode_info *ei = EXT4_I(inode);
1487
1488	if (!to_free)
1489		return;		/* Nothing to release, exit */
1490
1491	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1492
1493	trace_ext4_da_release_space(inode, to_free);
1494	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1495		/*
1496		 * if there aren't enough reserved blocks, then the
1497		 * counter is messed up somewhere.  Since this
1498		 * function is called from invalidate page, it's
1499		 * harmless to return without any action.
1500		 */
1501		ext4_warning(inode->i_sb, "ext4_da_release_space: "
1502			 "ino %lu, to_free %d with only %d reserved "
1503			 "data blocks", inode->i_ino, to_free,
1504			 ei->i_reserved_data_blocks);
1505		WARN_ON(1);
1506		to_free = ei->i_reserved_data_blocks;
1507	}
1508	ei->i_reserved_data_blocks -= to_free;
1509
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1510	/* update fs dirty data blocks counter */
1511	percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1512
1513	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1514
1515	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1516}
1517
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1518/*
1519 * Delayed allocation stuff
1520 */
1521
1522struct mpage_da_data {
1523	/* These are input fields for ext4_do_writepages() */
1524	struct inode *inode;
1525	struct writeback_control *wbc;
1526	unsigned int can_map:1;	/* Can writepages call map blocks? */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1527
1528	/* These are internal state of ext4_do_writepages() */
1529	pgoff_t first_page;	/* The first page to write */
1530	pgoff_t next_page;	/* Current page to examine */
1531	pgoff_t last_page;	/* Last page to examine */
1532	/*
1533	 * Extent to map - this can be after first_page because that can be
1534	 * fully mapped. We somewhat abuse m_flags to store whether the extent
1535	 * is delalloc or unwritten.
1536	 */
1537	struct ext4_map_blocks map;
1538	struct ext4_io_submit io_submit;	/* IO submission data */
1539	unsigned int do_map:1;
1540	unsigned int scanned_until_end:1;
1541	unsigned int journalled_more_data:1;
1542};
1543
1544static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1545				       bool invalidate)
1546{
1547	unsigned nr, i;
1548	pgoff_t index, end;
1549	struct folio_batch fbatch;
1550	struct inode *inode = mpd->inode;
1551	struct address_space *mapping = inode->i_mapping;
1552
1553	/* This is necessary when next_page == 0. */
1554	if (mpd->first_page >= mpd->next_page)
1555		return;
1556
1557	mpd->scanned_until_end = 0;
1558	index = mpd->first_page;
1559	end   = mpd->next_page - 1;
1560	if (invalidate) {
1561		ext4_lblk_t start, last;
1562		start = index << (PAGE_SHIFT - inode->i_blkbits);
1563		last = end << (PAGE_SHIFT - inode->i_blkbits);
1564
1565		/*
1566		 * avoid racing with extent status tree scans made by
1567		 * ext4_insert_delayed_block()
1568		 */
1569		down_write(&EXT4_I(inode)->i_data_sem);
1570		ext4_es_remove_extent(inode, start, last - start + 1);
1571		up_write(&EXT4_I(inode)->i_data_sem);
1572	}
1573
1574	folio_batch_init(&fbatch);
1575	while (index <= end) {
1576		nr = filemap_get_folios(mapping, &index, end, &fbatch);
1577		if (nr == 0)
1578			break;
1579		for (i = 0; i < nr; i++) {
1580			struct folio *folio = fbatch.folios[i];
1581
1582			if (folio->index < mpd->first_page)
1583				continue;
1584			if (folio_next_index(folio) - 1 > end)
1585				continue;
1586			BUG_ON(!folio_test_locked(folio));
1587			BUG_ON(folio_test_writeback(folio));
1588			if (invalidate) {
1589				if (folio_mapped(folio))
1590					folio_clear_dirty_for_io(folio);
1591				block_invalidate_folio(folio, 0,
1592						folio_size(folio));
1593				folio_clear_uptodate(folio);
1594			}
1595			folio_unlock(folio);
1596		}
1597		folio_batch_release(&fbatch);
 
1598	}
 
1599}
1600
1601static void ext4_print_free_blocks(struct inode *inode)
1602{
1603	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1604	struct super_block *sb = inode->i_sb;
1605	struct ext4_inode_info *ei = EXT4_I(inode);
1606
1607	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1608	       EXT4_C2B(EXT4_SB(inode->i_sb),
1609			ext4_count_free_clusters(sb)));
1610	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1611	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1612	       (long long) EXT4_C2B(EXT4_SB(sb),
1613		percpu_counter_sum(&sbi->s_freeclusters_counter)));
1614	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1615	       (long long) EXT4_C2B(EXT4_SB(sb),
1616		percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1617	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1618	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1619		 ei->i_reserved_data_blocks);
 
 
1620	return;
1621}
1622
1623/*
1624 * ext4_insert_delayed_block - adds a delayed block to the extents status
1625 *                             tree, incrementing the reserved cluster/block
1626 *                             count or making a pending reservation
1627 *                             where needed
1628 *
1629 * @inode - file containing the newly added block
1630 * @lblk - logical block to be added
 
1631 *
1632 * Returns 0 on success, negative error code on failure.
1633 */
1634static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1635{
1636	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1637	int ret;
1638	bool allocated = false;
 
 
 
1639
1640	/*
1641	 * If the cluster containing lblk is shared with a delayed,
1642	 * written, or unwritten extent in a bigalloc file system, it's
1643	 * already been accounted for and does not need to be reserved.
1644	 * A pending reservation must be made for the cluster if it's
1645	 * shared with a written or unwritten extent and doesn't already
1646	 * have one.  Written and unwritten extents can be purged from the
1647	 * extents status tree if the system is under memory pressure, so
1648	 * it's necessary to examine the extent tree if a search of the
1649	 * extents status tree doesn't get a match.
1650	 */
1651	if (sbi->s_cluster_ratio == 1) {
1652		ret = ext4_da_reserve_space(inode);
1653		if (ret != 0)   /* ENOSPC */
1654			return ret;
1655	} else {   /* bigalloc */
1656		if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1657			if (!ext4_es_scan_clu(inode,
1658					      &ext4_es_is_mapped, lblk)) {
1659				ret = ext4_clu_mapped(inode,
1660						      EXT4_B2C(sbi, lblk));
1661				if (ret < 0)
1662					return ret;
1663				if (ret == 0) {
1664					ret = ext4_da_reserve_space(inode);
1665					if (ret != 0)   /* ENOSPC */
1666						return ret;
1667				} else {
1668					allocated = true;
1669				}
1670			} else {
1671				allocated = true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1672			}
1673		}
1674	}
1675
1676	ext4_es_insert_delayed_block(inode, lblk, allocated);
1677	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1678}
1679
1680/*
1681 * This function is grabs code from the very beginning of
1682 * ext4_map_blocks, but assumes that the caller is from delayed write
1683 * time. This function looks up the requested blocks and sets the
1684 * buffer delay bit under the protection of i_data_sem.
1685 */
1686static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1687			      struct ext4_map_blocks *map,
1688			      struct buffer_head *bh)
1689{
1690	struct extent_status es;
1691	int retval;
1692	sector_t invalid_block = ~((sector_t) 0xffff);
1693#ifdef ES_AGGRESSIVE_TEST
1694	struct ext4_map_blocks orig_map;
1695
1696	memcpy(&orig_map, map, sizeof(*map));
1697#endif
1698
1699	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1700		invalid_block = ~0;
1701
1702	map->m_flags = 0;
1703	ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
 
1704		  (unsigned long) map->m_lblk);
1705
1706	/* Lookup extent status tree firstly */
1707	if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1708		if (ext4_es_is_hole(&es))
1709			goto add_delayed;
1710
1711		/*
1712		 * Delayed extent could be allocated by fallocate.
1713		 * So we need to check it.
1714		 */
1715		if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1716			map_bh(bh, inode->i_sb, invalid_block);
1717			set_buffer_new(bh);
1718			set_buffer_delay(bh);
1719			return 0;
1720		}
1721
1722		map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1723		retval = es.es_len - (iblock - es.es_lblk);
1724		if (retval > map->m_len)
1725			retval = map->m_len;
1726		map->m_len = retval;
1727		if (ext4_es_is_written(&es))
1728			map->m_flags |= EXT4_MAP_MAPPED;
1729		else if (ext4_es_is_unwritten(&es))
1730			map->m_flags |= EXT4_MAP_UNWRITTEN;
1731		else
1732			BUG();
1733
1734#ifdef ES_AGGRESSIVE_TEST
1735		ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1736#endif
1737		return retval;
1738	}
1739
1740	/*
1741	 * Try to see if we can get the block without requesting a new
1742	 * file system block.
1743	 */
1744	down_read(&EXT4_I(inode)->i_data_sem);
1745	if (ext4_has_inline_data(inode))
1746		retval = 0;
1747	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1748		retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1749	else
1750		retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1751	if (retval < 0) {
1752		up_read(&EXT4_I(inode)->i_data_sem);
1753		return retval;
1754	}
1755	if (retval > 0) {
1756		unsigned int status;
1757
1758		if (unlikely(retval != map->m_len)) {
1759			ext4_warning(inode->i_sb,
1760				     "ES len assertion failed for inode "
1761				     "%lu: retval %d != map->m_len %d",
1762				     inode->i_ino, retval, map->m_len);
1763			WARN_ON(1);
 
 
 
 
 
 
1764		}
1765
1766		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1767				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1768		ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1769				      map->m_pblk, status);
1770		up_read(&EXT4_I(inode)->i_data_sem);
1771		return retval;
 
 
1772	}
1773	up_read(&EXT4_I(inode)->i_data_sem);
1774
1775add_delayed:
1776	down_write(&EXT4_I(inode)->i_data_sem);
1777	retval = ext4_insert_delayed_block(inode, map->m_lblk);
1778	up_write(&EXT4_I(inode)->i_data_sem);
1779	if (retval)
1780		return retval;
1781
1782	map_bh(bh, inode->i_sb, invalid_block);
1783	set_buffer_new(bh);
1784	set_buffer_delay(bh);
1785	return retval;
1786}
1787
1788/*
1789 * This is a special get_block_t callback which is used by
1790 * ext4_da_write_begin().  It will either return mapped block or
1791 * reserve space for a single block.
1792 *
1793 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1794 * We also have b_blocknr = -1 and b_bdev initialized properly
1795 *
1796 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1797 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1798 * initialized properly.
1799 */
1800int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1801			   struct buffer_head *bh, int create)
1802{
1803	struct ext4_map_blocks map;
1804	int ret = 0;
1805
1806	BUG_ON(create == 0);
1807	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1808
1809	map.m_lblk = iblock;
1810	map.m_len = 1;
1811
1812	/*
1813	 * first, we need to know whether the block is allocated already
1814	 * preallocated blocks are unmapped but should treated
1815	 * the same as allocated blocks.
1816	 */
1817	ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1818	if (ret <= 0)
1819		return ret;
1820
1821	map_bh(bh, inode->i_sb, map.m_pblk);
1822	ext4_update_bh_state(bh, map.m_flags);
1823
1824	if (buffer_unwritten(bh)) {
1825		/* A delayed write to unwritten bh should be marked
1826		 * new and mapped.  Mapped ensures that we don't do
1827		 * get_block multiple times when we write to the same
1828		 * offset and new ensures that we do proper zero out
1829		 * for partial write.
1830		 */
1831		set_buffer_new(bh);
1832		set_buffer_mapped(bh);
1833	}
1834	return 0;
1835}
1836
1837static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1838{
1839	mpd->first_page += folio_nr_pages(folio);
1840	folio_unlock(folio);
1841}
1842
1843static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
1844{
1845	size_t len;
1846	loff_t size;
1847	int err;
1848
1849	BUG_ON(folio->index != mpd->first_page);
1850	folio_clear_dirty_for_io(folio);
1851	/*
1852	 * We have to be very careful here!  Nothing protects writeback path
1853	 * against i_size changes and the page can be writeably mapped into
1854	 * page tables. So an application can be growing i_size and writing
1855	 * data through mmap while writeback runs. folio_clear_dirty_for_io()
1856	 * write-protects our page in page tables and the page cannot get
1857	 * written to again until we release folio lock. So only after
1858	 * folio_clear_dirty_for_io() we are safe to sample i_size for
1859	 * ext4_bio_write_folio() to zero-out tail of the written page. We rely
1860	 * on the barrier provided by folio_test_clear_dirty() in
1861	 * folio_clear_dirty_for_io() to make sure i_size is really sampled only
1862	 * after page tables are updated.
1863	 */
1864	size = i_size_read(mpd->inode);
1865	len = folio_size(folio);
1866	if (folio_pos(folio) + len > size &&
1867	    !ext4_verity_in_progress(mpd->inode))
1868		len = size & ~PAGE_MASK;
1869	err = ext4_bio_write_folio(&mpd->io_submit, folio, len);
1870	if (!err)
1871		mpd->wbc->nr_to_write--;
1872
1873	return err;
1874}
1875
1876#define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
1877
1878/*
1879 * mballoc gives us at most this number of blocks...
1880 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1881 * The rest of mballoc seems to handle chunks up to full group size.
1882 */
1883#define MAX_WRITEPAGES_EXTENT_LEN 2048
1884
1885/*
1886 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1887 *
1888 * @mpd - extent of blocks
1889 * @lblk - logical number of the block in the file
1890 * @bh - buffer head we want to add to the extent
1891 *
1892 * The function is used to collect contig. blocks in the same state. If the
1893 * buffer doesn't require mapping for writeback and we haven't started the
1894 * extent of buffers to map yet, the function returns 'true' immediately - the
1895 * caller can write the buffer right away. Otherwise the function returns true
1896 * if the block has been added to the extent, false if the block couldn't be
1897 * added.
1898 */
1899static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1900				   struct buffer_head *bh)
1901{
1902	struct ext4_map_blocks *map = &mpd->map;
1903
1904	/* Buffer that doesn't need mapping for writeback? */
1905	if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1906	    (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1907		/* So far no extent to map => we write the buffer right away */
1908		if (map->m_len == 0)
1909			return true;
1910		return false;
1911	}
1912
1913	/* First block in the extent? */
1914	if (map->m_len == 0) {
1915		/* We cannot map unless handle is started... */
1916		if (!mpd->do_map)
1917			return false;
1918		map->m_lblk = lblk;
1919		map->m_len = 1;
1920		map->m_flags = bh->b_state & BH_FLAGS;
1921		return true;
1922	}
1923
1924	/* Don't go larger than mballoc is willing to allocate */
1925	if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1926		return false;
1927
1928	/* Can we merge the block to our big extent? */
1929	if (lblk == map->m_lblk + map->m_len &&
1930	    (bh->b_state & BH_FLAGS) == map->m_flags) {
1931		map->m_len++;
1932		return true;
1933	}
1934	return false;
1935}
1936
1937/*
1938 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1939 *
1940 * @mpd - extent of blocks for mapping
1941 * @head - the first buffer in the page
1942 * @bh - buffer we should start processing from
1943 * @lblk - logical number of the block in the file corresponding to @bh
1944 *
1945 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1946 * the page for IO if all buffers in this page were mapped and there's no
1947 * accumulated extent of buffers to map or add buffers in the page to the
1948 * extent of buffers to map. The function returns 1 if the caller can continue
1949 * by processing the next page, 0 if it should stop adding buffers to the
1950 * extent to map because we cannot extend it anymore. It can also return value
1951 * < 0 in case of error during IO submission.
1952 */
1953static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1954				   struct buffer_head *head,
1955				   struct buffer_head *bh,
1956				   ext4_lblk_t lblk)
1957{
1958	struct inode *inode = mpd->inode;
 
 
 
 
1959	int err;
1960	ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
1961							>> inode->i_blkbits;
1962
1963	if (ext4_verity_in_progress(inode))
1964		blocks = EXT_MAX_BLOCKS;
 
 
 
 
 
1965
1966	do {
1967		BUG_ON(buffer_locked(bh));
 
 
 
1968
1969		if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1970			/* Found extent to map? */
1971			if (mpd->map.m_len)
1972				return 0;
1973			/* Buffer needs mapping and handle is not started? */
1974			if (!mpd->do_map)
1975				return 0;
1976			/* Everything mapped so far and we hit EOF */
1977			break;
1978		}
1979	} while (lblk++, (bh = bh->b_this_page) != head);
1980	/* So far everything mapped? Submit the page for IO. */
1981	if (mpd->map.m_len == 0) {
1982		err = mpage_submit_folio(mpd, head->b_folio);
1983		if (err < 0)
1984			return err;
1985		mpage_folio_done(mpd, head->b_folio);
1986	}
1987	if (lblk >= blocks) {
1988		mpd->scanned_until_end = 1;
1989		return 0;
1990	}
1991	return 1;
1992}
1993
1994/*
1995 * mpage_process_folio - update folio buffers corresponding to changed extent
1996 *			 and may submit fully mapped page for IO
1997 * @mpd: description of extent to map, on return next extent to map
1998 * @folio: Contains these buffers.
1999 * @m_lblk: logical block mapping.
2000 * @m_pblk: corresponding physical mapping.
2001 * @map_bh: determines on return whether this page requires any further
2002 *		  mapping or not.
2003 *
2004 * Scan given folio buffers corresponding to changed extent and update buffer
2005 * state according to new extent state.
2006 * We map delalloc buffers to their physical location, clear unwritten bits.
2007 * If the given folio is not fully mapped, we update @mpd to the next extent in
2008 * the given folio that needs mapping & return @map_bh as true.
2009 */
2010static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2011			      ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2012			      bool *map_bh)
2013{
2014	struct buffer_head *head, *bh;
2015	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2016	ext4_lblk_t lblk = *m_lblk;
2017	ext4_fsblk_t pblock = *m_pblk;
2018	int err = 0;
2019	int blkbits = mpd->inode->i_blkbits;
2020	ssize_t io_end_size = 0;
2021	struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2022
2023	bh = head = folio_buffers(folio);
2024	do {
2025		if (lblk < mpd->map.m_lblk)
2026			continue;
2027		if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2028			/*
2029			 * Buffer after end of mapped extent.
2030			 * Find next buffer in the folio to map.
2031			 */
2032			mpd->map.m_len = 0;
2033			mpd->map.m_flags = 0;
2034			io_end_vec->size += io_end_size;
2035
2036			err = mpage_process_page_bufs(mpd, head, bh, lblk);
2037			if (err > 0)
2038				err = 0;
2039			if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2040				io_end_vec = ext4_alloc_io_end_vec(io_end);
2041				if (IS_ERR(io_end_vec)) {
2042					err = PTR_ERR(io_end_vec);
2043					goto out;
2044				}
2045				io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2046			}
2047			*map_bh = true;
2048			goto out;
2049		}
2050		if (buffer_delay(bh)) {
2051			clear_buffer_delay(bh);
2052			bh->b_blocknr = pblock++;
2053		}
2054		clear_buffer_unwritten(bh);
2055		io_end_size += (1 << blkbits);
2056	} while (lblk++, (bh = bh->b_this_page) != head);
2057
2058	io_end_vec->size += io_end_size;
2059	*map_bh = false;
2060out:
2061	*m_lblk = lblk;
2062	*m_pblk = pblock;
2063	return err;
2064}
2065
 
 
 
2066/*
2067 * mpage_map_buffers - update buffers corresponding to changed extent and
2068 *		       submit fully mapped pages for IO
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2069 *
2070 * @mpd - description of extent to map, on return next extent to map
2071 *
2072 * Scan buffers corresponding to changed extent (we expect corresponding pages
2073 * to be already locked) and update buffer state according to new extent state.
2074 * We map delalloc buffers to their physical location, clear unwritten bits,
2075 * and mark buffers as uninit when we perform writes to unwritten extents
2076 * and do extent conversion after IO is finished. If the last page is not fully
2077 * mapped, we update @map to the next extent in the last page that needs
2078 * mapping. Otherwise we submit the page for IO.
2079 */
2080static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
 
2081{
2082	struct folio_batch fbatch;
2083	unsigned nr, i;
2084	struct inode *inode = mpd->inode;
2085	int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2086	pgoff_t start, end;
2087	ext4_lblk_t lblk;
2088	ext4_fsblk_t pblock;
2089	int err;
2090	bool map_bh = false;
2091
2092	start = mpd->map.m_lblk >> bpp_bits;
2093	end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2094	lblk = start << bpp_bits;
2095	pblock = mpd->map.m_pblk;
2096
2097	folio_batch_init(&fbatch);
2098	while (start <= end) {
2099		nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2100		if (nr == 0)
2101			break;
2102		for (i = 0; i < nr; i++) {
2103			struct folio *folio = fbatch.folios[i];
2104
2105			err = mpage_process_folio(mpd, folio, &lblk, &pblock,
2106						 &map_bh);
2107			/*
2108			 * If map_bh is true, means page may require further bh
2109			 * mapping, or maybe the page was submitted for IO.
2110			 * So we return to call further extent mapping.
2111			 */
2112			if (err < 0 || map_bh)
2113				goto out;
2114			/* Page fully mapped - let IO run! */
2115			err = mpage_submit_folio(mpd, folio);
2116			if (err < 0)
2117				goto out;
2118			mpage_folio_done(mpd, folio);
2119		}
2120		folio_batch_release(&fbatch);
2121	}
2122	/* Extent fully mapped and matches with page boundary. We are done. */
2123	mpd->map.m_len = 0;
2124	mpd->map.m_flags = 0;
2125	return 0;
2126out:
2127	folio_batch_release(&fbatch);
2128	return err;
2129}
2130
2131static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2132{
2133	struct inode *inode = mpd->inode;
2134	struct ext4_map_blocks *map = &mpd->map;
2135	int get_blocks_flags;
2136	int err, dioread_nolock;
2137
2138	trace_ext4_da_write_pages_extent(inode, map);
2139	/*
2140	 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2141	 * to convert an unwritten extent to be initialized (in the case
2142	 * where we have written into one or more preallocated blocks).  It is
2143	 * possible that we're going to need more metadata blocks than
2144	 * previously reserved. However we must not fail because we're in
2145	 * writeback and there is nothing we can do about it so it might result
2146	 * in data loss.  So use reserved blocks to allocate metadata if
2147	 * possible.
2148	 *
2149	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2150	 * the blocks in question are delalloc blocks.  This indicates
2151	 * that the blocks and quotas has already been checked when
2152	 * the data was copied into the page cache.
2153	 */
2154	get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2155			   EXT4_GET_BLOCKS_METADATA_NOFAIL |
2156			   EXT4_GET_BLOCKS_IO_SUBMIT;
2157	dioread_nolock = ext4_should_dioread_nolock(inode);
2158	if (dioread_nolock)
2159		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2160	if (map->m_flags & BIT(BH_Delay))
2161		get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2162
2163	err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2164	if (err < 0)
2165		return err;
2166	if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2167		if (!mpd->io_submit.io_end->handle &&
2168		    ext4_handle_valid(handle)) {
2169			mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2170			handle->h_rsv_handle = NULL;
2171		}
2172		ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2173	}
2174
2175	BUG_ON(map->m_len == 0);
2176	return 0;
2177}
2178
2179/*
2180 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2181 *				 mpd->len and submit pages underlying it for IO
2182 *
2183 * @handle - handle for journal operations
2184 * @mpd - extent to map
2185 * @give_up_on_write - we set this to true iff there is a fatal error and there
2186 *                     is no hope of writing the data. The caller should discard
2187 *                     dirty pages to avoid infinite loops.
2188 *
2189 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2190 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2191 * them to initialized or split the described range from larger unwritten
2192 * extent. Note that we need not map all the described range since allocation
2193 * can return less blocks or the range is covered by more unwritten extents. We
2194 * cannot map more because we are limited by reserved transaction credits. On
2195 * the other hand we always make sure that the last touched page is fully
2196 * mapped so that it can be written out (and thus forward progress is
2197 * guaranteed). After mapping we submit all mapped pages for IO.
2198 */
2199static int mpage_map_and_submit_extent(handle_t *handle,
2200				       struct mpage_da_data *mpd,
2201				       bool *give_up_on_write)
2202{
2203	struct inode *inode = mpd->inode;
2204	struct ext4_map_blocks *map = &mpd->map;
2205	int err;
2206	loff_t disksize;
2207	int progress = 0;
2208	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2209	struct ext4_io_end_vec *io_end_vec;
2210
2211	io_end_vec = ext4_alloc_io_end_vec(io_end);
2212	if (IS_ERR(io_end_vec))
2213		return PTR_ERR(io_end_vec);
2214	io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2215	do {
2216		err = mpage_map_one_extent(handle, mpd);
2217		if (err < 0) {
2218			struct super_block *sb = inode->i_sb;
2219
2220			if (ext4_forced_shutdown(sb))
2221				goto invalidate_dirty_pages;
2222			/*
2223			 * Let the uper layers retry transient errors.
2224			 * In the case of ENOSPC, if ext4_count_free_blocks()
2225			 * is non-zero, a commit should free up blocks.
2226			 */
2227			if ((err == -ENOMEM) ||
2228			    (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2229				if (progress)
2230					goto update_disksize;
2231				return err;
2232			}
2233			ext4_msg(sb, KERN_CRIT,
2234				 "Delayed block allocation failed for "
2235				 "inode %lu at logical offset %llu with"
2236				 " max blocks %u with error %d",
2237				 inode->i_ino,
2238				 (unsigned long long)map->m_lblk,
2239				 (unsigned)map->m_len, -err);
2240			ext4_msg(sb, KERN_CRIT,
2241				 "This should not happen!! Data will "
2242				 "be lost\n");
2243			if (err == -ENOSPC)
2244				ext4_print_free_blocks(inode);
2245		invalidate_dirty_pages:
2246			*give_up_on_write = true;
2247			return err;
2248		}
2249		progress = 1;
2250		/*
2251		 * Update buffer state, submit mapped pages, and get us new
2252		 * extent to map
2253		 */
2254		err = mpage_map_and_submit_buffers(mpd);
2255		if (err < 0)
2256			goto update_disksize;
2257	} while (map->m_len);
 
 
 
 
 
2258
2259update_disksize:
2260	/*
2261	 * Update on-disk size after IO is submitted.  Races with
2262	 * truncate are avoided by checking i_size under i_data_sem.
2263	 */
2264	disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2265	if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2266		int err2;
2267		loff_t i_size;
2268
2269		down_write(&EXT4_I(inode)->i_data_sem);
2270		i_size = i_size_read(inode);
2271		if (disksize > i_size)
2272			disksize = i_size;
2273		if (disksize > EXT4_I(inode)->i_disksize)
2274			EXT4_I(inode)->i_disksize = disksize;
2275		up_write(&EXT4_I(inode)->i_data_sem);
2276		err2 = ext4_mark_inode_dirty(handle, inode);
2277		if (err2) {
2278			ext4_error_err(inode->i_sb, -err2,
2279				       "Failed to mark inode %lu dirty",
2280				       inode->i_ino);
2281		}
2282		if (!err)
2283			err = err2;
2284	}
2285	return err;
2286}
2287
2288/*
2289 * Calculate the total number of credits to reserve for one writepages
2290 * iteration. This is called from ext4_writepages(). We map an extent of
2291 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2292 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2293 * bpp - 1 blocks in bpp different extents.
2294 */
 
2295static int ext4_da_writepages_trans_blocks(struct inode *inode)
2296{
2297	int bpp = ext4_journal_blocks_per_page(inode);
2298
2299	return ext4_meta_trans_blocks(inode,
2300				MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2301}
2302
2303static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio,
2304				     size_t len)
2305{
2306	struct buffer_head *page_bufs = folio_buffers(folio);
2307	struct inode *inode = folio->mapping->host;
2308	int ret, err;
2309
2310	ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2311				     NULL, do_journal_get_write_access);
2312	err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2313				     NULL, write_end_fn);
2314	if (ret == 0)
2315		ret = err;
2316	err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len);
2317	if (ret == 0)
2318		ret = err;
2319	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2320
2321	return ret;
2322}
2323
2324static int mpage_journal_page_buffers(handle_t *handle,
2325				      struct mpage_da_data *mpd,
2326				      struct folio *folio)
2327{
2328	struct inode *inode = mpd->inode;
2329	loff_t size = i_size_read(inode);
2330	size_t len = folio_size(folio);
2331
2332	folio_clear_checked(folio);
2333	mpd->wbc->nr_to_write--;
2334
2335	if (folio_pos(folio) + len > size &&
2336	    !ext4_verity_in_progress(inode))
2337		len = size - folio_pos(folio);
2338
2339	return ext4_journal_folio_buffers(handle, folio, len);
2340}
2341
2342/*
2343 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2344 * 				 needing mapping, submit mapped pages
2345 *
2346 * @mpd - where to look for pages
2347 *
2348 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2349 * IO immediately. If we cannot map blocks, we submit just already mapped
2350 * buffers in the page for IO and keep page dirty. When we can map blocks and
2351 * we find a page which isn't mapped we start accumulating extent of buffers
2352 * underlying these pages that needs mapping (formed by either delayed or
2353 * unwritten buffers). We also lock the pages containing these buffers. The
2354 * extent found is returned in @mpd structure (starting at mpd->lblk with
2355 * length mpd->len blocks).
2356 *
2357 * Note that this function can attach bios to one io_end structure which are
2358 * neither logically nor physically contiguous. Although it may seem as an
2359 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2360 * case as we need to track IO to all buffers underlying a page in one io_end.
2361 */
2362static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2363{
2364	struct address_space *mapping = mpd->inode->i_mapping;
2365	struct folio_batch fbatch;
2366	unsigned int nr_folios;
2367	pgoff_t index = mpd->first_page;
2368	pgoff_t end = mpd->last_page;
2369	xa_mark_t tag;
2370	int i, err = 0;
2371	int blkbits = mpd->inode->i_blkbits;
2372	ext4_lblk_t lblk;
2373	struct buffer_head *head;
2374	handle_t *handle = NULL;
2375	int bpp = ext4_journal_blocks_per_page(mpd->inode);
2376
2377	if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2378		tag = PAGECACHE_TAG_TOWRITE;
2379	else
2380		tag = PAGECACHE_TAG_DIRTY;
2381
2382	mpd->map.m_len = 0;
2383	mpd->next_page = index;
2384	if (ext4_should_journal_data(mpd->inode)) {
2385		handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2386					    bpp);
2387		if (IS_ERR(handle))
2388			return PTR_ERR(handle);
2389	}
2390	folio_batch_init(&fbatch);
2391	while (index <= end) {
2392		nr_folios = filemap_get_folios_tag(mapping, &index, end,
2393				tag, &fbatch);
2394		if (nr_folios == 0)
2395			break;
2396
2397		for (i = 0; i < nr_folios; i++) {
2398			struct folio *folio = fbatch.folios[i];
2399
2400			/*
2401			 * Accumulated enough dirty pages? This doesn't apply
2402			 * to WB_SYNC_ALL mode. For integrity sync we have to
2403			 * keep going because someone may be concurrently
2404			 * dirtying pages, and we might have synced a lot of
2405			 * newly appeared dirty pages, but have not synced all
2406			 * of the old dirty pages.
2407			 */
2408			if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2409			    mpd->wbc->nr_to_write <=
2410			    mpd->map.m_len >> (PAGE_SHIFT - blkbits))
2411				goto out;
2412
2413			/* If we can't merge this page, we are done. */
2414			if (mpd->map.m_len > 0 && mpd->next_page != folio->index)
2415				goto out;
2416
2417			if (handle) {
2418				err = ext4_journal_ensure_credits(handle, bpp,
2419								  0);
2420				if (err < 0)
2421					goto out;
2422			}
2423
2424			folio_lock(folio);
2425			/*
2426			 * If the page is no longer dirty, or its mapping no
2427			 * longer corresponds to inode we are writing (which
2428			 * means it has been truncated or invalidated), or the
2429			 * page is already under writeback and we are not doing
2430			 * a data integrity writeback, skip the page
2431			 */
2432			if (!folio_test_dirty(folio) ||
2433			    (folio_test_writeback(folio) &&
2434			     (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2435			    unlikely(folio->mapping != mapping)) {
2436				folio_unlock(folio);
2437				continue;
2438			}
2439
2440			folio_wait_writeback(folio);
2441			BUG_ON(folio_test_writeback(folio));
2442
2443			/*
2444			 * Should never happen but for buggy code in
2445			 * other subsystems that call
2446			 * set_page_dirty() without properly warning
2447			 * the file system first.  See [1] for more
2448			 * information.
2449			 *
2450			 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2451			 */
2452			if (!folio_buffers(folio)) {
2453				ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2454				folio_clear_dirty(folio);
2455				folio_unlock(folio);
 
2456				continue;
2457			}
2458
2459			if (mpd->map.m_len == 0)
2460				mpd->first_page = folio->index;
2461			mpd->next_page = folio_next_index(folio);
2462			/*
2463			 * Writeout when we cannot modify metadata is simple.
2464			 * Just submit the page. For data=journal mode we
2465			 * first handle writeout of the page for checkpoint and
2466			 * only after that handle delayed page dirtying. This
2467			 * makes sure current data is checkpointed to the final
2468			 * location before possibly journalling it again which
2469			 * is desirable when the page is frequently dirtied
2470			 * through a pin.
2471			 */
2472			if (!mpd->can_map) {
2473				err = mpage_submit_folio(mpd, folio);
2474				if (err < 0)
2475					goto out;
2476				/* Pending dirtying of journalled data? */
2477				if (folio_test_checked(folio)) {
2478					err = mpage_journal_page_buffers(handle,
2479						mpd, folio);
2480					if (err < 0)
2481						goto out;
2482					mpd->journalled_more_data = 1;
2483				}
2484				mpage_folio_done(mpd, folio);
2485			} else {
2486				/* Add all dirty buffers to mpd */
2487				lblk = ((ext4_lblk_t)folio->index) <<
2488					(PAGE_SHIFT - blkbits);
2489				head = folio_buffers(folio);
2490				err = mpage_process_page_bufs(mpd, head, head,
2491						lblk);
2492				if (err <= 0)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2493					goto out;
2494				err = 0;
2495			}
2496		}
2497		folio_batch_release(&fbatch);
2498		cond_resched();
2499	}
2500	mpd->scanned_until_end = 1;
2501	if (handle)
2502		ext4_journal_stop(handle);
2503	return 0;
 
 
2504out:
2505	folio_batch_release(&fbatch);
2506	if (handle)
2507		ext4_journal_stop(handle);
2508	return err;
2509}
2510
2511static int ext4_do_writepages(struct mpage_da_data *mpd)
 
 
2512{
2513	struct writeback_control *wbc = mpd->wbc;
2514	pgoff_t	writeback_index = 0;
2515	long nr_to_write = wbc->nr_to_write;
2516	int range_whole = 0;
2517	int cycled = 1;
2518	handle_t *handle = NULL;
2519	struct inode *inode = mpd->inode;
2520	struct address_space *mapping = inode->i_mapping;
2521	int needed_blocks, rsv_blocks = 0, ret = 0;
 
 
 
 
 
2522	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
 
 
2523	struct blk_plug plug;
2524	bool give_up_on_write = false;
2525
2526	trace_ext4_writepages(inode, wbc);
2527
2528	/*
2529	 * No pages to write? This is mainly a kludge to avoid starting
2530	 * a transaction for special inodes like journal inode on last iput()
2531	 * because that could violate lock ordering on umount
2532	 */
2533	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2534		goto out_writepages;
2535
2536	/*
2537	 * If the filesystem has aborted, it is read-only, so return
2538	 * right away instead of dumping stack traces later on that
2539	 * will obscure the real source of the problem.  We test
2540	 * fs shutdown state instead of sb->s_flag's SB_RDONLY because
2541	 * the latter could be true if the filesystem is mounted
2542	 * read-only, and in that case, ext4_writepages should
2543	 * *never* be called, so if that ever happens, we would want
2544	 * the stack trace.
2545	 */
2546	if (unlikely(ext4_forced_shutdown(mapping->host->i_sb))) {
2547		ret = -EROFS;
2548		goto out_writepages;
2549	}
2550
2551	/*
2552	 * If we have inline data and arrive here, it means that
2553	 * we will soon create the block for the 1st page, so
2554	 * we'd better clear the inline data here.
2555	 */
2556	if (ext4_has_inline_data(inode)) {
2557		/* Just inode will be modified... */
2558		handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2559		if (IS_ERR(handle)) {
2560			ret = PTR_ERR(handle);
2561			goto out_writepages;
2562		}
2563		BUG_ON(ext4_test_inode_state(inode,
2564				EXT4_STATE_MAY_INLINE_DATA));
2565		ext4_destroy_inline_data(handle, inode);
2566		ext4_journal_stop(handle);
2567	}
2568
2569	/*
2570	 * data=journal mode does not do delalloc so we just need to writeout /
2571	 * journal already mapped buffers. On the other hand we need to commit
2572	 * transaction to make data stable. We expect all the data to be
2573	 * already in the journal (the only exception are DMA pinned pages
2574	 * dirtied behind our back) so we commit transaction here and run the
2575	 * writeback loop to checkpoint them. The checkpointing is not actually
2576	 * necessary to make data persistent *but* quite a few places (extent
2577	 * shifting operations, fsverity, ...) depend on being able to drop
2578	 * pagecache pages after calling filemap_write_and_wait() and for that
2579	 * checkpointing needs to happen.
2580	 */
2581	if (ext4_should_journal_data(inode)) {
2582		mpd->can_map = 0;
2583		if (wbc->sync_mode == WB_SYNC_ALL)
2584			ext4_fc_commit(sbi->s_journal,
2585				       EXT4_I(inode)->i_datasync_tid);
2586	}
2587	mpd->journalled_more_data = 0;
2588
2589	if (ext4_should_dioread_nolock(inode)) {
2590		/*
2591		 * We may need to convert up to one extent per block in
2592		 * the page and we may dirty the inode.
2593		 */
2594		rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2595						PAGE_SIZE >> inode->i_blkbits);
2596	}
2597
2598	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2599		range_whole = 1;
2600
 
2601	if (wbc->range_cyclic) {
2602		writeback_index = mapping->writeback_index;
2603		if (writeback_index)
2604			cycled = 0;
2605		mpd->first_page = writeback_index;
2606		mpd->last_page = -1;
 
 
2607	} else {
2608		mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2609		mpd->last_page = wbc->range_end >> PAGE_SHIFT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2610	}
2611
2612	ext4_io_submit_init(&mpd->io_submit, wbc);
2613retry:
2614	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2615		tag_pages_for_writeback(mapping, mpd->first_page,
2616					mpd->last_page);
2617	blk_start_plug(&plug);
 
2618
2619	/*
2620	 * First writeback pages that don't need mapping - we can avoid
2621	 * starting a transaction unnecessarily and also avoid being blocked
2622	 * in the block layer on device congestion while having transaction
2623	 * started.
2624	 */
2625	mpd->do_map = 0;
2626	mpd->scanned_until_end = 0;
2627	mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2628	if (!mpd->io_submit.io_end) {
2629		ret = -ENOMEM;
2630		goto unplug;
2631	}
2632	ret = mpage_prepare_extent_to_map(mpd);
2633	/* Unlock pages we didn't use */
2634	mpage_release_unused_pages(mpd, false);
2635	/* Submit prepared bio */
2636	ext4_io_submit(&mpd->io_submit);
2637	ext4_put_io_end_defer(mpd->io_submit.io_end);
2638	mpd->io_submit.io_end = NULL;
2639	if (ret < 0)
2640		goto unplug;
2641
2642	while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2643		/* For each extent of pages we use new io_end */
2644		mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2645		if (!mpd->io_submit.io_end) {
2646			ret = -ENOMEM;
2647			break;
2648		}
2649
2650		WARN_ON_ONCE(!mpd->can_map);
2651		/*
2652		 * We have two constraints: We find one extent to map and we
2653		 * must always write out whole page (makes a difference when
2654		 * blocksize < pagesize) so that we don't block on IO when we
2655		 * try to write out the rest of the page. Journalled mode is
2656		 * not supported by delalloc.
2657		 */
2658		BUG_ON(ext4_should_journal_data(inode));
2659		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2660
2661		/* start a new transaction */
2662		handle = ext4_journal_start_with_reserve(inode,
2663				EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2664		if (IS_ERR(handle)) {
2665			ret = PTR_ERR(handle);
2666			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2667			       "%ld pages, ino %lu; err %d", __func__,
2668				wbc->nr_to_write, inode->i_ino, ret);
2669			/* Release allocated io_end */
2670			ext4_put_io_end(mpd->io_submit.io_end);
2671			mpd->io_submit.io_end = NULL;
2672			break;
2673		}
2674		mpd->do_map = 1;
2675
2676		trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2677		ret = mpage_prepare_extent_to_map(mpd);
2678		if (!ret && mpd->map.m_len)
2679			ret = mpage_map_and_submit_extent(handle, mpd,
2680					&give_up_on_write);
2681		/*
2682		 * Caution: If the handle is synchronous,
2683		 * ext4_journal_stop() can wait for transaction commit
2684		 * to finish which may depend on writeback of pages to
2685		 * complete or on page lock to be released.  In that
2686		 * case, we have to wait until after we have
2687		 * submitted all the IO, released page locks we hold,
2688		 * and dropped io_end reference (for extent conversion
2689		 * to be able to complete) before stopping the handle.
 
2690		 */
2691		if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2692			ext4_journal_stop(handle);
2693			handle = NULL;
2694			mpd->do_map = 0;
2695		}
2696		/* Unlock pages we didn't use */
2697		mpage_release_unused_pages(mpd, give_up_on_write);
2698		/* Submit prepared bio */
2699		ext4_io_submit(&mpd->io_submit);
2700
2701		/*
2702		 * Drop our io_end reference we got from init. We have
2703		 * to be careful and use deferred io_end finishing if
2704		 * we are still holding the transaction as we can
2705		 * release the last reference to io_end which may end
2706		 * up doing unwritten extent conversion.
2707		 */
2708		if (handle) {
2709			ext4_put_io_end_defer(mpd->io_submit.io_end);
2710			ext4_journal_stop(handle);
2711		} else
2712			ext4_put_io_end(mpd->io_submit.io_end);
2713		mpd->io_submit.io_end = NULL;
2714
2715		if (ret == -ENOSPC && sbi->s_journal) {
2716			/*
2717			 * Commit the transaction which would
2718			 * free blocks released in the transaction
2719			 * and try again
2720			 */
2721			jbd2_journal_force_commit_nested(sbi->s_journal);
2722			ret = 0;
2723			continue;
2724		}
2725		/* Fatal error - ENOMEM, EIO... */
2726		if (ret)
 
 
 
 
 
 
 
 
 
 
 
2727			break;
2728	}
2729unplug:
2730	blk_finish_plug(&plug);
2731	if (!ret && !cycled && wbc->nr_to_write > 0) {
2732		cycled = 1;
2733		mpd->last_page = writeback_index - 1;
2734		mpd->first_page = 0;
 
2735		goto retry;
2736	}
2737
2738	/* Update index */
 
2739	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2740		/*
2741		 * Set the writeback_index so that range_cyclic
2742		 * mode will write it back later
2743		 */
2744		mapping->writeback_index = mpd->first_page;
2745
2746out_writepages:
2747	trace_ext4_writepages_result(inode, wbc, ret,
2748				     nr_to_write - wbc->nr_to_write);
2749	return ret;
2750}
2751
2752static int ext4_writepages(struct address_space *mapping,
2753			   struct writeback_control *wbc)
2754{
2755	struct super_block *sb = mapping->host->i_sb;
2756	struct mpage_da_data mpd = {
2757		.inode = mapping->host,
2758		.wbc = wbc,
2759		.can_map = 1,
2760	};
2761	int ret;
2762	int alloc_ctx;
2763
2764	if (unlikely(ext4_forced_shutdown(sb)))
2765		return -EIO;
2766
2767	alloc_ctx = ext4_writepages_down_read(sb);
2768	ret = ext4_do_writepages(&mpd);
2769	/*
2770	 * For data=journal writeback we could have come across pages marked
2771	 * for delayed dirtying (PageChecked) which were just added to the
2772	 * running transaction. Try once more to get them to stable storage.
2773	 */
2774	if (!ret && mpd.journalled_more_data)
2775		ret = ext4_do_writepages(&mpd);
2776	ext4_writepages_up_read(sb, alloc_ctx);
2777
2778	return ret;
2779}
2780
2781int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2782{
2783	struct writeback_control wbc = {
2784		.sync_mode = WB_SYNC_ALL,
2785		.nr_to_write = LONG_MAX,
2786		.range_start = jinode->i_dirty_start,
2787		.range_end = jinode->i_dirty_end,
2788	};
2789	struct mpage_da_data mpd = {
2790		.inode = jinode->i_vfs_inode,
2791		.wbc = &wbc,
2792		.can_map = 0,
2793	};
2794	return ext4_do_writepages(&mpd);
2795}
2796
2797static int ext4_dax_writepages(struct address_space *mapping,
2798			       struct writeback_control *wbc)
2799{
2800	int ret;
2801	long nr_to_write = wbc->nr_to_write;
2802	struct inode *inode = mapping->host;
2803	int alloc_ctx;
2804
2805	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2806		return -EIO;
2807
2808	alloc_ctx = ext4_writepages_down_read(inode->i_sb);
2809	trace_ext4_writepages(inode, wbc);
2810
2811	ret = dax_writeback_mapping_range(mapping,
2812					  EXT4_SB(inode->i_sb)->s_daxdev, wbc);
2813	trace_ext4_writepages_result(inode, wbc, ret,
2814				     nr_to_write - wbc->nr_to_write);
2815	ext4_writepages_up_read(inode->i_sb, alloc_ctx);
2816	return ret;
2817}
2818
 
2819static int ext4_nonda_switch(struct super_block *sb)
2820{
2821	s64 free_clusters, dirty_clusters;
2822	struct ext4_sb_info *sbi = EXT4_SB(sb);
2823
2824	/*
2825	 * switch to non delalloc mode if we are running low
2826	 * on free block. The free block accounting via percpu
2827	 * counters can get slightly wrong with percpu_counter_batch getting
2828	 * accumulated on each CPU without updating global counters
2829	 * Delalloc need an accurate free block accounting. So switch
2830	 * to non delalloc when we are near to error range.
2831	 */
2832	free_clusters =
2833		percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2834	dirty_clusters =
2835		percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2836	/*
2837	 * Start pushing delalloc when 1/2 of free blocks are dirty.
2838	 */
2839	if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2840		try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2841
2842	if (2 * free_clusters < 3 * dirty_clusters ||
2843	    free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2844		/*
2845		 * free block count is less than 150% of dirty blocks
2846		 * or free blocks is less than watermark
2847		 */
2848		return 1;
2849	}
 
 
 
 
 
 
 
2850	return 0;
2851}
2852
2853static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2854			       loff_t pos, unsigned len,
2855			       struct page **pagep, void **fsdata)
2856{
2857	int ret, retries = 0;
2858	struct folio *folio;
2859	pgoff_t index;
2860	struct inode *inode = mapping->host;
 
2861
2862	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2863		return -EIO;
2864
2865	index = pos >> PAGE_SHIFT;
2866
2867	if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2868		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2869		return ext4_write_begin(file, mapping, pos,
2870					len, pagep, fsdata);
2871	}
2872	*fsdata = (void *)0;
2873	trace_ext4_da_write_begin(inode, pos, len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2874
2875	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2876		ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
2877						      pagep, fsdata);
2878		if (ret < 0)
2879			return ret;
2880		if (ret == 1)
2881			return 0;
2882	}
 
2883
2884retry:
2885	folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
2886			mapping_gfp_mask(mapping));
2887	if (IS_ERR(folio))
2888		return PTR_ERR(folio);
2889
2890#ifdef CONFIG_FS_ENCRYPTION
2891	ret = ext4_block_write_begin(folio, pos, len, ext4_da_get_block_prep);
2892#else
2893	ret = __block_write_begin(&folio->page, pos, len, ext4_da_get_block_prep);
2894#endif
2895	if (ret < 0) {
2896		folio_unlock(folio);
2897		folio_put(folio);
 
2898		/*
2899		 * block_write_begin may have instantiated a few blocks
2900		 * outside i_size.  Trim these off again. Don't need
2901		 * i_size_read because we hold inode lock.
2902		 */
2903		if (pos + len > inode->i_size)
2904			ext4_truncate_failed_write(inode);
2905
2906		if (ret == -ENOSPC &&
2907		    ext4_should_retry_alloc(inode->i_sb, &retries))
2908			goto retry;
2909		return ret;
2910	}
2911
2912	*pagep = &folio->page;
 
 
2913	return ret;
2914}
2915
2916/*
2917 * Check if we should update i_disksize
2918 * when write to the end of file but not require block allocation
2919 */
2920static int ext4_da_should_update_i_disksize(struct folio *folio,
2921					    unsigned long offset)
2922{
2923	struct buffer_head *bh;
2924	struct inode *inode = folio->mapping->host;
2925	unsigned int idx;
2926	int i;
2927
2928	bh = folio_buffers(folio);
2929	idx = offset >> inode->i_blkbits;
2930
2931	for (i = 0; i < idx; i++)
2932		bh = bh->b_this_page;
2933
2934	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2935		return 0;
2936	return 1;
2937}
2938
2939static int ext4_da_do_write_end(struct address_space *mapping,
2940			loff_t pos, unsigned len, unsigned copied,
2941			struct folio *folio)
 
2942{
2943	struct inode *inode = mapping->host;
2944	loff_t old_size = inode->i_size;
2945	bool disksize_changed = false;
2946	loff_t new_i_size;
 
 
2947
2948	/*
2949	 * block_write_end() will mark the inode as dirty with I_DIRTY_PAGES
2950	 * flag, which all that's needed to trigger page writeback.
2951	 */
2952	copied = block_write_end(NULL, mapping, pos, len, copied,
2953			&folio->page, NULL);
2954	new_i_size = pos + copied;
2955
2956	/*
2957	 * It's important to update i_size while still holding folio lock,
2958	 * because folio writeout could otherwise come in and zero beyond
2959	 * i_size.
2960	 *
2961	 * Since we are holding inode lock, we are sure i_disksize <=
2962	 * i_size. We also know that if i_disksize < i_size, there are
2963	 * delalloc writes pending in the range up to i_size. If the end of
2964	 * the current write is <= i_size, there's no need to touch
2965	 * i_disksize since writeback will push i_disksize up to i_size
2966	 * eventually. If the end of the current write is > i_size and
2967	 * inside an allocated block which ext4_da_should_update_i_disksize()
2968	 * checked, we need to update i_disksize here as certain
2969	 * ext4_writepages() paths not allocating blocks and update i_disksize.
2970	 */
2971	if (new_i_size > inode->i_size) {
2972		unsigned long end;
2973
2974		i_size_write(inode, new_i_size);
2975		end = (new_i_size - 1) & (PAGE_SIZE - 1);
2976		if (copied && ext4_da_should_update_i_disksize(folio, end)) {
2977			ext4_update_i_disksize(inode, new_i_size);
2978			disksize_changed = true;
2979		}
2980	}
2981
2982	folio_unlock(folio);
2983	folio_put(folio);
 
2984
2985	if (old_size < pos)
2986		pagecache_isize_extended(inode, old_size, pos);
 
 
 
2987
2988	if (disksize_changed) {
2989		handle_t *handle;
 
 
 
 
 
 
 
 
 
 
2990
2991		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
2992		if (IS_ERR(handle))
2993			return PTR_ERR(handle);
2994		ext4_mark_inode_dirty(handle, inode);
2995		ext4_journal_stop(handle);
 
 
 
 
2996	}
 
 
 
 
 
 
 
 
2997
2998	return copied;
2999}
3000
3001static int ext4_da_write_end(struct file *file,
3002			     struct address_space *mapping,
3003			     loff_t pos, unsigned len, unsigned copied,
3004			     struct page *page, void *fsdata)
3005{
3006	struct inode *inode = mapping->host;
3007	int write_mode = (int)(unsigned long)fsdata;
3008	struct folio *folio = page_folio(page);
 
 
 
3009
3010	if (write_mode == FALL_BACK_TO_NONDELALLOC)
3011		return ext4_write_end(file, mapping, pos,
3012				      len, copied, &folio->page, fsdata);
3013
3014	trace_ext4_da_write_end(inode, pos, len, copied);
 
3015
3016	if (write_mode != CONVERT_INLINE_DATA &&
3017	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3018	    ext4_has_inline_data(inode))
3019		return ext4_write_inline_data_end(inode, pos, len, copied,
3020						  folio);
3021
3022	if (unlikely(copied < len) && !folio_test_uptodate(folio))
3023		copied = 0;
3024
3025	return ext4_da_do_write_end(mapping, pos, len, copied, folio);
3026}
3027
3028/*
3029 * Force all delayed allocation blocks to be allocated for a given inode.
3030 */
3031int ext4_alloc_da_blocks(struct inode *inode)
3032{
3033	trace_ext4_alloc_da_blocks(inode);
3034
3035	if (!EXT4_I(inode)->i_reserved_data_blocks)
 
3036		return 0;
3037
3038	/*
3039	 * We do something simple for now.  The filemap_flush() will
3040	 * also start triggering a write of the data blocks, which is
3041	 * not strictly speaking necessary (and for users of
3042	 * laptop_mode, not even desirable).  However, to do otherwise
3043	 * would require replicating code paths in:
3044	 *
3045	 * ext4_writepages() ->
3046	 *    write_cache_pages() ---> (via passed in callback function)
3047	 *        __mpage_da_writepage() -->
3048	 *           mpage_add_bh_to_extent()
3049	 *           mpage_da_map_blocks()
3050	 *
3051	 * The problem is that write_cache_pages(), located in
3052	 * mm/page-writeback.c, marks pages clean in preparation for
3053	 * doing I/O, which is not desirable if we're not planning on
3054	 * doing I/O at all.
3055	 *
3056	 * We could call write_cache_pages(), and then redirty all of
3057	 * the pages by calling redirty_page_for_writepage() but that
3058	 * would be ugly in the extreme.  So instead we would need to
3059	 * replicate parts of the code in the above functions,
3060	 * simplifying them because we wouldn't actually intend to
3061	 * write out the pages, but rather only collect contiguous
3062	 * logical block extents, call the multi-block allocator, and
3063	 * then update the buffer heads with the block allocations.
3064	 *
3065	 * For now, though, we'll cheat by calling filemap_flush(),
3066	 * which will map the blocks, and start the I/O, but not
3067	 * actually wait for the I/O to complete.
3068	 */
3069	return filemap_flush(inode->i_mapping);
3070}
3071
3072/*
3073 * bmap() is special.  It gets used by applications such as lilo and by
3074 * the swapper to find the on-disk block of a specific piece of data.
3075 *
3076 * Naturally, this is dangerous if the block concerned is still in the
3077 * journal.  If somebody makes a swapfile on an ext4 data-journaling
3078 * filesystem and enables swap, then they may get a nasty shock when the
3079 * data getting swapped to that swapfile suddenly gets overwritten by
3080 * the original zero's written out previously to the journal and
3081 * awaiting writeback in the kernel's buffer cache.
3082 *
3083 * So, if we see any bmap calls here on a modified, data-journaled file,
3084 * take extra steps to flush any blocks which might be in the cache.
3085 */
3086static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3087{
3088	struct inode *inode = mapping->host;
3089	sector_t ret = 0;
3090
3091	inode_lock_shared(inode);
3092	/*
3093	 * We can get here for an inline file via the FIBMAP ioctl
3094	 */
3095	if (ext4_has_inline_data(inode))
3096		goto out;
3097
3098	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3099	    (test_opt(inode->i_sb, DELALLOC) ||
3100	     ext4_should_journal_data(inode))) {
3101		/*
3102		 * With delalloc or journalled data we want to sync the file so
3103		 * that we can make sure we allocate blocks for file and data
3104		 * is in place for the user to see it
3105		 */
3106		filemap_write_and_wait(mapping);
3107	}
3108
3109	ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3110
3111out:
3112	inode_unlock_shared(inode);
3113	return ret;
 
 
3114}
3115
3116static int ext4_read_folio(struct file *file, struct folio *folio)
3117{
3118	int ret = -EAGAIN;
3119	struct inode *inode = folio->mapping->host;
3120
3121	trace_ext4_read_folio(inode, folio);
3122
3123	if (ext4_has_inline_data(inode))
3124		ret = ext4_readpage_inline(inode, folio);
3125
3126	if (ret == -EAGAIN)
3127		return ext4_mpage_readpages(inode, NULL, folio);
3128
3129	return ret;
3130}
3131
3132static void ext4_readahead(struct readahead_control *rac)
 
 
3133{
3134	struct inode *inode = rac->mapping->host;
3135
3136	/* If the file has inline data, no need to do readahead. */
3137	if (ext4_has_inline_data(inode))
3138		return;
3139
3140	ext4_mpage_readpages(inode, rac, NULL);
3141}
3142
3143static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3144				size_t length)
3145{
3146	trace_ext4_invalidate_folio(folio, offset, length);
 
3147
3148	/* No journalling happens on data buffers when this function is used */
3149	WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3150
3151	block_invalidate_folio(folio, offset, length);
 
 
 
 
 
 
 
 
 
3152}
3153
3154static int __ext4_journalled_invalidate_folio(struct folio *folio,
3155					    size_t offset, size_t length)
3156{
3157	journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3158
3159	trace_ext4_journalled_invalidate_folio(folio, offset, length);
3160
3161	/*
 
 
 
 
 
3162	 * If it's a full truncate we just forget about the pending dirtying
3163	 */
3164	if (offset == 0 && length == folio_size(folio))
3165		folio_clear_checked(folio);
3166
3167	return jbd2_journal_invalidate_folio(journal, folio, offset, length);
 
 
 
3168}
3169
3170/* Wrapper for aops... */
3171static void ext4_journalled_invalidate_folio(struct folio *folio,
3172					   size_t offset,
3173					   size_t length)
3174{
3175	WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3176}
3177
3178static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3179{
3180	struct inode *inode = folio->mapping->host;
3181	journal_t *journal = EXT4_JOURNAL(inode);
3182
3183	trace_ext4_release_folio(inode, folio);
3184
3185	/* Page has dirty journalled data -> cannot release */
3186	if (folio_test_checked(folio))
3187		return false;
3188	if (journal)
3189		return jbd2_journal_try_to_free_buffers(journal, folio);
3190	else
3191		return try_to_free_buffers(folio);
3192}
3193
3194static bool ext4_inode_datasync_dirty(struct inode *inode)
 
 
 
 
 
 
3195{
3196	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 
 
 
 
3197
3198	if (journal) {
3199		if (jbd2_transaction_committed(journal,
3200			EXT4_I(inode)->i_datasync_tid))
3201			return false;
3202		if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3203			return !list_empty(&EXT4_I(inode)->i_fc_list);
3204		return true;
3205	}
 
3206
3207	/* Any metadata buffers to write? */
3208	if (!list_empty(&inode->i_mapping->i_private_list))
3209		return true;
3210	return inode->i_state & I_DIRTY_DATASYNC;
3211}
3212
3213static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3214			   struct ext4_map_blocks *map, loff_t offset,
3215			   loff_t length, unsigned int flags)
3216{
3217	u8 blkbits = inode->i_blkbits;
 
 
 
 
 
 
 
 
 
 
 
3218
3219	/*
3220	 * Writes that span EOF might trigger an I/O size update on completion,
3221	 * so consider them to be dirty for the purpose of O_DSYNC, even if
3222	 * there is no other metadata changes being made or are pending.
3223	 */
3224	iomap->flags = 0;
3225	if (ext4_inode_datasync_dirty(inode) ||
3226	    offset + length > i_size_read(inode))
3227		iomap->flags |= IOMAP_F_DIRTY;
3228
3229	if (map->m_flags & EXT4_MAP_NEW)
3230		iomap->flags |= IOMAP_F_NEW;
 
 
 
3231
3232	if (flags & IOMAP_DAX)
3233		iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3234	else
3235		iomap->bdev = inode->i_sb->s_bdev;
3236	iomap->offset = (u64) map->m_lblk << blkbits;
3237	iomap->length = (u64) map->m_len << blkbits;
3238
3239	if ((map->m_flags & EXT4_MAP_MAPPED) &&
3240	    !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3241		iomap->flags |= IOMAP_F_MERGED;
3242
3243	/*
3244	 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3245	 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3246	 * set. In order for any allocated unwritten extents to be converted
3247	 * into written extents correctly within the ->end_io() handler, we
3248	 * need to ensure that the iomap->type is set appropriately. Hence, the
3249	 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3250	 * been set first.
3251	 */
3252	if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3253		iomap->type = IOMAP_UNWRITTEN;
3254		iomap->addr = (u64) map->m_pblk << blkbits;
3255		if (flags & IOMAP_DAX)
3256			iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3257	} else if (map->m_flags & EXT4_MAP_MAPPED) {
3258		iomap->type = IOMAP_MAPPED;
3259		iomap->addr = (u64) map->m_pblk << blkbits;
3260		if (flags & IOMAP_DAX)
3261			iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3262	} else if (map->m_flags & EXT4_MAP_DELAYED) {
3263		iomap->type = IOMAP_DELALLOC;
3264		iomap->addr = IOMAP_NULL_ADDR;
3265	} else {
3266		iomap->type = IOMAP_HOLE;
3267		iomap->addr = IOMAP_NULL_ADDR;
3268	}
3269}
3270
3271static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3272			    unsigned int flags)
3273{
3274	handle_t *handle;
3275	u8 blkbits = inode->i_blkbits;
3276	int ret, dio_credits, m_flags = 0, retries = 0;
 
 
 
 
3277
3278	/*
3279	 * Trim the mapping request to the maximum value that we can map at
3280	 * once for direct I/O.
3281	 */
3282	if (map->m_len > DIO_MAX_BLOCKS)
3283		map->m_len = DIO_MAX_BLOCKS;
3284	dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3285
3286retry:
3287	/*
3288	 * Either we allocate blocks and then don't get an unwritten extent, so
3289	 * in that case we have reserved enough credits. Or, the blocks are
3290	 * already allocated and unwritten. In that case, the extent conversion
3291	 * fits into the credits as well.
3292	 */
3293	handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3294	if (IS_ERR(handle))
3295		return PTR_ERR(handle);
3296
3297	/*
3298	 * DAX and direct I/O are the only two operations that are currently
3299	 * supported with IOMAP_WRITE.
3300	 */
3301	WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3302	if (flags & IOMAP_DAX)
3303		m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3304	/*
3305	 * We use i_size instead of i_disksize here because delalloc writeback
3306	 * can complete at any point during the I/O and subsequently push the
3307	 * i_disksize out to i_size. This could be beyond where direct I/O is
3308	 * happening and thus expose allocated blocks to direct I/O reads.
3309	 */
3310	else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3311		m_flags = EXT4_GET_BLOCKS_CREATE;
3312	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3313		m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3314
3315	ret = ext4_map_blocks(handle, inode, map, m_flags);
 
 
 
 
 
3316
 
 
 
 
 
 
 
 
 
3317	/*
3318	 * We cannot fill holes in indirect tree based inodes as that could
3319	 * expose stale data in the case of a crash. Use the magic error code
3320	 * to fallback to buffered I/O.
3321	 */
3322	if (!m_flags && !ret)
3323		ret = -ENOTBLK;
3324
3325	ext4_journal_stop(handle);
3326	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3327		goto retry;
3328
3329	return ret;
3330}
3331
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3332
3333static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3334		unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3335{
3336	int ret;
3337	struct ext4_map_blocks map;
3338	u8 blkbits = inode->i_blkbits;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3339
3340	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3341		return -EINVAL;
3342
3343	if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3344		return -ERANGE;
3345
3346	/*
3347	 * Calculate the first and last logical blocks respectively.
3348	 */
3349	map.m_lblk = offset >> blkbits;
3350	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3351			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3352
3353	if (flags & IOMAP_WRITE) {
3354		/*
3355		 * We check here if the blocks are already allocated, then we
3356		 * don't need to start a journal txn and we can directly return
3357		 * the mapping information. This could boost performance
3358		 * especially in multi-threaded overwrite requests.
3359		 */
3360		if (offset + length <= i_size_read(inode)) {
3361			ret = ext4_map_blocks(NULL, inode, &map, 0);
3362			if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3363				goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3364		}
3365		ret = ext4_iomap_alloc(inode, &map, flags);
3366	} else {
3367		ret = ext4_map_blocks(NULL, inode, &map, 0);
3368	}
3369
3370	if (ret < 0)
3371		return ret;
3372out:
3373	/*
3374	 * When inline encryption is enabled, sometimes I/O to an encrypted file
3375	 * has to be broken up to guarantee DUN contiguity.  Handle this by
3376	 * limiting the length of the mapping returned.
3377	 */
3378	map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3379
3380	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3381
3382	return 0;
3383}
3384
3385static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3386		loff_t length, unsigned flags, struct iomap *iomap,
3387		struct iomap *srcmap)
3388{
3389	int ret;
 
 
3390
3391	/*
3392	 * Even for writes we don't need to allocate blocks, so just pretend
3393	 * we are reading to save overhead of starting a transaction.
3394	 */
3395	flags &= ~IOMAP_WRITE;
3396	ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3397	WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
 
 
 
 
 
 
 
3398	return ret;
3399}
3400
3401static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3402			  ssize_t written, unsigned flags, struct iomap *iomap)
 
 
 
 
 
 
 
 
 
 
 
 
3403{
3404	/*
3405	 * Check to see whether an error occurred while writing out the data to
3406	 * the allocated blocks. If so, return the magic error code so that we
3407	 * fallback to buffered I/O and attempt to complete the remainder of
3408	 * the I/O. Any blocks that may have been allocated in preparation for
3409	 * the direct I/O will be reused during buffered I/O.
3410	 */
3411	if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3412		return -ENOTBLK;
3413
3414	return 0;
3415}
3416
3417const struct iomap_ops ext4_iomap_ops = {
3418	.iomap_begin		= ext4_iomap_begin,
3419	.iomap_end		= ext4_iomap_end,
3420};
3421
3422const struct iomap_ops ext4_iomap_overwrite_ops = {
3423	.iomap_begin		= ext4_iomap_overwrite_begin,
3424	.iomap_end		= ext4_iomap_end,
 
 
 
 
 
3425};
3426
3427static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3428				   loff_t length, unsigned int flags,
3429				   struct iomap *iomap, struct iomap *srcmap)
3430{
3431	int ret;
3432	struct ext4_map_blocks map;
3433	u8 blkbits = inode->i_blkbits;
3434
3435	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3436		return -EINVAL;
3437
3438	if (ext4_has_inline_data(inode)) {
3439		ret = ext4_inline_data_iomap(inode, iomap);
3440		if (ret != -EAGAIN) {
3441			if (ret == 0 && offset >= iomap->length)
3442				ret = -ENOENT;
3443			return ret;
3444		}
3445	}
3446
3447	/*
3448	 * Calculate the first and last logical block respectively.
3449	 */
3450	map.m_lblk = offset >> blkbits;
3451	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3452			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3453
3454	/*
3455	 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3456	 * So handle it here itself instead of querying ext4_map_blocks().
3457	 * Since ext4_map_blocks() will warn about it and will return
3458	 * -EIO error.
3459	 */
3460	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3461		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3462
3463		if (offset >= sbi->s_bitmap_maxbytes) {
3464			map.m_flags = 0;
3465			goto set_iomap;
3466		}
3467	}
3468
3469	ret = ext4_map_blocks(NULL, inode, &map, 0);
3470	if (ret < 0)
3471		return ret;
3472set_iomap:
3473	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3474
3475	return 0;
3476}
3477
3478const struct iomap_ops ext4_iomap_report_ops = {
3479	.iomap_begin = ext4_iomap_begin_report,
3480};
3481
3482/*
3483 * For data=journal mode, folio should be marked dirty only when it was
3484 * writeably mapped. When that happens, it was already attached to the
3485 * transaction and marked as jbddirty (we take care of this in
3486 * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3487 * so we should have nothing to do here, except for the case when someone
3488 * had the page pinned and dirtied the page through this pin (e.g. by doing
3489 * direct IO to it). In that case we'd need to attach buffers here to the
3490 * transaction but we cannot due to lock ordering.  We cannot just dirty the
3491 * folio and leave attached buffers clean, because the buffers' dirty state is
3492 * "definitive".  We cannot just set the buffers dirty or jbddirty because all
3493 * the journalling code will explode.  So what we do is to mark the folio
3494 * "pending dirty" and next time ext4_writepages() is called, attach buffers
3495 * to the transaction appropriately.
3496 */
3497static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3498		struct folio *folio)
3499{
3500	WARN_ON_ONCE(!folio_buffers(folio));
3501	if (folio_maybe_dma_pinned(folio))
3502		folio_set_checked(folio);
3503	return filemap_dirty_folio(mapping, folio);
3504}
3505
3506static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3507{
3508	WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3509	WARN_ON_ONCE(!folio_buffers(folio));
3510	return block_dirty_folio(mapping, folio);
3511}
3512
3513static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3514				    struct file *file, sector_t *span)
3515{
3516	return iomap_swapfile_activate(sis, file, span,
3517				       &ext4_iomap_report_ops);
3518}
3519
3520static const struct address_space_operations ext4_aops = {
3521	.read_folio		= ext4_read_folio,
3522	.readahead		= ext4_readahead,
3523	.writepages		= ext4_writepages,
3524	.write_begin		= ext4_write_begin,
3525	.write_end		= ext4_write_end,
3526	.dirty_folio		= ext4_dirty_folio,
3527	.bmap			= ext4_bmap,
3528	.invalidate_folio	= ext4_invalidate_folio,
3529	.release_folio		= ext4_release_folio,
3530	.direct_IO		= noop_direct_IO,
3531	.migrate_folio		= buffer_migrate_folio,
3532	.is_partially_uptodate  = block_is_partially_uptodate,
3533	.error_remove_folio	= generic_error_remove_folio,
3534	.swap_activate		= ext4_iomap_swap_activate,
3535};
3536
3537static const struct address_space_operations ext4_journalled_aops = {
3538	.read_folio		= ext4_read_folio,
3539	.readahead		= ext4_readahead,
3540	.writepages		= ext4_writepages,
3541	.write_begin		= ext4_write_begin,
3542	.write_end		= ext4_journalled_write_end,
3543	.dirty_folio		= ext4_journalled_dirty_folio,
3544	.bmap			= ext4_bmap,
3545	.invalidate_folio	= ext4_journalled_invalidate_folio,
3546	.release_folio		= ext4_release_folio,
3547	.direct_IO		= noop_direct_IO,
3548	.migrate_folio		= buffer_migrate_folio_norefs,
3549	.is_partially_uptodate  = block_is_partially_uptodate,
3550	.error_remove_folio	= generic_error_remove_folio,
3551	.swap_activate		= ext4_iomap_swap_activate,
3552};
3553
3554static const struct address_space_operations ext4_da_aops = {
3555	.read_folio		= ext4_read_folio,
3556	.readahead		= ext4_readahead,
3557	.writepages		= ext4_writepages,
 
3558	.write_begin		= ext4_da_write_begin,
3559	.write_end		= ext4_da_write_end,
3560	.dirty_folio		= ext4_dirty_folio,
3561	.bmap			= ext4_bmap,
3562	.invalidate_folio	= ext4_invalidate_folio,
3563	.release_folio		= ext4_release_folio,
3564	.direct_IO		= noop_direct_IO,
3565	.migrate_folio		= buffer_migrate_folio,
3566	.is_partially_uptodate  = block_is_partially_uptodate,
3567	.error_remove_folio	= generic_error_remove_folio,
3568	.swap_activate		= ext4_iomap_swap_activate,
3569};
3570
3571static const struct address_space_operations ext4_dax_aops = {
3572	.writepages		= ext4_dax_writepages,
3573	.direct_IO		= noop_direct_IO,
3574	.dirty_folio		= noop_dirty_folio,
3575	.bmap			= ext4_bmap,
3576	.swap_activate		= ext4_iomap_swap_activate,
3577};
3578
3579void ext4_set_aops(struct inode *inode)
3580{
3581	switch (ext4_inode_journal_mode(inode)) {
3582	case EXT4_INODE_ORDERED_DATA_MODE:
 
 
 
 
 
3583	case EXT4_INODE_WRITEBACK_DATA_MODE:
 
 
 
 
3584		break;
3585	case EXT4_INODE_JOURNAL_DATA_MODE:
3586		inode->i_mapping->a_ops = &ext4_journalled_aops;
3587		return;
3588	default:
3589		BUG();
3590	}
3591	if (IS_DAX(inode))
3592		inode->i_mapping->a_ops = &ext4_dax_aops;
3593	else if (test_opt(inode->i_sb, DELALLOC))
3594		inode->i_mapping->a_ops = &ext4_da_aops;
3595	else
3596		inode->i_mapping->a_ops = &ext4_aops;
3597}
3598
 
3599/*
3600 * Here we can't skip an unwritten buffer even though it usually reads zero
3601 * because it might have data in pagecache (eg, if called from ext4_zero_range,
3602 * ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a
3603 * racing writeback can come later and flush the stale pagecache to disk.
3604 */
3605static int __ext4_block_zero_page_range(handle_t *handle,
3606		struct address_space *mapping, loff_t from, loff_t length)
 
 
 
3607{
3608	ext4_fsblk_t index = from >> PAGE_SHIFT;
3609	unsigned offset = from & (PAGE_SIZE-1);
3610	unsigned blocksize, pos;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3611	ext4_lblk_t iblock;
3612	struct inode *inode = mapping->host;
3613	struct buffer_head *bh;
3614	struct folio *folio;
3615	int err = 0;
3616
3617	folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT,
3618				    FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3619				    mapping_gfp_constraint(mapping, ~__GFP_FS));
3620	if (IS_ERR(folio))
3621		return PTR_ERR(folio);
3622
3623	blocksize = inode->i_sb->s_blocksize;
 
 
 
 
 
3624
3625	iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3626
3627	bh = folio_buffers(folio);
3628	if (!bh)
3629		bh = create_empty_buffers(folio, blocksize, 0);
3630
3631	/* Find the buffer that contains "offset" */
 
3632	pos = blocksize;
3633	while (offset >= pos) {
3634		bh = bh->b_this_page;
3635		iblock++;
3636		pos += blocksize;
3637	}
3638	if (buffer_freed(bh)) {
3639		BUFFER_TRACE(bh, "freed: skip");
3640		goto unlock;
3641	}
3642	if (!buffer_mapped(bh)) {
3643		BUFFER_TRACE(bh, "unmapped");
3644		ext4_get_block(inode, iblock, bh, 0);
3645		/* unmapped? It's a hole - nothing to do */
3646		if (!buffer_mapped(bh)) {
3647			BUFFER_TRACE(bh, "still unmapped");
3648			goto unlock;
3649		}
3650	}
3651
3652	/* Ok, it's mapped. Make sure it's up-to-date */
3653	if (folio_test_uptodate(folio))
3654		set_buffer_uptodate(bh);
3655
3656	if (!buffer_uptodate(bh)) {
3657		err = ext4_read_bh_lock(bh, 0, true);
3658		if (err)
3659			goto unlock;
3660		if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3661			/* We expect the key to be set. */
3662			BUG_ON(!fscrypt_has_encryption_key(inode));
3663			err = fscrypt_decrypt_pagecache_blocks(folio,
3664							       blocksize,
3665							       bh_offset(bh));
3666			if (err) {
3667				clear_buffer_uptodate(bh);
3668				goto unlock;
3669			}
3670		}
3671	}
3672	if (ext4_should_journal_data(inode)) {
3673		BUFFER_TRACE(bh, "get write access");
3674		err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3675						    EXT4_JTR_NONE);
3676		if (err)
3677			goto unlock;
3678	}
3679	folio_zero_range(folio, offset, length);
3680	BUFFER_TRACE(bh, "zeroed end of block");
3681
3682	if (ext4_should_journal_data(inode)) {
3683		err = ext4_dirty_journalled_data(handle, bh);
3684	} else {
3685		err = 0;
3686		mark_buffer_dirty(bh);
3687		if (ext4_should_order_data(inode))
3688			err = ext4_jbd2_inode_add_write(handle, inode, from,
3689					length);
3690	}
3691
3692unlock:
3693	folio_unlock(folio);
3694	folio_put(folio);
3695	return err;
3696}
3697
3698/*
3699 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3700 * starting from file offset 'from'.  The range to be zero'd must
3701 * be contained with in one block.  If the specified range exceeds
3702 * the end of the block it will be shortened to end of the block
3703 * that corresponds to 'from'
3704 */
3705static int ext4_block_zero_page_range(handle_t *handle,
3706		struct address_space *mapping, loff_t from, loff_t length)
3707{
3708	struct inode *inode = mapping->host;
3709	unsigned offset = from & (PAGE_SIZE-1);
3710	unsigned blocksize = inode->i_sb->s_blocksize;
3711	unsigned max = blocksize - (offset & (blocksize - 1));
3712
3713	/*
3714	 * correct length if it does not fall between
3715	 * 'from' and the end of the block
3716	 */
3717	if (length > max || length < 0)
3718		length = max;
3719
3720	if (IS_DAX(inode)) {
3721		return dax_zero_range(inode, from, length, NULL,
3722				      &ext4_iomap_ops);
3723	}
3724	return __ext4_block_zero_page_range(handle, mapping, from, length);
3725}
3726
3727/*
3728 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3729 * up to the end of the block which corresponds to `from'.
3730 * This required during truncate. We need to physically zero the tail end
3731 * of that block so it doesn't yield old data if the file is later grown.
3732 */
3733static int ext4_block_truncate_page(handle_t *handle,
3734		struct address_space *mapping, loff_t from)
3735{
3736	unsigned offset = from & (PAGE_SIZE-1);
3737	unsigned length;
3738	unsigned blocksize;
3739	struct inode *inode = mapping->host;
 
 
 
 
 
 
 
 
 
3740
3741	/* If we are processing an encrypted inode during orphan list handling */
3742	if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3743		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3744
3745	blocksize = inode->i_sb->s_blocksize;
3746	length = blocksize - (offset & (blocksize - 1));
 
3747
3748	return ext4_block_zero_page_range(handle, mapping, from, length);
3749}
 
 
 
 
 
 
3750
3751int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3752			     loff_t lstart, loff_t length)
3753{
3754	struct super_block *sb = inode->i_sb;
3755	struct address_space *mapping = inode->i_mapping;
3756	unsigned partial_start, partial_end;
3757	ext4_fsblk_t start, end;
3758	loff_t byte_end = (lstart + length - 1);
3759	int err = 0;
3760
3761	partial_start = lstart & (sb->s_blocksize - 1);
3762	partial_end = byte_end & (sb->s_blocksize - 1);
3763
3764	start = lstart >> sb->s_blocksize_bits;
3765	end = byte_end >> sb->s_blocksize_bits;
 
 
 
3766
3767	/* Handle partial zero within the single block */
3768	if (start == end &&
3769	    (partial_start || (partial_end != sb->s_blocksize - 1))) {
3770		err = ext4_block_zero_page_range(handle, mapping,
3771						 lstart, length);
3772		return err;
3773	}
3774	/* Handle partial zero out on the start of the range */
3775	if (partial_start) {
3776		err = ext4_block_zero_page_range(handle, mapping,
3777						 lstart, sb->s_blocksize);
3778		if (err)
3779			return err;
3780	}
3781	/* Handle partial zero out on the end of the range */
3782	if (partial_end != sb->s_blocksize - 1)
3783		err = ext4_block_zero_page_range(handle, mapping,
3784						 byte_end - partial_end,
3785						 partial_end + 1);
3786	return err;
3787}
3788
3789int ext4_can_truncate(struct inode *inode)
3790{
3791	if (S_ISREG(inode->i_mode))
3792		return 1;
3793	if (S_ISDIR(inode->i_mode))
3794		return 1;
3795	if (S_ISLNK(inode->i_mode))
3796		return !ext4_inode_is_fast_symlink(inode);
3797	return 0;
3798}
3799
3800/*
3801 * We have to make sure i_disksize gets properly updated before we truncate
3802 * page cache due to hole punching or zero range. Otherwise i_disksize update
3803 * can get lost as it may have been postponed to submission of writeback but
3804 * that will never happen after we truncate page cache.
3805 */
3806int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3807				      loff_t len)
3808{
3809	handle_t *handle;
3810	int ret;
3811
3812	loff_t size = i_size_read(inode);
3813
3814	WARN_ON(!inode_is_locked(inode));
3815	if (offset > size || offset + len < size)
3816		return 0;
3817
3818	if (EXT4_I(inode)->i_disksize >= size)
3819		return 0;
3820
3821	handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3822	if (IS_ERR(handle))
3823		return PTR_ERR(handle);
3824	ext4_update_i_disksize(inode, size);
3825	ret = ext4_mark_inode_dirty(handle, inode);
3826	ext4_journal_stop(handle);
3827
3828	return ret;
3829}
3830
3831static void ext4_wait_dax_page(struct inode *inode)
3832{
3833	filemap_invalidate_unlock(inode->i_mapping);
3834	schedule();
3835	filemap_invalidate_lock(inode->i_mapping);
3836}
3837
3838int ext4_break_layouts(struct inode *inode)
3839{
3840	struct page *page;
3841	int error;
3842
3843	if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3844		return -EINVAL;
3845
3846	do {
3847		page = dax_layout_busy_page(inode->i_mapping);
3848		if (!page)
3849			return 0;
3850
3851		error = ___wait_var_event(&page->_refcount,
3852				atomic_read(&page->_refcount) == 1,
3853				TASK_INTERRUPTIBLE, 0, 0,
3854				ext4_wait_dax_page(inode));
3855	} while (error == 0);
3856
3857	return error;
3858}
3859
3860/*
3861 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3862 * associated with the given offset and length
3863 *
3864 * @inode:  File inode
3865 * @offset: The offset where the hole will begin
3866 * @len:    The length of the hole
3867 *
3868 * Returns: 0 on success or negative on failure
3869 */
3870
3871int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3872{
3873	struct inode *inode = file_inode(file);
3874	struct super_block *sb = inode->i_sb;
3875	ext4_lblk_t first_block, stop_block;
3876	struct address_space *mapping = inode->i_mapping;
3877	loff_t first_block_offset, last_block_offset, max_length;
3878	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3879	handle_t *handle;
3880	unsigned int credits;
3881	int ret = 0, ret2 = 0;
3882
3883	trace_ext4_punch_hole(inode, offset, length, 0);
3884
3885	/*
3886	 * Write out all dirty pages to avoid race conditions
3887	 * Then release them.
3888	 */
3889	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3890		ret = filemap_write_and_wait_range(mapping, offset,
3891						   offset + length - 1);
3892		if (ret)
3893			return ret;
3894	}
3895
3896	inode_lock(inode);
3897
3898	/* No need to punch hole beyond i_size */
3899	if (offset >= inode->i_size)
3900		goto out_mutex;
3901
3902	/*
3903	 * If the hole extends beyond i_size, set the hole
3904	 * to end after the page that contains i_size
3905	 */
3906	if (offset + length > inode->i_size) {
3907		length = inode->i_size +
3908		   PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3909		   offset;
3910	}
3911
3912	/*
3913	 * For punch hole the length + offset needs to be within one block
3914	 * before last range. Adjust the length if it goes beyond that limit.
3915	 */
3916	max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
3917	if (offset + length > max_length)
3918		length = max_length - offset;
3919
3920	if (offset & (sb->s_blocksize - 1) ||
3921	    (offset + length) & (sb->s_blocksize - 1)) {
3922		/*
3923		 * Attach jinode to inode for jbd2 if we do any zeroing of
3924		 * partial block
3925		 */
3926		ret = ext4_inode_attach_jinode(inode);
3927		if (ret < 0)
3928			goto out_mutex;
3929
3930	}
3931
3932	/* Wait all existing dio workers, newcomers will block on i_rwsem */
3933	inode_dio_wait(inode);
3934
3935	ret = file_modified(file);
3936	if (ret)
3937		goto out_mutex;
3938
3939	/*
3940	 * Prevent page faults from reinstantiating pages we have released from
3941	 * page cache.
3942	 */
3943	filemap_invalidate_lock(mapping);
3944
3945	ret = ext4_break_layouts(inode);
3946	if (ret)
3947		goto out_dio;
3948
3949	first_block_offset = round_up(offset, sb->s_blocksize);
3950	last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3951
3952	/* Now release the pages and zero block aligned part of pages*/
3953	if (last_block_offset > first_block_offset) {
3954		ret = ext4_update_disksize_before_punch(inode, offset, length);
3955		if (ret)
3956			goto out_dio;
3957		truncate_pagecache_range(inode, first_block_offset,
3958					 last_block_offset);
3959	}
3960
3961	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3962		credits = ext4_writepage_trans_blocks(inode);
3963	else
3964		credits = ext4_blocks_for_truncate(inode);
3965	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3966	if (IS_ERR(handle)) {
3967		ret = PTR_ERR(handle);
3968		ext4_std_error(sb, ret);
3969		goto out_dio;
3970	}
3971
3972	ret = ext4_zero_partial_blocks(handle, inode, offset,
3973				       length);
3974	if (ret)
3975		goto out_stop;
3976
3977	first_block = (offset + sb->s_blocksize - 1) >>
3978		EXT4_BLOCK_SIZE_BITS(sb);
3979	stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3980
3981	/* If there are blocks to remove, do it */
3982	if (stop_block > first_block) {
3983		ext4_lblk_t hole_len = stop_block - first_block;
3984
3985		down_write(&EXT4_I(inode)->i_data_sem);
3986		ext4_discard_preallocations(inode);
3987
3988		ext4_es_remove_extent(inode, first_block, hole_len);
3989
3990		if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3991			ret = ext4_ext_remove_space(inode, first_block,
3992						    stop_block - 1);
3993		else
3994			ret = ext4_ind_remove_space(handle, inode, first_block,
3995						    stop_block);
3996
3997		ext4_es_insert_extent(inode, first_block, hole_len, ~0,
3998				      EXTENT_STATUS_HOLE);
3999		up_write(&EXT4_I(inode)->i_data_sem);
4000	}
4001	ext4_fc_track_range(handle, inode, first_block, stop_block);
4002	if (IS_SYNC(inode))
4003		ext4_handle_sync(handle);
4004
4005	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4006	ret2 = ext4_mark_inode_dirty(handle, inode);
4007	if (unlikely(ret2))
4008		ret = ret2;
4009	if (ret >= 0)
4010		ext4_update_inode_fsync_trans(handle, inode, 1);
4011out_stop:
4012	ext4_journal_stop(handle);
4013out_dio:
4014	filemap_invalidate_unlock(mapping);
4015out_mutex:
4016	inode_unlock(inode);
4017	return ret;
4018}
4019
4020int ext4_inode_attach_jinode(struct inode *inode)
4021{
4022	struct ext4_inode_info *ei = EXT4_I(inode);
4023	struct jbd2_inode *jinode;
4024
4025	if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4026		return 0;
4027
4028	jinode = jbd2_alloc_inode(GFP_KERNEL);
4029	spin_lock(&inode->i_lock);
4030	if (!ei->jinode) {
4031		if (!jinode) {
4032			spin_unlock(&inode->i_lock);
4033			return -ENOMEM;
4034		}
4035		ei->jinode = jinode;
4036		jbd2_journal_init_jbd_inode(ei->jinode, inode);
4037		jinode = NULL;
4038	}
4039	spin_unlock(&inode->i_lock);
4040	if (unlikely(jinode != NULL))
4041		jbd2_free_inode(jinode);
4042	return 0;
4043}
4044
4045/*
4046 * ext4_truncate()
4047 *
4048 * We block out ext4_get_block() block instantiations across the entire
4049 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4050 * simultaneously on behalf of the same inode.
4051 *
4052 * As we work through the truncate and commit bits of it to the journal there
4053 * is one core, guiding principle: the file's tree must always be consistent on
4054 * disk.  We must be able to restart the truncate after a crash.
4055 *
4056 * The file's tree may be transiently inconsistent in memory (although it
4057 * probably isn't), but whenever we close off and commit a journal transaction,
4058 * the contents of (the filesystem + the journal) must be consistent and
4059 * restartable.  It's pretty simple, really: bottom up, right to left (although
4060 * left-to-right works OK too).
4061 *
4062 * Note that at recovery time, journal replay occurs *before* the restart of
4063 * truncate against the orphan inode list.
4064 *
4065 * The committed inode has the new, desired i_size (which is the same as
4066 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4067 * that this inode's truncate did not complete and it will again call
4068 * ext4_truncate() to have another go.  So there will be instantiated blocks
4069 * to the right of the truncation point in a crashed ext4 filesystem.  But
4070 * that's fine - as long as they are linked from the inode, the post-crash
4071 * ext4_truncate() run will find them and release them.
4072 */
4073int ext4_truncate(struct inode *inode)
4074{
4075	struct ext4_inode_info *ei = EXT4_I(inode);
4076	unsigned int credits;
4077	int err = 0, err2;
4078	handle_t *handle;
4079	struct address_space *mapping = inode->i_mapping;
4080
4081	/*
4082	 * There is a possibility that we're either freeing the inode
4083	 * or it's a completely new inode. In those cases we might not
4084	 * have i_rwsem locked because it's not necessary.
4085	 */
4086	if (!(inode->i_state & (I_NEW|I_FREEING)))
4087		WARN_ON(!inode_is_locked(inode));
4088	trace_ext4_truncate_enter(inode);
4089
4090	if (!ext4_can_truncate(inode))
4091		goto out_trace;
 
 
4092
4093	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4094		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4095
4096	if (ext4_has_inline_data(inode)) {
4097		int has_inline = 1;
4098
4099		err = ext4_inline_data_truncate(inode, &has_inline);
4100		if (err || has_inline)
4101			goto out_trace;
4102	}
4103
4104	/* If we zero-out tail of the page, we have to create jinode for jbd2 */
4105	if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4106		err = ext4_inode_attach_jinode(inode);
4107		if (err)
4108			goto out_trace;
4109	}
4110
4111	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4112		credits = ext4_writepage_trans_blocks(inode);
4113	else
4114		credits = ext4_blocks_for_truncate(inode);
4115
4116	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4117	if (IS_ERR(handle)) {
4118		err = PTR_ERR(handle);
4119		goto out_trace;
4120	}
4121
4122	if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4123		ext4_block_truncate_page(handle, mapping, inode->i_size);
4124
4125	/*
4126	 * We add the inode to the orphan list, so that if this
4127	 * truncate spans multiple transactions, and we crash, we will
4128	 * resume the truncate when the filesystem recovers.  It also
4129	 * marks the inode dirty, to catch the new size.
4130	 *
4131	 * Implication: the file must always be in a sane, consistent
4132	 * truncatable state while each transaction commits.
4133	 */
4134	err = ext4_orphan_add(handle, inode);
4135	if (err)
4136		goto out_stop;
4137
4138	down_write(&EXT4_I(inode)->i_data_sem);
4139
4140	ext4_discard_preallocations(inode);
4141
4142	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4143		err = ext4_ext_truncate(handle, inode);
4144	else
4145		ext4_ind_truncate(handle, inode);
4146
4147	up_write(&ei->i_data_sem);
4148	if (err)
4149		goto out_stop;
4150
4151	if (IS_SYNC(inode))
4152		ext4_handle_sync(handle);
4153
4154out_stop:
4155	/*
4156	 * If this was a simple ftruncate() and the file will remain alive,
4157	 * then we need to clear up the orphan record which we created above.
4158	 * However, if this was a real unlink then we were called by
4159	 * ext4_evict_inode(), and we allow that function to clean up the
4160	 * orphan info for us.
4161	 */
4162	if (inode->i_nlink)
4163		ext4_orphan_del(handle, inode);
4164
4165	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4166	err2 = ext4_mark_inode_dirty(handle, inode);
4167	if (unlikely(err2 && !err))
4168		err = err2;
4169	ext4_journal_stop(handle);
4170
4171out_trace:
4172	trace_ext4_truncate_exit(inode);
4173	return err;
4174}
4175
4176static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4177{
4178	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4179		return inode_peek_iversion_raw(inode);
4180	else
4181		return inode_peek_iversion(inode);
4182}
4183
4184static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4185				 struct ext4_inode_info *ei)
4186{
4187	struct inode *inode = &(ei->vfs_inode);
4188	u64 i_blocks = READ_ONCE(inode->i_blocks);
4189	struct super_block *sb = inode->i_sb;
4190
4191	if (i_blocks <= ~0U) {
4192		/*
4193		 * i_blocks can be represented in a 32 bit variable
4194		 * as multiple of 512 bytes
4195		 */
4196		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4197		raw_inode->i_blocks_high = 0;
4198		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4199		return 0;
4200	}
4201
4202	/*
4203	 * This should never happen since sb->s_maxbytes should not have
4204	 * allowed this, sb->s_maxbytes was set according to the huge_file
4205	 * feature in ext4_fill_super().
4206	 */
4207	if (!ext4_has_feature_huge_file(sb))
4208		return -EFSCORRUPTED;
4209
4210	if (i_blocks <= 0xffffffffffffULL) {
4211		/*
4212		 * i_blocks can be represented in a 48 bit variable
4213		 * as multiple of 512 bytes
4214		 */
4215		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4216		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4217		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4218	} else {
4219		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4220		/* i_block is stored in file system block size */
4221		i_blocks = i_blocks >> (inode->i_blkbits - 9);
4222		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4223		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4224	}
4225	return 0;
4226}
4227
4228static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4229{
4230	struct ext4_inode_info *ei = EXT4_I(inode);
4231	uid_t i_uid;
4232	gid_t i_gid;
4233	projid_t i_projid;
4234	int block;
4235	int err;
4236
4237	err = ext4_inode_blocks_set(raw_inode, ei);
4238
4239	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4240	i_uid = i_uid_read(inode);
4241	i_gid = i_gid_read(inode);
4242	i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4243	if (!(test_opt(inode->i_sb, NO_UID32))) {
4244		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4245		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4246		/*
4247		 * Fix up interoperability with old kernels. Otherwise,
4248		 * old inodes get re-used with the upper 16 bits of the
4249		 * uid/gid intact.
4250		 */
4251		if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4252			raw_inode->i_uid_high = 0;
4253			raw_inode->i_gid_high = 0;
4254		} else {
4255			raw_inode->i_uid_high =
4256				cpu_to_le16(high_16_bits(i_uid));
4257			raw_inode->i_gid_high =
4258				cpu_to_le16(high_16_bits(i_gid));
4259		}
4260	} else {
4261		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4262		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4263		raw_inode->i_uid_high = 0;
4264		raw_inode->i_gid_high = 0;
4265	}
4266	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4267
4268	EXT4_INODE_SET_CTIME(inode, raw_inode);
4269	EXT4_INODE_SET_MTIME(inode, raw_inode);
4270	EXT4_INODE_SET_ATIME(inode, raw_inode);
4271	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4272
4273	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4274	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4275	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4276		raw_inode->i_file_acl_high =
4277			cpu_to_le16(ei->i_file_acl >> 32);
4278	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4279	ext4_isize_set(raw_inode, ei->i_disksize);
4280
4281	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4282	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4283		if (old_valid_dev(inode->i_rdev)) {
4284			raw_inode->i_block[0] =
4285				cpu_to_le32(old_encode_dev(inode->i_rdev));
4286			raw_inode->i_block[1] = 0;
4287		} else {
4288			raw_inode->i_block[0] = 0;
4289			raw_inode->i_block[1] =
4290				cpu_to_le32(new_encode_dev(inode->i_rdev));
4291			raw_inode->i_block[2] = 0;
4292		}
4293	} else if (!ext4_has_inline_data(inode)) {
4294		for (block = 0; block < EXT4_N_BLOCKS; block++)
4295			raw_inode->i_block[block] = ei->i_data[block];
4296	}
4297
4298	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4299		u64 ivers = ext4_inode_peek_iversion(inode);
4300
4301		raw_inode->i_disk_version = cpu_to_le32(ivers);
4302		if (ei->i_extra_isize) {
4303			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4304				raw_inode->i_version_hi =
4305					cpu_to_le32(ivers >> 32);
4306			raw_inode->i_extra_isize =
4307				cpu_to_le16(ei->i_extra_isize);
4308		}
4309	}
4310
4311	if (i_projid != EXT4_DEF_PROJID &&
4312	    !ext4_has_feature_project(inode->i_sb))
4313		err = err ?: -EFSCORRUPTED;
4314
4315	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4316	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4317		raw_inode->i_projid = cpu_to_le32(i_projid);
4318
4319	ext4_inode_csum_set(inode, raw_inode, ei);
4320	return err;
4321}
4322
4323/*
4324 * ext4_get_inode_loc returns with an extra refcount against the inode's
4325 * underlying buffer_head on success. If we pass 'inode' and it does not
4326 * have in-inode xattr, we have all inode data in memory that is needed
4327 * to recreate the on-disk version of this inode.
4328 */
4329static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4330				struct inode *inode, struct ext4_iloc *iloc,
4331				ext4_fsblk_t *ret_block)
4332{
4333	struct ext4_group_desc	*gdp;
4334	struct buffer_head	*bh;
 
4335	ext4_fsblk_t		block;
4336	struct blk_plug		plug;
4337	int			inodes_per_block, inode_offset;
4338
4339	iloc->bh = NULL;
4340	if (ino < EXT4_ROOT_INO ||
4341	    ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4342		return -EFSCORRUPTED;
4343
4344	iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4345	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4346	if (!gdp)
4347		return -EIO;
4348
4349	/*
4350	 * Figure out the offset within the block group inode table
4351	 */
4352	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4353	inode_offset = ((ino - 1) %
4354			EXT4_INODES_PER_GROUP(sb));
 
4355	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4356
4357	block = ext4_inode_table(sb, gdp);
4358	if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4359	    (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4360		ext4_error(sb, "Invalid inode table block %llu in "
4361			   "block_group %u", block, iloc->block_group);
4362		return -EFSCORRUPTED;
4363	}
4364	block += (inode_offset / inodes_per_block);
4365
4366	bh = sb_getblk(sb, block);
4367	if (unlikely(!bh))
4368		return -ENOMEM;
4369	if (ext4_buffer_uptodate(bh))
4370		goto has_buffer;
4371
4372	lock_buffer(bh);
4373	if (ext4_buffer_uptodate(bh)) {
4374		/* Someone brought it uptodate while we waited */
4375		unlock_buffer(bh);
4376		goto has_buffer;
4377	}
 
 
4378
4379	/*
4380	 * If we have all information of the inode in memory and this
4381	 * is the only valid inode in the block, we need not read the
4382	 * block.
4383	 */
4384	if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4385		struct buffer_head *bitmap_bh;
4386		int i, start;
4387
4388		start = inode_offset & ~(inodes_per_block - 1);
4389
4390		/* Is the inode bitmap in cache? */
4391		bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4392		if (unlikely(!bitmap_bh))
4393			goto make_io;
4394
4395		/*
4396		 * If the inode bitmap isn't in cache then the
4397		 * optimisation may end up performing two reads instead
4398		 * of one, so skip it.
4399		 */
4400		if (!buffer_uptodate(bitmap_bh)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4401			brelse(bitmap_bh);
4402			goto make_io;
4403		}
4404		for (i = start; i < start + inodes_per_block; i++) {
4405			if (i == inode_offset)
4406				continue;
4407			if (ext4_test_bit(i, bitmap_bh->b_data))
4408				break;
4409		}
4410		brelse(bitmap_bh);
4411		if (i == start + inodes_per_block) {
4412			struct ext4_inode *raw_inode =
4413				(struct ext4_inode *) (bh->b_data + iloc->offset);
4414
4415			/* all other inodes are free, so skip I/O */
4416			memset(bh->b_data, 0, bh->b_size);
4417			if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4418				ext4_fill_raw_inode(inode, raw_inode);
4419			set_buffer_uptodate(bh);
4420			unlock_buffer(bh);
4421			goto has_buffer;
4422		}
4423	}
4424
4425make_io:
4426	/*
4427	 * If we need to do any I/O, try to pre-readahead extra
4428	 * blocks from the inode table.
4429	 */
4430	blk_start_plug(&plug);
4431	if (EXT4_SB(sb)->s_inode_readahead_blks) {
4432		ext4_fsblk_t b, end, table;
4433		unsigned num;
4434		__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4435
4436		table = ext4_inode_table(sb, gdp);
4437		/* s_inode_readahead_blks is always a power of 2 */
4438		b = block & ~((ext4_fsblk_t) ra_blks - 1);
4439		if (table > b)
4440			b = table;
4441		end = b + ra_blks;
4442		num = EXT4_INODES_PER_GROUP(sb);
4443		if (ext4_has_group_desc_csum(sb))
4444			num -= ext4_itable_unused_count(sb, gdp);
4445		table += num / inodes_per_block;
4446		if (end > table)
4447			end = table;
4448		while (b <= end)
4449			ext4_sb_breadahead_unmovable(sb, b++);
4450	}
4451
4452	/*
4453	 * There are other valid inodes in the buffer, this inode
4454	 * has in-inode xattrs, or we don't have this inode in memory.
4455	 * Read the block from disk.
4456	 */
4457	trace_ext4_load_inode(sb, ino);
4458	ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4459	blk_finish_plug(&plug);
4460	wait_on_buffer(bh);
4461	ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4462	if (!buffer_uptodate(bh)) {
4463		if (ret_block)
4464			*ret_block = block;
4465		brelse(bh);
4466		return -EIO;
 
4467	}
4468has_buffer:
4469	iloc->bh = bh;
4470	return 0;
4471}
4472
4473static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4474					struct ext4_iloc *iloc)
4475{
4476	ext4_fsblk_t err_blk = 0;
4477	int ret;
4478
4479	ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4480					&err_blk);
4481
4482	if (ret == -EIO)
4483		ext4_error_inode_block(inode, err_blk, EIO,
4484					"unable to read itable block");
4485
4486	return ret;
4487}
4488
4489int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4490{
4491	ext4_fsblk_t err_blk = 0;
4492	int ret;
4493
4494	ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4495					&err_blk);
4496
4497	if (ret == -EIO)
4498		ext4_error_inode_block(inode, err_blk, EIO,
4499					"unable to read itable block");
4500
4501	return ret;
4502}
4503
4504
4505int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4506			  struct ext4_iloc *iloc)
4507{
4508	return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4509}
4510
4511static bool ext4_should_enable_dax(struct inode *inode)
4512{
4513	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4514
4515	if (test_opt2(inode->i_sb, DAX_NEVER))
4516		return false;
4517	if (!S_ISREG(inode->i_mode))
4518		return false;
4519	if (ext4_should_journal_data(inode))
4520		return false;
4521	if (ext4_has_inline_data(inode))
4522		return false;
4523	if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4524		return false;
4525	if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4526		return false;
4527	if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4528		return false;
4529	if (test_opt(inode->i_sb, DAX_ALWAYS))
4530		return true;
4531
4532	return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4533}
4534
4535void ext4_set_inode_flags(struct inode *inode, bool init)
4536{
4537	unsigned int flags = EXT4_I(inode)->i_flags;
4538	unsigned int new_fl = 0;
4539
4540	WARN_ON_ONCE(IS_DAX(inode) && init);
4541
 
4542	if (flags & EXT4_SYNC_FL)
4543		new_fl |= S_SYNC;
4544	if (flags & EXT4_APPEND_FL)
4545		new_fl |= S_APPEND;
4546	if (flags & EXT4_IMMUTABLE_FL)
4547		new_fl |= S_IMMUTABLE;
4548	if (flags & EXT4_NOATIME_FL)
4549		new_fl |= S_NOATIME;
4550	if (flags & EXT4_DIRSYNC_FL)
4551		new_fl |= S_DIRSYNC;
 
4552
4553	/* Because of the way inode_set_flags() works we must preserve S_DAX
4554	 * here if already set. */
4555	new_fl |= (inode->i_flags & S_DAX);
4556	if (init && ext4_should_enable_dax(inode))
4557		new_fl |= S_DAX;
4558
4559	if (flags & EXT4_ENCRYPT_FL)
4560		new_fl |= S_ENCRYPTED;
4561	if (flags & EXT4_CASEFOLD_FL)
4562		new_fl |= S_CASEFOLD;
4563	if (flags & EXT4_VERITY_FL)
4564		new_fl |= S_VERITY;
4565	inode_set_flags(inode, new_fl,
4566			S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4567			S_ENCRYPTED|S_CASEFOLD|S_VERITY);
 
 
 
 
 
 
 
 
4568}
4569
4570static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4571				  struct ext4_inode_info *ei)
4572{
4573	blkcnt_t i_blocks ;
4574	struct inode *inode = &(ei->vfs_inode);
4575	struct super_block *sb = inode->i_sb;
4576
4577	if (ext4_has_feature_huge_file(sb)) {
 
4578		/* we are using combined 48 bit field */
4579		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4580					le32_to_cpu(raw_inode->i_blocks_lo);
4581		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4582			/* i_blocks represent file system block size */
4583			return i_blocks  << (inode->i_blkbits - 9);
4584		} else {
4585			return i_blocks;
4586		}
4587	} else {
4588		return le32_to_cpu(raw_inode->i_blocks_lo);
4589	}
4590}
4591
4592static inline int ext4_iget_extra_inode(struct inode *inode,
4593					 struct ext4_inode *raw_inode,
4594					 struct ext4_inode_info *ei)
4595{
4596	__le32 *magic = (void *)raw_inode +
4597			EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4598
4599	if (EXT4_INODE_HAS_XATTR_SPACE(inode)  &&
4600	    *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4601		int err;
4602
4603		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4604		err = ext4_find_inline_data_nolock(inode);
4605		if (!err && ext4_has_inline_data(inode))
4606			ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4607		return err;
4608	} else
4609		EXT4_I(inode)->i_inline_off = 0;
4610	return 0;
4611}
4612
4613int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4614{
4615	if (!ext4_has_feature_project(inode->i_sb))
4616		return -EOPNOTSUPP;
4617	*projid = EXT4_I(inode)->i_projid;
4618	return 0;
4619}
4620
4621/*
4622 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4623 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4624 * set.
4625 */
4626static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4627{
4628	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4629		inode_set_iversion_raw(inode, val);
4630	else
4631		inode_set_iversion_queried(inode, val);
4632}
4633
4634static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags)
4635
4636{
4637	if (flags & EXT4_IGET_EA_INODE) {
4638		if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4639			return "missing EA_INODE flag";
4640		if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4641		    EXT4_I(inode)->i_file_acl)
4642			return "ea_inode with extended attributes";
4643	} else {
4644		if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4645			return "unexpected EA_INODE flag";
4646	}
4647	if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD))
4648		return "unexpected bad inode w/o EXT4_IGET_BAD";
4649	return NULL;
4650}
4651
4652struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4653			  ext4_iget_flags flags, const char *function,
4654			  unsigned int line)
4655{
4656	struct ext4_iloc iloc;
4657	struct ext4_inode *raw_inode;
4658	struct ext4_inode_info *ei;
4659	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4660	struct inode *inode;
4661	const char *err_str;
4662	journal_t *journal = EXT4_SB(sb)->s_journal;
4663	long ret;
4664	loff_t size;
4665	int block;
4666	uid_t i_uid;
4667	gid_t i_gid;
4668	projid_t i_projid;
4669
4670	if ((!(flags & EXT4_IGET_SPECIAL) &&
4671	     ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4672	      ino == le32_to_cpu(es->s_usr_quota_inum) ||
4673	      ino == le32_to_cpu(es->s_grp_quota_inum) ||
4674	      ino == le32_to_cpu(es->s_prj_quota_inum) ||
4675	      ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4676	    (ino < EXT4_ROOT_INO) ||
4677	    (ino > le32_to_cpu(es->s_inodes_count))) {
4678		if (flags & EXT4_IGET_HANDLE)
4679			return ERR_PTR(-ESTALE);
4680		__ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4681			     "inode #%lu: comm %s: iget: illegal inode #",
4682			     ino, current->comm);
4683		return ERR_PTR(-EFSCORRUPTED);
4684	}
4685
4686	inode = iget_locked(sb, ino);
4687	if (!inode)
4688		return ERR_PTR(-ENOMEM);
4689	if (!(inode->i_state & I_NEW)) {
4690		if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4691			ext4_error_inode(inode, function, line, 0, err_str);
4692			iput(inode);
4693			return ERR_PTR(-EFSCORRUPTED);
4694		}
4695		return inode;
4696	}
4697
4698	ei = EXT4_I(inode);
4699	iloc.bh = NULL;
4700
4701	ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4702	if (ret < 0)
4703		goto bad_inode;
4704	raw_inode = ext4_raw_inode(&iloc);
4705
4706	if ((flags & EXT4_IGET_HANDLE) &&
4707	    (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4708		ret = -ESTALE;
4709		goto bad_inode;
4710	}
4711
4712	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4713		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4714		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4715			EXT4_INODE_SIZE(inode->i_sb) ||
4716		    (ei->i_extra_isize & 3)) {
4717			ext4_error_inode(inode, function, line, 0,
4718					 "iget: bad extra_isize %u "
4719					 "(inode size %u)",
4720					 ei->i_extra_isize,
4721					 EXT4_INODE_SIZE(inode->i_sb));
4722			ret = -EFSCORRUPTED;
4723			goto bad_inode;
4724		}
4725	} else
4726		ei->i_extra_isize = 0;
4727
4728	/* Precompute checksum seed for inode metadata */
4729	if (ext4_has_metadata_csum(sb)) {
 
4730		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4731		__u32 csum;
4732		__le32 inum = cpu_to_le32(inode->i_ino);
4733		__le32 gen = raw_inode->i_generation;
4734		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4735				   sizeof(inum));
4736		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4737					      sizeof(gen));
4738	}
4739
4740	if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4741	    ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4742	     (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4743		ext4_error_inode_err(inode, function, line, 0,
4744				EFSBADCRC, "iget: checksum invalid");
4745		ret = -EFSBADCRC;
4746		goto bad_inode;
4747	}
4748
4749	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4750	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4751	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4752	if (ext4_has_feature_project(sb) &&
4753	    EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4754	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4755		i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4756	else
4757		i_projid = EXT4_DEF_PROJID;
4758
4759	if (!(test_opt(inode->i_sb, NO_UID32))) {
4760		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4761		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4762	}
4763	i_uid_write(inode, i_uid);
4764	i_gid_write(inode, i_gid);
4765	ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4766	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4767
4768	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
4769	ei->i_inline_off = 0;
4770	ei->i_dir_start_lookup = 0;
4771	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4772	/* We now have enough fields to check if the inode was active or not.
4773	 * This is needed because nfsd might try to access dead inodes
4774	 * the test is that same one that e2fsck uses
4775	 * NeilBrown 1999oct15
4776	 */
4777	if (inode->i_nlink == 0) {
4778		if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4779		     !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4780		    ino != EXT4_BOOT_LOADER_INO) {
4781			/* this inode is deleted or unallocated */
4782			if (flags & EXT4_IGET_SPECIAL) {
4783				ext4_error_inode(inode, function, line, 0,
4784						 "iget: special inode unallocated");
4785				ret = -EFSCORRUPTED;
4786			} else
4787				ret = -ESTALE;
4788			goto bad_inode;
4789		}
4790		/* The only unlinked inodes we let through here have
4791		 * valid i_mode and are being read by the orphan
4792		 * recovery code: that's fine, we're about to complete
4793		 * the process of deleting those.
4794		 * OR it is the EXT4_BOOT_LOADER_INO which is
4795		 * not initialized on a new filesystem. */
4796	}
4797	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4798	ext4_set_inode_flags(inode, true);
4799	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4800	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4801	if (ext4_has_feature_64bit(sb))
4802		ei->i_file_acl |=
4803			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4804	inode->i_size = ext4_isize(sb, raw_inode);
4805	if ((size = i_size_read(inode)) < 0) {
4806		ext4_error_inode(inode, function, line, 0,
4807				 "iget: bad i_size value: %lld", size);
4808		ret = -EFSCORRUPTED;
4809		goto bad_inode;
4810	}
4811	/*
4812	 * If dir_index is not enabled but there's dir with INDEX flag set,
4813	 * we'd normally treat htree data as empty space. But with metadata
4814	 * checksumming that corrupts checksums so forbid that.
4815	 */
4816	if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4817	    ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4818		ext4_error_inode(inode, function, line, 0,
4819			 "iget: Dir with htree data on filesystem without dir_index feature.");
4820		ret = -EFSCORRUPTED;
4821		goto bad_inode;
4822	}
4823	ei->i_disksize = inode->i_size;
4824#ifdef CONFIG_QUOTA
4825	ei->i_reserved_quota = 0;
4826#endif
4827	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4828	ei->i_block_group = iloc.block_group;
4829	ei->i_last_alloc_group = ~0;
4830	/*
4831	 * NOTE! The in-memory inode i_data array is in little-endian order
4832	 * even on big-endian machines: we do NOT byteswap the block numbers!
4833	 */
4834	for (block = 0; block < EXT4_N_BLOCKS; block++)
4835		ei->i_data[block] = raw_inode->i_block[block];
4836	INIT_LIST_HEAD(&ei->i_orphan);
4837	ext4_fc_init_inode(&ei->vfs_inode);
4838
4839	/*
4840	 * Set transaction id's of transactions that have to be committed
4841	 * to finish f[data]sync. We set them to currently running transaction
4842	 * as we cannot be sure that the inode or some of its metadata isn't
4843	 * part of the transaction - the inode could have been reclaimed and
4844	 * now it is reread from disk.
4845	 */
4846	if (journal) {
4847		transaction_t *transaction;
4848		tid_t tid;
4849
4850		read_lock(&journal->j_state_lock);
4851		if (journal->j_running_transaction)
4852			transaction = journal->j_running_transaction;
4853		else
4854			transaction = journal->j_committing_transaction;
4855		if (transaction)
4856			tid = transaction->t_tid;
4857		else
4858			tid = journal->j_commit_sequence;
4859		read_unlock(&journal->j_state_lock);
4860		ei->i_sync_tid = tid;
4861		ei->i_datasync_tid = tid;
4862	}
4863
4864	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4865		if (ei->i_extra_isize == 0) {
4866			/* The extra space is currently unused. Use it. */
4867			BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4868			ei->i_extra_isize = sizeof(struct ext4_inode) -
4869					    EXT4_GOOD_OLD_INODE_SIZE;
4870		} else {
4871			ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4872			if (ret)
4873				goto bad_inode;
 
 
4874		}
4875	}
4876
4877	EXT4_INODE_GET_CTIME(inode, raw_inode);
4878	EXT4_INODE_GET_ATIME(inode, raw_inode);
4879	EXT4_INODE_GET_MTIME(inode, raw_inode);
4880	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4881
4882	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4883		u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4884
4885		if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4886			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4887				ivers |=
4888		    (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4889		}
4890		ext4_inode_set_iversion_queried(inode, ivers);
4891	}
4892
4893	ret = 0;
4894	if (ei->i_file_acl &&
4895	    !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4896		ext4_error_inode(inode, function, line, 0,
4897				 "iget: bad extended attribute block %llu",
4898				 ei->i_file_acl);
4899		ret = -EFSCORRUPTED;
4900		goto bad_inode;
4901	} else if (!ext4_has_inline_data(inode)) {
4902		/* validate the block references in the inode */
4903		if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4904			(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4905			(S_ISLNK(inode->i_mode) &&
4906			!ext4_inode_is_fast_symlink(inode)))) {
4907			if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4908				ret = ext4_ext_check_inode(inode);
4909			else
4910				ret = ext4_ind_check_inode(inode);
4911		}
4912	}
4913	if (ret)
4914		goto bad_inode;
4915
4916	if (S_ISREG(inode->i_mode)) {
4917		inode->i_op = &ext4_file_inode_operations;
4918		inode->i_fop = &ext4_file_operations;
4919		ext4_set_aops(inode);
4920	} else if (S_ISDIR(inode->i_mode)) {
4921		inode->i_op = &ext4_dir_inode_operations;
4922		inode->i_fop = &ext4_dir_operations;
4923	} else if (S_ISLNK(inode->i_mode)) {
4924		/* VFS does not allow setting these so must be corruption */
4925		if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4926			ext4_error_inode(inode, function, line, 0,
4927					 "iget: immutable or append flags "
4928					 "not allowed on symlinks");
4929			ret = -EFSCORRUPTED;
4930			goto bad_inode;
4931		}
4932		if (IS_ENCRYPTED(inode)) {
4933			inode->i_op = &ext4_encrypted_symlink_inode_operations;
4934		} else if (ext4_inode_is_fast_symlink(inode)) {
4935			inode->i_link = (char *)ei->i_data;
4936			inode->i_op = &ext4_fast_symlink_inode_operations;
4937			nd_terminate_link(ei->i_data, inode->i_size,
4938				sizeof(ei->i_data) - 1);
4939		} else {
4940			inode->i_op = &ext4_symlink_inode_operations;
 
4941		}
4942	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4943	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4944		inode->i_op = &ext4_special_inode_operations;
4945		if (raw_inode->i_block[0])
4946			init_special_inode(inode, inode->i_mode,
4947			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4948		else
4949			init_special_inode(inode, inode->i_mode,
4950			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4951	} else if (ino == EXT4_BOOT_LOADER_INO) {
4952		make_bad_inode(inode);
4953	} else {
4954		ret = -EFSCORRUPTED;
4955		ext4_error_inode(inode, function, line, 0,
4956				 "iget: bogus i_mode (%o)", inode->i_mode);
4957		goto bad_inode;
4958	}
4959	if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) {
4960		ext4_error_inode(inode, function, line, 0,
4961				 "casefold flag without casefold feature");
4962		ret = -EFSCORRUPTED;
4963		goto bad_inode;
4964	}
4965	if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4966		ext4_error_inode(inode, function, line, 0, err_str);
4967		ret = -EFSCORRUPTED;
4968		goto bad_inode;
4969	}
4970
4971	brelse(iloc.bh);
 
4972	unlock_new_inode(inode);
4973	return inode;
4974
4975bad_inode:
4976	brelse(iloc.bh);
4977	iget_failed(inode);
4978	return ERR_PTR(ret);
4979}
4980
4981static void __ext4_update_other_inode_time(struct super_block *sb,
4982					   unsigned long orig_ino,
4983					   unsigned long ino,
4984					   struct ext4_inode *raw_inode)
4985{
4986	struct inode *inode;
 
 
4987
4988	inode = find_inode_by_ino_rcu(sb, ino);
4989	if (!inode)
4990		return;
4991
4992	if (!inode_is_dirtytime_only(inode))
4993		return;
4994
4995	spin_lock(&inode->i_lock);
4996	if (inode_is_dirtytime_only(inode)) {
4997		struct ext4_inode_info	*ei = EXT4_I(inode);
4998
4999		inode->i_state &= ~I_DIRTY_TIME;
5000		spin_unlock(&inode->i_lock);
5001
5002		spin_lock(&ei->i_raw_lock);
5003		EXT4_INODE_SET_CTIME(inode, raw_inode);
5004		EXT4_INODE_SET_MTIME(inode, raw_inode);
5005		EXT4_INODE_SET_ATIME(inode, raw_inode);
5006		ext4_inode_csum_set(inode, raw_inode, ei);
5007		spin_unlock(&ei->i_raw_lock);
5008		trace_ext4_other_inode_update_time(inode, orig_ino);
5009		return;
5010	}
5011	spin_unlock(&inode->i_lock);
5012}
5013
5014/*
5015 * Opportunistically update the other time fields for other inodes in
5016 * the same inode table block.
5017 */
5018static void ext4_update_other_inodes_time(struct super_block *sb,
5019					  unsigned long orig_ino, char *buf)
5020{
5021	unsigned long ino;
5022	int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5023	int inode_size = EXT4_INODE_SIZE(sb);
5024
5025	/*
5026	 * Calculate the first inode in the inode table block.  Inode
5027	 * numbers are one-based.  That is, the first inode in a block
5028	 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5029	 */
5030	ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5031	rcu_read_lock();
5032	for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5033		if (ino == orig_ino)
5034			continue;
5035		__ext4_update_other_inode_time(sb, orig_ino, ino,
5036					       (struct ext4_inode *)buf);
5037	}
5038	rcu_read_unlock();
5039}
5040
5041/*
5042 * Post the struct inode info into an on-disk inode location in the
5043 * buffer-cache.  This gobbles the caller's reference to the
5044 * buffer_head in the inode location struct.
5045 *
5046 * The caller must have write access to iloc->bh.
5047 */
5048static int ext4_do_update_inode(handle_t *handle,
5049				struct inode *inode,
5050				struct ext4_iloc *iloc)
5051{
5052	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5053	struct ext4_inode_info *ei = EXT4_I(inode);
5054	struct buffer_head *bh = iloc->bh;
5055	struct super_block *sb = inode->i_sb;
5056	int err;
5057	int need_datasync = 0, set_large_file = 0;
5058
5059	spin_lock(&ei->i_raw_lock);
5060
5061	/*
5062	 * For fields not tracked in the in-memory inode, initialise them
5063	 * to zero for new inodes.
5064	 */
5065	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5066		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5067
5068	if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5069		need_datasync = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5070	if (ei->i_disksize > 0x7fffffffULL) {
5071		if (!ext4_has_feature_large_file(sb) ||
5072		    EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5073			set_large_file = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5074	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5075
5076	err = ext4_fill_raw_inode(inode, raw_inode);
5077	spin_unlock(&ei->i_raw_lock);
5078	if (err) {
5079		EXT4_ERROR_INODE(inode, "corrupted inode contents");
5080		goto out_brelse;
 
5081	}
5082
5083	if (inode->i_sb->s_flags & SB_LAZYTIME)
5084		ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5085					      bh->b_data);
5086
5087	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5088	err = ext4_handle_dirty_metadata(handle, NULL, bh);
5089	if (err)
5090		goto out_error;
5091	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5092	if (set_large_file) {
5093		BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5094		err = ext4_journal_get_write_access(handle, sb,
5095						    EXT4_SB(sb)->s_sbh,
5096						    EXT4_JTR_NONE);
5097		if (err)
5098			goto out_error;
5099		lock_buffer(EXT4_SB(sb)->s_sbh);
5100		ext4_set_feature_large_file(sb);
5101		ext4_superblock_csum_set(sb);
5102		unlock_buffer(EXT4_SB(sb)->s_sbh);
5103		ext4_handle_sync(handle);
5104		err = ext4_handle_dirty_metadata(handle, NULL,
5105						 EXT4_SB(sb)->s_sbh);
5106	}
5107	ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5108out_error:
5109	ext4_std_error(inode->i_sb, err);
5110out_brelse:
5111	brelse(bh);
 
5112	return err;
5113}
5114
5115/*
5116 * ext4_write_inode()
5117 *
5118 * We are called from a few places:
5119 *
5120 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5121 *   Here, there will be no transaction running. We wait for any running
5122 *   transaction to commit.
5123 *
5124 * - Within flush work (sys_sync(), kupdate and such).
5125 *   We wait on commit, if told to.
5126 *
5127 * - Within iput_final() -> write_inode_now()
5128 *   We wait on commit, if told to.
 
5129 *
5130 * In all cases it is actually safe for us to return without doing anything,
5131 * because the inode has been copied into a raw inode buffer in
5132 * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
5133 * writeback.
5134 *
5135 * Note that we are absolutely dependent upon all inode dirtiers doing the
5136 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5137 * which we are interested.
5138 *
5139 * It would be a bug for them to not do this.  The code:
5140 *
5141 *	mark_inode_dirty(inode)
5142 *	stuff();
5143 *	inode->i_size = expr;
5144 *
5145 * is in error because write_inode() could occur while `stuff()' is running,
5146 * and the new i_size will be lost.  Plus the inode will no longer be on the
5147 * superblock's dirty inode list.
5148 */
5149int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5150{
5151	int err;
5152
5153	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5154		return 0;
5155
5156	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5157		return -EIO;
5158
5159	if (EXT4_SB(inode->i_sb)->s_journal) {
5160		if (ext4_journal_current_handle()) {
5161			ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5162			dump_stack();
5163			return -EIO;
5164		}
5165
5166		/*
5167		 * No need to force transaction in WB_SYNC_NONE mode. Also
5168		 * ext4_sync_fs() will force the commit after everything is
5169		 * written.
5170		 */
5171		if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5172			return 0;
5173
5174		err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5175						EXT4_I(inode)->i_sync_tid);
5176	} else {
5177		struct ext4_iloc iloc;
5178
5179		err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5180		if (err)
5181			return err;
5182		/*
5183		 * sync(2) will flush the whole buffer cache. No need to do
5184		 * it here separately for each inode.
5185		 */
5186		if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5187			sync_dirty_buffer(iloc.bh);
5188		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5189			ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5190					       "IO error syncing inode");
5191			err = -EIO;
5192		}
5193		brelse(iloc.bh);
5194	}
5195	return err;
5196}
5197
5198/*
5199 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5200 * buffers that are attached to a folio straddling i_size and are undergoing
5201 * commit. In that case we have to wait for commit to finish and try again.
5202 */
5203static void ext4_wait_for_tail_page_commit(struct inode *inode)
5204{
5205	unsigned offset;
5206	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5207	tid_t commit_tid = 0;
5208	int ret;
5209
5210	offset = inode->i_size & (PAGE_SIZE - 1);
5211	/*
5212	 * If the folio is fully truncated, we don't need to wait for any commit
5213	 * (and we even should not as __ext4_journalled_invalidate_folio() may
5214	 * strip all buffers from the folio but keep the folio dirty which can then
5215	 * confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5216	 * buffers). Also we don't need to wait for any commit if all buffers in
5217	 * the folio remain valid. This is most beneficial for the common case of
5218	 * blocksize == PAGESIZE.
5219	 */
5220	if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5221		return;
5222	while (1) {
5223		struct folio *folio = filemap_lock_folio(inode->i_mapping,
5224				      inode->i_size >> PAGE_SHIFT);
5225		if (IS_ERR(folio))
5226			return;
5227		ret = __ext4_journalled_invalidate_folio(folio, offset,
5228						folio_size(folio) - offset);
5229		folio_unlock(folio);
5230		folio_put(folio);
5231		if (ret != -EBUSY)
5232			return;
5233		commit_tid = 0;
5234		read_lock(&journal->j_state_lock);
5235		if (journal->j_committing_transaction)
5236			commit_tid = journal->j_committing_transaction->t_tid;
5237		read_unlock(&journal->j_state_lock);
5238		if (commit_tid)
5239			jbd2_log_wait_commit(journal, commit_tid);
5240	}
5241}
5242
5243/*
5244 * ext4_setattr()
5245 *
5246 * Called from notify_change.
5247 *
5248 * We want to trap VFS attempts to truncate the file as soon as
5249 * possible.  In particular, we want to make sure that when the VFS
5250 * shrinks i_size, we put the inode on the orphan list and modify
5251 * i_disksize immediately, so that during the subsequent flushing of
5252 * dirty pages and freeing of disk blocks, we can guarantee that any
5253 * commit will leave the blocks being flushed in an unused state on
5254 * disk.  (On recovery, the inode will get truncated and the blocks will
5255 * be freed, so we have a strong guarantee that no future commit will
5256 * leave these blocks visible to the user.)
5257 *
5258 * Another thing we have to assure is that if we are in ordered mode
5259 * and inode is still attached to the committing transaction, we must
5260 * we start writeout of all the dirty pages which are being truncated.
5261 * This way we are sure that all the data written in the previous
5262 * transaction are already on disk (truncate waits for pages under
5263 * writeback).
5264 *
5265 * Called with inode->i_rwsem down.
5266 */
5267int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5268		 struct iattr *attr)
5269{
5270	struct inode *inode = d_inode(dentry);
5271	int error, rc = 0;
5272	int orphan = 0;
5273	const unsigned int ia_valid = attr->ia_valid;
5274	bool inc_ivers = true;
5275
5276	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5277		return -EIO;
5278
5279	if (unlikely(IS_IMMUTABLE(inode)))
5280		return -EPERM;
5281
5282	if (unlikely(IS_APPEND(inode) &&
5283		     (ia_valid & (ATTR_MODE | ATTR_UID |
5284				  ATTR_GID | ATTR_TIMES_SET))))
5285		return -EPERM;
5286
5287	error = setattr_prepare(idmap, dentry, attr);
5288	if (error)
5289		return error;
5290
5291	error = fscrypt_prepare_setattr(dentry, attr);
5292	if (error)
5293		return error;
5294
5295	error = fsverity_prepare_setattr(dentry, attr);
5296	if (error)
5297		return error;
5298
5299	if (is_quota_modification(idmap, inode, attr)) {
5300		error = dquot_initialize(inode);
5301		if (error)
5302			return error;
5303	}
5304
5305	if (i_uid_needs_update(idmap, attr, inode) ||
5306	    i_gid_needs_update(idmap, attr, inode)) {
5307		handle_t *handle;
5308
5309		/* (user+group)*(old+new) structure, inode write (sb,
5310		 * inode block, ? - but truncate inode update has it) */
5311		handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5312			(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5313			 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5314		if (IS_ERR(handle)) {
5315			error = PTR_ERR(handle);
5316			goto err_out;
5317		}
5318
5319		/* dquot_transfer() calls back ext4_get_inode_usage() which
5320		 * counts xattr inode references.
5321		 */
5322		down_read(&EXT4_I(inode)->xattr_sem);
5323		error = dquot_transfer(idmap, inode, attr);
5324		up_read(&EXT4_I(inode)->xattr_sem);
5325
5326		if (error) {
5327			ext4_journal_stop(handle);
5328			return error;
5329		}
5330		/* Update corresponding info in inode so that everything is in
5331		 * one transaction */
5332		i_uid_update(idmap, attr, inode);
5333		i_gid_update(idmap, attr, inode);
 
 
5334		error = ext4_mark_inode_dirty(handle, inode);
5335		ext4_journal_stop(handle);
5336		if (unlikely(error)) {
5337			return error;
5338		}
5339	}
5340
5341	if (attr->ia_valid & ATTR_SIZE) {
5342		handle_t *handle;
5343		loff_t oldsize = inode->i_size;
5344		loff_t old_disksize;
5345		int shrink = (attr->ia_size < inode->i_size);
5346
5347		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5348			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5349
5350			if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5351				return -EFBIG;
5352			}
5353		}
5354		if (!S_ISREG(inode->i_mode)) {
5355			return -EINVAL;
 
 
 
 
 
 
 
 
 
5356		}
 
 
 
 
 
 
 
 
 
5357
5358		if (attr->ia_size == inode->i_size)
5359			inc_ivers = false;
5360
5361		if (shrink) {
5362			if (ext4_should_order_data(inode)) {
5363				error = ext4_begin_ordered_truncate(inode,
5364							    attr->ia_size);
5365				if (error)
 
 
 
 
5366					goto err_out;
 
 
 
 
 
5367			}
5368			/*
5369			 * Blocks are going to be removed from the inode. Wait
5370			 * for dio in flight.
5371			 */
5372			inode_dio_wait(inode);
5373		}
 
5374
5375		filemap_invalidate_lock(inode->i_mapping);
 
 
 
 
5376
5377		rc = ext4_break_layouts(inode);
5378		if (rc) {
5379			filemap_invalidate_unlock(inode->i_mapping);
5380			goto err_out;
5381		}
5382
5383		if (attr->ia_size != inode->i_size) {
5384			handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5385			if (IS_ERR(handle)) {
5386				error = PTR_ERR(handle);
5387				goto out_mmap_sem;
5388			}
5389			if (ext4_handle_valid(handle) && shrink) {
5390				error = ext4_orphan_add(handle, inode);
5391				orphan = 1;
5392			}
5393			/*
5394			 * Update c/mtime on truncate up, ext4_truncate() will
5395			 * update c/mtime in shrink case below
5396			 */
5397			if (!shrink)
5398				inode_set_mtime_to_ts(inode,
5399						      inode_set_ctime_current(inode));
5400
5401			if (shrink)
5402				ext4_fc_track_range(handle, inode,
5403					(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5404					inode->i_sb->s_blocksize_bits,
5405					EXT_MAX_BLOCKS - 1);
5406			else
5407				ext4_fc_track_range(
5408					handle, inode,
5409					(oldsize > 0 ? oldsize - 1 : oldsize) >>
5410					inode->i_sb->s_blocksize_bits,
5411					(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5412					inode->i_sb->s_blocksize_bits);
5413
5414			down_write(&EXT4_I(inode)->i_data_sem);
5415			old_disksize = EXT4_I(inode)->i_disksize;
5416			EXT4_I(inode)->i_disksize = attr->ia_size;
5417			rc = ext4_mark_inode_dirty(handle, inode);
5418			if (!error)
5419				error = rc;
5420			/*
5421			 * We have to update i_size under i_data_sem together
5422			 * with i_disksize to avoid races with writeback code
5423			 * running ext4_wb_update_i_disksize().
5424			 */
5425			if (!error)
5426				i_size_write(inode, attr->ia_size);
5427			else
5428				EXT4_I(inode)->i_disksize = old_disksize;
5429			up_write(&EXT4_I(inode)->i_data_sem);
5430			ext4_journal_stop(handle);
5431			if (error)
5432				goto out_mmap_sem;
5433			if (!shrink) {
5434				pagecache_isize_extended(inode, oldsize,
5435							 inode->i_size);
5436			} else if (ext4_should_journal_data(inode)) {
5437				ext4_wait_for_tail_page_commit(inode);
5438			}
5439		}
5440
5441		/*
5442		 * Truncate pagecache after we've waited for commit
5443		 * in data=journal mode to make pages freeable.
5444		 */
5445		truncate_pagecache(inode, inode->i_size);
5446		/*
5447		 * Call ext4_truncate() even if i_size didn't change to
5448		 * truncate possible preallocated blocks.
5449		 */
5450		if (attr->ia_size <= oldsize) {
5451			rc = ext4_truncate(inode);
5452			if (rc)
5453				error = rc;
5454		}
5455out_mmap_sem:
5456		filemap_invalidate_unlock(inode->i_mapping);
5457	}
5458
5459	if (!error) {
5460		if (inc_ivers)
5461			inode_inc_iversion(inode);
5462		setattr_copy(idmap, inode, attr);
5463		mark_inode_dirty(inode);
5464	}
5465
5466	/*
5467	 * If the call to ext4_truncate failed to get a transaction handle at
5468	 * all, we need to clean up the in-core orphan list manually.
5469	 */
5470	if (orphan && inode->i_nlink)
5471		ext4_orphan_del(NULL, inode);
5472
5473	if (!error && (ia_valid & ATTR_MODE))
5474		rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
5475
5476err_out:
5477	if  (error)
5478		ext4_std_error(inode->i_sb, error);
5479	if (!error)
5480		error = rc;
5481	return error;
5482}
5483
5484u32 ext4_dio_alignment(struct inode *inode)
 
5485{
5486	if (fsverity_active(inode))
5487		return 0;
5488	if (ext4_should_journal_data(inode))
5489		return 0;
5490	if (ext4_has_inline_data(inode))
5491		return 0;
5492	if (IS_ENCRYPTED(inode)) {
5493		if (!fscrypt_dio_supported(inode))
5494			return 0;
5495		return i_blocksize(inode);
5496	}
5497	return 1; /* use the iomap defaults */
5498}
5499
5500int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
5501		 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5502{
5503	struct inode *inode = d_inode(path->dentry);
5504	struct ext4_inode *raw_inode;
5505	struct ext4_inode_info *ei = EXT4_I(inode);
5506	unsigned int flags;
5507
5508	if ((request_mask & STATX_BTIME) &&
5509	    EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5510		stat->result_mask |= STATX_BTIME;
5511		stat->btime.tv_sec = ei->i_crtime.tv_sec;
5512		stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5513	}
5514
5515	/*
5516	 * Return the DIO alignment restrictions if requested.  We only return
5517	 * this information when requested, since on encrypted files it might
5518	 * take a fair bit of work to get if the file wasn't opened recently.
5519	 */
5520	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5521		u32 dio_align = ext4_dio_alignment(inode);
5522
5523		stat->result_mask |= STATX_DIOALIGN;
5524		if (dio_align == 1) {
5525			struct block_device *bdev = inode->i_sb->s_bdev;
5526
5527			/* iomap defaults */
5528			stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5529			stat->dio_offset_align = bdev_logical_block_size(bdev);
5530		} else {
5531			stat->dio_mem_align = dio_align;
5532			stat->dio_offset_align = dio_align;
5533		}
5534	}
5535
5536	flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5537	if (flags & EXT4_APPEND_FL)
5538		stat->attributes |= STATX_ATTR_APPEND;
5539	if (flags & EXT4_COMPR_FL)
5540		stat->attributes |= STATX_ATTR_COMPRESSED;
5541	if (flags & EXT4_ENCRYPT_FL)
5542		stat->attributes |= STATX_ATTR_ENCRYPTED;
5543	if (flags & EXT4_IMMUTABLE_FL)
5544		stat->attributes |= STATX_ATTR_IMMUTABLE;
5545	if (flags & EXT4_NODUMP_FL)
5546		stat->attributes |= STATX_ATTR_NODUMP;
5547	if (flags & EXT4_VERITY_FL)
5548		stat->attributes |= STATX_ATTR_VERITY;
5549
5550	stat->attributes_mask |= (STATX_ATTR_APPEND |
5551				  STATX_ATTR_COMPRESSED |
5552				  STATX_ATTR_ENCRYPTED |
5553				  STATX_ATTR_IMMUTABLE |
5554				  STATX_ATTR_NODUMP |
5555				  STATX_ATTR_VERITY);
5556
5557	generic_fillattr(idmap, request_mask, inode, stat);
5558	return 0;
5559}
5560
5561int ext4_file_getattr(struct mnt_idmap *idmap,
5562		      const struct path *path, struct kstat *stat,
5563		      u32 request_mask, unsigned int query_flags)
5564{
5565	struct inode *inode = d_inode(path->dentry);
5566	u64 delalloc_blocks;
5567
5568	ext4_getattr(idmap, path, stat, request_mask, query_flags);
5569
5570	/*
5571	 * If there is inline data in the inode, the inode will normally not
5572	 * have data blocks allocated (it may have an external xattr block).
5573	 * Report at least one sector for such files, so tools like tar, rsync,
5574	 * others don't incorrectly think the file is completely sparse.
5575	 */
5576	if (unlikely(ext4_has_inline_data(inode)))
5577		stat->blocks += (stat->size + 511) >> 9;
5578
5579	/*
5580	 * We can't update i_blocks if the block allocation is delayed
5581	 * otherwise in the case of system crash before the real block
5582	 * allocation is done, we will have i_blocks inconsistent with
5583	 * on-disk file blocks.
5584	 * We always keep i_blocks updated together with real
5585	 * allocation. But to not confuse with user, stat
5586	 * will return the blocks that include the delayed allocation
5587	 * blocks for this file.
5588	 */
5589	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5590				   EXT4_I(inode)->i_reserved_data_blocks);
5591	stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
 
5592	return 0;
5593}
5594
5595static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5596				   int pextents)
5597{
5598	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5599		return ext4_ind_trans_blocks(inode, lblocks);
5600	return ext4_ext_index_trans_blocks(inode, pextents);
5601}
5602
5603/*
5604 * Account for index blocks, block groups bitmaps and block group
5605 * descriptor blocks if modify datablocks and index blocks
5606 * worse case, the indexs blocks spread over different block groups
5607 *
5608 * If datablocks are discontiguous, they are possible to spread over
5609 * different block groups too. If they are contiguous, with flexbg,
5610 * they could still across block group boundary.
5611 *
5612 * Also account for superblock, inode, quota and xattr blocks
5613 */
5614static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5615				  int pextents)
5616{
5617	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5618	int gdpblocks;
5619	int idxblocks;
5620	int ret;
5621
5622	/*
5623	 * How many index blocks need to touch to map @lblocks logical blocks
5624	 * to @pextents physical extents?
 
 
 
 
5625	 */
5626	idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5627
5628	ret = idxblocks;
5629
5630	/*
5631	 * Now let's see how many group bitmaps and group descriptors need
5632	 * to account
5633	 */
5634	groups = idxblocks + pextents;
 
 
 
 
 
5635	gdpblocks = groups;
5636	if (groups > ngroups)
5637		groups = ngroups;
5638	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5639		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5640
5641	/* bitmaps and block group descriptor blocks */
5642	ret += groups + gdpblocks;
5643
5644	/* Blocks for super block, inode, quota and xattr blocks */
5645	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5646
5647	return ret;
5648}
5649
5650/*
5651 * Calculate the total number of credits to reserve to fit
5652 * the modification of a single pages into a single transaction,
5653 * which may include multiple chunks of block allocations.
5654 *
5655 * This could be called via ext4_write_begin()
5656 *
5657 * We need to consider the worse case, when
5658 * one new block per extent.
5659 */
5660int ext4_writepage_trans_blocks(struct inode *inode)
5661{
5662	int bpp = ext4_journal_blocks_per_page(inode);
5663	int ret;
5664
5665	ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5666
5667	/* Account for data blocks for journalled mode */
5668	if (ext4_should_journal_data(inode))
5669		ret += bpp;
5670	return ret;
5671}
5672
5673/*
5674 * Calculate the journal credits for a chunk of data modification.
5675 *
5676 * This is called from DIO, fallocate or whoever calling
5677 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5678 *
5679 * journal buffers for data blocks are not included here, as DIO
5680 * and fallocate do no need to journal data buffers.
5681 */
5682int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5683{
5684	return ext4_meta_trans_blocks(inode, nrblocks, 1);
5685}
5686
5687/*
5688 * The caller must have previously called ext4_reserve_inode_write().
5689 * Give this, we know that the caller already has write access to iloc->bh.
5690 */
5691int ext4_mark_iloc_dirty(handle_t *handle,
5692			 struct inode *inode, struct ext4_iloc *iloc)
5693{
5694	int err = 0;
5695
5696	if (unlikely(ext4_forced_shutdown(inode->i_sb))) {
5697		put_bh(iloc->bh);
5698		return -EIO;
5699	}
5700	ext4_fc_track_inode(handle, inode);
5701
5702	/* the do_update_inode consumes one bh->b_count */
5703	get_bh(iloc->bh);
5704
5705	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5706	err = ext4_do_update_inode(handle, inode, iloc);
5707	put_bh(iloc->bh);
5708	return err;
5709}
5710
5711/*
5712 * On success, We end up with an outstanding reference count against
5713 * iloc->bh.  This _must_ be cleaned up later.
5714 */
5715
5716int
5717ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5718			 struct ext4_iloc *iloc)
5719{
5720	int err;
5721
5722	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5723		return -EIO;
5724
5725	err = ext4_get_inode_loc(inode, iloc);
5726	if (!err) {
5727		BUFFER_TRACE(iloc->bh, "get_write_access");
5728		err = ext4_journal_get_write_access(handle, inode->i_sb,
5729						    iloc->bh, EXT4_JTR_NONE);
5730		if (err) {
5731			brelse(iloc->bh);
5732			iloc->bh = NULL;
5733		}
5734	}
5735	ext4_std_error(inode->i_sb, err);
5736	return err;
5737}
5738
5739static int __ext4_expand_extra_isize(struct inode *inode,
5740				     unsigned int new_extra_isize,
5741				     struct ext4_iloc *iloc,
5742				     handle_t *handle, int *no_expand)
 
 
 
 
5743{
5744	struct ext4_inode *raw_inode;
5745	struct ext4_xattr_ibody_header *header;
5746	unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5747	struct ext4_inode_info *ei = EXT4_I(inode);
5748	int error;
5749
5750	/* this was checked at iget time, but double check for good measure */
5751	if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5752	    (ei->i_extra_isize & 3)) {
5753		EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5754				 ei->i_extra_isize,
5755				 EXT4_INODE_SIZE(inode->i_sb));
5756		return -EFSCORRUPTED;
5757	}
5758	if ((new_extra_isize < ei->i_extra_isize) ||
5759	    (new_extra_isize < 4) ||
5760	    (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5761		return -EINVAL;	/* Should never happen */
5762
5763	raw_inode = ext4_raw_inode(iloc);
5764
5765	header = IHDR(inode, raw_inode);
5766
5767	/* No extended attributes present */
5768	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5769	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5770		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5771		       EXT4_I(inode)->i_extra_isize, 0,
5772		       new_extra_isize - EXT4_I(inode)->i_extra_isize);
5773		EXT4_I(inode)->i_extra_isize = new_extra_isize;
5774		return 0;
5775	}
5776
5777	/*
5778	 * We may need to allocate external xattr block so we need quotas
5779	 * initialized. Here we can be called with various locks held so we
5780	 * cannot affort to initialize quotas ourselves. So just bail.
5781	 */
5782	if (dquot_initialize_needed(inode))
5783		return -EAGAIN;
5784
5785	/* try to expand with EAs present */
5786	error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5787					   raw_inode, handle);
5788	if (error) {
5789		/*
5790		 * Inode size expansion failed; don't try again
5791		 */
5792		*no_expand = 1;
5793	}
5794
5795	return error;
5796}
5797
5798/*
5799 * Expand an inode by new_extra_isize bytes.
5800 * Returns 0 on success or negative error number on failure.
5801 */
5802static int ext4_try_to_expand_extra_isize(struct inode *inode,
5803					  unsigned int new_extra_isize,
5804					  struct ext4_iloc iloc,
5805					  handle_t *handle)
5806{
5807	int no_expand;
5808	int error;
5809
5810	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5811		return -EOVERFLOW;
5812
5813	/*
5814	 * In nojournal mode, we can immediately attempt to expand
5815	 * the inode.  When journaled, we first need to obtain extra
5816	 * buffer credits since we may write into the EA block
5817	 * with this same handle. If journal_extend fails, then it will
5818	 * only result in a minor loss of functionality for that inode.
5819	 * If this is felt to be critical, then e2fsck should be run to
5820	 * force a large enough s_min_extra_isize.
5821	 */
5822	if (ext4_journal_extend(handle,
5823				EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5824		return -ENOSPC;
5825
5826	if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5827		return -EBUSY;
5828
5829	error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5830					  handle, &no_expand);
5831	ext4_write_unlock_xattr(inode, &no_expand);
5832
5833	return error;
5834}
5835
5836int ext4_expand_extra_isize(struct inode *inode,
5837			    unsigned int new_extra_isize,
5838			    struct ext4_iloc *iloc)
5839{
5840	handle_t *handle;
5841	int no_expand;
5842	int error, rc;
5843
5844	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5845		brelse(iloc->bh);
5846		return -EOVERFLOW;
5847	}
5848
5849	handle = ext4_journal_start(inode, EXT4_HT_INODE,
5850				    EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5851	if (IS_ERR(handle)) {
5852		error = PTR_ERR(handle);
5853		brelse(iloc->bh);
5854		return error;
5855	}
5856
5857	ext4_write_lock_xattr(inode, &no_expand);
5858
5859	BUFFER_TRACE(iloc->bh, "get_write_access");
5860	error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5861					      EXT4_JTR_NONE);
5862	if (error) {
5863		brelse(iloc->bh);
5864		goto out_unlock;
5865	}
5866
5867	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5868					  handle, &no_expand);
5869
5870	rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5871	if (!error)
5872		error = rc;
5873
5874out_unlock:
5875	ext4_write_unlock_xattr(inode, &no_expand);
5876	ext4_journal_stop(handle);
5877	return error;
5878}
5879
5880/*
5881 * What we do here is to mark the in-core inode as clean with respect to inode
5882 * dirtiness (it may still be data-dirty).
5883 * This means that the in-core inode may be reaped by prune_icache
5884 * without having to perform any I/O.  This is a very good thing,
5885 * because *any* task may call prune_icache - even ones which
5886 * have a transaction open against a different journal.
5887 *
5888 * Is this cheating?  Not really.  Sure, we haven't written the
5889 * inode out, but prune_icache isn't a user-visible syncing function.
5890 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5891 * we start and wait on commits.
 
 
 
 
 
 
 
 
5892 */
5893int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5894				const char *func, unsigned int line)
5895{
5896	struct ext4_iloc iloc;
5897	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5898	int err;
 
5899
5900	might_sleep();
5901	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5902	err = ext4_reserve_inode_write(handle, inode, &iloc);
5903	if (err)
5904		goto out;
5905
5906	if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5907		ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5908					       iloc, handle);
5909
5910	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5911out:
5912	if (unlikely(err))
5913		ext4_error_inode_err(inode, func, line, 0, err,
5914					"mark_inode_dirty error");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5915	return err;
5916}
5917
5918/*
5919 * ext4_dirty_inode() is called from __mark_inode_dirty()
5920 *
5921 * We're really interested in the case where a file is being extended.
5922 * i_size has been changed by generic_commit_write() and we thus need
5923 * to include the updated inode in the current transaction.
5924 *
5925 * Also, dquot_alloc_block() will always dirty the inode when blocks
5926 * are allocated to the file.
5927 *
5928 * If the inode is marked synchronous, we don't honour that here - doing
5929 * so would cause a commit on atime updates, which we don't bother doing.
5930 * We handle synchronous inodes at the highest possible level.
5931 */
5932void ext4_dirty_inode(struct inode *inode, int flags)
5933{
5934	handle_t *handle;
5935
5936	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5937	if (IS_ERR(handle))
5938		return;
 
5939	ext4_mark_inode_dirty(handle, inode);
 
5940	ext4_journal_stop(handle);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5941}
 
5942
5943int ext4_change_inode_journal_flag(struct inode *inode, int val)
5944{
5945	journal_t *journal;
5946	handle_t *handle;
5947	int err;
5948	int alloc_ctx;
5949
5950	/*
5951	 * We have to be very careful here: changing a data block's
5952	 * journaling status dynamically is dangerous.  If we write a
5953	 * data block to the journal, change the status and then delete
5954	 * that block, we risk forgetting to revoke the old log record
5955	 * from the journal and so a subsequent replay can corrupt data.
5956	 * So, first we make sure that the journal is empty and that
5957	 * nobody is changing anything.
5958	 */
5959
5960	journal = EXT4_JOURNAL(inode);
5961	if (!journal)
5962		return 0;
5963	if (is_journal_aborted(journal))
5964		return -EROFS;
5965
5966	/* Wait for all existing dio workers */
5967	inode_dio_wait(inode);
5968
5969	/*
5970	 * Before flushing the journal and switching inode's aops, we have
5971	 * to flush all dirty data the inode has. There can be outstanding
5972	 * delayed allocations, there can be unwritten extents created by
5973	 * fallocate or buffered writes in dioread_nolock mode covered by
5974	 * dirty data which can be converted only after flushing the dirty
5975	 * data (and journalled aops don't know how to handle these cases).
5976	 */
5977	if (val) {
5978		filemap_invalidate_lock(inode->i_mapping);
5979		err = filemap_write_and_wait(inode->i_mapping);
5980		if (err < 0) {
5981			filemap_invalidate_unlock(inode->i_mapping);
5982			return err;
5983		}
5984	}
5985
5986	alloc_ctx = ext4_writepages_down_write(inode->i_sb);
5987	jbd2_journal_lock_updates(journal);
5988
5989	/*
5990	 * OK, there are no updates running now, and all cached data is
5991	 * synced to disk.  We are now in a completely consistent state
5992	 * which doesn't have anything in the journal, and we know that
5993	 * no filesystem updates are running, so it is safe to modify
5994	 * the inode's in-core data-journaling state flag now.
5995	 */
5996
5997	if (val)
5998		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5999	else {
6000		err = jbd2_journal_flush(journal, 0);
6001		if (err < 0) {
6002			jbd2_journal_unlock_updates(journal);
6003			ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6004			return err;
6005		}
6006		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6007	}
6008	ext4_set_aops(inode);
6009
6010	jbd2_journal_unlock_updates(journal);
6011	ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6012
6013	if (val)
6014		filemap_invalidate_unlock(inode->i_mapping);
6015
6016	/* Finally we can mark the inode as dirty. */
6017
6018	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6019	if (IS_ERR(handle))
6020		return PTR_ERR(handle);
6021
6022	ext4_fc_mark_ineligible(inode->i_sb,
6023		EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6024	err = ext4_mark_inode_dirty(handle, inode);
6025	ext4_handle_sync(handle);
6026	ext4_journal_stop(handle);
6027	ext4_std_error(inode->i_sb, err);
6028
6029	return err;
6030}
6031
6032static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6033			    struct buffer_head *bh)
6034{
6035	return !buffer_mapped(bh);
6036}
6037
6038vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6039{
6040	struct vm_area_struct *vma = vmf->vma;
6041	struct folio *folio = page_folio(vmf->page);
6042	loff_t size;
6043	unsigned long len;
6044	int err;
6045	vm_fault_t ret;
6046	struct file *file = vma->vm_file;
6047	struct inode *inode = file_inode(file);
6048	struct address_space *mapping = inode->i_mapping;
6049	handle_t *handle;
6050	get_block_t *get_block;
6051	int retries = 0;
6052
6053	if (unlikely(IS_IMMUTABLE(inode)))
6054		return VM_FAULT_SIGBUS;
6055
6056	sb_start_pagefault(inode->i_sb);
6057	file_update_time(vma->vm_file);
6058
6059	filemap_invalidate_lock_shared(mapping);
6060
6061	err = ext4_convert_inline_data(inode);
6062	if (err)
6063		goto out_ret;
6064
6065	/*
6066	 * On data journalling we skip straight to the transaction handle:
6067	 * there's no delalloc; page truncated will be checked later; the
6068	 * early return w/ all buffers mapped (calculates size/len) can't
6069	 * be used; and there's no dioread_nolock, so only ext4_get_block.
6070	 */
6071	if (ext4_should_journal_data(inode))
6072		goto retry_alloc;
6073
6074	/* Delalloc case is easy... */
6075	if (test_opt(inode->i_sb, DELALLOC) &&
 
6076	    !ext4_nonda_switch(inode->i_sb)) {
6077		do {
6078			err = block_page_mkwrite(vma, vmf,
6079						   ext4_da_get_block_prep);
6080		} while (err == -ENOSPC &&
6081		       ext4_should_retry_alloc(inode->i_sb, &retries));
6082		goto out_ret;
6083	}
6084
6085	folio_lock(folio);
6086	size = i_size_read(inode);
6087	/* Page got truncated from under us? */
6088	if (folio->mapping != mapping || folio_pos(folio) > size) {
6089		folio_unlock(folio);
6090		ret = VM_FAULT_NOPAGE;
6091		goto out;
6092	}
6093
6094	len = folio_size(folio);
6095	if (folio_pos(folio) + len > size)
6096		len = size - folio_pos(folio);
 
6097	/*
6098	 * Return if we have all the buffers mapped. This avoids the need to do
6099	 * journal_start/journal_stop which can block and take a long time
6100	 *
6101	 * This cannot be done for data journalling, as we have to add the
6102	 * inode to the transaction's list to writeprotect pages on commit.
6103	 */
6104	if (folio_buffers(folio)) {
6105		if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6106					    0, len, NULL,
6107					    ext4_bh_unmapped)) {
6108			/* Wait so that we don't change page under IO */
6109			folio_wait_stable(folio);
6110			ret = VM_FAULT_LOCKED;
6111			goto out;
6112		}
6113	}
6114	folio_unlock(folio);
6115	/* OK, we need to fill the hole... */
6116	if (ext4_should_dioread_nolock(inode))
6117		get_block = ext4_get_block_unwritten;
6118	else
6119		get_block = ext4_get_block;
6120retry_alloc:
6121	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6122				    ext4_writepage_trans_blocks(inode));
6123	if (IS_ERR(handle)) {
6124		ret = VM_FAULT_SIGBUS;
6125		goto out;
6126	}
6127	/*
6128	 * Data journalling can't use block_page_mkwrite() because it
6129	 * will set_buffer_dirty() before do_journal_get_write_access()
6130	 * thus might hit warning messages for dirty metadata buffers.
6131	 */
6132	if (!ext4_should_journal_data(inode)) {
6133		err = block_page_mkwrite(vma, vmf, get_block);
6134	} else {
6135		folio_lock(folio);
6136		size = i_size_read(inode);
6137		/* Page got truncated from under us? */
6138		if (folio->mapping != mapping || folio_pos(folio) > size) {
6139			ret = VM_FAULT_NOPAGE;
6140			goto out_error;
6141		}
6142
6143		len = folio_size(folio);
6144		if (folio_pos(folio) + len > size)
6145			len = size - folio_pos(folio);
6146
6147		err = __block_write_begin(&folio->page, 0, len, ext4_get_block);
6148		if (!err) {
6149			ret = VM_FAULT_SIGBUS;
6150			if (ext4_journal_folio_buffers(handle, folio, len))
6151				goto out_error;
6152		} else {
6153			folio_unlock(folio);
6154		}
 
6155	}
6156	ext4_journal_stop(handle);
6157	if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6158		goto retry_alloc;
6159out_ret:
6160	ret = vmf_fs_error(err);
6161out:
6162	filemap_invalidate_unlock_shared(mapping);
6163	sb_end_pagefault(inode->i_sb);
6164	return ret;
6165out_error:
6166	folio_unlock(folio);
6167	ext4_journal_stop(handle);
6168	goto out;
6169}