Linux Audio

Check our new training course

Loading...
v3.1
 
   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/module.h>
  22#include <linux/fs.h>
 
  23#include <linux/time.h>
  24#include <linux/jbd2.h>
  25#include <linux/highuid.h>
  26#include <linux/pagemap.h>
 
  27#include <linux/quotaops.h>
  28#include <linux/string.h>
  29#include <linux/buffer_head.h>
  30#include <linux/writeback.h>
  31#include <linux/pagevec.h>
  32#include <linux/mpage.h>
  33#include <linux/namei.h>
  34#include <linux/uio.h>
  35#include <linux/bio.h>
  36#include <linux/workqueue.h>
  37#include <linux/kernel.h>
  38#include <linux/printk.h>
  39#include <linux/slab.h>
  40#include <linux/ratelimit.h>
 
 
  41
  42#include "ext4_jbd2.h"
  43#include "xattr.h"
  44#include "acl.h"
  45#include "ext4_extents.h"
  46#include "truncate.h"
  47
  48#include <trace/events/ext4.h>
  49
  50#define MPAGE_DA_EXTENT_TAIL 0x01
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  51
  52static inline int ext4_begin_ordered_truncate(struct inode *inode,
  53					      loff_t new_size)
  54{
  55	trace_ext4_begin_ordered_truncate(inode, new_size);
  56	/*
  57	 * If jinode is zero, then we never opened the file for
  58	 * writing, so there's no need to call
  59	 * jbd2_journal_begin_ordered_truncate() since there's no
  60	 * outstanding writes we need to flush.
  61	 */
  62	if (!EXT4_I(inode)->jinode)
  63		return 0;
  64	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
  65						   EXT4_I(inode)->jinode,
  66						   new_size);
  67}
  68
  69static void ext4_invalidatepage(struct page *page, unsigned long offset);
  70static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
  71				   struct buffer_head *bh_result, int create);
  72static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
  73static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
  74static int __ext4_journalled_writepage(struct page *page, unsigned int len);
  75static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
 
  76
  77/*
  78 * Test whether an inode is a fast symlink.
 
  79 */
  80static int ext4_inode_is_fast_symlink(struct inode *inode)
  81{
  82	int ea_blocks = EXT4_I(inode)->i_file_acl ?
  83		(inode->i_sb->s_blocksize >> 9) : 0;
  84
  85	return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
  86}
  87
  88/*
  89 * Restart the transaction associated with *handle.  This does a commit,
  90 * so before we call here everything must be consistently dirtied against
  91 * this transaction.
  92 */
  93int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
  94				 int nblocks)
  95{
  96	int ret;
 
 
  97
  98	/*
  99	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
 100	 * moment, get_block can be called only for blocks inside i_size since
 101	 * page cache has been already dropped and writes are blocked by
 102	 * i_mutex. So we can safely drop the i_data_sem here.
 103	 */
 104	BUG_ON(EXT4_JOURNAL(inode) == NULL);
 105	jbd_debug(2, "restarting handle %p\n", handle);
 106	up_write(&EXT4_I(inode)->i_data_sem);
 107	ret = ext4_journal_restart(handle, nblocks);
 108	down_write(&EXT4_I(inode)->i_data_sem);
 109	ext4_discard_preallocations(inode);
 110
 111	return ret;
 
 
 
 112}
 113
 114/*
 115 * Called at the last iput() if i_nlink is zero.
 116 */
 117void ext4_evict_inode(struct inode *inode)
 118{
 119	handle_t *handle;
 120	int err;
 
 
 
 
 
 
 
 
 121
 122	trace_ext4_evict_inode(inode);
 123
 124	ext4_ioend_wait(inode);
 125
 126	if (inode->i_nlink) {
 127		/*
 128		 * When journalling data dirty buffers are tracked only in the
 129		 * journal. So although mm thinks everything is clean and
 130		 * ready for reaping the inode might still have some pages to
 131		 * write in the running transaction or waiting to be
 132		 * checkpointed. Thus calling jbd2_journal_invalidatepage()
 133		 * (via truncate_inode_pages()) to discard these buffers can
 134		 * cause data loss. Also even if we did not discard these
 135		 * buffers, we would have no way to find them after the inode
 136		 * is reaped and thus user could see stale data if he tries to
 137		 * read them before the transaction is checkpointed. So be
 138		 * careful and force everything to disk here... We use
 139		 * ei->i_datasync_tid to store the newest transaction
 140		 * containing inode's data.
 141		 *
 142		 * Note that directories do not have this problem because they
 143		 * don't use page cache.
 144		 */
 145		if (ext4_should_journal_data(inode) &&
 146		    (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
 
 147			journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 148			tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
 149
 150			jbd2_log_start_commit(journal, commit_tid);
 151			jbd2_log_wait_commit(journal, commit_tid);
 152			filemap_write_and_wait(&inode->i_data);
 153		}
 154		truncate_inode_pages(&inode->i_data, 0);
 
 155		goto no_delete;
 156	}
 157
 158	if (!is_bad_inode(inode))
 159		dquot_initialize(inode);
 
 160
 161	if (ext4_should_order_data(inode))
 162		ext4_begin_ordered_truncate(inode, 0);
 163	truncate_inode_pages(&inode->i_data, 0);
 164
 165	if (is_bad_inode(inode))
 166		goto no_delete;
 
 
 
 
 
 
 
 167
 168	handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 169	if (IS_ERR(handle)) {
 170		ext4_std_error(inode->i_sb, PTR_ERR(handle));
 171		/*
 172		 * If we're going to skip the normal cleanup, we still need to
 173		 * make sure that the in-core orphan linked list is properly
 174		 * cleaned up.
 175		 */
 176		ext4_orphan_del(NULL, inode);
 
 
 177		goto no_delete;
 178	}
 179
 180	if (IS_SYNC(inode))
 181		ext4_handle_sync(handle);
 
 
 
 
 
 
 
 
 
 
 182	inode->i_size = 0;
 183	err = ext4_mark_inode_dirty(handle, inode);
 184	if (err) {
 185		ext4_warning(inode->i_sb,
 186			     "couldn't mark inode dirty (err %d)", err);
 187		goto stop_handle;
 188	}
 189	if (inode->i_blocks)
 190		ext4_truncate(inode);
 191
 192	/*
 193	 * ext4_ext_truncate() doesn't reserve any slop when it
 194	 * restarts journal transactions; therefore there may not be
 195	 * enough credits left in the handle to remove the inode from
 196	 * the orphan list and set the dtime field.
 197	 */
 198	if (!ext4_handle_has_enough_credits(handle, 3)) {
 199		err = ext4_journal_extend(handle, 3);
 200		if (err > 0)
 201			err = ext4_journal_restart(handle, 3);
 202		if (err != 0) {
 203			ext4_warning(inode->i_sb,
 204				     "couldn't extend journal (err %d)", err);
 205		stop_handle:
 206			ext4_journal_stop(handle);
 207			ext4_orphan_del(NULL, inode);
 208			goto no_delete;
 209		}
 210	}
 211
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 212	/*
 213	 * Kill off the orphan record which ext4_truncate created.
 214	 * AKPM: I think this can be inside the above `if'.
 215	 * Note that ext4_orphan_del() has to be able to cope with the
 216	 * deletion of a non-existent orphan - this is because we don't
 217	 * know if ext4_truncate() actually created an orphan record.
 218	 * (Well, we could do this if we need to, but heck - it works)
 219	 */
 220	ext4_orphan_del(handle, inode);
 221	EXT4_I(inode)->i_dtime	= get_seconds();
 222
 223	/*
 224	 * One subtle ordering requirement: if anything has gone wrong
 225	 * (transaction abort, IO errors, whatever), then we can still
 226	 * do these next steps (the fs will already have been marked as
 227	 * having errors), but we can't free the inode if the mark_dirty
 228	 * fails.
 229	 */
 230	if (ext4_mark_inode_dirty(handle, inode))
 231		/* If that failed, just do the required in-core inode clear. */
 232		ext4_clear_inode(inode);
 233	else
 234		ext4_free_inode(handle, inode);
 235	ext4_journal_stop(handle);
 
 
 
 236	return;
 237no_delete:
 
 
 
 
 
 
 
 
 238	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
 239}
 240
 241#ifdef CONFIG_QUOTA
 242qsize_t *ext4_get_reserved_space(struct inode *inode)
 243{
 244	return &EXT4_I(inode)->i_reserved_quota;
 245}
 246#endif
 247
 248/*
 249 * Calculate the number of metadata blocks need to reserve
 250 * to allocate a block located at @lblock
 251 */
 252static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 253{
 254	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 255		return ext4_ext_calc_metadata_amount(inode, lblock);
 256
 257	return ext4_ind_calc_metadata_amount(inode, lblock);
 258}
 259
 260/*
 261 * Called with i_data_sem down, which is important since we can call
 262 * ext4_discard_preallocations() from here.
 263 */
 264void ext4_da_update_reserve_space(struct inode *inode,
 265					int used, int quota_claim)
 266{
 267	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 268	struct ext4_inode_info *ei = EXT4_I(inode);
 269
 270	spin_lock(&ei->i_block_reservation_lock);
 271	trace_ext4_da_update_reserve_space(inode, used);
 272	if (unlikely(used > ei->i_reserved_data_blocks)) {
 273		ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
 274			 "with only %d reserved data blocks\n",
 275			 __func__, inode->i_ino, used,
 276			 ei->i_reserved_data_blocks);
 277		WARN_ON(1);
 278		used = ei->i_reserved_data_blocks;
 279	}
 280
 281	/* Update per-inode reservations */
 282	ei->i_reserved_data_blocks -= used;
 283	ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
 284	percpu_counter_sub(&sbi->s_dirtyblocks_counter,
 285			   used + ei->i_allocated_meta_blocks);
 286	ei->i_allocated_meta_blocks = 0;
 287
 288	if (ei->i_reserved_data_blocks == 0) {
 289		/*
 290		 * We can release all of the reserved metadata blocks
 291		 * only when we have written all of the delayed
 292		 * allocation blocks.
 293		 */
 294		percpu_counter_sub(&sbi->s_dirtyblocks_counter,
 295				   ei->i_reserved_meta_blocks);
 296		ei->i_reserved_meta_blocks = 0;
 297		ei->i_da_metadata_calc_len = 0;
 298	}
 299	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 300
 301	/* Update quota subsystem for data blocks */
 302	if (quota_claim)
 303		dquot_claim_block(inode, used);
 304	else {
 305		/*
 306		 * We did fallocate with an offset that is already delayed
 307		 * allocated. So on delayed allocated writeback we should
 308		 * not re-claim the quota for fallocated blocks.
 309		 */
 310		dquot_release_reservation_block(inode, used);
 311	}
 312
 313	/*
 314	 * If we have done all the pending block allocations and if
 315	 * there aren't any writers on the inode, we can discard the
 316	 * inode's preallocations.
 317	 */
 318	if ((ei->i_reserved_data_blocks == 0) &&
 319	    (atomic_read(&inode->i_writecount) == 0))
 320		ext4_discard_preallocations(inode);
 321}
 322
 323static int __check_block_validity(struct inode *inode, const char *func,
 324				unsigned int line,
 325				struct ext4_map_blocks *map)
 326{
 327	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
 328				   map->m_len)) {
 
 
 
 329		ext4_error_inode(inode, func, line, map->m_pblk,
 330				 "lblock %lu mapped to illegal pblock "
 331				 "(length %d)", (unsigned long) map->m_lblk,
 332				 map->m_len);
 333		return -EIO;
 334	}
 335	return 0;
 336}
 337
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 338#define check_block_validity(inode, map)	\
 339	__check_block_validity((inode), __func__, __LINE__, (map))
 340
 341/*
 342 * Return the number of contiguous dirty pages in a given inode
 343 * starting at page frame idx.
 344 */
 345static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
 346				    unsigned int max_pages)
 347{
 348	struct address_space *mapping = inode->i_mapping;
 349	pgoff_t	index;
 350	struct pagevec pvec;
 351	pgoff_t num = 0;
 352	int i, nr_pages, done = 0;
 353
 354	if (max_pages == 0)
 355		return 0;
 356	pagevec_init(&pvec, 0);
 357	while (!done) {
 358		index = idx;
 359		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 360					      PAGECACHE_TAG_DIRTY,
 361					      (pgoff_t)PAGEVEC_SIZE);
 362		if (nr_pages == 0)
 363			break;
 364		for (i = 0; i < nr_pages; i++) {
 365			struct page *page = pvec.pages[i];
 366			struct buffer_head *bh, *head;
 
 
 367
 368			lock_page(page);
 369			if (unlikely(page->mapping != mapping) ||
 370			    !PageDirty(page) ||
 371			    PageWriteback(page) ||
 372			    page->index != idx) {
 373				done = 1;
 374				unlock_page(page);
 375				break;
 376			}
 377			if (page_has_buffers(page)) {
 378				bh = head = page_buffers(page);
 379				do {
 380					if (!buffer_delay(bh) &&
 381					    !buffer_unwritten(bh))
 382						done = 1;
 383					bh = bh->b_this_page;
 384				} while (!done && (bh != head));
 385			}
 386			unlock_page(page);
 387			if (done)
 388				break;
 389			idx++;
 390			num++;
 391			if (num >= max_pages) {
 392				done = 1;
 393				break;
 394			}
 395		}
 396		pagevec_release(&pvec);
 397	}
 398	return num;
 399}
 
 400
 401/*
 402 * The ext4_map_blocks() function tries to look up the requested blocks,
 403 * and returns if the blocks are already mapped.
 404 *
 405 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 406 * and store the allocated blocks in the result buffer head and mark it
 407 * mapped.
 408 *
 409 * If file type is extents based, it will call ext4_ext_map_blocks(),
 410 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 411 * based files
 412 *
 413 * On success, it returns the number of blocks being mapped or allocate.
 414 * if create==0 and the blocks are pre-allocated and uninitialized block,
 415 * the result buffer head is unmapped. If the create ==1, it will make sure
 416 * the buffer head is mapped.
 417 *
 418 * It returns 0 if plain look up failed (blocks have not been allocated), in
 419 * that casem, buffer head is unmapped
 
 420 *
 421 * It returns the error in case of allocation failure.
 422 */
 423int ext4_map_blocks(handle_t *handle, struct inode *inode,
 424		    struct ext4_map_blocks *map, int flags)
 425{
 
 426	int retval;
 
 
 
 
 
 
 427
 428	map->m_flags = 0;
 429	ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
 430		  "logical block %lu\n", inode->i_ino, flags, map->m_len,
 431		  (unsigned long) map->m_lblk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 432	/*
 433	 * Try to see if we can get the block without requesting a new
 434	 * file system block.
 435	 */
 436	down_read((&EXT4_I(inode)->i_data_sem));
 437	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 438		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 439	} else {
 440		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 441	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 442	up_read((&EXT4_I(inode)->i_data_sem));
 443
 
 444	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 445		int ret = check_block_validity(inode, map);
 446		if (ret != 0)
 447			return ret;
 448	}
 449
 450	/* If it is only a block(s) look up */
 451	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 452		return retval;
 453
 454	/*
 455	 * Returns if the blocks have already allocated
 456	 *
 457	 * Note that if blocks have been preallocated
 458	 * ext4_ext_get_block() returns th create = 0
 459	 * with buffer head unmapped.
 460	 */
 461	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 462		return retval;
 
 
 
 
 
 
 463
 464	/*
 465	 * When we call get_blocks without the create flag, the
 466	 * BH_Unwritten flag could have gotten set if the blocks
 467	 * requested were part of a uninitialized extent.  We need to
 468	 * clear this flag now that we are committed to convert all or
 469	 * part of the uninitialized extent to be an initialized
 470	 * extent.  This is because we need to avoid the combination
 471	 * of BH_Unwritten and BH_Mapped flags being simultaneously
 472	 * set on the buffer_head.
 473	 */
 474	map->m_flags &= ~EXT4_MAP_UNWRITTEN;
 475
 476	/*
 477	 * New blocks allocate and/or writing to uninitialized extent
 478	 * will possibly result in updating i_data, so we take
 479	 * the write lock of i_data_sem, and call get_blocks()
 480	 * with create == 1 flag.
 481	 */
 482	down_write((&EXT4_I(inode)->i_data_sem));
 483
 484	/*
 485	 * if the caller is from delayed allocation writeout path
 486	 * we have already reserved fs blocks for allocation
 487	 * let the underlying get_block() function know to
 488	 * avoid double accounting
 489	 */
 490	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 491		ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
 492	/*
 493	 * We need to check for EXT4 here because migrate
 494	 * could have changed the inode type in between
 495	 */
 496	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 497		retval = ext4_ext_map_blocks(handle, inode, map, flags);
 498	} else {
 499		retval = ext4_ind_map_blocks(handle, inode, map, flags);
 500
 501		if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
 502			/*
 503			 * We allocated new blocks which will result in
 504			 * i_data's format changing.  Force the migrate
 505			 * to fail by clearing migrate flags
 506			 */
 507			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 508		}
 509
 510		/*
 511		 * Update reserved blocks/metadata blocks after successful
 512		 * block allocation which had been deferred till now. We don't
 513		 * support fallocate for non extent files. So we can update
 514		 * reserve space here.
 515		 */
 516		if ((retval > 0) &&
 517			(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
 518			ext4_da_update_reserve_space(inode, retval, 1);
 519	}
 520	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 521		ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
 522
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 523	up_write((&EXT4_I(inode)->i_data_sem));
 524	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 525		int ret = check_block_validity(inode, map);
 526		if (ret != 0)
 527			return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 528	}
 
 
 
 
 
 
 529	return retval;
 530}
 531
 532/* Maximum number of blocks we map for direct IO at once. */
 533#define DIO_MAX_BLOCKS 4096
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 534
 535static int _ext4_get_block(struct inode *inode, sector_t iblock,
 536			   struct buffer_head *bh, int flags)
 537{
 538	handle_t *handle = ext4_journal_current_handle();
 539	struct ext4_map_blocks map;
 540	int ret = 0, started = 0;
 541	int dio_credits;
 
 
 542
 543	map.m_lblk = iblock;
 544	map.m_len = bh->b_size >> inode->i_blkbits;
 545
 546	if (flags && !handle) {
 547		/* Direct IO write... */
 548		if (map.m_len > DIO_MAX_BLOCKS)
 549			map.m_len = DIO_MAX_BLOCKS;
 550		dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
 551		handle = ext4_journal_start(inode, dio_credits);
 552		if (IS_ERR(handle)) {
 553			ret = PTR_ERR(handle);
 554			return ret;
 555		}
 556		started = 1;
 557	}
 558
 559	ret = ext4_map_blocks(handle, inode, &map, flags);
 560	if (ret > 0) {
 561		map_bh(bh, inode->i_sb, map.m_pblk);
 562		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
 563		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 564		ret = 0;
 
 
 
 565	}
 566	if (started)
 567		ext4_journal_stop(handle);
 568	return ret;
 569}
 570
 571int ext4_get_block(struct inode *inode, sector_t iblock,
 572		   struct buffer_head *bh, int create)
 573{
 574	return _ext4_get_block(inode, iblock, bh,
 575			       create ? EXT4_GET_BLOCKS_CREATE : 0);
 576}
 577
 578/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 579 * `handle' can be NULL if create is zero
 580 */
 581struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 582				ext4_lblk_t block, int create, int *errp)
 583{
 584	struct ext4_map_blocks map;
 585	struct buffer_head *bh;
 586	int fatal = 0, err;
 
 
 587
 588	J_ASSERT(handle != NULL || create == 0);
 
 
 589
 590	map.m_lblk = block;
 591	map.m_len = 1;
 592	err = ext4_map_blocks(handle, inode, &map,
 593			      create ? EXT4_GET_BLOCKS_CREATE : 0);
 594
 
 
 595	if (err < 0)
 596		*errp = err;
 597	if (err <= 0)
 598		return NULL;
 599	*errp = 0;
 600
 601	bh = sb_getblk(inode->i_sb, map.m_pblk);
 602	if (!bh) {
 603		*errp = -EIO;
 604		return NULL;
 605	}
 606	if (map.m_flags & EXT4_MAP_NEW) {
 607		J_ASSERT(create != 0);
 608		J_ASSERT(handle != NULL);
 
 609
 610		/*
 611		 * Now that we do not always journal data, we should
 612		 * keep in mind whether this should always journal the
 613		 * new buffer as metadata.  For now, regular file
 614		 * writes use ext4_get_block instead, so it's not a
 615		 * problem.
 616		 */
 617		lock_buffer(bh);
 618		BUFFER_TRACE(bh, "call get_create_access");
 619		fatal = ext4_journal_get_create_access(handle, bh);
 620		if (!fatal && !buffer_uptodate(bh)) {
 
 
 
 
 
 621			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 622			set_buffer_uptodate(bh);
 623		}
 624		unlock_buffer(bh);
 625		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 626		err = ext4_handle_dirty_metadata(handle, inode, bh);
 627		if (!fatal)
 628			fatal = err;
 629	} else {
 630		BUFFER_TRACE(bh, "not a new buffer");
 631	}
 632	if (fatal) {
 633		*errp = fatal;
 634		brelse(bh);
 635		bh = NULL;
 636	}
 637	return bh;
 
 
 
 638}
 639
 640struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 641			       ext4_lblk_t block, int create, int *err)
 642{
 643	struct buffer_head *bh;
 
 644
 645	bh = ext4_getblk(handle, inode, block, create, err);
 646	if (!bh)
 647		return bh;
 648	if (buffer_uptodate(bh))
 649		return bh;
 650	ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
 651	wait_on_buffer(bh);
 652	if (buffer_uptodate(bh))
 653		return bh;
 654	put_bh(bh);
 655	*err = -EIO;
 656	return NULL;
 657}
 658
 659static int walk_page_buffers(handle_t *handle,
 660			     struct buffer_head *head,
 661			     unsigned from,
 662			     unsigned to,
 663			     int *partial,
 664			     int (*fn)(handle_t *handle,
 665				       struct buffer_head *bh))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 666{
 667	struct buffer_head *bh;
 668	unsigned block_start, block_end;
 669	unsigned blocksize = head->b_size;
 670	int err, ret = 0;
 671	struct buffer_head *next;
 672
 673	for (bh = head, block_start = 0;
 674	     ret == 0 && (bh != head || !block_start);
 675	     block_start = block_end, bh = next) {
 676		next = bh->b_this_page;
 677		block_end = block_start + blocksize;
 678		if (block_end <= from || block_start >= to) {
 679			if (partial && !buffer_uptodate(bh))
 680				*partial = 1;
 681			continue;
 682		}
 683		err = (*fn)(handle, bh);
 684		if (!ret)
 685			ret = err;
 686	}
 687	return ret;
 688}
 689
 690/*
 691 * To preserve ordering, it is essential that the hole instantiation and
 692 * the data write be encapsulated in a single transaction.  We cannot
 693 * close off a transaction and start a new one between the ext4_get_block()
 694 * and the commit_write().  So doing the jbd2_journal_start at the start of
 695 * prepare_write() is the right place.
 696 *
 697 * Also, this function can nest inside ext4_writepage() ->
 698 * block_write_full_page(). In that case, we *know* that ext4_writepage()
 699 * has generated enough buffer credits to do the whole page.  So we won't
 700 * block on the journal in that case, which is good, because the caller may
 701 * be PF_MEMALLOC.
 702 *
 703 * By accident, ext4 can be reentered when a transaction is open via
 704 * quota file writes.  If we were to commit the transaction while thus
 705 * reentered, there can be a deadlock - we would be holding a quota
 706 * lock, and the commit would never complete if another thread had a
 707 * transaction open and was blocking on the quota lock - a ranking
 708 * violation.
 709 *
 710 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
 711 * will _not_ run commit under these circumstances because handle->h_ref
 712 * is elevated.  We'll still have enough credits for the tiny quotafile
 713 * write.
 714 */
 715static int do_journal_get_write_access(handle_t *handle,
 716				       struct buffer_head *bh)
 717{
 718	int dirty = buffer_dirty(bh);
 719	int ret;
 720
 721	if (!buffer_mapped(bh) || buffer_freed(bh))
 722		return 0;
 723	/*
 724	 * __block_write_begin() could have dirtied some buffers. Clean
 725	 * the dirty bit as jbd2_journal_get_write_access() could complain
 726	 * otherwise about fs integrity issues. Setting of the dirty bit
 727	 * by __block_write_begin() isn't a real problem here as we clear
 728	 * the bit before releasing a page lock and thus writeback cannot
 729	 * ever write the buffer.
 730	 */
 731	if (dirty)
 732		clear_buffer_dirty(bh);
 733	ret = ext4_journal_get_write_access(handle, bh);
 
 
 734	if (!ret && dirty)
 735		ret = ext4_handle_dirty_metadata(handle, NULL, bh);
 736	return ret;
 737}
 738
 739static int ext4_get_block_write(struct inode *inode, sector_t iblock,
 740		   struct buffer_head *bh_result, int create);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 741static int ext4_write_begin(struct file *file, struct address_space *mapping,
 742			    loff_t pos, unsigned len, unsigned flags,
 743			    struct page **pagep, void **fsdata)
 744{
 745	struct inode *inode = mapping->host;
 746	int ret, needed_blocks;
 747	handle_t *handle;
 748	int retries = 0;
 749	struct page *page;
 750	pgoff_t index;
 751	unsigned from, to;
 752
 753	trace_ext4_write_begin(inode, pos, len, flags);
 
 
 
 754	/*
 755	 * Reserve one block more for addition to orphan list in case
 756	 * we allocate blocks but write fails for some reason
 757	 */
 758	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
 759	index = pos >> PAGE_CACHE_SHIFT;
 760	from = pos & (PAGE_CACHE_SIZE - 1);
 761	to = from + len;
 762
 763retry:
 764	handle = ext4_journal_start(inode, needed_blocks);
 765	if (IS_ERR(handle)) {
 766		ret = PTR_ERR(handle);
 767		goto out;
 
 
 768	}
 769
 770	/* We cannot recurse into the filesystem as the transaction is already
 771	 * started */
 772	flags |= AOP_FLAG_NOFS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 773
 774	page = grab_cache_page_write_begin(mapping, index, flags);
 775	if (!page) {
 
 
 
 
 
 
 
 
 
 
 
 
 776		ext4_journal_stop(handle);
 777		ret = -ENOMEM;
 778		goto out;
 779	}
 780	*pagep = page;
 
 781
 
 782	if (ext4_should_dioread_nolock(inode))
 783		ret = __block_write_begin(page, pos, len, ext4_get_block_write);
 
 
 
 
 
 
 
 
 784	else
 785		ret = __block_write_begin(page, pos, len, ext4_get_block);
 786
 787	if (!ret && ext4_should_journal_data(inode)) {
 788		ret = walk_page_buffers(handle, page_buffers(page),
 789				from, to, NULL, do_journal_get_write_access);
 
 790	}
 791
 792	if (ret) {
 
 
 
 793		unlock_page(page);
 794		page_cache_release(page);
 795		/*
 796		 * __block_write_begin may have instantiated a few blocks
 797		 * outside i_size.  Trim these off again. Don't need
 798		 * i_size_read because we hold i_mutex.
 799		 *
 800		 * Add inode to orphan list in case we crash before
 801		 * truncate finishes
 802		 */
 803		if (pos + len > inode->i_size && ext4_can_truncate(inode))
 804			ext4_orphan_add(handle, inode);
 805
 806		ext4_journal_stop(handle);
 807		if (pos + len > inode->i_size) {
 808			ext4_truncate_failed_write(inode);
 809			/*
 810			 * If truncate failed early the inode might
 811			 * still be on the orphan list; we need to
 812			 * make sure the inode is removed from the
 813			 * orphan list in that case.
 814			 */
 815			if (inode->i_nlink)
 816				ext4_orphan_del(NULL, inode);
 817		}
 818	}
 819
 820	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 821		goto retry;
 822out:
 
 
 
 
 823	return ret;
 824}
 825
 826/* For write_end() in data=journal mode */
 827static int write_end_fn(handle_t *handle, struct buffer_head *bh)
 
 828{
 
 829	if (!buffer_mapped(bh) || buffer_freed(bh))
 830		return 0;
 831	set_buffer_uptodate(bh);
 832	return ext4_handle_dirty_metadata(handle, NULL, bh);
 
 
 
 833}
 834
 835static int ext4_generic_write_end(struct file *file,
 836				  struct address_space *mapping,
 837				  loff_t pos, unsigned len, unsigned copied,
 838				  struct page *page, void *fsdata)
 
 
 
 
 
 
 
 839{
 840	int i_size_changed = 0;
 841	struct inode *inode = mapping->host;
 842	handle_t *handle = ext4_journal_current_handle();
 
 
 
 
 
 843
 844	copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 845
 
 
 
 
 
 846	/*
 847	 * No need to use i_size_read() here, the i_size
 848	 * cannot change under us because we hold i_mutex.
 849	 *
 850	 * But it's important to update i_size while still holding page lock:
 851	 * page writeout could otherwise come in and zero beyond i_size.
 
 
 
 852	 */
 853	if (pos + copied > inode->i_size) {
 854		i_size_write(inode, pos + copied);
 855		i_size_changed = 1;
 856	}
 857
 858	if (pos + copied >  EXT4_I(inode)->i_disksize) {
 859		/* We need to mark inode dirty even if
 860		 * new_i_size is less that inode->i_size
 861		 * bu greater than i_disksize.(hint delalloc)
 862		 */
 863		ext4_update_i_disksize(inode, (pos + copied));
 864		i_size_changed = 1;
 865	}
 866	unlock_page(page);
 867	page_cache_release(page);
 868
 
 
 869	/*
 870	 * Don't mark the inode dirty under page lock. First, it unnecessarily
 871	 * makes the holding time of page lock longer. Second, it forces lock
 872	 * ordering of page lock and transaction start for journaling
 873	 * filesystems.
 874	 */
 875	if (i_size_changed)
 876		ext4_mark_inode_dirty(handle, inode);
 877
 878	return copied;
 879}
 880
 881/*
 882 * We need to pick up the new inode size which generic_commit_write gave us
 883 * `file' can be NULL - eg, when called from page_symlink().
 884 *
 885 * ext4 never places buffers on inode->i_mapping->private_list.  metadata
 886 * buffers are managed internally.
 887 */
 888static int ext4_ordered_write_end(struct file *file,
 889				  struct address_space *mapping,
 890				  loff_t pos, unsigned len, unsigned copied,
 891				  struct page *page, void *fsdata)
 892{
 893	handle_t *handle = ext4_journal_current_handle();
 894	struct inode *inode = mapping->host;
 895	int ret = 0, ret2;
 896
 897	trace_ext4_ordered_write_end(inode, pos, len, copied);
 898	ret = ext4_jbd2_file_inode(handle, inode);
 899
 900	if (ret == 0) {
 901		ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
 902							page, fsdata);
 903		copied = ret2;
 904		if (pos + len > inode->i_size && ext4_can_truncate(inode))
 905			/* if we have allocated more blocks and copied
 906			 * less. We will have blocks allocated outside
 907			 * inode->i_size. So truncate them
 908			 */
 909			ext4_orphan_add(handle, inode);
 910		if (ret2 < 0)
 911			ret = ret2;
 912	}
 913	ret2 = ext4_journal_stop(handle);
 914	if (!ret)
 915		ret = ret2;
 916
 917	if (pos + len > inode->i_size) {
 918		ext4_truncate_failed_write(inode);
 919		/*
 920		 * If truncate failed early the inode might still be
 921		 * on the orphan list; we need to make sure the inode
 922		 * is removed from the orphan list in that case.
 923		 */
 924		if (inode->i_nlink)
 925			ext4_orphan_del(NULL, inode);
 926	}
 927
 928
 929	return ret ? ret : copied;
 930}
 931
 932static int ext4_writeback_write_end(struct file *file,
 933				    struct address_space *mapping,
 934				    loff_t pos, unsigned len, unsigned copied,
 935				    struct page *page, void *fsdata)
 
 
 
 
 
 936{
 937	handle_t *handle = ext4_journal_current_handle();
 938	struct inode *inode = mapping->host;
 939	int ret = 0, ret2;
 940
 941	trace_ext4_writeback_write_end(inode, pos, len, copied);
 942	ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
 943							page, fsdata);
 944	copied = ret2;
 945	if (pos + len > inode->i_size && ext4_can_truncate(inode))
 946		/* if we have allocated more blocks and copied
 947		 * less. We will have blocks allocated outside
 948		 * inode->i_size. So truncate them
 949		 */
 950		ext4_orphan_add(handle, inode);
 951
 952	if (ret2 < 0)
 953		ret = ret2;
 954
 955	ret2 = ext4_journal_stop(handle);
 956	if (!ret)
 957		ret = ret2;
 
 
 
 
 958
 959	if (pos + len > inode->i_size) {
 960		ext4_truncate_failed_write(inode);
 961		/*
 962		 * If truncate failed early the inode might still be
 963		 * on the orphan list; we need to make sure the inode
 964		 * is removed from the orphan list in that case.
 965		 */
 966		if (inode->i_nlink)
 967			ext4_orphan_del(NULL, inode);
 968	}
 969
 970	return ret ? ret : copied;
 
 
 
 
 
 
 
 
 971}
 972
 973static int ext4_journalled_write_end(struct file *file,
 974				     struct address_space *mapping,
 975				     loff_t pos, unsigned len, unsigned copied,
 976				     struct page *page, void *fsdata)
 977{
 978	handle_t *handle = ext4_journal_current_handle();
 979	struct inode *inode = mapping->host;
 
 980	int ret = 0, ret2;
 981	int partial = 0;
 982	unsigned from, to;
 983	loff_t new_i_size;
 
 984
 985	trace_ext4_journalled_write_end(inode, pos, len, copied);
 986	from = pos & (PAGE_CACHE_SIZE - 1);
 987	to = from + len;
 988
 989	BUG_ON(!ext4_handle_valid(handle));
 990
 991	if (copied < len) {
 992		if (!PageUptodate(page))
 993			copied = 0;
 994		page_zero_new_buffers(page, from+copied, to);
 995	}
 996
 997	ret = walk_page_buffers(handle, page_buffers(page), from,
 998				to, &partial, write_end_fn);
 999	if (!partial)
1000		SetPageUptodate(page);
1001	new_i_size = pos + copied;
1002	if (new_i_size > inode->i_size)
1003		i_size_write(inode, pos+copied);
 
 
 
 
 
 
 
 
1004	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1005	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1006	if (new_i_size > EXT4_I(inode)->i_disksize) {
1007		ext4_update_i_disksize(inode, new_i_size);
 
 
 
 
 
1008		ret2 = ext4_mark_inode_dirty(handle, inode);
1009		if (!ret)
1010			ret = ret2;
1011	}
1012
1013	unlock_page(page);
1014	page_cache_release(page);
1015	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1016		/* if we have allocated more blocks and copied
1017		 * less. We will have blocks allocated outside
1018		 * inode->i_size. So truncate them
1019		 */
1020		ext4_orphan_add(handle, inode);
1021
1022	ret2 = ext4_journal_stop(handle);
1023	if (!ret)
1024		ret = ret2;
1025	if (pos + len > inode->i_size) {
1026		ext4_truncate_failed_write(inode);
1027		/*
1028		 * If truncate failed early the inode might still be
1029		 * on the orphan list; we need to make sure the inode
1030		 * is removed from the orphan list in that case.
1031		 */
1032		if (inode->i_nlink)
1033			ext4_orphan_del(NULL, inode);
1034	}
1035
1036	return ret ? ret : copied;
1037}
1038
1039/*
1040 * Reserve a single block located at lblock
1041 */
1042static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1043{
1044	int retries = 0;
1045	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1046	struct ext4_inode_info *ei = EXT4_I(inode);
1047	unsigned long md_needed;
1048	int ret;
1049
1050	/*
1051	 * recalculate the amount of metadata blocks to reserve
1052	 * in order to allocate nrblocks
1053	 * worse case is one extent per block
1054	 */
1055repeat:
1056	spin_lock(&ei->i_block_reservation_lock);
1057	md_needed = ext4_calc_metadata_amount(inode, lblock);
1058	trace_ext4_da_reserve_space(inode, md_needed);
1059	spin_unlock(&ei->i_block_reservation_lock);
1060
1061	/*
1062	 * We will charge metadata quota at writeout time; this saves
1063	 * us from metadata over-estimation, though we may go over by
1064	 * a small amount in the end.  Here we just reserve for data.
1065	 */
1066	ret = dquot_reserve_block(inode, 1);
1067	if (ret)
1068		return ret;
1069	/*
1070	 * We do still charge estimated metadata to the sb though;
1071	 * we cannot afford to run out of free blocks.
1072	 */
1073	if (ext4_claim_free_blocks(sbi, md_needed + 1, 0)) {
1074		dquot_release_reservation_block(inode, 1);
1075		if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1076			yield();
1077			goto repeat;
1078		}
1079		return -ENOSPC;
1080	}
1081	spin_lock(&ei->i_block_reservation_lock);
1082	ei->i_reserved_data_blocks++;
1083	ei->i_reserved_meta_blocks += md_needed;
1084	spin_unlock(&ei->i_block_reservation_lock);
1085
1086	return 0;       /* success */
1087}
1088
1089static void ext4_da_release_space(struct inode *inode, int to_free)
1090{
1091	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1092	struct ext4_inode_info *ei = EXT4_I(inode);
1093
1094	if (!to_free)
1095		return;		/* Nothing to release, exit */
1096
1097	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1098
1099	trace_ext4_da_release_space(inode, to_free);
1100	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1101		/*
1102		 * if there aren't enough reserved blocks, then the
1103		 * counter is messed up somewhere.  Since this
1104		 * function is called from invalidate page, it's
1105		 * harmless to return without any action.
1106		 */
1107		ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1108			 "ino %lu, to_free %d with only %d reserved "
1109			 "data blocks\n", inode->i_ino, to_free,
1110			 ei->i_reserved_data_blocks);
1111		WARN_ON(1);
1112		to_free = ei->i_reserved_data_blocks;
1113	}
1114	ei->i_reserved_data_blocks -= to_free;
1115
1116	if (ei->i_reserved_data_blocks == 0) {
1117		/*
1118		 * We can release all of the reserved metadata blocks
1119		 * only when we have written all of the delayed
1120		 * allocation blocks.
1121		 */
1122		percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1123				   ei->i_reserved_meta_blocks);
1124		ei->i_reserved_meta_blocks = 0;
1125		ei->i_da_metadata_calc_len = 0;
1126	}
1127
1128	/* update fs dirty data blocks counter */
1129	percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1130
1131	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1132
1133	dquot_release_reservation_block(inode, to_free);
1134}
1135
1136static void ext4_da_page_release_reservation(struct page *page,
1137					     unsigned long offset)
1138{
1139	int to_release = 0;
1140	struct buffer_head *head, *bh;
1141	unsigned int curr_off = 0;
1142
1143	head = page_buffers(page);
1144	bh = head;
1145	do {
1146		unsigned int next_off = curr_off + bh->b_size;
1147
1148		if ((offset <= curr_off) && (buffer_delay(bh))) {
1149			to_release++;
1150			clear_buffer_delay(bh);
1151		}
1152		curr_off = next_off;
1153	} while ((bh = bh->b_this_page) != head);
1154	ext4_da_release_space(page->mapping->host, to_release);
1155}
1156
1157/*
1158 * Delayed allocation stuff
1159 */
1160
1161/*
1162 * mpage_da_submit_io - walks through extent of pages and try to write
1163 * them with writepage() call back
1164 *
1165 * @mpd->inode: inode
1166 * @mpd->first_page: first page of the extent
1167 * @mpd->next_page: page after the last page of the extent
1168 *
1169 * By the time mpage_da_submit_io() is called we expect all blocks
1170 * to be allocated. this may be wrong if allocation failed.
1171 *
1172 * As pages are already locked by write_cache_pages(), we can't use it
1173 */
1174static int mpage_da_submit_io(struct mpage_da_data *mpd,
1175			      struct ext4_map_blocks *map)
1176{
1177	struct pagevec pvec;
1178	unsigned long index, end;
1179	int ret = 0, err, nr_pages, i;
1180	struct inode *inode = mpd->inode;
1181	struct address_space *mapping = inode->i_mapping;
1182	loff_t size = i_size_read(inode);
1183	unsigned int len, block_start;
1184	struct buffer_head *bh, *page_bufs = NULL;
1185	int journal_data = ext4_should_journal_data(inode);
1186	sector_t pblock = 0, cur_logical = 0;
1187	struct ext4_io_submit io_submit;
1188
1189	BUG_ON(mpd->next_page <= mpd->first_page);
1190	memset(&io_submit, 0, sizeof(io_submit));
1191	/*
1192	 * We need to start from the first_page to the next_page - 1
1193	 * to make sure we also write the mapped dirty buffer_heads.
1194	 * If we look at mpd->b_blocknr we would only be looking
1195	 * at the currently mapped buffer_heads.
 
1196	 */
1197	index = mpd->first_page;
1198	end = mpd->next_page - 1;
1199
1200	pagevec_init(&pvec, 0);
1201	while (index <= end) {
1202		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1203		if (nr_pages == 0)
1204			break;
1205		for (i = 0; i < nr_pages; i++) {
1206			int commit_write = 0, skip_page = 0;
1207			struct page *page = pvec.pages[i];
1208
1209			index = page->index;
1210			if (index > end)
1211				break;
1212
1213			if (index == size >> PAGE_CACHE_SHIFT)
1214				len = size & ~PAGE_CACHE_MASK;
1215			else
1216				len = PAGE_CACHE_SIZE;
1217			if (map) {
1218				cur_logical = index << (PAGE_CACHE_SHIFT -
1219							inode->i_blkbits);
1220				pblock = map->m_pblk + (cur_logical -
1221							map->m_lblk);
1222			}
1223			index++;
1224
1225			BUG_ON(!PageLocked(page));
1226			BUG_ON(PageWriteback(page));
1227
1228			/*
1229			 * If the page does not have buffers (for
1230			 * whatever reason), try to create them using
1231			 * __block_write_begin.  If this fails,
1232			 * skip the page and move on.
1233			 */
1234			if (!page_has_buffers(page)) {
1235				if (__block_write_begin(page, 0, len,
1236						noalloc_get_block_write)) {
1237				skip_page:
1238					unlock_page(page);
1239					continue;
1240				}
1241				commit_write = 1;
1242			}
1243
1244			bh = page_bufs = page_buffers(page);
1245			block_start = 0;
1246			do {
1247				if (!bh)
1248					goto skip_page;
1249				if (map && (cur_logical >= map->m_lblk) &&
1250				    (cur_logical <= (map->m_lblk +
1251						     (map->m_len - 1)))) {
1252					if (buffer_delay(bh)) {
1253						clear_buffer_delay(bh);
1254						bh->b_blocknr = pblock;
1255					}
1256					if (buffer_unwritten(bh) ||
1257					    buffer_mapped(bh))
1258						BUG_ON(bh->b_blocknr != pblock);
1259					if (map->m_flags & EXT4_MAP_UNINIT)
1260						set_buffer_uninit(bh);
1261					clear_buffer_unwritten(bh);
1262				}
1263
1264				/* skip page if block allocation undone */
1265				if (buffer_delay(bh) || buffer_unwritten(bh))
1266					skip_page = 1;
1267				bh = bh->b_this_page;
1268				block_start += bh->b_size;
1269				cur_logical++;
1270				pblock++;
1271			} while (bh != page_bufs);
1272
1273			if (skip_page)
1274				goto skip_page;
1275
1276			if (commit_write)
1277				/* mark the buffer_heads as dirty & uptodate */
1278				block_commit_write(page, 0, len);
1279
1280			clear_page_dirty_for_io(page);
1281			/*
1282			 * Delalloc doesn't support data journalling,
1283			 * but eventually maybe we'll lift this
1284			 * restriction.
1285			 */
1286			if (unlikely(journal_data && PageChecked(page)))
1287				err = __ext4_journalled_writepage(page, len);
1288			else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1289				err = ext4_bio_write_page(&io_submit, page,
1290							  len, mpd->wbc);
1291			else if (buffer_uninit(page_bufs)) {
1292				ext4_set_bh_endio(page_bufs, inode);
1293				err = block_write_full_page_endio(page,
1294					noalloc_get_block_write,
1295					mpd->wbc, ext4_end_io_buffer_write);
1296			} else
1297				err = block_write_full_page(page,
1298					noalloc_get_block_write, mpd->wbc);
1299
1300			if (!err)
1301				mpd->pages_written++;
1302			/*
1303			 * In error case, we have to continue because
1304			 * remaining pages are still locked
1305			 */
1306			if (ret == 0)
1307				ret = err;
1308		}
1309		pagevec_release(&pvec);
1310	}
1311	ext4_io_submit(&io_submit);
1312	return ret;
1313}
1314
1315static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
 
1316{
1317	int nr_pages, i;
1318	pgoff_t index, end;
1319	struct pagevec pvec;
1320	struct inode *inode = mpd->inode;
1321	struct address_space *mapping = inode->i_mapping;
1322
 
 
 
 
 
1323	index = mpd->first_page;
1324	end   = mpd->next_page - 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1325	while (index <= end) {
1326		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1327		if (nr_pages == 0)
1328			break;
1329		for (i = 0; i < nr_pages; i++) {
1330			struct page *page = pvec.pages[i];
1331			if (page->index > end)
1332				break;
1333			BUG_ON(!PageLocked(page));
1334			BUG_ON(PageWriteback(page));
1335			block_invalidatepage(page, 0);
1336			ClearPageUptodate(page);
1337			unlock_page(page);
 
 
 
 
 
 
 
 
1338		}
1339		index = pvec.pages[nr_pages - 1]->index + 1;
1340		pagevec_release(&pvec);
1341	}
1342	return;
1343}
1344
1345static void ext4_print_free_blocks(struct inode *inode)
1346{
1347	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1348	printk(KERN_CRIT "Total free blocks count %lld\n",
1349	       ext4_count_free_blocks(inode->i_sb));
1350	printk(KERN_CRIT "Free/Dirty block details\n");
1351	printk(KERN_CRIT "free_blocks=%lld\n",
1352	       (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
1353	printk(KERN_CRIT "dirty_blocks=%lld\n",
1354	       (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1355	printk(KERN_CRIT "Block reservation details\n");
1356	printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1357	       EXT4_I(inode)->i_reserved_data_blocks);
1358	printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1359	       EXT4_I(inode)->i_reserved_meta_blocks);
 
 
 
 
1360	return;
1361}
1362
 
 
 
 
 
 
1363/*
1364 * mpage_da_map_and_submit - go through given space, map them
1365 *       if necessary, and then submit them for I/O
 
 
1366 *
1367 * @mpd - bh describing space
1368 *
1369 * The function skips space we know is already mapped to disk blocks.
1370 *
 
1371 */
1372static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1373{
1374	int err, blks, get_blocks_flags;
1375	struct ext4_map_blocks map, *mapp = NULL;
1376	sector_t next = mpd->b_blocknr;
1377	unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1378	loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1379	handle_t *handle = NULL;
1380
1381	/*
1382	 * If the blocks are mapped already, or we couldn't accumulate
1383	 * any blocks, then proceed immediately to the submission stage.
1384	 */
1385	if ((mpd->b_size == 0) ||
1386	    ((mpd->b_state  & (1 << BH_Mapped)) &&
1387	     !(mpd->b_state & (1 << BH_Delay)) &&
1388	     !(mpd->b_state & (1 << BH_Unwritten))))
1389		goto submit_io;
1390
1391	handle = ext4_journal_current_handle();
1392	BUG_ON(!handle);
1393
1394	/*
1395	 * Call ext4_map_blocks() to allocate any delayed allocation
1396	 * blocks, or to convert an uninitialized extent to be
1397	 * initialized (in the case where we have written into
1398	 * one or more preallocated blocks).
1399	 *
1400	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1401	 * indicate that we are on the delayed allocation path.  This
1402	 * affects functions in many different parts of the allocation
1403	 * call path.  This flag exists primarily because we don't
1404	 * want to change *many* call functions, so ext4_map_blocks()
1405	 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1406	 * inode's allocation semaphore is taken.
1407	 *
1408	 * If the blocks in questions were delalloc blocks, set
1409	 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1410	 * variables are updated after the blocks have been allocated.
1411	 */
1412	map.m_lblk = next;
1413	map.m_len = max_blocks;
1414	get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1415	if (ext4_should_dioread_nolock(mpd->inode))
1416		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1417	if (mpd->b_state & (1 << BH_Delay))
1418		get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1419
1420	blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1421	if (blks < 0) {
1422		struct super_block *sb = mpd->inode->i_sb;
1423
1424		err = blks;
1425		/*
1426		 * If get block returns EAGAIN or ENOSPC and there
1427		 * appears to be free blocks we will just let
1428		 * mpage_da_submit_io() unlock all of the pages.
1429		 */
1430		if (err == -EAGAIN)
1431			goto submit_io;
1432
1433		if (err == -ENOSPC &&
1434		    ext4_count_free_blocks(sb)) {
1435			mpd->retval = err;
1436			goto submit_io;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1437		}
1438
1439		/*
1440		 * get block failure will cause us to loop in
1441		 * writepages, because a_ops->writepage won't be able
1442		 * to make progress. The page will be redirtied by
1443		 * writepage and writepages will again try to write
1444		 * the same.
1445		 */
1446		if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1447			ext4_msg(sb, KERN_CRIT,
1448				 "delayed block allocation failed for inode %lu "
1449				 "at logical offset %llu with max blocks %zd "
1450				 "with error %d", mpd->inode->i_ino,
1451				 (unsigned long long) next,
1452				 mpd->b_size >> mpd->inode->i_blkbits, err);
1453			ext4_msg(sb, KERN_CRIT,
1454				"This should not happen!! Data will be lost\n");
1455			if (err == -ENOSPC)
1456				ext4_print_free_blocks(mpd->inode);
1457		}
1458		/* invalidate all the pages */
1459		ext4_da_block_invalidatepages(mpd);
1460
1461		/* Mark this page range as having been completed */
1462		mpd->io_done = 1;
1463		return;
1464	}
1465	BUG_ON(blks == 0);
1466
1467	mapp = &map;
1468	if (map.m_flags & EXT4_MAP_NEW) {
1469		struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1470		int i;
1471
1472		for (i = 0; i < map.m_len; i++)
1473			unmap_underlying_metadata(bdev, map.m_pblk + i);
1474	}
 
 
 
1475
1476	if (ext4_should_order_data(mpd->inode)) {
1477		err = ext4_jbd2_file_inode(handle, mpd->inode);
1478		if (err)
1479			/* This only happens if the journal is aborted */
1480			return;
1481	}
1482
1483	/*
1484	 * Update on-disk size along with block allocation.
 
1485	 */
1486	disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1487	if (disksize > i_size_read(mpd->inode))
1488		disksize = i_size_read(mpd->inode);
1489	if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1490		ext4_update_i_disksize(mpd->inode, disksize);
1491		err = ext4_mark_inode_dirty(handle, mpd->inode);
1492		if (err)
1493			ext4_error(mpd->inode->i_sb,
1494				   "Failed to mark inode %lu dirty",
1495				   mpd->inode->i_ino);
1496	}
1497
1498submit_io:
1499	mpage_da_submit_io(mpd, mapp);
1500	mpd->io_done = 1;
1501}
1502
1503#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1504		(1 << BH_Delay) | (1 << BH_Unwritten))
 
 
1505
1506/*
1507 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1508 *
1509 * @mpd->lbh - extent of blocks
1510 * @logical - logical number of the block in the file
1511 * @bh - bh of the block (used to access block's state)
1512 *
1513 * the function is used to collect contig. blocks in same state
1514 */
1515static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1516				   sector_t logical, size_t b_size,
1517				   unsigned long b_state)
1518{
1519	sector_t next;
1520	int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1521
1522	/*
1523	 * XXX Don't go larger than mballoc is willing to allocate
1524	 * This is a stopgap solution.  We eventually need to fold
1525	 * mpage_da_submit_io() into this function and then call
1526	 * ext4_map_blocks() multiple times in a loop
1527	 */
1528	if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1529		goto flush_it;
1530
1531	/* check if thereserved journal credits might overflow */
1532	if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1533		if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1534			/*
1535			 * With non-extent format we are limited by the journal
1536			 * credit available.  Total credit needed to insert
1537			 * nrblocks contiguous blocks is dependent on the
1538			 * nrblocks.  So limit nrblocks.
1539			 */
1540			goto flush_it;
1541		} else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1542				EXT4_MAX_TRANS_DATA) {
1543			/*
1544			 * Adding the new buffer_head would make it cross the
1545			 * allowed limit for which we have journal credit
1546			 * reserved. So limit the new bh->b_size
1547			 */
1548			b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1549						mpd->inode->i_blkbits;
1550			/* we will do mpage_da_submit_io in the next loop */
1551		}
1552	}
1553	/*
1554	 * First block in the extent
1555	 */
1556	if (mpd->b_size == 0) {
1557		mpd->b_blocknr = logical;
1558		mpd->b_size = b_size;
1559		mpd->b_state = b_state & BH_FLAGS;
1560		return;
1561	}
1562
1563	next = mpd->b_blocknr + nrblocks;
1564	/*
1565	 * Can we merge the block to our big extent?
1566	 */
1567	if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1568		mpd->b_size += b_size;
1569		return;
1570	}
1571
1572flush_it:
1573	/*
1574	 * We couldn't merge the block to our extent, so we
1575	 * need to flush current  extent and start new one
1576	 */
1577	mpage_da_map_and_submit(mpd);
1578	return;
1579}
1580
1581static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1582{
1583	return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1584}
1585
1586/*
1587 * This is a special get_blocks_t callback which is used by
1588 * ext4_da_write_begin().  It will either return mapped block or
1589 * reserve space for a single block.
1590 *
1591 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1592 * We also have b_blocknr = -1 and b_bdev initialized properly
1593 *
1594 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1595 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1596 * initialized properly.
1597 */
1598static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1599				  struct buffer_head *bh, int create)
1600{
1601	struct ext4_map_blocks map;
1602	int ret = 0;
1603	sector_t invalid_block = ~((sector_t) 0xffff);
1604
1605	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1606		invalid_block = ~0;
1607
1608	BUG_ON(create == 0);
1609	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1610
1611	map.m_lblk = iblock;
1612	map.m_len = 1;
1613
1614	/*
1615	 * first, we need to know whether the block is allocated already
1616	 * preallocated blocks are unmapped but should treated
1617	 * the same as allocated blocks.
1618	 */
1619	ret = ext4_map_blocks(NULL, inode, &map, 0);
1620	if (ret < 0)
1621		return ret;
1622	if (ret == 0) {
1623		if (buffer_delay(bh))
1624			return 0; /* Not sure this could or should happen */
1625		/*
1626		 * XXX: __block_write_begin() unmaps passed block, is it OK?
1627		 */
1628		ret = ext4_da_reserve_space(inode, iblock);
1629		if (ret)
1630			/* not enough space to reserve */
1631			return ret;
1632
1633		map_bh(bh, inode->i_sb, invalid_block);
1634		set_buffer_new(bh);
1635		set_buffer_delay(bh);
1636		return 0;
1637	}
1638
1639	map_bh(bh, inode->i_sb, map.m_pblk);
1640	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1641
1642	if (buffer_unwritten(bh)) {
1643		/* A delayed write to unwritten bh should be marked
1644		 * new and mapped.  Mapped ensures that we don't do
1645		 * get_block multiple times when we write to the same
1646		 * offset and new ensures that we do proper zero out
1647		 * for partial write.
1648		 */
1649		set_buffer_new(bh);
1650		set_buffer_mapped(bh);
1651	}
1652	return 0;
1653}
1654
1655/*
1656 * This function is used as a standard get_block_t calback function
1657 * when there is no desire to allocate any blocks.  It is used as a
1658 * callback function for block_write_begin() and block_write_full_page().
1659 * These functions should only try to map a single block at a time.
1660 *
1661 * Since this function doesn't do block allocations even if the caller
1662 * requests it by passing in create=1, it is critically important that
1663 * any caller checks to make sure that any buffer heads are returned
1664 * by this function are either all already mapped or marked for
1665 * delayed allocation before calling  block_write_full_page().  Otherwise,
1666 * b_blocknr could be left unitialized, and the page write functions will
1667 * be taken by surprise.
1668 */
1669static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1670				   struct buffer_head *bh_result, int create)
1671{
1672	BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1673	return _ext4_get_block(inode, iblock, bh_result, 0);
1674}
1675
1676static int bget_one(handle_t *handle, struct buffer_head *bh)
1677{
1678	get_bh(bh);
1679	return 0;
1680}
1681
1682static int bput_one(handle_t *handle, struct buffer_head *bh)
1683{
1684	put_bh(bh);
1685	return 0;
1686}
1687
1688static int __ext4_journalled_writepage(struct page *page,
1689				       unsigned int len)
1690{
1691	struct address_space *mapping = page->mapping;
1692	struct inode *inode = mapping->host;
1693	struct buffer_head *page_bufs;
1694	handle_t *handle = NULL;
1695	int ret = 0;
1696	int err;
 
 
1697
1698	ClearPageChecked(page);
1699	page_bufs = page_buffers(page);
1700	BUG_ON(!page_bufs);
1701	walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1702	/* As soon as we unlock the page, it can go away, but we have
1703	 * references to buffers so we are safe */
 
 
 
 
 
 
 
 
 
1704	unlock_page(page);
1705
1706	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
 
1707	if (IS_ERR(handle)) {
1708		ret = PTR_ERR(handle);
 
 
 
 
 
 
 
 
 
 
 
 
1709		goto out;
1710	}
1711
1712	BUG_ON(!ext4_handle_valid(handle));
 
 
 
 
 
 
 
 
1713
1714	ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1715				do_journal_get_write_access);
1716
1717	err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1718				write_end_fn);
 
 
 
 
1719	if (ret == 0)
1720		ret = err;
1721	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1722	err = ext4_journal_stop(handle);
1723	if (!ret)
1724		ret = err;
1725
1726	walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1727	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1728out:
 
 
 
1729	return ret;
1730}
1731
1732static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1733static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1734
1735/*
1736 * Note that we don't need to start a transaction unless we're journaling data
1737 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1738 * need to file the inode to the transaction's list in ordered mode because if
1739 * we are writing back data added by write(), the inode is already there and if
1740 * we are writing back data modified via mmap(), no one guarantees in which
1741 * transaction the data will hit the disk. In case we are journaling data, we
1742 * cannot start transaction directly because transaction start ranks above page
1743 * lock so we have to do some magic.
1744 *
1745 * This function can get called via...
1746 *   - ext4_da_writepages after taking page lock (have journal handle)
1747 *   - journal_submit_inode_data_buffers (no journal handle)
1748 *   - shrink_page_list via pdflush (no journal handle)
1749 *   - grab_page_cache when doing write_begin (have journal handle)
1750 *
1751 * We don't do any block allocation in this function. If we have page with
1752 * multiple blocks we need to write those buffer_heads that are mapped. This
1753 * is important for mmaped based write. So if we do with blocksize 1K
1754 * truncate(f, 1024);
1755 * a = mmap(f, 0, 4096);
1756 * a[0] = 'a';
1757 * truncate(f, 4096);
1758 * we have in the page first buffer_head mapped via page_mkwrite call back
1759 * but other bufer_heads would be unmapped but dirty(dirty done via the
1760 * do_wp_page). So writepage should write the first block. If we modify
1761 * the mmap area beyond 1024 we will again get a page_fault and the
1762 * page_mkwrite callback will do the block allocation and mark the
1763 * buffer_heads mapped.
1764 *
1765 * We redirty the page if we have any buffer_heads that is either delay or
1766 * unwritten in the page.
1767 *
1768 * We can get recursively called as show below.
1769 *
1770 *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1771 *		ext4_writepage()
1772 *
1773 * But since we don't do any block allocation we should not deadlock.
1774 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1775 */
1776static int ext4_writepage(struct page *page,
1777			  struct writeback_control *wbc)
1778{
1779	int ret = 0, commit_write = 0;
 
1780	loff_t size;
1781	unsigned int len;
1782	struct buffer_head *page_bufs = NULL;
1783	struct inode *inode = page->mapping->host;
 
 
 
 
 
 
 
1784
1785	trace_ext4_writepage(page);
1786	size = i_size_read(inode);
1787	if (page->index == size >> PAGE_CACHE_SHIFT)
1788		len = size & ~PAGE_CACHE_MASK;
 
1789	else
1790		len = PAGE_CACHE_SIZE;
1791
1792	/*
1793	 * If the page does not have buffers (for whatever reason),
1794	 * try to create them using __block_write_begin.  If this
1795	 * fails, redirty the page and move on.
1796	 */
1797	if (!page_has_buffers(page)) {
1798		if (__block_write_begin(page, 0, len,
1799					noalloc_get_block_write)) {
1800		redirty_page:
1801			redirty_page_for_writepage(wbc, page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1802			unlock_page(page);
1803			return 0;
1804		}
1805		commit_write = 1;
1806	}
1807	page_bufs = page_buffers(page);
1808	if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1809			      ext4_bh_delay_or_unwritten)) {
1810		/*
1811		 * We don't want to do block allocation, so redirty
1812		 * the page and return.  We may reach here when we do
1813		 * a journal commit via journal_submit_inode_data_buffers.
1814		 * We can also reach here via shrink_page_list
1815		 */
1816		goto redirty_page;
1817	}
1818	if (commit_write)
1819		/* now mark the buffer_heads as dirty and uptodate */
1820		block_commit_write(page, 0, len);
1821
1822	if (PageChecked(page) && ext4_should_journal_data(inode))
1823		/*
1824		 * It's mmapped pagecache.  Add buffers and journal it.  There
1825		 * doesn't seem much point in redirtying the page here.
1826		 */
1827		return __ext4_journalled_writepage(page, len);
1828
1829	if (buffer_uninit(page_bufs)) {
1830		ext4_set_bh_endio(page_bufs, inode);
1831		ret = block_write_full_page_endio(page, noalloc_get_block_write,
1832					    wbc, ext4_end_io_buffer_write);
1833	} else
1834		ret = block_write_full_page(page, noalloc_get_block_write,
1835					    wbc);
1836
 
 
 
1837	return ret;
1838}
1839
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1840/*
1841 * This is called via ext4_da_writepages() to
1842 * calculate the total number of credits to reserve to fit
1843 * a single extent allocation into a single transaction,
1844 * ext4_da_writpeages() will loop calling this before
1845 * the block allocation.
 
 
 
 
 
 
 
1846 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1847
 
 
 
 
 
 
 
1848static int ext4_da_writepages_trans_blocks(struct inode *inode)
1849{
1850	int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
 
 
 
 
 
 
 
 
 
1851
1852	/*
1853	 * With non-extent format the journal credit needed to
1854	 * insert nrblocks contiguous block is dependent on
1855	 * number of contiguous block. So we will limit
1856	 * number of contiguous block to a sane value
1857	 */
1858	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
1859	    (max_blocks > EXT4_MAX_TRANS_DATA))
1860		max_blocks = EXT4_MAX_TRANS_DATA;
1861
1862	return ext4_chunk_trans_blocks(inode, max_blocks);
1863}
1864
1865/*
1866 * write_cache_pages_da - walk the list of dirty pages of the given
1867 * address space and accumulate pages that need writing, and call
1868 * mpage_da_map_and_submit to map a single contiguous memory region
1869 * and then write them.
1870 */
1871static int write_cache_pages_da(struct address_space *mapping,
1872				struct writeback_control *wbc,
1873				struct mpage_da_data *mpd,
1874				pgoff_t *done_index)
1875{
1876	struct buffer_head	*bh, *head;
1877	struct inode		*inode = mapping->host;
1878	struct pagevec		pvec;
1879	unsigned int		nr_pages;
1880	sector_t		logical;
1881	pgoff_t			index, end;
1882	long			nr_to_write = wbc->nr_to_write;
1883	int			i, tag, ret = 0;
1884
1885	memset(mpd, 0, sizeof(struct mpage_da_data));
1886	mpd->wbc = wbc;
1887	mpd->inode = inode;
1888	pagevec_init(&pvec, 0);
1889	index = wbc->range_start >> PAGE_CACHE_SHIFT;
1890	end = wbc->range_end >> PAGE_CACHE_SHIFT;
 
 
 
 
 
 
 
1891
1892	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1893		tag = PAGECACHE_TAG_TOWRITE;
1894	else
1895		tag = PAGECACHE_TAG_DIRTY;
1896
1897	*done_index = index;
 
 
1898	while (index <= end) {
1899		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1900			      min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1901		if (nr_pages == 0)
1902			return 0;
1903
1904		for (i = 0; i < nr_pages; i++) {
1905			struct page *page = pvec.pages[i];
1906
1907			/*
1908			 * At this point, the page may be truncated or
1909			 * invalidated (changing page->mapping to NULL), or
1910			 * even swizzled back from swapper_space to tmpfs file
1911			 * mapping. However, page->index will not change
1912			 * because we have a reference on the page.
 
1913			 */
1914			if (page->index > end)
1915				goto out;
1916
1917			*done_index = page->index + 1;
1918
1919			/*
1920			 * If we can't merge this page, and we have
1921			 * accumulated an contiguous region, write it
1922			 */
1923			if ((mpd->next_page != page->index) &&
1924			    (mpd->next_page != mpd->first_page)) {
1925				mpage_da_map_and_submit(mpd);
1926				goto ret_extent_tail;
1927			}
1928
1929			lock_page(page);
1930
1931			/*
1932			 * If the page is no longer dirty, or its
1933			 * mapping no longer corresponds to inode we
1934			 * are writing (which means it has been
1935			 * truncated or invalidated), or the page is
1936			 * already under writeback and we are not
1937			 * doing a data integrity writeback, skip the page
1938			 */
1939			if (!PageDirty(page) ||
1940			    (PageWriteback(page) &&
1941			     (wbc->sync_mode == WB_SYNC_NONE)) ||
1942			    unlikely(page->mapping != mapping)) {
1943				unlock_page(page);
1944				continue;
1945			}
1946
1947			wait_on_page_writeback(page);
1948			BUG_ON(PageWriteback(page));
1949
1950			if (mpd->next_page != page->index)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1951				mpd->first_page = page->index;
1952			mpd->next_page = page->index + 1;
1953			logical = (sector_t) page->index <<
1954				(PAGE_CACHE_SHIFT - inode->i_blkbits);
1955
1956			if (!page_has_buffers(page)) {
1957				mpage_add_bh_to_extent(mpd, logical,
1958						       PAGE_CACHE_SIZE,
1959						       (1 << BH_Dirty) | (1 << BH_Uptodate));
1960				if (mpd->io_done)
1961					goto ret_extent_tail;
 
 
 
 
1962			} else {
1963				/*
1964				 * Page with regular buffer heads,
1965				 * just add all dirty ones
1966				 */
1967				head = page_buffers(page);
1968				bh = head;
1969				do {
1970					BUG_ON(buffer_locked(bh));
1971					/*
1972					 * We need to try to allocate
1973					 * unmapped blocks in the same page.
1974					 * Otherwise we won't make progress
1975					 * with the page in ext4_writepage
1976					 */
1977					if (ext4_bh_delay_or_unwritten(NULL, bh)) {
1978						mpage_add_bh_to_extent(mpd, logical,
1979								       bh->b_size,
1980								       bh->b_state);
1981						if (mpd->io_done)
1982							goto ret_extent_tail;
1983					} else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
1984						/*
1985						 * mapped dirty buffer. We need
1986						 * to update the b_state
1987						 * because we look at b_state
1988						 * in mpage_da_map_blocks.  We
1989						 * don't update b_size because
1990						 * if we find an unmapped
1991						 * buffer_head later we need to
1992						 * use the b_state flag of that
1993						 * buffer_head.
1994						 */
1995						if (mpd->b_size == 0)
1996							mpd->b_state = bh->b_state & BH_FLAGS;
1997					}
1998					logical++;
1999				} while ((bh = bh->b_this_page) != head);
2000			}
2001
2002			if (nr_to_write > 0) {
2003				nr_to_write--;
2004				if (nr_to_write == 0 &&
2005				    wbc->sync_mode == WB_SYNC_NONE)
2006					/*
2007					 * We stop writing back only if we are
2008					 * not doing integrity sync. In case of
2009					 * integrity sync we have to keep going
2010					 * because someone may be concurrently
2011					 * dirtying pages, and we might have
2012					 * synced a lot of newly appeared dirty
2013					 * pages, but have not synced all of the
2014					 * old dirty pages.
2015					 */
2016					goto out;
 
2017			}
 
2018		}
2019		pagevec_release(&pvec);
2020		cond_resched();
2021	}
 
2022	return 0;
2023ret_extent_tail:
2024	ret = MPAGE_DA_EXTENT_TAIL;
2025out:
2026	pagevec_release(&pvec);
2027	cond_resched();
2028	return ret;
2029}
2030
 
 
 
 
 
2031
2032static int ext4_da_writepages(struct address_space *mapping,
2033			      struct writeback_control *wbc)
2034{
2035	pgoff_t	index;
 
 
2036	int range_whole = 0;
 
2037	handle_t *handle = NULL;
2038	struct mpage_da_data mpd;
2039	struct inode *inode = mapping->host;
2040	int pages_written = 0;
2041	unsigned int max_pages;
2042	int range_cyclic, cycled = 1, io_done = 0;
2043	int needed_blocks, ret = 0;
2044	long desired_nr_to_write, nr_to_writebump = 0;
2045	loff_t range_start = wbc->range_start;
2046	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2047	pgoff_t done_index = 0;
2048	pgoff_t end;
2049
2050	trace_ext4_da_writepages(inode, wbc);
2051
2052	/*
2053	 * No pages to write? This is mainly a kludge to avoid starting
2054	 * a transaction for special inodes like journal inode on last iput()
2055	 * because that could violate lock ordering on umount
2056	 */
2057	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2058		return 0;
 
 
 
 
 
 
 
2059
2060	/*
2061	 * If the filesystem has aborted, it is read-only, so return
2062	 * right away instead of dumping stack traces later on that
2063	 * will obscure the real source of the problem.  We test
2064	 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2065	 * the latter could be true if the filesystem is mounted
2066	 * read-only, and in that case, ext4_da_writepages should
2067	 * *never* be called, so if that ever happens, we would want
2068	 * the stack trace.
2069	 */
2070	if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2071		return -EROFS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2072
2073	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2074		range_whole = 1;
2075
2076	range_cyclic = wbc->range_cyclic;
2077	if (wbc->range_cyclic) {
2078		index = mapping->writeback_index;
2079		if (index)
2080			cycled = 0;
2081		wbc->range_start = index << PAGE_CACHE_SHIFT;
2082		wbc->range_end  = LLONG_MAX;
2083		wbc->range_cyclic = 0;
2084		end = -1;
2085	} else {
2086		index = wbc->range_start >> PAGE_CACHE_SHIFT;
2087		end = wbc->range_end >> PAGE_CACHE_SHIFT;
2088	}
2089
2090	/*
2091	 * This works around two forms of stupidity.  The first is in
2092	 * the writeback code, which caps the maximum number of pages
2093	 * written to be 1024 pages.  This is wrong on multiple
2094	 * levels; different architectues have a different page size,
2095	 * which changes the maximum amount of data which gets
2096	 * written.  Secondly, 4 megabytes is way too small.  XFS
2097	 * forces this value to be 16 megabytes by multiplying
2098	 * nr_to_write parameter by four, and then relies on its
2099	 * allocator to allocate larger extents to make them
2100	 * contiguous.  Unfortunately this brings us to the second
2101	 * stupidity, which is that ext4's mballoc code only allocates
2102	 * at most 2048 blocks.  So we force contiguous writes up to
2103	 * the number of dirty blocks in the inode, or
2104	 * sbi->max_writeback_mb_bump whichever is smaller.
2105	 */
2106	max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2107	if (!range_cyclic && range_whole) {
2108		if (wbc->nr_to_write == LONG_MAX)
2109			desired_nr_to_write = wbc->nr_to_write;
2110		else
2111			desired_nr_to_write = wbc->nr_to_write * 8;
2112	} else
2113		desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2114							   max_pages);
2115	if (desired_nr_to_write > max_pages)
2116		desired_nr_to_write = max_pages;
2117
2118	if (wbc->nr_to_write < desired_nr_to_write) {
2119		nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2120		wbc->nr_to_write = desired_nr_to_write;
2121	}
2122
 
2123retry:
2124	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2125		tag_pages_for_writeback(mapping, index, end);
 
 
2126
2127	while (!ret && wbc->nr_to_write > 0) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2128
 
2129		/*
2130		 * we  insert one extent at a time. So we need
2131		 * credit needed for single extent allocation.
2132		 * journalled mode is currently not supported
2133		 * by delalloc
 
2134		 */
2135		BUG_ON(ext4_should_journal_data(inode));
2136		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2137
2138		/* start a new transaction*/
2139		handle = ext4_journal_start(inode, needed_blocks);
 
2140		if (IS_ERR(handle)) {
2141			ret = PTR_ERR(handle);
2142			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2143			       "%ld pages, ino %lu; err %d", __func__,
2144				wbc->nr_to_write, inode->i_ino, ret);
2145			goto out_writepages;
 
 
 
2146		}
 
2147
 
 
 
 
 
2148		/*
2149		 * Now call write_cache_pages_da() to find the next
2150		 * contiguous region of logical blocks that need
2151		 * blocks to be allocated by ext4 and submit them.
 
 
 
 
 
2152		 */
2153		ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2154		/*
2155		 * If we have a contiguous extent of pages and we
2156		 * haven't done the I/O yet, map the blocks and submit
2157		 * them for I/O.
2158		 */
2159		if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2160			mpage_da_map_and_submit(&mpd);
2161			ret = MPAGE_DA_EXTENT_TAIL;
2162		}
2163		trace_ext4_da_write_pages(inode, &mpd);
2164		wbc->nr_to_write -= mpd.pages_written;
 
 
2165
2166		ext4_journal_stop(handle);
 
 
 
 
 
 
 
 
 
 
 
 
2167
2168		if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2169			/* commit the transaction which would
 
2170			 * free blocks released in the transaction
2171			 * and try again
2172			 */
2173			jbd2_journal_force_commit_nested(sbi->s_journal);
2174			ret = 0;
2175		} else if (ret == MPAGE_DA_EXTENT_TAIL) {
2176			/*
2177			 * got one extent now try with
2178			 * rest of the pages
2179			 */
2180			pages_written += mpd.pages_written;
2181			ret = 0;
2182			io_done = 1;
2183		} else if (wbc->nr_to_write)
2184			/*
2185			 * There is no more writeout needed
2186			 * or we requested for a noblocking writeout
2187			 * and we found the device congested
2188			 */
2189			break;
2190	}
2191	if (!io_done && !cycled) {
 
 
2192		cycled = 1;
2193		index = 0;
2194		wbc->range_start = index << PAGE_CACHE_SHIFT;
2195		wbc->range_end  = mapping->writeback_index - 1;
2196		goto retry;
2197	}
2198
2199	/* Update index */
2200	wbc->range_cyclic = range_cyclic;
2201	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2202		/*
2203		 * set the writeback_index so that range_cyclic
2204		 * mode will write it back later
2205		 */
2206		mapping->writeback_index = done_index;
2207
2208out_writepages:
2209	wbc->nr_to_write -= nr_to_writebump;
2210	wbc->range_start = range_start;
2211	trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2212	return ret;
2213}
2214
2215#define FALL_BACK_TO_NONDELALLOC 1
2216static int ext4_nonda_switch(struct super_block *sb)
2217{
2218	s64 free_blocks, dirty_blocks;
2219	struct ext4_sb_info *sbi = EXT4_SB(sb);
2220
2221	/*
2222	 * switch to non delalloc mode if we are running low
2223	 * on free block. The free block accounting via percpu
2224	 * counters can get slightly wrong with percpu_counter_batch getting
2225	 * accumulated on each CPU without updating global counters
2226	 * Delalloc need an accurate free block accounting. So switch
2227	 * to non delalloc when we are near to error range.
2228	 */
2229	free_blocks  = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2230	dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2231	if (2 * free_blocks < 3 * dirty_blocks ||
2232		free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
 
 
 
 
 
 
 
 
2233		/*
2234		 * free block count is less than 150% of dirty blocks
2235		 * or free blocks is less than watermark
2236		 */
2237		return 1;
2238	}
2239	/*
2240	 * Even if we don't switch but are nearing capacity,
2241	 * start pushing delalloc when 1/2 of free blocks are dirty.
2242	 */
2243	if (free_blocks < 2 * dirty_blocks)
2244		writeback_inodes_sb_if_idle(sb);
2245
2246	return 0;
2247}
2248
2249static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2250			       loff_t pos, unsigned len, unsigned flags,
2251			       struct page **pagep, void **fsdata)
2252{
2253	int ret, retries = 0;
2254	struct page *page;
2255	pgoff_t index;
2256	struct inode *inode = mapping->host;
2257	handle_t *handle;
2258
2259	index = pos >> PAGE_CACHE_SHIFT;
 
 
 
2260
2261	if (ext4_nonda_switch(inode->i_sb)) {
2262		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2263		return ext4_write_begin(file, mapping, pos,
2264					len, flags, pagep, fsdata);
2265	}
2266	*fsdata = (void *)0;
2267	trace_ext4_da_write_begin(inode, pos, len, flags);
2268retry:
2269	/*
2270	 * With delayed allocation, we don't log the i_disksize update
2271	 * if there is delayed block allocation. But we still need
2272	 * to journalling the i_disksize update if writes to the end
2273	 * of file which has an already mapped buffer.
2274	 */
2275	handle = ext4_journal_start(inode, 1);
2276	if (IS_ERR(handle)) {
2277		ret = PTR_ERR(handle);
2278		goto out;
2279	}
2280	/* We cannot recurse into the filesystem as the transaction is already
2281	 * started */
2282	flags |= AOP_FLAG_NOFS;
2283
2284	page = grab_cache_page_write_begin(mapping, index, flags);
2285	if (!page) {
2286		ext4_journal_stop(handle);
2287		ret = -ENOMEM;
2288		goto out;
 
 
2289	}
2290	*pagep = page;
2291
 
 
 
 
 
 
 
 
 
 
 
 
2292	ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
 
2293	if (ret < 0) {
2294		unlock_page(page);
2295		ext4_journal_stop(handle);
2296		page_cache_release(page);
2297		/*
2298		 * block_write_begin may have instantiated a few blocks
2299		 * outside i_size.  Trim these off again. Don't need
2300		 * i_size_read because we hold i_mutex.
2301		 */
2302		if (pos + len > inode->i_size)
2303			ext4_truncate_failed_write(inode);
 
 
 
 
 
2304	}
2305
2306	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2307		goto retry;
2308out:
2309	return ret;
2310}
2311
2312/*
2313 * Check if we should update i_disksize
2314 * when write to the end of file but not require block allocation
2315 */
2316static int ext4_da_should_update_i_disksize(struct page *page,
2317					    unsigned long offset)
2318{
2319	struct buffer_head *bh;
2320	struct inode *inode = page->mapping->host;
2321	unsigned int idx;
2322	int i;
2323
2324	bh = page_buffers(page);
2325	idx = offset >> inode->i_blkbits;
2326
2327	for (i = 0; i < idx; i++)
2328		bh = bh->b_this_page;
2329
2330	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2331		return 0;
2332	return 1;
2333}
2334
2335static int ext4_da_write_end(struct file *file,
2336			     struct address_space *mapping,
2337			     loff_t pos, unsigned len, unsigned copied,
2338			     struct page *page, void *fsdata)
2339{
2340	struct inode *inode = mapping->host;
2341	int ret = 0, ret2;
2342	handle_t *handle = ext4_journal_current_handle();
2343	loff_t new_i_size;
2344	unsigned long start, end;
2345	int write_mode = (int)(unsigned long)fsdata;
2346
2347	if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2348		if (ext4_should_order_data(inode)) {
2349			return ext4_ordered_write_end(file, mapping, pos,
2350					len, copied, page, fsdata);
2351		} else if (ext4_should_writeback_data(inode)) {
2352			return ext4_writeback_write_end(file, mapping, pos,
2353					len, copied, page, fsdata);
2354		} else {
2355			BUG();
2356		}
2357	}
2358
2359	trace_ext4_da_write_end(inode, pos, len, copied);
2360	start = pos & (PAGE_CACHE_SIZE - 1);
2361	end = start + copied - 1;
2362
2363	/*
2364	 * generic_write_end() will run mark_inode_dirty() if i_size
2365	 * changes.  So let's piggyback the i_disksize mark_inode_dirty
2366	 * into that.
2367	 */
2368
2369	new_i_size = pos + copied;
2370	if (new_i_size > EXT4_I(inode)->i_disksize) {
2371		if (ext4_da_should_update_i_disksize(page, end)) {
2372			down_write(&EXT4_I(inode)->i_data_sem);
2373			if (new_i_size > EXT4_I(inode)->i_disksize) {
2374				/*
2375				 * Updating i_disksize when extending file
2376				 * without needing block allocation
2377				 */
2378				if (ext4_should_order_data(inode))
2379					ret = ext4_jbd2_file_inode(handle,
2380								   inode);
2381
2382				EXT4_I(inode)->i_disksize = new_i_size;
2383			}
2384			up_write(&EXT4_I(inode)->i_data_sem);
2385			/* We need to mark inode dirty even if
2386			 * new_i_size is less that inode->i_size
2387			 * bu greater than i_disksize.(hint delalloc)
2388			 */
2389			ext4_mark_inode_dirty(handle, inode);
2390		}
2391	}
2392	ret2 = generic_write_end(file, mapping, pos, len, copied,
2393							page, fsdata);
2394	copied = ret2;
2395	if (ret2 < 0)
2396		ret = ret2;
2397	ret2 = ext4_journal_stop(handle);
2398	if (!ret)
2399		ret = ret2;
2400
2401	return ret ? ret : copied;
2402}
2403
2404static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2405{
2406	/*
2407	 * Drop reserved blocks
 
 
 
 
 
 
 
 
 
 
 
 
2408	 */
2409	BUG_ON(!PageLocked(page));
2410	if (!page_has_buffers(page))
2411		goto out;
2412
2413	ext4_da_page_release_reservation(page, offset);
2414
2415out:
2416	ext4_invalidatepage(page, offset);
2417
2418	return;
2419}
2420
2421/*
2422 * Force all delayed allocation blocks to be allocated for a given inode.
2423 */
2424int ext4_alloc_da_blocks(struct inode *inode)
2425{
2426	trace_ext4_alloc_da_blocks(inode);
2427
2428	if (!EXT4_I(inode)->i_reserved_data_blocks &&
2429	    !EXT4_I(inode)->i_reserved_meta_blocks)
2430		return 0;
2431
2432	/*
2433	 * We do something simple for now.  The filemap_flush() will
2434	 * also start triggering a write of the data blocks, which is
2435	 * not strictly speaking necessary (and for users of
2436	 * laptop_mode, not even desirable).  However, to do otherwise
2437	 * would require replicating code paths in:
2438	 *
2439	 * ext4_da_writepages() ->
2440	 *    write_cache_pages() ---> (via passed in callback function)
2441	 *        __mpage_da_writepage() -->
2442	 *           mpage_add_bh_to_extent()
2443	 *           mpage_da_map_blocks()
2444	 *
2445	 * The problem is that write_cache_pages(), located in
2446	 * mm/page-writeback.c, marks pages clean in preparation for
2447	 * doing I/O, which is not desirable if we're not planning on
2448	 * doing I/O at all.
2449	 *
2450	 * We could call write_cache_pages(), and then redirty all of
2451	 * the pages by calling redirty_page_for_writepage() but that
2452	 * would be ugly in the extreme.  So instead we would need to
2453	 * replicate parts of the code in the above functions,
2454	 * simplifying them because we wouldn't actually intend to
2455	 * write out the pages, but rather only collect contiguous
2456	 * logical block extents, call the multi-block allocator, and
2457	 * then update the buffer heads with the block allocations.
2458	 *
2459	 * For now, though, we'll cheat by calling filemap_flush(),
2460	 * which will map the blocks, and start the I/O, but not
2461	 * actually wait for the I/O to complete.
2462	 */
2463	return filemap_flush(inode->i_mapping);
2464}
2465
2466/*
2467 * bmap() is special.  It gets used by applications such as lilo and by
2468 * the swapper to find the on-disk block of a specific piece of data.
2469 *
2470 * Naturally, this is dangerous if the block concerned is still in the
2471 * journal.  If somebody makes a swapfile on an ext4 data-journaling
2472 * filesystem and enables swap, then they may get a nasty shock when the
2473 * data getting swapped to that swapfile suddenly gets overwritten by
2474 * the original zero's written out previously to the journal and
2475 * awaiting writeback in the kernel's buffer cache.
2476 *
2477 * So, if we see any bmap calls here on a modified, data-journaled file,
2478 * take extra steps to flush any blocks which might be in the cache.
2479 */
2480static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2481{
2482	struct inode *inode = mapping->host;
2483	journal_t *journal;
 
2484	int err;
2485
 
 
 
 
 
 
 
2486	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2487			test_opt(inode->i_sb, DELALLOC)) {
2488		/*
2489		 * With delalloc we want to sync the file
2490		 * so that we can make sure we allocate
2491		 * blocks for file
2492		 */
2493		filemap_write_and_wait(mapping);
2494	}
2495
2496	if (EXT4_JOURNAL(inode) &&
2497	    ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2498		/*
2499		 * This is a REALLY heavyweight approach, but the use of
2500		 * bmap on dirty files is expected to be extremely rare:
2501		 * only if we run lilo or swapon on a freshly made file
2502		 * do we expect this to happen.
2503		 *
2504		 * (bmap requires CAP_SYS_RAWIO so this does not
2505		 * represent an unprivileged user DOS attack --- we'd be
2506		 * in trouble if mortal users could trigger this path at
2507		 * will.)
2508		 *
2509		 * NB. EXT4_STATE_JDATA is not set on files other than
2510		 * regular files.  If somebody wants to bmap a directory
2511		 * or symlink and gets confused because the buffer
2512		 * hasn't yet been flushed to disk, they deserve
2513		 * everything they get.
2514		 */
2515
2516		ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2517		journal = EXT4_JOURNAL(inode);
2518		jbd2_journal_lock_updates(journal);
2519		err = jbd2_journal_flush(journal);
2520		jbd2_journal_unlock_updates(journal);
2521
2522		if (err)
2523			return 0;
2524	}
2525
2526	return generic_block_bmap(mapping, block, ext4_get_block);
 
 
 
 
2527}
2528
2529static int ext4_readpage(struct file *file, struct page *page)
2530{
 
 
 
 
2531	trace_ext4_readpage(page);
2532	return mpage_readpage(page, ext4_get_block);
 
 
 
 
 
 
 
2533}
2534
2535static int
2536ext4_readpages(struct file *file, struct address_space *mapping,
2537		struct list_head *pages, unsigned nr_pages)
2538{
2539	return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
 
 
 
 
 
 
2540}
2541
2542static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
 
2543{
2544	struct buffer_head *head, *bh;
2545	unsigned int curr_off = 0;
2546
2547	if (!page_has_buffers(page))
2548		return;
2549	head = bh = page_buffers(page);
2550	do {
2551		if (offset <= curr_off && test_clear_buffer_uninit(bh)
2552					&& bh->b_private) {
2553			ext4_free_io_end(bh->b_private);
2554			bh->b_private = NULL;
2555			bh->b_end_io = NULL;
2556		}
2557		curr_off = curr_off + bh->b_size;
2558		bh = bh->b_this_page;
2559	} while (bh != head);
2560}
2561
2562static void ext4_invalidatepage(struct page *page, unsigned long offset)
 
2563{
2564	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2565
2566	trace_ext4_invalidatepage(page, offset);
2567
2568	/*
2569	 * free any io_end structure allocated for buffers to be discarded
2570	 */
2571	if (ext4_should_dioread_nolock(page->mapping->host))
2572		ext4_invalidatepage_free_endio(page, offset);
2573	/*
2574	 * If it's a full truncate we just forget about the pending dirtying
2575	 */
2576	if (offset == 0)
2577		ClearPageChecked(page);
2578
2579	if (journal)
2580		jbd2_journal_invalidatepage(journal, page, offset);
2581	else
2582		block_invalidatepage(page, offset);
 
 
 
 
 
2583}
2584
2585static int ext4_releasepage(struct page *page, gfp_t wait)
2586{
2587	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2588
2589	trace_ext4_releasepage(page);
2590
2591	WARN_ON(PageChecked(page));
2592	if (!page_has_buffers(page))
2593		return 0;
2594	if (journal)
2595		return jbd2_journal_try_to_free_buffers(journal, page, wait);
2596	else
2597		return try_to_free_buffers(page);
2598}
2599
2600/*
2601 * ext4_get_block used when preparing for a DIO write or buffer write.
2602 * We allocate an uinitialized extent if blocks haven't been allocated.
2603 * The extent will be converted to initialized after the IO is complete.
2604 */
2605static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2606		   struct buffer_head *bh_result, int create)
2607{
2608	ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2609		   inode->i_ino, create);
2610	return _ext4_get_block(inode, iblock, bh_result,
2611			       EXT4_GET_BLOCKS_IO_CREATE_EXT);
 
 
 
 
 
 
 
 
 
 
 
2612}
2613
2614static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2615			    ssize_t size, void *private, int ret,
2616			    bool is_async)
2617{
2618	struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2619        ext4_io_end_t *io_end = iocb->private;
2620	struct workqueue_struct *wq;
2621	unsigned long flags;
2622	struct ext4_inode_info *ei;
2623
2624	/* if not async direct IO or dio with 0 bytes write, just return */
2625	if (!io_end || !size)
2626		goto out;
 
 
 
 
 
 
2627
2628	ext_debug("ext4_end_io_dio(): io_end 0x%p"
2629		  "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2630 		  iocb->private, io_end->inode->i_ino, iocb, offset,
2631		  size);
2632
2633	/* if not aio dio with unwritten extents, just free io and return */
2634	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2635		ext4_free_io_end(io_end);
2636		iocb->private = NULL;
2637out:
2638		if (is_async)
2639			aio_complete(iocb, ret, 0);
2640		inode_dio_done(inode);
2641		return;
2642	}
2643
2644	io_end->offset = offset;
2645	io_end->size = size;
2646	if (is_async) {
2647		io_end->iocb = iocb;
2648		io_end->result = ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2649	}
2650	wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2651
2652	/* Add the io_end to per-inode completed aio dio list*/
2653	ei = EXT4_I(io_end->inode);
2654	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2655	list_add_tail(&io_end->list, &ei->i_completed_io_list);
2656	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 
 
 
 
 
 
 
 
 
 
 
 
2657
2658	/* queue the work to convert unwritten extents to written */
2659	queue_work(wq, &io_end->work);
2660	iocb->private = NULL;
2661
2662	/* XXX: probably should move into the real I/O completion handler */
2663	inode_dio_done(inode);
 
 
 
 
 
 
 
 
 
 
 
2664}
2665
2666static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
 
 
2667{
2668	ext4_io_end_t *io_end = bh->b_private;
2669	struct workqueue_struct *wq;
2670	struct inode *inode;
2671	unsigned long flags;
2672
2673	if (!test_clear_buffer_uninit(bh) || !io_end)
2674		goto out;
2675
2676	if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2677		printk("sb umounted, discard end_io request for inode %lu\n",
2678			io_end->inode->i_ino);
2679		ext4_free_io_end(io_end);
2680		goto out;
2681	}
2682
2683	/*
2684	 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2685	 * but being more careful is always safe for the future change.
2686	 */
2687	inode = io_end->inode;
2688	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2689		io_end->flag |= EXT4_IO_END_UNWRITTEN;
2690		atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
2691	}
2692
2693	/* Add the io_end to per-inode completed io list*/
2694	spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2695	list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2696	spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2697
2698	wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2699	/* queue the work to convert unwritten extents to written */
2700	queue_work(wq, &io_end->work);
 
 
 
 
 
 
 
 
 
2701out:
2702	bh->b_private = NULL;
2703	bh->b_end_io = NULL;
2704	clear_buffer_uninit(bh);
2705	end_buffer_async_write(bh, uptodate);
 
 
 
 
 
 
2706}
2707
2708static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2709{
2710	ext4_io_end_t *io_end;
2711	struct page *page = bh->b_page;
2712	loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2713	size_t size = bh->b_size;
2714
2715retry:
2716	io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2717	if (!io_end) {
2718		pr_warn_ratelimited("%s: allocation fail\n", __func__);
2719		schedule();
2720		goto retry;
2721	}
2722	io_end->offset = offset;
2723	io_end->size = size;
2724	/*
2725	 * We need to hold a reference to the page to make sure it
2726	 * doesn't get evicted before ext4_end_io_work() has a chance
2727	 * to convert the extent from written to unwritten.
2728	 */
2729	io_end->page = page;
2730	get_page(io_end->page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2731
2732	bh->b_private = io_end;
2733	bh->b_end_io = ext4_end_io_buffer_write;
2734	return 0;
2735}
2736
2737/*
2738 * For ext4 extent files, ext4 will do direct-io write to holes,
2739 * preallocated extents, and those write extend the file, no need to
2740 * fall back to buffered IO.
2741 *
2742 * For holes, we fallocate those blocks, mark them as uninitialized
2743 * If those blocks were preallocated, we mark sure they are splited, but
2744 * still keep the range to write as uninitialized.
2745 *
2746 * The unwrritten extents will be converted to written when DIO is completed.
2747 * For async direct IO, since the IO may still pending when return, we
2748 * set up an end_io call back function, which will do the conversion
2749 * when async direct IO completed.
2750 *
2751 * If the O_DIRECT write will extend the file then add this inode to the
2752 * orphan list.  So recovery will truncate it back to the original size
2753 * if the machine crashes during the write.
2754 *
2755 */
2756static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2757			      const struct iovec *iov, loff_t offset,
2758			      unsigned long nr_segs)
2759{
2760	struct file *file = iocb->ki_filp;
2761	struct inode *inode = file->f_mapping->host;
2762	ssize_t ret;
2763	size_t count = iov_length(iov, nr_segs);
2764
2765	loff_t final_size = offset + count;
2766	if (rw == WRITE && final_size <= inode->i_size) {
2767		/*
2768 		 * We could direct write to holes and fallocate.
2769		 *
2770 		 * Allocated blocks to fill the hole are marked as uninitialized
2771 		 * to prevent parallel buffered read to expose the stale data
2772 		 * before DIO complete the data IO.
2773		 *
2774 		 * As to previously fallocated extents, ext4 get_block
2775 		 * will just simply mark the buffer mapped but still
2776 		 * keep the extents uninitialized.
2777 		 *
2778		 * for non AIO case, we will convert those unwritten extents
2779		 * to written after return back from blockdev_direct_IO.
2780		 *
2781		 * for async DIO, the conversion needs to be defered when
2782		 * the IO is completed. The ext4 end_io callback function
2783		 * will be called to take care of the conversion work.
2784		 * Here for async case, we allocate an io_end structure to
2785		 * hook to the iocb.
2786 		 */
2787		iocb->private = NULL;
2788		EXT4_I(inode)->cur_aio_dio = NULL;
2789		if (!is_sync_kiocb(iocb)) {
2790			iocb->private = ext4_init_io_end(inode, GFP_NOFS);
2791			if (!iocb->private)
2792				return -ENOMEM;
2793			/*
2794			 * we save the io structure for current async
2795			 * direct IO, so that later ext4_map_blocks()
2796			 * could flag the io structure whether there
2797			 * is a unwritten extents needs to be converted
2798			 * when IO is completed.
2799			 */
2800			EXT4_I(inode)->cur_aio_dio = iocb->private;
2801		}
2802
2803		ret = __blockdev_direct_IO(rw, iocb, inode,
2804					 inode->i_sb->s_bdev, iov,
2805					 offset, nr_segs,
2806					 ext4_get_block_write,
2807					 ext4_end_io_dio,
2808					 NULL,
2809					 DIO_LOCKING | DIO_SKIP_HOLES);
2810		if (iocb->private)
2811			EXT4_I(inode)->cur_aio_dio = NULL;
2812		/*
2813		 * The io_end structure takes a reference to the inode,
2814		 * that structure needs to be destroyed and the
2815		 * reference to the inode need to be dropped, when IO is
2816		 * complete, even with 0 byte write, or failed.
2817		 *
2818		 * In the successful AIO DIO case, the io_end structure will be
2819		 * desctroyed and the reference to the inode will be dropped
2820		 * after the end_io call back function is called.
2821		 *
2822		 * In the case there is 0 byte write, or error case, since
2823		 * VFS direct IO won't invoke the end_io call back function,
2824		 * we need to free the end_io structure here.
2825		 */
2826		if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
2827			ext4_free_io_end(iocb->private);
2828			iocb->private = NULL;
2829		} else if (ret > 0 && ext4_test_inode_state(inode,
2830						EXT4_STATE_DIO_UNWRITTEN)) {
2831			int err;
2832			/*
2833			 * for non AIO case, since the IO is already
2834			 * completed, we could do the conversion right here
2835			 */
2836			err = ext4_convert_unwritten_extents(inode,
2837							     offset, ret);
2838			if (err < 0)
2839				ret = err;
2840			ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
2841		}
2842		return ret;
2843	}
2844
2845	/* for write the the end of file case, we fall back to old way */
2846	return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
 
 
2847}
2848
2849static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2850			      const struct iovec *iov, loff_t offset,
2851			      unsigned long nr_segs)
2852{
2853	struct file *file = iocb->ki_filp;
2854	struct inode *inode = file->f_mapping->host;
2855	ssize_t ret;
 
2856
2857	trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
2858	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
2859		ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
2860	else
2861		ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2862	trace_ext4_direct_IO_exit(inode, offset,
2863				iov_length(iov, nr_segs), rw, ret);
2864	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2865}
2866
 
 
 
 
2867/*
2868 * Pages can be marked dirty completely asynchronously from ext4's journalling
2869 * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
2870 * much here because ->set_page_dirty is called under VFS locks.  The page is
2871 * not necessarily locked.
 
2872 *
2873 * We cannot just dirty the page and leave attached buffers clean, because the
2874 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
2875 * or jbddirty because all the journalling code will explode.
2876 *
2877 * So what we do is to mark the page "pending dirty" and next time writepage
2878 * is called, propagate that into the buffers appropriately.
2879 */
2880static int ext4_journalled_set_page_dirty(struct page *page)
 
2881{
2882	SetPageChecked(page);
2883	return __set_page_dirty_nobuffers(page);
 
2884}
2885
2886static const struct address_space_operations ext4_ordered_aops = {
2887	.readpage		= ext4_readpage,
2888	.readpages		= ext4_readpages,
2889	.writepage		= ext4_writepage,
2890	.write_begin		= ext4_write_begin,
2891	.write_end		= ext4_ordered_write_end,
2892	.bmap			= ext4_bmap,
2893	.invalidatepage		= ext4_invalidatepage,
2894	.releasepage		= ext4_releasepage,
2895	.direct_IO		= ext4_direct_IO,
2896	.migratepage		= buffer_migrate_page,
2897	.is_partially_uptodate  = block_is_partially_uptodate,
2898	.error_remove_page	= generic_error_remove_page,
2899};
2900
2901static const struct address_space_operations ext4_writeback_aops = {
2902	.readpage		= ext4_readpage,
2903	.readpages		= ext4_readpages,
2904	.writepage		= ext4_writepage,
 
 
 
 
 
 
 
2905	.write_begin		= ext4_write_begin,
2906	.write_end		= ext4_writeback_write_end,
 
2907	.bmap			= ext4_bmap,
2908	.invalidatepage		= ext4_invalidatepage,
2909	.releasepage		= ext4_releasepage,
2910	.direct_IO		= ext4_direct_IO,
2911	.migratepage		= buffer_migrate_page,
2912	.is_partially_uptodate  = block_is_partially_uptodate,
2913	.error_remove_page	= generic_error_remove_page,
 
2914};
2915
2916static const struct address_space_operations ext4_journalled_aops = {
2917	.readpage		= ext4_readpage,
2918	.readpages		= ext4_readpages,
2919	.writepage		= ext4_writepage,
2920	.write_begin		= ext4_write_begin,
2921	.write_end		= ext4_journalled_write_end,
2922	.set_page_dirty		= ext4_journalled_set_page_dirty,
2923	.bmap			= ext4_bmap,
2924	.invalidatepage		= ext4_invalidatepage,
2925	.releasepage		= ext4_releasepage,
 
 
2926	.is_partially_uptodate  = block_is_partially_uptodate,
2927	.error_remove_page	= generic_error_remove_page,
 
2928};
2929
2930static const struct address_space_operations ext4_da_aops = {
2931	.readpage		= ext4_readpage,
2932	.readpages		= ext4_readpages,
2933	.writepage		= ext4_writepage,
2934	.writepages		= ext4_da_writepages,
2935	.write_begin		= ext4_da_write_begin,
2936	.write_end		= ext4_da_write_end,
 
2937	.bmap			= ext4_bmap,
2938	.invalidatepage		= ext4_da_invalidatepage,
2939	.releasepage		= ext4_releasepage,
2940	.direct_IO		= ext4_direct_IO,
2941	.migratepage		= buffer_migrate_page,
2942	.is_partially_uptodate  = block_is_partially_uptodate,
2943	.error_remove_page	= generic_error_remove_page,
 
 
 
 
 
 
 
 
 
2944};
2945
2946void ext4_set_aops(struct inode *inode)
2947{
2948	if (ext4_should_order_data(inode) &&
2949		test_opt(inode->i_sb, DELALLOC))
2950		inode->i_mapping->a_ops = &ext4_da_aops;
2951	else if (ext4_should_order_data(inode))
2952		inode->i_mapping->a_ops = &ext4_ordered_aops;
2953	else if (ext4_should_writeback_data(inode) &&
2954		 test_opt(inode->i_sb, DELALLOC))
 
 
 
 
 
 
2955		inode->i_mapping->a_ops = &ext4_da_aops;
2956	else if (ext4_should_writeback_data(inode))
2957		inode->i_mapping->a_ops = &ext4_writeback_aops;
2958	else
2959		inode->i_mapping->a_ops = &ext4_journalled_aops;
2960}
2961
2962/*
2963 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2964 * up to the end of the block which corresponds to `from'.
2965 * This required during truncate. We need to physically zero the tail end
2966 * of that block so it doesn't yield old data if the file is later grown.
2967 */
2968int ext4_block_truncate_page(handle_t *handle,
2969		struct address_space *mapping, loff_t from)
2970{
2971	unsigned offset = from & (PAGE_CACHE_SIZE-1);
2972	unsigned length;
2973	unsigned blocksize;
2974	struct inode *inode = mapping->host;
2975
2976	blocksize = inode->i_sb->s_blocksize;
2977	length = blocksize - (offset & (blocksize - 1));
2978
2979	return ext4_block_zero_page_range(handle, mapping, from, length);
2980}
2981
2982/*
2983 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
2984 * starting from file offset 'from'.  The range to be zero'd must
2985 * be contained with in one block.  If the specified range exceeds
2986 * the end of the block it will be shortened to end of the block
2987 * that cooresponds to 'from'
2988 */
2989int ext4_block_zero_page_range(handle_t *handle,
2990		struct address_space *mapping, loff_t from, loff_t length)
2991{
2992	ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2993	unsigned offset = from & (PAGE_CACHE_SIZE-1);
2994	unsigned blocksize, max, pos;
2995	ext4_lblk_t iblock;
2996	struct inode *inode = mapping->host;
2997	struct buffer_head *bh;
2998	struct page *page;
2999	int err = 0;
3000
3001	page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3002				   mapping_gfp_mask(mapping) & ~__GFP_FS);
3003	if (!page)
3004		return -EINVAL;
3005
3006	blocksize = inode->i_sb->s_blocksize;
3007	max = blocksize - (offset & (blocksize - 1));
3008
3009	/*
3010	 * correct length if it does not fall between
3011	 * 'from' and the end of the block
3012	 */
3013	if (length > max || length < 0)
3014		length = max;
3015
3016	iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3017
3018	if (!page_has_buffers(page))
3019		create_empty_buffers(page, blocksize, 0);
3020
3021	/* Find the buffer that contains "offset" */
3022	bh = page_buffers(page);
3023	pos = blocksize;
3024	while (offset >= pos) {
3025		bh = bh->b_this_page;
3026		iblock++;
3027		pos += blocksize;
3028	}
3029
3030	err = 0;
3031	if (buffer_freed(bh)) {
3032		BUFFER_TRACE(bh, "freed: skip");
3033		goto unlock;
3034	}
3035
3036	if (!buffer_mapped(bh)) {
3037		BUFFER_TRACE(bh, "unmapped");
3038		ext4_get_block(inode, iblock, bh, 0);
3039		/* unmapped? It's a hole - nothing to do */
3040		if (!buffer_mapped(bh)) {
3041			BUFFER_TRACE(bh, "still unmapped");
3042			goto unlock;
3043		}
3044	}
3045
3046	/* Ok, it's mapped. Make sure it's up-to-date */
3047	if (PageUptodate(page))
3048		set_buffer_uptodate(bh);
3049
3050	if (!buffer_uptodate(bh)) {
3051		err = -EIO;
3052		ll_rw_block(READ, 1, &bh);
3053		wait_on_buffer(bh);
3054		/* Uhhuh. Read error. Complain and punt. */
3055		if (!buffer_uptodate(bh))
3056			goto unlock;
 
 
 
 
 
 
 
 
 
 
3057	}
3058
3059	if (ext4_should_journal_data(inode)) {
3060		BUFFER_TRACE(bh, "get write access");
3061		err = ext4_journal_get_write_access(handle, bh);
 
3062		if (err)
3063			goto unlock;
3064	}
3065
3066	zero_user(page, offset, length);
3067
3068	BUFFER_TRACE(bh, "zeroed end of block");
3069
3070	err = 0;
3071	if (ext4_should_journal_data(inode)) {
3072		err = ext4_handle_dirty_metadata(handle, inode, bh);
3073	} else {
3074		if (ext4_should_order_data(inode) && EXT4_I(inode)->jinode)
3075			err = ext4_jbd2_file_inode(handle, inode);
3076		mark_buffer_dirty(bh);
 
 
 
3077	}
3078
3079unlock:
3080	unlock_page(page);
3081	page_cache_release(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3082	return err;
3083}
3084
3085int ext4_can_truncate(struct inode *inode)
3086{
3087	if (S_ISREG(inode->i_mode))
3088		return 1;
3089	if (S_ISDIR(inode->i_mode))
3090		return 1;
3091	if (S_ISLNK(inode->i_mode))
3092		return !ext4_inode_is_fast_symlink(inode);
3093	return 0;
3094}
3095
3096/*
3097 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3098 * associated with the given offset and length
3099 *
3100 * @inode:  File inode
3101 * @offset: The offset where the hole will begin
3102 * @len:    The length of the hole
3103 *
3104 * Returns: 0 on sucess or negative on failure
3105 */
3106
3107int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3108{
3109	struct inode *inode = file->f_path.dentry->d_inode;
3110	if (!S_ISREG(inode->i_mode))
3111		return -ENOTSUPP;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3112
3113	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3114		/* TODO: Add support for non extent hole punching */
3115		return -ENOTSUPP;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3116	}
3117
3118	return ext4_ext_punch_hole(file, offset, length);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3119}
3120
3121/*
3122 * ext4_truncate()
3123 *
3124 * We block out ext4_get_block() block instantiations across the entire
3125 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3126 * simultaneously on behalf of the same inode.
3127 *
3128 * As we work through the truncate and commmit bits of it to the journal there
3129 * is one core, guiding principle: the file's tree must always be consistent on
3130 * disk.  We must be able to restart the truncate after a crash.
3131 *
3132 * The file's tree may be transiently inconsistent in memory (although it
3133 * probably isn't), but whenever we close off and commit a journal transaction,
3134 * the contents of (the filesystem + the journal) must be consistent and
3135 * restartable.  It's pretty simple, really: bottom up, right to left (although
3136 * left-to-right works OK too).
3137 *
3138 * Note that at recovery time, journal replay occurs *before* the restart of
3139 * truncate against the orphan inode list.
3140 *
3141 * The committed inode has the new, desired i_size (which is the same as
3142 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
3143 * that this inode's truncate did not complete and it will again call
3144 * ext4_truncate() to have another go.  So there will be instantiated blocks
3145 * to the right of the truncation point in a crashed ext4 filesystem.  But
3146 * that's fine - as long as they are linked from the inode, the post-crash
3147 * ext4_truncate() run will find them and release them.
3148 */
3149void ext4_truncate(struct inode *inode)
3150{
 
 
 
 
 
 
 
 
 
 
 
 
 
3151	trace_ext4_truncate_enter(inode);
3152
3153	if (!ext4_can_truncate(inode))
3154		return;
3155
3156	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3157
3158	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3159		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3160
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3161	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3162		ext4_ext_truncate(inode);
3163	else
3164		ext4_ind_truncate(inode);
 
 
 
 
3165
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3166	trace_ext4_truncate_exit(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3167}
3168
3169/*
3170 * ext4_get_inode_loc returns with an extra refcount against the inode's
3171 * underlying buffer_head on success. If 'in_mem' is true, we have all
3172 * data in memory that is needed to recreate the on-disk version of this
3173 * inode.
3174 */
3175static int __ext4_get_inode_loc(struct inode *inode,
3176				struct ext4_iloc *iloc, int in_mem)
 
3177{
3178	struct ext4_group_desc	*gdp;
3179	struct buffer_head	*bh;
3180	struct super_block	*sb = inode->i_sb;
3181	ext4_fsblk_t		block;
 
3182	int			inodes_per_block, inode_offset;
3183
3184	iloc->bh = NULL;
3185	if (!ext4_valid_inum(sb, inode->i_ino))
3186		return -EIO;
 
3187
3188	iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3189	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3190	if (!gdp)
3191		return -EIO;
3192
3193	/*
3194	 * Figure out the offset within the block group inode table
3195	 */
3196	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3197	inode_offset = ((inode->i_ino - 1) %
3198			EXT4_INODES_PER_GROUP(sb));
3199	block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3200	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3201
 
 
 
 
 
 
 
 
 
3202	bh = sb_getblk(sb, block);
3203	if (!bh) {
3204		EXT4_ERROR_INODE_BLOCK(inode, block,
3205				       "unable to read itable block");
3206		return -EIO;
 
 
 
 
 
 
3207	}
3208	if (!buffer_uptodate(bh)) {
3209		lock_buffer(bh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3210
3211		/*
3212		 * If the buffer has the write error flag, we have failed
3213		 * to write out another inode in the same block.  In this
3214		 * case, we don't have to read the block because we may
3215		 * read the old inode data successfully.
3216		 */
3217		if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3218			set_buffer_uptodate(bh);
3219
3220		if (buffer_uptodate(bh)) {
3221			/* someone brought it uptodate while we waited */
3222			unlock_buffer(bh);
3223			goto has_buffer;
3224		}
3225
3226		/*
3227		 * If we have all information of the inode in memory and this
3228		 * is the only valid inode in the block, we need not read the
3229		 * block.
3230		 */
3231		if (in_mem) {
3232			struct buffer_head *bitmap_bh;
3233			int i, start;
3234
3235			start = inode_offset & ~(inodes_per_block - 1);
3236
3237			/* Is the inode bitmap in cache? */
3238			bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3239			if (!bitmap_bh)
3240				goto make_io;
3241
3242			/*
3243			 * If the inode bitmap isn't in cache then the
3244			 * optimisation may end up performing two reads instead
3245			 * of one, so skip it.
3246			 */
3247			if (!buffer_uptodate(bitmap_bh)) {
3248				brelse(bitmap_bh);
3249				goto make_io;
3250			}
3251			for (i = start; i < start + inodes_per_block; i++) {
3252				if (i == inode_offset)
3253					continue;
3254				if (ext4_test_bit(i, bitmap_bh->b_data))
3255					break;
3256			}
3257			brelse(bitmap_bh);
3258			if (i == start + inodes_per_block) {
3259				/* all other inodes are free, so skip I/O */
3260				memset(bh->b_data, 0, bh->b_size);
3261				set_buffer_uptodate(bh);
3262				unlock_buffer(bh);
3263				goto has_buffer;
3264			}
3265		}
3266
3267make_io:
3268		/*
3269		 * If we need to do any I/O, try to pre-readahead extra
3270		 * blocks from the inode table.
3271		 */
3272		if (EXT4_SB(sb)->s_inode_readahead_blks) {
3273			ext4_fsblk_t b, end, table;
3274			unsigned num;
3275
3276			table = ext4_inode_table(sb, gdp);
3277			/* s_inode_readahead_blks is always a power of 2 */
3278			b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3279			if (table > b)
3280				b = table;
3281			end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3282			num = EXT4_INODES_PER_GROUP(sb);
3283			if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3284				       EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3285				num -= ext4_itable_unused_count(sb, gdp);
3286			table += num / inodes_per_block;
3287			if (end > table)
3288				end = table;
3289			while (b <= end)
3290				sb_breadahead(sb, b++);
3291		}
 
3292
3293		/*
3294		 * There are other valid inodes in the buffer, this inode
3295		 * has in-inode xattrs, or we don't have this inode in memory.
3296		 * Read the block from disk.
3297		 */
3298		trace_ext4_load_inode(inode);
3299		get_bh(bh);
3300		bh->b_end_io = end_buffer_read_sync;
3301		submit_bh(READ | REQ_META | REQ_PRIO, bh);
3302		wait_on_buffer(bh);
3303		if (!buffer_uptodate(bh)) {
3304			EXT4_ERROR_INODE_BLOCK(inode, block,
3305					       "unable to read itable block");
3306			brelse(bh);
3307			return -EIO;
3308		}
3309	}
3310has_buffer:
3311	iloc->bh = bh;
3312	return 0;
3313}
3314
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3315int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3316{
3317	/* We have all inode data except xattrs in memory here. */
3318	return __ext4_get_inode_loc(inode, iloc,
3319		!ext4_test_inode_state(inode, EXT4_STATE_XATTR));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3320}
3321
3322void ext4_set_inode_flags(struct inode *inode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3323{
3324	unsigned int flags = EXT4_I(inode)->i_flags;
 
 
 
3325
3326	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3327	if (flags & EXT4_SYNC_FL)
3328		inode->i_flags |= S_SYNC;
3329	if (flags & EXT4_APPEND_FL)
3330		inode->i_flags |= S_APPEND;
3331	if (flags & EXT4_IMMUTABLE_FL)
3332		inode->i_flags |= S_IMMUTABLE;
3333	if (flags & EXT4_NOATIME_FL)
3334		inode->i_flags |= S_NOATIME;
3335	if (flags & EXT4_DIRSYNC_FL)
3336		inode->i_flags |= S_DIRSYNC;
3337}
3338
3339/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3340void ext4_get_inode_flags(struct ext4_inode_info *ei)
3341{
3342	unsigned int vfs_fl;
3343	unsigned long old_fl, new_fl;
3344
3345	do {
3346		vfs_fl = ei->vfs_inode.i_flags;
3347		old_fl = ei->i_flags;
3348		new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3349				EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3350				EXT4_DIRSYNC_FL);
3351		if (vfs_fl & S_SYNC)
3352			new_fl |= EXT4_SYNC_FL;
3353		if (vfs_fl & S_APPEND)
3354			new_fl |= EXT4_APPEND_FL;
3355		if (vfs_fl & S_IMMUTABLE)
3356			new_fl |= EXT4_IMMUTABLE_FL;
3357		if (vfs_fl & S_NOATIME)
3358			new_fl |= EXT4_NOATIME_FL;
3359		if (vfs_fl & S_DIRSYNC)
3360			new_fl |= EXT4_DIRSYNC_FL;
3361	} while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3362}
3363
3364static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3365				  struct ext4_inode_info *ei)
3366{
3367	blkcnt_t i_blocks ;
3368	struct inode *inode = &(ei->vfs_inode);
3369	struct super_block *sb = inode->i_sb;
3370
3371	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3372				EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3373		/* we are using combined 48 bit field */
3374		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3375					le32_to_cpu(raw_inode->i_blocks_lo);
3376		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3377			/* i_blocks represent file system block size */
3378			return i_blocks  << (inode->i_blkbits - 9);
3379		} else {
3380			return i_blocks;
3381		}
3382	} else {
3383		return le32_to_cpu(raw_inode->i_blocks_lo);
3384	}
3385}
3386
3387struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3388{
3389	struct ext4_iloc iloc;
3390	struct ext4_inode *raw_inode;
3391	struct ext4_inode_info *ei;
 
3392	struct inode *inode;
3393	journal_t *journal = EXT4_SB(sb)->s_journal;
3394	long ret;
 
3395	int block;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3396
3397	inode = iget_locked(sb, ino);
3398	if (!inode)
3399		return ERR_PTR(-ENOMEM);
3400	if (!(inode->i_state & I_NEW))
3401		return inode;
3402
3403	ei = EXT4_I(inode);
3404	iloc.bh = NULL;
3405
3406	ret = __ext4_get_inode_loc(inode, &iloc, 0);
3407	if (ret < 0)
3408		goto bad_inode;
3409	raw_inode = ext4_raw_inode(&iloc);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3410	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3411	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3412	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
 
 
 
 
 
 
 
3413	if (!(test_opt(inode->i_sb, NO_UID32))) {
3414		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3415		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3416	}
3417	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 
 
 
3418
3419	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
 
3420	ei->i_dir_start_lookup = 0;
3421	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3422	/* We now have enough fields to check if the inode was active or not.
3423	 * This is needed because nfsd might try to access dead inodes
3424	 * the test is that same one that e2fsck uses
3425	 * NeilBrown 1999oct15
3426	 */
3427	if (inode->i_nlink == 0) {
3428		if (inode->i_mode == 0 ||
3429		    !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
 
3430			/* this inode is deleted */
3431			ret = -ESTALE;
3432			goto bad_inode;
3433		}
3434		/* The only unlinked inodes we let through here have
3435		 * valid i_mode and are being read by the orphan
3436		 * recovery code: that's fine, we're about to complete
3437		 * the process of deleting those. */
 
 
3438	}
3439	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
 
3440	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3441	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3442	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3443		ei->i_file_acl |=
3444			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3445	inode->i_size = ext4_isize(raw_inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3446	ei->i_disksize = inode->i_size;
3447#ifdef CONFIG_QUOTA
3448	ei->i_reserved_quota = 0;
3449#endif
3450	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3451	ei->i_block_group = iloc.block_group;
3452	ei->i_last_alloc_group = ~0;
3453	/*
3454	 * NOTE! The in-memory inode i_data array is in little-endian order
3455	 * even on big-endian machines: we do NOT byteswap the block numbers!
3456	 */
3457	for (block = 0; block < EXT4_N_BLOCKS; block++)
3458		ei->i_data[block] = raw_inode->i_block[block];
3459	INIT_LIST_HEAD(&ei->i_orphan);
 
3460
3461	/*
3462	 * Set transaction id's of transactions that have to be committed
3463	 * to finish f[data]sync. We set them to currently running transaction
3464	 * as we cannot be sure that the inode or some of its metadata isn't
3465	 * part of the transaction - the inode could have been reclaimed and
3466	 * now it is reread from disk.
3467	 */
3468	if (journal) {
3469		transaction_t *transaction;
3470		tid_t tid;
3471
3472		read_lock(&journal->j_state_lock);
3473		if (journal->j_running_transaction)
3474			transaction = journal->j_running_transaction;
3475		else
3476			transaction = journal->j_committing_transaction;
3477		if (transaction)
3478			tid = transaction->t_tid;
3479		else
3480			tid = journal->j_commit_sequence;
3481		read_unlock(&journal->j_state_lock);
3482		ei->i_sync_tid = tid;
3483		ei->i_datasync_tid = tid;
3484	}
3485
3486	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3487		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3488		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3489		    EXT4_INODE_SIZE(inode->i_sb)) {
3490			ret = -EIO;
3491			goto bad_inode;
3492		}
3493		if (ei->i_extra_isize == 0) {
3494			/* The extra space is currently unused. Use it. */
 
3495			ei->i_extra_isize = sizeof(struct ext4_inode) -
3496					    EXT4_GOOD_OLD_INODE_SIZE;
3497		} else {
3498			__le32 *magic = (void *)raw_inode +
3499					EXT4_GOOD_OLD_INODE_SIZE +
3500					ei->i_extra_isize;
3501			if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3502				ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3503		}
3504	} else
3505		ei->i_extra_isize = 0;
3506
3507	EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3508	EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3509	EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3510	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3511
3512	inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3513	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3514		if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3515			inode->i_version |=
3516			(__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
 
 
 
 
3517	}
3518
3519	ret = 0;
3520	if (ei->i_file_acl &&
3521	    !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3522		EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
 
3523				 ei->i_file_acl);
3524		ret = -EIO;
3525		goto bad_inode;
3526	} else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3527		if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3528		    (S_ISLNK(inode->i_mode) &&
3529		     !ext4_inode_is_fast_symlink(inode)))
3530			/* Validate extent which is part of inode */
3531			ret = ext4_ext_check_inode(inode);
3532	} else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3533		   (S_ISLNK(inode->i_mode) &&
3534		    !ext4_inode_is_fast_symlink(inode))) {
3535		/* Validate block references which are part of inode */
3536		ret = ext4_ind_check_inode(inode);
3537	}
3538	if (ret)
3539		goto bad_inode;
3540
3541	if (S_ISREG(inode->i_mode)) {
3542		inode->i_op = &ext4_file_inode_operations;
3543		inode->i_fop = &ext4_file_operations;
3544		ext4_set_aops(inode);
3545	} else if (S_ISDIR(inode->i_mode)) {
3546		inode->i_op = &ext4_dir_inode_operations;
3547		inode->i_fop = &ext4_dir_operations;
3548	} else if (S_ISLNK(inode->i_mode)) {
3549		if (ext4_inode_is_fast_symlink(inode)) {
 
 
 
 
 
 
 
 
 
 
 
3550			inode->i_op = &ext4_fast_symlink_inode_operations;
3551			nd_terminate_link(ei->i_data, inode->i_size,
3552				sizeof(ei->i_data) - 1);
3553		} else {
3554			inode->i_op = &ext4_symlink_inode_operations;
3555			ext4_set_aops(inode);
3556		}
3557	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3558	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3559		inode->i_op = &ext4_special_inode_operations;
3560		if (raw_inode->i_block[0])
3561			init_special_inode(inode, inode->i_mode,
3562			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3563		else
3564			init_special_inode(inode, inode->i_mode,
3565			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
 
 
3566	} else {
3567		ret = -EIO;
3568		EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
 
3569		goto bad_inode;
3570	}
 
 
 
 
 
 
 
 
 
 
3571	brelse(iloc.bh);
3572	ext4_set_inode_flags(inode);
3573	unlock_new_inode(inode);
3574	return inode;
3575
3576bad_inode:
3577	brelse(iloc.bh);
3578	iget_failed(inode);
3579	return ERR_PTR(ret);
3580}
3581
3582static int ext4_inode_blocks_set(handle_t *handle,
3583				struct ext4_inode *raw_inode,
3584				struct ext4_inode_info *ei)
 
3585{
3586	struct inode *inode = &(ei->vfs_inode);
3587	u64 i_blocks = inode->i_blocks;
3588	struct super_block *sb = inode->i_sb;
3589
3590	if (i_blocks <= ~0U) {
3591		/*
3592		 * i_blocks can be represnted in a 32 bit variable
3593		 * as multiple of 512 bytes
3594		 */
3595		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3596		raw_inode->i_blocks_high = 0;
3597		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3598		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
3599	}
3600	if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3601		return -EFBIG;
3602
3603	if (i_blocks <= 0xffffffffffffULL) {
3604		/*
3605		 * i_blocks can be represented in a 48 bit variable
3606		 * as multiple of 512 bytes
3607		 */
3608		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3609		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3610		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3611	} else {
3612		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3613		/* i_block is stored in file system block size */
3614		i_blocks = i_blocks >> (inode->i_blkbits - 9);
3615		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3616		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
 
 
 
 
 
 
 
 
 
3617	}
3618	return 0;
3619}
3620
3621/*
3622 * Post the struct inode info into an on-disk inode location in the
3623 * buffer-cache.  This gobbles the caller's reference to the
3624 * buffer_head in the inode location struct.
3625 *
3626 * The caller must have write access to iloc->bh.
3627 */
3628static int ext4_do_update_inode(handle_t *handle,
3629				struct inode *inode,
3630				struct ext4_iloc *iloc)
3631{
3632	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3633	struct ext4_inode_info *ei = EXT4_I(inode);
3634	struct buffer_head *bh = iloc->bh;
3635	int err = 0, rc, block;
 
 
 
 
3636
3637	/* For fields not not tracking in the in-memory inode,
3638	 * initialise them to zero for new inodes. */
 
 
3639	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
3640		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3641
3642	ext4_get_inode_flags(ei);
3643	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3644	if (!(test_opt(inode->i_sb, NO_UID32))) {
3645		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3646		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3647/*
3648 * Fix up interoperability with old kernels. Otherwise, old inodes get
3649 * re-used with the upper 16 bits of the uid/gid intact
3650 */
3651		if (!ei->i_dtime) {
3652			raw_inode->i_uid_high =
3653				cpu_to_le16(high_16_bits(inode->i_uid));
3654			raw_inode->i_gid_high =
3655				cpu_to_le16(high_16_bits(inode->i_gid));
3656		} else {
3657			raw_inode->i_uid_high = 0;
3658			raw_inode->i_gid_high = 0;
3659		}
3660	} else {
3661		raw_inode->i_uid_low =
3662			cpu_to_le16(fs_high2lowuid(inode->i_uid));
3663		raw_inode->i_gid_low =
3664			cpu_to_le16(fs_high2lowgid(inode->i_gid));
3665		raw_inode->i_uid_high = 0;
3666		raw_inode->i_gid_high = 0;
3667	}
3668	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3669
3670	EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
3671	EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
3672	EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
3673	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
3674
3675	if (ext4_inode_blocks_set(handle, raw_inode, ei))
 
 
 
3676		goto out_brelse;
3677	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3678	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
3679	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3680	    cpu_to_le32(EXT4_OS_HURD))
3681		raw_inode->i_file_acl_high =
3682			cpu_to_le16(ei->i_file_acl >> 32);
3683	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
3684	ext4_isize_set(raw_inode, ei->i_disksize);
3685	if (ei->i_disksize > 0x7fffffffULL) {
3686		struct super_block *sb = inode->i_sb;
3687		if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
3688				EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
3689				EXT4_SB(sb)->s_es->s_rev_level ==
3690				cpu_to_le32(EXT4_GOOD_OLD_REV)) {
3691			/* If this is the first large file
3692			 * created, add a flag to the superblock.
3693			 */
3694			err = ext4_journal_get_write_access(handle,
3695					EXT4_SB(sb)->s_sbh);
3696			if (err)
3697				goto out_brelse;
3698			ext4_update_dynamic_rev(sb);
3699			EXT4_SET_RO_COMPAT_FEATURE(sb,
3700					EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
3701			sb->s_dirt = 1;
3702			ext4_handle_sync(handle);
3703			err = ext4_handle_dirty_metadata(handle, NULL,
3704					EXT4_SB(sb)->s_sbh);
3705		}
3706	}
3707	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3708	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3709		if (old_valid_dev(inode->i_rdev)) {
3710			raw_inode->i_block[0] =
3711				cpu_to_le32(old_encode_dev(inode->i_rdev));
3712			raw_inode->i_block[1] = 0;
3713		} else {
3714			raw_inode->i_block[0] = 0;
3715			raw_inode->i_block[1] =
3716				cpu_to_le32(new_encode_dev(inode->i_rdev));
3717			raw_inode->i_block[2] = 0;
3718		}
3719	} else
3720		for (block = 0; block < EXT4_N_BLOCKS; block++)
3721			raw_inode->i_block[block] = ei->i_data[block];
3722
3723	raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
3724	if (ei->i_extra_isize) {
3725		if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3726			raw_inode->i_version_hi =
3727			cpu_to_le32(inode->i_version >> 32);
3728		raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3729	}
3730
3731	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3732	rc = ext4_handle_dirty_metadata(handle, NULL, bh);
3733	if (!err)
3734		err = rc;
3735	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
3736
3737	ext4_update_inode_fsync_trans(handle, inode, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3738out_brelse:
3739	brelse(bh);
3740	ext4_std_error(inode->i_sb, err);
3741	return err;
3742}
3743
3744/*
3745 * ext4_write_inode()
3746 *
3747 * We are called from a few places:
3748 *
3749 * - Within generic_file_write() for O_SYNC files.
3750 *   Here, there will be no transaction running. We wait for any running
3751 *   trasnaction to commit.
3752 *
3753 * - Within sys_sync(), kupdate and such.
3754 *   We wait on commit, if tol to.
3755 *
3756 * - Within prune_icache() (PF_MEMALLOC == true)
3757 *   Here we simply return.  We can't afford to block kswapd on the
3758 *   journal commit.
3759 *
3760 * In all cases it is actually safe for us to return without doing anything,
3761 * because the inode has been copied into a raw inode buffer in
3762 * ext4_mark_inode_dirty().  This is a correctness thing for O_SYNC and for
3763 * knfsd.
3764 *
3765 * Note that we are absolutely dependent upon all inode dirtiers doing the
3766 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3767 * which we are interested.
3768 *
3769 * It would be a bug for them to not do this.  The code:
3770 *
3771 *	mark_inode_dirty(inode)
3772 *	stuff();
3773 *	inode->i_size = expr;
3774 *
3775 * is in error because a kswapd-driven write_inode() could occur while
3776 * `stuff()' is running, and the new i_size will be lost.  Plus the inode
3777 * will no longer be on the superblock's dirty inode list.
3778 */
3779int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
3780{
3781	int err;
3782
3783	if (current->flags & PF_MEMALLOC)
 
3784		return 0;
3785
 
 
 
3786	if (EXT4_SB(inode->i_sb)->s_journal) {
3787		if (ext4_journal_current_handle()) {
3788			jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3789			dump_stack();
3790			return -EIO;
3791		}
3792
3793		if (wbc->sync_mode != WB_SYNC_ALL)
 
 
 
 
 
3794			return 0;
3795
3796		err = ext4_force_commit(inode->i_sb);
 
3797	} else {
3798		struct ext4_iloc iloc;
3799
3800		err = __ext4_get_inode_loc(inode, &iloc, 0);
3801		if (err)
3802			return err;
3803		if (wbc->sync_mode == WB_SYNC_ALL)
 
 
 
 
3804			sync_dirty_buffer(iloc.bh);
3805		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
3806			EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
3807					 "IO error syncing inode");
3808			err = -EIO;
3809		}
3810		brelse(iloc.bh);
3811	}
3812	return err;
3813}
3814
3815/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3816 * ext4_setattr()
3817 *
3818 * Called from notify_change.
3819 *
3820 * We want to trap VFS attempts to truncate the file as soon as
3821 * possible.  In particular, we want to make sure that when the VFS
3822 * shrinks i_size, we put the inode on the orphan list and modify
3823 * i_disksize immediately, so that during the subsequent flushing of
3824 * dirty pages and freeing of disk blocks, we can guarantee that any
3825 * commit will leave the blocks being flushed in an unused state on
3826 * disk.  (On recovery, the inode will get truncated and the blocks will
3827 * be freed, so we have a strong guarantee that no future commit will
3828 * leave these blocks visible to the user.)
3829 *
3830 * Another thing we have to assure is that if we are in ordered mode
3831 * and inode is still attached to the committing transaction, we must
3832 * we start writeout of all the dirty pages which are being truncated.
3833 * This way we are sure that all the data written in the previous
3834 * transaction are already on disk (truncate waits for pages under
3835 * writeback).
3836 *
3837 * Called with inode->i_mutex down.
3838 */
3839int ext4_setattr(struct dentry *dentry, struct iattr *attr)
 
3840{
3841	struct inode *inode = dentry->d_inode;
3842	int error, rc = 0;
3843	int orphan = 0;
3844	const unsigned int ia_valid = attr->ia_valid;
 
 
 
 
 
 
 
 
 
 
 
 
3845
3846	error = inode_change_ok(inode, attr);
3847	if (error)
3848		return error;
3849
3850	if (is_quota_modification(inode, attr))
3851		dquot_initialize(inode);
3852	if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3853		(ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
 
 
 
 
 
 
 
 
 
 
 
 
3854		handle_t *handle;
3855
3856		/* (user+group)*(old+new) structure, inode write (sb,
3857		 * inode block, ? - but truncate inode update has it) */
3858		handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3859					EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
 
3860		if (IS_ERR(handle)) {
3861			error = PTR_ERR(handle);
3862			goto err_out;
3863		}
3864		error = dquot_transfer(inode, attr);
 
 
 
 
 
 
 
3865		if (error) {
3866			ext4_journal_stop(handle);
3867			return error;
3868		}
3869		/* Update corresponding info in inode so that everything is in
3870		 * one transaction */
3871		if (attr->ia_valid & ATTR_UID)
3872			inode->i_uid = attr->ia_uid;
3873		if (attr->ia_valid & ATTR_GID)
3874			inode->i_gid = attr->ia_gid;
3875		error = ext4_mark_inode_dirty(handle, inode);
3876		ext4_journal_stop(handle);
 
 
 
3877	}
3878
3879	if (attr->ia_valid & ATTR_SIZE) {
3880		inode_dio_wait(inode);
 
 
 
3881
3882		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3883			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3884
3885			if (attr->ia_size > sbi->s_bitmap_maxbytes)
3886				return -EFBIG;
 
 
 
 
3887		}
3888	}
3889
3890	if (S_ISREG(inode->i_mode) &&
3891	    attr->ia_valid & ATTR_SIZE &&
3892	    (attr->ia_size < inode->i_size)) {
3893		handle_t *handle;
3894
3895		handle = ext4_journal_start(inode, 3);
3896		if (IS_ERR(handle)) {
3897			error = PTR_ERR(handle);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3898			goto err_out;
3899		}
3900		if (ext4_handle_valid(handle)) {
3901			error = ext4_orphan_add(handle, inode);
3902			orphan = 1;
3903		}
3904		EXT4_I(inode)->i_disksize = attr->ia_size;
3905		rc = ext4_mark_inode_dirty(handle, inode);
3906		if (!error)
3907			error = rc;
3908		ext4_journal_stop(handle);
3909
3910		if (ext4_should_order_data(inode)) {
3911			error = ext4_begin_ordered_truncate(inode,
3912							    attr->ia_size);
3913			if (error) {
3914				/* Do as much error cleanup as possible */
3915				handle = ext4_journal_start(inode, 3);
3916				if (IS_ERR(handle)) {
3917					ext4_orphan_del(NULL, inode);
3918					goto err_out;
3919				}
3920				ext4_orphan_del(handle, inode);
3921				orphan = 0;
3922				ext4_journal_stop(handle);
3923				goto err_out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3924			}
3925		}
3926	}
3927
3928	if (attr->ia_valid & ATTR_SIZE) {
3929		if (attr->ia_size != i_size_read(inode)) {
3930			truncate_setsize(inode, attr->ia_size);
3931			ext4_truncate(inode);
3932		} else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3933			ext4_truncate(inode);
 
 
 
 
 
 
 
 
 
 
3934	}
3935
3936	if (!rc) {
3937		setattr_copy(inode, attr);
 
 
3938		mark_inode_dirty(inode);
3939	}
3940
3941	/*
3942	 * If the call to ext4_truncate failed to get a transaction handle at
3943	 * all, we need to clean up the in-core orphan list manually.
3944	 */
3945	if (orphan && inode->i_nlink)
3946		ext4_orphan_del(NULL, inode);
3947
3948	if (!rc && (ia_valid & ATTR_MODE))
3949		rc = ext4_acl_chmod(inode);
3950
3951err_out:
3952	ext4_std_error(inode->i_sb, error);
 
3953	if (!error)
3954		error = rc;
3955	return error;
3956}
3957
3958int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
3959		 struct kstat *stat)
3960{
3961	struct inode *inode;
3962	unsigned long delalloc_blocks;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3963
3964	inode = dentry->d_inode;
3965	generic_fillattr(inode, stat);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3966
3967	/*
3968	 * We can't update i_blocks if the block allocation is delayed
3969	 * otherwise in the case of system crash before the real block
3970	 * allocation is done, we will have i_blocks inconsistent with
3971	 * on-disk file blocks.
3972	 * We always keep i_blocks updated together with real
3973	 * allocation. But to not confuse with user, stat
3974	 * will return the blocks that include the delayed allocation
3975	 * blocks for this file.
3976	 */
3977	delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
3978
3979	stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
3980	return 0;
3981}
3982
3983static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 
3984{
3985	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3986		return ext4_ind_trans_blocks(inode, nrblocks, chunk);
3987	return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
3988}
3989
3990/*
3991 * Account for index blocks, block groups bitmaps and block group
3992 * descriptor blocks if modify datablocks and index blocks
3993 * worse case, the indexs blocks spread over different block groups
3994 *
3995 * If datablocks are discontiguous, they are possible to spread over
3996 * different block groups too. If they are contiuguous, with flexbg,
3997 * they could still across block group boundary.
3998 *
3999 * Also account for superblock, inode, quota and xattr blocks
4000 */
4001static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 
4002{
4003	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4004	int gdpblocks;
4005	int idxblocks;
4006	int ret = 0;
4007
4008	/*
4009	 * How many index blocks need to touch to modify nrblocks?
4010	 * The "Chunk" flag indicating whether the nrblocks is
4011	 * physically contiguous on disk
4012	 *
4013	 * For Direct IO and fallocate, they calls get_block to allocate
4014	 * one single extent at a time, so they could set the "Chunk" flag
4015	 */
4016	idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4017
4018	ret = idxblocks;
4019
4020	/*
4021	 * Now let's see how many group bitmaps and group descriptors need
4022	 * to account
4023	 */
4024	groups = idxblocks;
4025	if (chunk)
4026		groups += 1;
4027	else
4028		groups += nrblocks;
4029
4030	gdpblocks = groups;
4031	if (groups > ngroups)
4032		groups = ngroups;
4033	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4034		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4035
4036	/* bitmaps and block group descriptor blocks */
4037	ret += groups + gdpblocks;
4038
4039	/* Blocks for super block, inode, quota and xattr blocks */
4040	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4041
4042	return ret;
4043}
4044
4045/*
4046 * Calculate the total number of credits to reserve to fit
4047 * the modification of a single pages into a single transaction,
4048 * which may include multiple chunks of block allocations.
4049 *
4050 * This could be called via ext4_write_begin()
4051 *
4052 * We need to consider the worse case, when
4053 * one new block per extent.
4054 */
4055int ext4_writepage_trans_blocks(struct inode *inode)
4056{
4057	int bpp = ext4_journal_blocks_per_page(inode);
4058	int ret;
4059
4060	ret = ext4_meta_trans_blocks(inode, bpp, 0);
4061
4062	/* Account for data blocks for journalled mode */
4063	if (ext4_should_journal_data(inode))
4064		ret += bpp;
4065	return ret;
4066}
4067
4068/*
4069 * Calculate the journal credits for a chunk of data modification.
4070 *
4071 * This is called from DIO, fallocate or whoever calling
4072 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4073 *
4074 * journal buffers for data blocks are not included here, as DIO
4075 * and fallocate do no need to journal data buffers.
4076 */
4077int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4078{
4079	return ext4_meta_trans_blocks(inode, nrblocks, 1);
4080}
4081
4082/*
4083 * The caller must have previously called ext4_reserve_inode_write().
4084 * Give this, we know that the caller already has write access to iloc->bh.
4085 */
4086int ext4_mark_iloc_dirty(handle_t *handle,
4087			 struct inode *inode, struct ext4_iloc *iloc)
4088{
4089	int err = 0;
4090
4091	if (test_opt(inode->i_sb, I_VERSION))
4092		inode_inc_iversion(inode);
 
 
 
4093
4094	/* the do_update_inode consumes one bh->b_count */
4095	get_bh(iloc->bh);
4096
4097	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4098	err = ext4_do_update_inode(handle, inode, iloc);
4099	put_bh(iloc->bh);
4100	return err;
4101}
4102
4103/*
4104 * On success, We end up with an outstanding reference count against
4105 * iloc->bh.  This _must_ be cleaned up later.
4106 */
4107
4108int
4109ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4110			 struct ext4_iloc *iloc)
4111{
4112	int err;
4113
 
 
 
4114	err = ext4_get_inode_loc(inode, iloc);
4115	if (!err) {
4116		BUFFER_TRACE(iloc->bh, "get_write_access");
4117		err = ext4_journal_get_write_access(handle, iloc->bh);
 
4118		if (err) {
4119			brelse(iloc->bh);
4120			iloc->bh = NULL;
4121		}
4122	}
4123	ext4_std_error(inode->i_sb, err);
4124	return err;
4125}
4126
4127/*
4128 * Expand an inode by new_extra_isize bytes.
4129 * Returns 0 on success or negative error number on failure.
4130 */
4131static int ext4_expand_extra_isize(struct inode *inode,
4132				   unsigned int new_extra_isize,
4133				   struct ext4_iloc iloc,
4134				   handle_t *handle)
4135{
4136	struct ext4_inode *raw_inode;
4137	struct ext4_xattr_ibody_header *header;
 
 
 
4138
4139	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4140		return 0;
 
 
 
 
 
 
 
 
 
 
4141
4142	raw_inode = ext4_raw_inode(&iloc);
4143
4144	header = IHDR(inode, raw_inode);
4145
4146	/* No extended attributes present */
4147	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4148	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4149		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4150			new_extra_isize);
 
4151		EXT4_I(inode)->i_extra_isize = new_extra_isize;
4152		return 0;
4153	}
4154
 
 
 
 
 
 
 
 
4155	/* try to expand with EAs present */
4156	return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4157					  raw_inode, handle);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4158}
4159
4160/*
4161 * What we do here is to mark the in-core inode as clean with respect to inode
4162 * dirtiness (it may still be data-dirty).
4163 * This means that the in-core inode may be reaped by prune_icache
4164 * without having to perform any I/O.  This is a very good thing,
4165 * because *any* task may call prune_icache - even ones which
4166 * have a transaction open against a different journal.
4167 *
4168 * Is this cheating?  Not really.  Sure, we haven't written the
4169 * inode out, but prune_icache isn't a user-visible syncing function.
4170 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4171 * we start and wait on commits.
4172 *
4173 * Is this efficient/effective?  Well, we're being nice to the system
4174 * by cleaning up our inodes proactively so they can be reaped
4175 * without I/O.  But we are potentially leaving up to five seconds'
4176 * worth of inodes floating about which prune_icache wants us to
4177 * write out.  One way to fix that would be to get prune_icache()
4178 * to do a write_super() to free up some memory.  It has the desired
4179 * effect.
4180 */
4181int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
 
4182{
4183	struct ext4_iloc iloc;
4184	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4185	static unsigned int mnt_count;
4186	int err, ret;
4187
4188	might_sleep();
4189	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4190	err = ext4_reserve_inode_write(handle, inode, &iloc);
4191	if (ext4_handle_valid(handle) &&
4192	    EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4193	    !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4194		/*
4195		 * We need extra buffer credits since we may write into EA block
4196		 * with this same handle. If journal_extend fails, then it will
4197		 * only result in a minor loss of functionality for that inode.
4198		 * If this is felt to be critical, then e2fsck should be run to
4199		 * force a large enough s_min_extra_isize.
4200		 */
4201		if ((jbd2_journal_extend(handle,
4202			     EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4203			ret = ext4_expand_extra_isize(inode,
4204						      sbi->s_want_extra_isize,
4205						      iloc, handle);
4206			if (ret) {
4207				ext4_set_inode_state(inode,
4208						     EXT4_STATE_NO_EXPAND);
4209				if (mnt_count !=
4210					le16_to_cpu(sbi->s_es->s_mnt_count)) {
4211					ext4_warning(inode->i_sb,
4212					"Unable to expand inode %lu. Delete"
4213					" some EAs or run e2fsck.",
4214					inode->i_ino);
4215					mnt_count =
4216					  le16_to_cpu(sbi->s_es->s_mnt_count);
4217				}
4218			}
4219		}
4220	}
4221	if (!err)
4222		err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4223	return err;
4224}
4225
4226/*
4227 * ext4_dirty_inode() is called from __mark_inode_dirty()
4228 *
4229 * We're really interested in the case where a file is being extended.
4230 * i_size has been changed by generic_commit_write() and we thus need
4231 * to include the updated inode in the current transaction.
4232 *
4233 * Also, dquot_alloc_block() will always dirty the inode when blocks
4234 * are allocated to the file.
4235 *
4236 * If the inode is marked synchronous, we don't honour that here - doing
4237 * so would cause a commit on atime updates, which we don't bother doing.
4238 * We handle synchronous inodes at the highest possible level.
4239 */
4240void ext4_dirty_inode(struct inode *inode, int flags)
4241{
4242	handle_t *handle;
4243
4244	handle = ext4_journal_start(inode, 2);
4245	if (IS_ERR(handle))
4246		goto out;
4247
4248	ext4_mark_inode_dirty(handle, inode);
4249
4250	ext4_journal_stop(handle);
4251out:
4252	return;
4253}
4254
4255#if 0
4256/*
4257 * Bind an inode's backing buffer_head into this transaction, to prevent
4258 * it from being flushed to disk early.  Unlike
4259 * ext4_reserve_inode_write, this leaves behind no bh reference and
4260 * returns no iloc structure, so the caller needs to repeat the iloc
4261 * lookup to mark the inode dirty later.
4262 */
4263static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4264{
4265	struct ext4_iloc iloc;
4266
4267	int err = 0;
4268	if (handle) {
4269		err = ext4_get_inode_loc(inode, &iloc);
4270		if (!err) {
4271			BUFFER_TRACE(iloc.bh, "get_write_access");
4272			err = jbd2_journal_get_write_access(handle, iloc.bh);
4273			if (!err)
4274				err = ext4_handle_dirty_metadata(handle,
4275								 NULL,
4276								 iloc.bh);
4277			brelse(iloc.bh);
4278		}
4279	}
4280	ext4_std_error(inode->i_sb, err);
4281	return err;
4282}
4283#endif
4284
4285int ext4_change_inode_journal_flag(struct inode *inode, int val)
4286{
4287	journal_t *journal;
4288	handle_t *handle;
4289	int err;
 
4290
4291	/*
4292	 * We have to be very careful here: changing a data block's
4293	 * journaling status dynamically is dangerous.  If we write a
4294	 * data block to the journal, change the status and then delete
4295	 * that block, we risk forgetting to revoke the old log record
4296	 * from the journal and so a subsequent replay can corrupt data.
4297	 * So, first we make sure that the journal is empty and that
4298	 * nobody is changing anything.
4299	 */
4300
4301	journal = EXT4_JOURNAL(inode);
4302	if (!journal)
4303		return 0;
4304	if (is_journal_aborted(journal))
4305		return -EROFS;
4306
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4307	jbd2_journal_lock_updates(journal);
4308	jbd2_journal_flush(journal);
4309
4310	/*
4311	 * OK, there are no updates running now, and all cached data is
4312	 * synced to disk.  We are now in a completely consistent state
4313	 * which doesn't have anything in the journal, and we know that
4314	 * no filesystem updates are running, so it is safe to modify
4315	 * the inode's in-core data-journaling state flag now.
4316	 */
4317
4318	if (val)
4319		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4320	else
 
 
 
 
 
 
4321		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
 
4322	ext4_set_aops(inode);
4323
4324	jbd2_journal_unlock_updates(journal);
 
 
 
 
4325
4326	/* Finally we can mark the inode as dirty. */
4327
4328	handle = ext4_journal_start(inode, 1);
4329	if (IS_ERR(handle))
4330		return PTR_ERR(handle);
4331
 
 
4332	err = ext4_mark_inode_dirty(handle, inode);
4333	ext4_handle_sync(handle);
4334	ext4_journal_stop(handle);
4335	ext4_std_error(inode->i_sb, err);
4336
4337	return err;
4338}
4339
4340static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
 
4341{
4342	return !buffer_mapped(bh);
4343}
4344
4345int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4346{
 
4347	struct page *page = vmf->page;
4348	loff_t size;
4349	unsigned long len;
4350	int ret;
 
4351	struct file *file = vma->vm_file;
4352	struct inode *inode = file->f_path.dentry->d_inode;
4353	struct address_space *mapping = inode->i_mapping;
4354	handle_t *handle;
4355	get_block_t *get_block;
4356	int retries = 0;
4357
 
 
 
 
 
 
 
 
 
 
 
 
4358	/*
4359	 * This check is racy but catches the common case. We rely on
4360	 * __block_page_mkwrite() to do a reliable check.
 
 
4361	 */
4362	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 
 
4363	/* Delalloc case is easy... */
4364	if (test_opt(inode->i_sb, DELALLOC) &&
4365	    !ext4_should_journal_data(inode) &&
4366	    !ext4_nonda_switch(inode->i_sb)) {
4367		do {
4368			ret = __block_page_mkwrite(vma, vmf,
4369						   ext4_da_get_block_prep);
4370		} while (ret == -ENOSPC &&
4371		       ext4_should_retry_alloc(inode->i_sb, &retries));
4372		goto out_ret;
4373	}
4374
4375	lock_page(page);
4376	size = i_size_read(inode);
4377	/* Page got truncated from under us? */
4378	if (page->mapping != mapping || page_offset(page) > size) {
4379		unlock_page(page);
4380		ret = VM_FAULT_NOPAGE;
4381		goto out;
4382	}
4383
4384	if (page->index == size >> PAGE_CACHE_SHIFT)
4385		len = size & ~PAGE_CACHE_MASK;
4386	else
4387		len = PAGE_CACHE_SIZE;
4388	/*
4389	 * Return if we have all the buffers mapped. This avoids the need to do
4390	 * journal_start/journal_stop which can block and take a long time
 
 
 
4391	 */
4392	if (page_has_buffers(page)) {
4393		if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4394					ext4_bh_unmapped)) {
 
4395			/* Wait so that we don't change page under IO */
4396			wait_on_page_writeback(page);
4397			ret = VM_FAULT_LOCKED;
4398			goto out;
4399		}
4400	}
4401	unlock_page(page);
4402	/* OK, we need to fill the hole... */
4403	if (ext4_should_dioread_nolock(inode))
4404		get_block = ext4_get_block_write;
4405	else
4406		get_block = ext4_get_block;
4407retry_alloc:
4408	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
 
4409	if (IS_ERR(handle)) {
4410		ret = VM_FAULT_SIGBUS;
4411		goto out;
4412	}
4413	ret = __block_page_mkwrite(vma, vmf, get_block);
4414	if (!ret && ext4_should_journal_data(inode)) {
4415		if (walk_page_buffers(handle, page_buffers(page), 0,
4416			  PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4417			unlock_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4418			ret = VM_FAULT_SIGBUS;
4419			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
4420		}
4421		ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4422	}
4423	ext4_journal_stop(handle);
4424	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4425		goto retry_alloc;
4426out_ret:
4427	ret = block_page_mkwrite_return(ret);
4428out:
 
 
4429	return ret;
 
 
 
 
4430}
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/ext4/inode.c
   4 *
   5 * Copyright (C) 1992, 1993, 1994, 1995
   6 * Remy Card (card@masi.ibp.fr)
   7 * Laboratoire MASI - Institut Blaise Pascal
   8 * Universite Pierre et Marie Curie (Paris VI)
   9 *
  10 *  from
  11 *
  12 *  linux/fs/minix/inode.c
  13 *
  14 *  Copyright (C) 1991, 1992  Linus Torvalds
  15 *
  16 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  17 *	(jj@sunsite.ms.mff.cuni.cz)
  18 *
  19 *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  20 */
  21
 
  22#include <linux/fs.h>
  23#include <linux/mount.h>
  24#include <linux/time.h>
 
  25#include <linux/highuid.h>
  26#include <linux/pagemap.h>
  27#include <linux/dax.h>
  28#include <linux/quotaops.h>
  29#include <linux/string.h>
  30#include <linux/buffer_head.h>
  31#include <linux/writeback.h>
  32#include <linux/pagevec.h>
  33#include <linux/mpage.h>
  34#include <linux/namei.h>
  35#include <linux/uio.h>
  36#include <linux/bio.h>
  37#include <linux/workqueue.h>
  38#include <linux/kernel.h>
  39#include <linux/printk.h>
  40#include <linux/slab.h>
  41#include <linux/bitops.h>
  42#include <linux/iomap.h>
  43#include <linux/iversion.h>
  44
  45#include "ext4_jbd2.h"
  46#include "xattr.h"
  47#include "acl.h"
 
  48#include "truncate.h"
  49
  50#include <trace/events/ext4.h>
  51
  52static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
  53			      struct ext4_inode_info *ei)
  54{
  55	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  56	__u32 csum;
  57	__u16 dummy_csum = 0;
  58	int offset = offsetof(struct ext4_inode, i_checksum_lo);
  59	unsigned int csum_size = sizeof(dummy_csum);
  60
  61	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
  62	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
  63	offset += csum_size;
  64	csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
  65			   EXT4_GOOD_OLD_INODE_SIZE - offset);
  66
  67	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
  68		offset = offsetof(struct ext4_inode, i_checksum_hi);
  69		csum = ext4_chksum(sbi, csum, (__u8 *)raw +
  70				   EXT4_GOOD_OLD_INODE_SIZE,
  71				   offset - EXT4_GOOD_OLD_INODE_SIZE);
  72		if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
  73			csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
  74					   csum_size);
  75			offset += csum_size;
  76		}
  77		csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
  78				   EXT4_INODE_SIZE(inode->i_sb) - offset);
  79	}
  80
  81	return csum;
  82}
  83
  84static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
  85				  struct ext4_inode_info *ei)
  86{
  87	__u32 provided, calculated;
  88
  89	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  90	    cpu_to_le32(EXT4_OS_LINUX) ||
  91	    !ext4_has_metadata_csum(inode->i_sb))
  92		return 1;
  93
  94	provided = le16_to_cpu(raw->i_checksum_lo);
  95	calculated = ext4_inode_csum(inode, raw, ei);
  96	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  97	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
  98		provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
  99	else
 100		calculated &= 0xFFFF;
 101
 102	return provided == calculated;
 103}
 104
 105void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
 106			 struct ext4_inode_info *ei)
 107{
 108	__u32 csum;
 109
 110	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
 111	    cpu_to_le32(EXT4_OS_LINUX) ||
 112	    !ext4_has_metadata_csum(inode->i_sb))
 113		return;
 114
 115	csum = ext4_inode_csum(inode, raw, ei);
 116	raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
 117	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
 118	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
 119		raw->i_checksum_hi = cpu_to_le16(csum >> 16);
 120}
 121
 122static inline int ext4_begin_ordered_truncate(struct inode *inode,
 123					      loff_t new_size)
 124{
 125	trace_ext4_begin_ordered_truncate(inode, new_size);
 126	/*
 127	 * If jinode is zero, then we never opened the file for
 128	 * writing, so there's no need to call
 129	 * jbd2_journal_begin_ordered_truncate() since there's no
 130	 * outstanding writes we need to flush.
 131	 */
 132	if (!EXT4_I(inode)->jinode)
 133		return 0;
 134	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
 135						   EXT4_I(inode)->jinode,
 136						   new_size);
 137}
 138
 
 
 
 
 
 139static int __ext4_journalled_writepage(struct page *page, unsigned int len);
 140static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
 141				  int pextents);
 142
 143/*
 144 * Test whether an inode is a fast symlink.
 145 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
 146 */
 147int ext4_inode_is_fast_symlink(struct inode *inode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 148{
 149	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 150		int ea_blocks = EXT4_I(inode)->i_file_acl ?
 151				EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
 152
 153		if (ext4_has_inline_data(inode))
 154			return 0;
 
 
 
 
 
 
 
 
 
 
 155
 156		return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
 157	}
 158	return S_ISLNK(inode->i_mode) && inode->i_size &&
 159	       (inode->i_size < EXT4_N_BLOCKS * 4);
 160}
 161
 162/*
 163 * Called at the last iput() if i_nlink is zero.
 164 */
 165void ext4_evict_inode(struct inode *inode)
 166{
 167	handle_t *handle;
 168	int err;
 169	/*
 170	 * Credits for final inode cleanup and freeing:
 171	 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
 172	 * (xattr block freeing), bitmap, group descriptor (inode freeing)
 173	 */
 174	int extra_credits = 6;
 175	struct ext4_xattr_inode_array *ea_inode_array = NULL;
 176	bool freeze_protected = false;
 177
 178	trace_ext4_evict_inode(inode);
 179
 180	if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
 181		ext4_evict_ea_inode(inode);
 182	if (inode->i_nlink) {
 183		/*
 184		 * When journalling data dirty buffers are tracked only in the
 185		 * journal. So although mm thinks everything is clean and
 186		 * ready for reaping the inode might still have some pages to
 187		 * write in the running transaction or waiting to be
 188		 * checkpointed. Thus calling jbd2_journal_invalidate_folio()
 189		 * (via truncate_inode_pages()) to discard these buffers can
 190		 * cause data loss. Also even if we did not discard these
 191		 * buffers, we would have no way to find them after the inode
 192		 * is reaped and thus user could see stale data if he tries to
 193		 * read them before the transaction is checkpointed. So be
 194		 * careful and force everything to disk here... We use
 195		 * ei->i_datasync_tid to store the newest transaction
 196		 * containing inode's data.
 197		 *
 198		 * Note that directories do not have this problem because they
 199		 * don't use page cache.
 200		 */
 201		if (inode->i_ino != EXT4_JOURNAL_INO &&
 202		    ext4_should_journal_data(inode) &&
 203		    S_ISREG(inode->i_mode) && inode->i_data.nrpages) {
 204			journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 205			tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
 206
 207			jbd2_complete_transaction(journal, commit_tid);
 
 208			filemap_write_and_wait(&inode->i_data);
 209		}
 210		truncate_inode_pages_final(&inode->i_data);
 211
 212		goto no_delete;
 213	}
 214
 215	if (is_bad_inode(inode))
 216		goto no_delete;
 217	dquot_initialize(inode);
 218
 219	if (ext4_should_order_data(inode))
 220		ext4_begin_ordered_truncate(inode, 0);
 221	truncate_inode_pages_final(&inode->i_data);
 222
 223	/*
 224	 * For inodes with journalled data, transaction commit could have
 225	 * dirtied the inode. And for inodes with dioread_nolock, unwritten
 226	 * extents converting worker could merge extents and also have dirtied
 227	 * the inode. Flush worker is ignoring it because of I_FREEING flag but
 228	 * we still need to remove the inode from the writeback lists.
 229	 */
 230	if (!list_empty_careful(&inode->i_io_list))
 231		inode_io_list_del(inode);
 232
 233	/*
 234	 * Protect us against freezing - iput() caller didn't have to have any
 235	 * protection against it. When we are in a running transaction though,
 236	 * we are already protected against freezing and we cannot grab further
 237	 * protection due to lock ordering constraints.
 238	 */
 239	if (!ext4_journal_current_handle()) {
 240		sb_start_intwrite(inode->i_sb);
 241		freeze_protected = true;
 242	}
 243
 244	if (!IS_NOQUOTA(inode))
 245		extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
 246
 247	/*
 248	 * Block bitmap, group descriptor, and inode are accounted in both
 249	 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
 250	 */
 251	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
 252			 ext4_blocks_for_truncate(inode) + extra_credits - 3);
 253	if (IS_ERR(handle)) {
 254		ext4_std_error(inode->i_sb, PTR_ERR(handle));
 255		/*
 256		 * If we're going to skip the normal cleanup, we still need to
 257		 * make sure that the in-core orphan linked list is properly
 258		 * cleaned up.
 259		 */
 260		ext4_orphan_del(NULL, inode);
 261		if (freeze_protected)
 262			sb_end_intwrite(inode->i_sb);
 263		goto no_delete;
 264	}
 265
 266	if (IS_SYNC(inode))
 267		ext4_handle_sync(handle);
 268
 269	/*
 270	 * Set inode->i_size to 0 before calling ext4_truncate(). We need
 271	 * special handling of symlinks here because i_size is used to
 272	 * determine whether ext4_inode_info->i_data contains symlink data or
 273	 * block mappings. Setting i_size to 0 will remove its fast symlink
 274	 * status. Erase i_data so that it becomes a valid empty block map.
 275	 */
 276	if (ext4_inode_is_fast_symlink(inode))
 277		memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
 278	inode->i_size = 0;
 279	err = ext4_mark_inode_dirty(handle, inode);
 280	if (err) {
 281		ext4_warning(inode->i_sb,
 282			     "couldn't mark inode dirty (err %d)", err);
 283		goto stop_handle;
 284	}
 285	if (inode->i_blocks) {
 286		err = ext4_truncate(inode);
 287		if (err) {
 288			ext4_error_err(inode->i_sb, -err,
 289				       "couldn't truncate inode %lu (err %d)",
 290				       inode->i_ino, err);
 291			goto stop_handle;
 
 
 
 
 
 
 
 
 
 
 
 
 
 292		}
 293	}
 294
 295	/* Remove xattr references. */
 296	err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
 297				      extra_credits);
 298	if (err) {
 299		ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
 300stop_handle:
 301		ext4_journal_stop(handle);
 302		ext4_orphan_del(NULL, inode);
 303		if (freeze_protected)
 304			sb_end_intwrite(inode->i_sb);
 305		ext4_xattr_inode_array_free(ea_inode_array);
 306		goto no_delete;
 307	}
 308
 309	/*
 310	 * Kill off the orphan record which ext4_truncate created.
 311	 * AKPM: I think this can be inside the above `if'.
 312	 * Note that ext4_orphan_del() has to be able to cope with the
 313	 * deletion of a non-existent orphan - this is because we don't
 314	 * know if ext4_truncate() actually created an orphan record.
 315	 * (Well, we could do this if we need to, but heck - it works)
 316	 */
 317	ext4_orphan_del(handle, inode);
 318	EXT4_I(inode)->i_dtime	= (__u32)ktime_get_real_seconds();
 319
 320	/*
 321	 * One subtle ordering requirement: if anything has gone wrong
 322	 * (transaction abort, IO errors, whatever), then we can still
 323	 * do these next steps (the fs will already have been marked as
 324	 * having errors), but we can't free the inode if the mark_dirty
 325	 * fails.
 326	 */
 327	if (ext4_mark_inode_dirty(handle, inode))
 328		/* If that failed, just do the required in-core inode clear. */
 329		ext4_clear_inode(inode);
 330	else
 331		ext4_free_inode(handle, inode);
 332	ext4_journal_stop(handle);
 333	if (freeze_protected)
 334		sb_end_intwrite(inode->i_sb);
 335	ext4_xattr_inode_array_free(ea_inode_array);
 336	return;
 337no_delete:
 338	/*
 339	 * Check out some where else accidentally dirty the evicting inode,
 340	 * which may probably cause inode use-after-free issues later.
 341	 */
 342	WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
 343
 344	if (!list_empty(&EXT4_I(inode)->i_fc_list))
 345		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
 346	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
 347}
 348
 349#ifdef CONFIG_QUOTA
 350qsize_t *ext4_get_reserved_space(struct inode *inode)
 351{
 352	return &EXT4_I(inode)->i_reserved_quota;
 353}
 354#endif
 355
 356/*
 
 
 
 
 
 
 
 
 
 
 
 
 357 * Called with i_data_sem down, which is important since we can call
 358 * ext4_discard_preallocations() from here.
 359 */
 360void ext4_da_update_reserve_space(struct inode *inode,
 361					int used, int quota_claim)
 362{
 363	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 364	struct ext4_inode_info *ei = EXT4_I(inode);
 365
 366	spin_lock(&ei->i_block_reservation_lock);
 367	trace_ext4_da_update_reserve_space(inode, used, quota_claim);
 368	if (unlikely(used > ei->i_reserved_data_blocks)) {
 369		ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
 370			 "with only %d reserved data blocks",
 371			 __func__, inode->i_ino, used,
 372			 ei->i_reserved_data_blocks);
 373		WARN_ON(1);
 374		used = ei->i_reserved_data_blocks;
 375	}
 376
 377	/* Update per-inode reservations */
 378	ei->i_reserved_data_blocks -= used;
 379	percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
 
 
 
 380
 381	spin_unlock(&ei->i_block_reservation_lock);
 
 
 
 
 
 
 
 
 
 
 
 382
 383	/* Update quota subsystem for data blocks */
 384	if (quota_claim)
 385		dquot_claim_block(inode, EXT4_C2B(sbi, used));
 386	else {
 387		/*
 388		 * We did fallocate with an offset that is already delayed
 389		 * allocated. So on delayed allocated writeback we should
 390		 * not re-claim the quota for fallocated blocks.
 391		 */
 392		dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
 393	}
 394
 395	/*
 396	 * If we have done all the pending block allocations and if
 397	 * there aren't any writers on the inode, we can discard the
 398	 * inode's preallocations.
 399	 */
 400	if ((ei->i_reserved_data_blocks == 0) &&
 401	    !inode_is_open_for_write(inode))
 402		ext4_discard_preallocations(inode, 0);
 403}
 404
 405static int __check_block_validity(struct inode *inode, const char *func,
 406				unsigned int line,
 407				struct ext4_map_blocks *map)
 408{
 409	if (ext4_has_feature_journal(inode->i_sb) &&
 410	    (inode->i_ino ==
 411	     le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
 412		return 0;
 413	if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
 414		ext4_error_inode(inode, func, line, map->m_pblk,
 415				 "lblock %lu mapped to illegal pblock %llu "
 416				 "(length %d)", (unsigned long) map->m_lblk,
 417				 map->m_pblk, map->m_len);
 418		return -EFSCORRUPTED;
 419	}
 420	return 0;
 421}
 422
 423int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
 424		       ext4_lblk_t len)
 425{
 426	int ret;
 427
 428	if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
 429		return fscrypt_zeroout_range(inode, lblk, pblk, len);
 430
 431	ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
 432	if (ret > 0)
 433		ret = 0;
 434
 435	return ret;
 436}
 437
 438#define check_block_validity(inode, map)	\
 439	__check_block_validity((inode), __func__, __LINE__, (map))
 440
 441#ifdef ES_AGGRESSIVE_TEST
 442static void ext4_map_blocks_es_recheck(handle_t *handle,
 443				       struct inode *inode,
 444				       struct ext4_map_blocks *es_map,
 445				       struct ext4_map_blocks *map,
 446				       int flags)
 447{
 448	int retval;
 
 
 
 
 449
 450	map->m_flags = 0;
 451	/*
 452	 * There is a race window that the result is not the same.
 453	 * e.g. xfstests #223 when dioread_nolock enables.  The reason
 454	 * is that we lookup a block mapping in extent status tree with
 455	 * out taking i_data_sem.  So at the time the unwritten extent
 456	 * could be converted.
 457	 */
 458	down_read(&EXT4_I(inode)->i_data_sem);
 459	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 460		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 461	} else {
 462		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 463	}
 464	up_read((&EXT4_I(inode)->i_data_sem));
 465
 466	/*
 467	 * We don't check m_len because extent will be collpased in status
 468	 * tree.  So the m_len might not equal.
 469	 */
 470	if (es_map->m_lblk != map->m_lblk ||
 471	    es_map->m_flags != map->m_flags ||
 472	    es_map->m_pblk != map->m_pblk) {
 473		printk("ES cache assertion failed for inode: %lu "
 474		       "es_cached ex [%d/%d/%llu/%x] != "
 475		       "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
 476		       inode->i_ino, es_map->m_lblk, es_map->m_len,
 477		       es_map->m_pblk, es_map->m_flags, map->m_lblk,
 478		       map->m_len, map->m_pblk, map->m_flags,
 479		       retval, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 480	}
 
 481}
 482#endif /* ES_AGGRESSIVE_TEST */
 483
 484/*
 485 * The ext4_map_blocks() function tries to look up the requested blocks,
 486 * and returns if the blocks are already mapped.
 487 *
 488 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 489 * and store the allocated blocks in the result buffer head and mark it
 490 * mapped.
 491 *
 492 * If file type is extents based, it will call ext4_ext_map_blocks(),
 493 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 494 * based files
 495 *
 496 * On success, it returns the number of blocks being mapped or allocated.  if
 497 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
 498 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
 
 499 *
 500 * It returns 0 if plain look up failed (blocks have not been allocated), in
 501 * that case, @map is returned as unmapped but we still do fill map->m_len to
 502 * indicate the length of a hole starting at map->m_lblk.
 503 *
 504 * It returns the error in case of allocation failure.
 505 */
 506int ext4_map_blocks(handle_t *handle, struct inode *inode,
 507		    struct ext4_map_blocks *map, int flags)
 508{
 509	struct extent_status es;
 510	int retval;
 511	int ret = 0;
 512#ifdef ES_AGGRESSIVE_TEST
 513	struct ext4_map_blocks orig_map;
 514
 515	memcpy(&orig_map, map, sizeof(*map));
 516#endif
 517
 518	map->m_flags = 0;
 519	ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
 520		  flags, map->m_len, (unsigned long) map->m_lblk);
 521
 522	/*
 523	 * ext4_map_blocks returns an int, and m_len is an unsigned int
 524	 */
 525	if (unlikely(map->m_len > INT_MAX))
 526		map->m_len = INT_MAX;
 527
 528	/* We can handle the block number less than EXT_MAX_BLOCKS */
 529	if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
 530		return -EFSCORRUPTED;
 531
 532	/* Lookup extent status tree firstly */
 533	if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
 534	    ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
 535		if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
 536			map->m_pblk = ext4_es_pblock(&es) +
 537					map->m_lblk - es.es_lblk;
 538			map->m_flags |= ext4_es_is_written(&es) ?
 539					EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
 540			retval = es.es_len - (map->m_lblk - es.es_lblk);
 541			if (retval > map->m_len)
 542				retval = map->m_len;
 543			map->m_len = retval;
 544		} else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
 545			map->m_pblk = 0;
 546			retval = es.es_len - (map->m_lblk - es.es_lblk);
 547			if (retval > map->m_len)
 548				retval = map->m_len;
 549			map->m_len = retval;
 550			retval = 0;
 551		} else {
 552			BUG();
 553		}
 554
 555		if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
 556			return retval;
 557#ifdef ES_AGGRESSIVE_TEST
 558		ext4_map_blocks_es_recheck(handle, inode, map,
 559					   &orig_map, flags);
 560#endif
 561		goto found;
 562	}
 563	/*
 564	 * In the query cache no-wait mode, nothing we can do more if we
 565	 * cannot find extent in the cache.
 566	 */
 567	if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
 568		return 0;
 569
 570	/*
 571	 * Try to see if we can get the block without requesting a new
 572	 * file system block.
 573	 */
 574	down_read(&EXT4_I(inode)->i_data_sem);
 575	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 576		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 577	} else {
 578		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 579	}
 580	if (retval > 0) {
 581		unsigned int status;
 582
 583		if (unlikely(retval != map->m_len)) {
 584			ext4_warning(inode->i_sb,
 585				     "ES len assertion failed for inode "
 586				     "%lu: retval %d != map->m_len %d",
 587				     inode->i_ino, retval, map->m_len);
 588			WARN_ON(1);
 589		}
 590
 591		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 592				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 593		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
 594		    !(status & EXTENT_STATUS_WRITTEN) &&
 595		    ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
 596				       map->m_lblk + map->m_len - 1))
 597			status |= EXTENT_STATUS_DELAYED;
 598		ret = ext4_es_insert_extent(inode, map->m_lblk,
 599					    map->m_len, map->m_pblk, status);
 600		if (ret < 0)
 601			retval = ret;
 602	}
 603	up_read((&EXT4_I(inode)->i_data_sem));
 604
 605found:
 606	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 607		ret = check_block_validity(inode, map);
 608		if (ret != 0)
 609			return ret;
 610	}
 611
 612	/* If it is only a block(s) look up */
 613	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 614		return retval;
 615
 616	/*
 617	 * Returns if the blocks have already allocated
 618	 *
 619	 * Note that if blocks have been preallocated
 620	 * ext4_ext_get_block() returns the create = 0
 621	 * with buffer head unmapped.
 622	 */
 623	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 624		/*
 625		 * If we need to convert extent to unwritten
 626		 * we continue and do the actual work in
 627		 * ext4_ext_map_blocks()
 628		 */
 629		if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
 630			return retval;
 631
 632	/*
 633	 * Here we clear m_flags because after allocating an new extent,
 634	 * it will be set again.
 
 
 
 
 
 
 635	 */
 636	map->m_flags &= ~EXT4_MAP_FLAGS;
 637
 638	/*
 639	 * New blocks allocate and/or writing to unwritten extent
 640	 * will possibly result in updating i_data, so we take
 641	 * the write lock of i_data_sem, and call get_block()
 642	 * with create == 1 flag.
 643	 */
 644	down_write(&EXT4_I(inode)->i_data_sem);
 645
 646	/*
 
 
 
 
 
 
 
 
 647	 * We need to check for EXT4 here because migrate
 648	 * could have changed the inode type in between
 649	 */
 650	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 651		retval = ext4_ext_map_blocks(handle, inode, map, flags);
 652	} else {
 653		retval = ext4_ind_map_blocks(handle, inode, map, flags);
 654
 655		if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
 656			/*
 657			 * We allocated new blocks which will result in
 658			 * i_data's format changing.  Force the migrate
 659			 * to fail by clearing migrate flags
 660			 */
 661			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 662		}
 663
 664		/*
 665		 * Update reserved blocks/metadata blocks after successful
 666		 * block allocation which had been deferred till now. We don't
 667		 * support fallocate for non extent files. So we can update
 668		 * reserve space here.
 669		 */
 670		if ((retval > 0) &&
 671			(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
 672			ext4_da_update_reserve_space(inode, retval, 1);
 673	}
 
 
 674
 675	if (retval > 0) {
 676		unsigned int status;
 677
 678		if (unlikely(retval != map->m_len)) {
 679			ext4_warning(inode->i_sb,
 680				     "ES len assertion failed for inode "
 681				     "%lu: retval %d != map->m_len %d",
 682				     inode->i_ino, retval, map->m_len);
 683			WARN_ON(1);
 684		}
 685
 686		/*
 687		 * We have to zeroout blocks before inserting them into extent
 688		 * status tree. Otherwise someone could look them up there and
 689		 * use them before they are really zeroed. We also have to
 690		 * unmap metadata before zeroing as otherwise writeback can
 691		 * overwrite zeros with stale data from block device.
 692		 */
 693		if (flags & EXT4_GET_BLOCKS_ZERO &&
 694		    map->m_flags & EXT4_MAP_MAPPED &&
 695		    map->m_flags & EXT4_MAP_NEW) {
 696			ret = ext4_issue_zeroout(inode, map->m_lblk,
 697						 map->m_pblk, map->m_len);
 698			if (ret) {
 699				retval = ret;
 700				goto out_sem;
 701			}
 702		}
 703
 704		/*
 705		 * If the extent has been zeroed out, we don't need to update
 706		 * extent status tree.
 707		 */
 708		if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
 709		    ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
 710			if (ext4_es_is_written(&es))
 711				goto out_sem;
 712		}
 713		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 714				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 715		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
 716		    !(status & EXTENT_STATUS_WRITTEN) &&
 717		    ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
 718				       map->m_lblk + map->m_len - 1))
 719			status |= EXTENT_STATUS_DELAYED;
 720		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
 721					    map->m_pblk, status);
 722		if (ret < 0) {
 723			retval = ret;
 724			goto out_sem;
 725		}
 726	}
 727
 728out_sem:
 729	up_write((&EXT4_I(inode)->i_data_sem));
 730	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 731		ret = check_block_validity(inode, map);
 732		if (ret != 0)
 733			return ret;
 734
 735		/*
 736		 * Inodes with freshly allocated blocks where contents will be
 737		 * visible after transaction commit must be on transaction's
 738		 * ordered data list.
 739		 */
 740		if (map->m_flags & EXT4_MAP_NEW &&
 741		    !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
 742		    !(flags & EXT4_GET_BLOCKS_ZERO) &&
 743		    !ext4_is_quota_file(inode) &&
 744		    ext4_should_order_data(inode)) {
 745			loff_t start_byte =
 746				(loff_t)map->m_lblk << inode->i_blkbits;
 747			loff_t length = (loff_t)map->m_len << inode->i_blkbits;
 748
 749			if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
 750				ret = ext4_jbd2_inode_add_wait(handle, inode,
 751						start_byte, length);
 752			else
 753				ret = ext4_jbd2_inode_add_write(handle, inode,
 754						start_byte, length);
 755			if (ret)
 756				return ret;
 757		}
 758	}
 759	if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
 760				map->m_flags & EXT4_MAP_MAPPED))
 761		ext4_fc_track_range(handle, inode, map->m_lblk,
 762					map->m_lblk + map->m_len - 1);
 763	if (retval < 0)
 764		ext_debug(inode, "failed with err %d\n", retval);
 765	return retval;
 766}
 767
 768/*
 769 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
 770 * we have to be careful as someone else may be manipulating b_state as well.
 771 */
 772static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
 773{
 774	unsigned long old_state;
 775	unsigned long new_state;
 776
 777	flags &= EXT4_MAP_FLAGS;
 778
 779	/* Dummy buffer_head? Set non-atomically. */
 780	if (!bh->b_page) {
 781		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
 782		return;
 783	}
 784	/*
 785	 * Someone else may be modifying b_state. Be careful! This is ugly but
 786	 * once we get rid of using bh as a container for mapping information
 787	 * to pass to / from get_block functions, this can go away.
 788	 */
 789	do {
 790		old_state = READ_ONCE(bh->b_state);
 791		new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
 792	} while (unlikely(
 793		 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
 794}
 795
 796static int _ext4_get_block(struct inode *inode, sector_t iblock,
 797			   struct buffer_head *bh, int flags)
 798{
 
 799	struct ext4_map_blocks map;
 800	int ret = 0;
 801
 802	if (ext4_has_inline_data(inode))
 803		return -ERANGE;
 804
 805	map.m_lblk = iblock;
 806	map.m_len = bh->b_size >> inode->i_blkbits;
 807
 808	ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
 809			      flags);
 
 
 
 
 
 
 
 
 
 
 
 
 810	if (ret > 0) {
 811		map_bh(bh, inode->i_sb, map.m_pblk);
 812		ext4_update_bh_state(bh, map.m_flags);
 813		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 814		ret = 0;
 815	} else if (ret == 0) {
 816		/* hole case, need to fill in bh->b_size */
 817		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 818	}
 
 
 819	return ret;
 820}
 821
 822int ext4_get_block(struct inode *inode, sector_t iblock,
 823		   struct buffer_head *bh, int create)
 824{
 825	return _ext4_get_block(inode, iblock, bh,
 826			       create ? EXT4_GET_BLOCKS_CREATE : 0);
 827}
 828
 829/*
 830 * Get block function used when preparing for buffered write if we require
 831 * creating an unwritten extent if blocks haven't been allocated.  The extent
 832 * will be converted to written after the IO is complete.
 833 */
 834int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
 835			     struct buffer_head *bh_result, int create)
 836{
 837	ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
 838		   inode->i_ino, create);
 839	return _ext4_get_block(inode, iblock, bh_result,
 840			       EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
 841}
 842
 843/* Maximum number of blocks we map for direct IO at once. */
 844#define DIO_MAX_BLOCKS 4096
 845
 846/*
 847 * `handle' can be NULL if create is zero
 848 */
 849struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 850				ext4_lblk_t block, int map_flags)
 851{
 852	struct ext4_map_blocks map;
 853	struct buffer_head *bh;
 854	int create = map_flags & EXT4_GET_BLOCKS_CREATE;
 855	bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
 856	int err;
 857
 858	ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 859		    || handle != NULL || create == 0);
 860	ASSERT(create == 0 || !nowait);
 861
 862	map.m_lblk = block;
 863	map.m_len = 1;
 864	err = ext4_map_blocks(handle, inode, &map, map_flags);
 
 865
 866	if (err == 0)
 867		return create ? ERR_PTR(-ENOSPC) : NULL;
 868	if (err < 0)
 869		return ERR_PTR(err);
 870
 871	if (nowait)
 872		return sb_find_get_block(inode->i_sb, map.m_pblk);
 873
 874	bh = sb_getblk(inode->i_sb, map.m_pblk);
 875	if (unlikely(!bh))
 876		return ERR_PTR(-ENOMEM);
 
 
 877	if (map.m_flags & EXT4_MAP_NEW) {
 878		ASSERT(create != 0);
 879		ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
 880			    || (handle != NULL));
 881
 882		/*
 883		 * Now that we do not always journal data, we should
 884		 * keep in mind whether this should always journal the
 885		 * new buffer as metadata.  For now, regular file
 886		 * writes use ext4_get_block instead, so it's not a
 887		 * problem.
 888		 */
 889		lock_buffer(bh);
 890		BUFFER_TRACE(bh, "call get_create_access");
 891		err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
 892						     EXT4_JTR_NONE);
 893		if (unlikely(err)) {
 894			unlock_buffer(bh);
 895			goto errout;
 896		}
 897		if (!buffer_uptodate(bh)) {
 898			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 899			set_buffer_uptodate(bh);
 900		}
 901		unlock_buffer(bh);
 902		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 903		err = ext4_handle_dirty_metadata(handle, inode, bh);
 904		if (unlikely(err))
 905			goto errout;
 906	} else
 907		BUFFER_TRACE(bh, "not a new buffer");
 
 
 
 
 
 
 908	return bh;
 909errout:
 910	brelse(bh);
 911	return ERR_PTR(err);
 912}
 913
 914struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 915			       ext4_lblk_t block, int map_flags)
 916{
 917	struct buffer_head *bh;
 918	int ret;
 919
 920	bh = ext4_getblk(handle, inode, block, map_flags);
 921	if (IS_ERR(bh))
 922		return bh;
 923	if (!bh || ext4_buffer_uptodate(bh))
 924		return bh;
 925
 926	ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
 927	if (ret) {
 928		put_bh(bh);
 929		return ERR_PTR(ret);
 930	}
 931	return bh;
 932}
 933
 934/* Read a contiguous batch of blocks. */
 935int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
 936		     bool wait, struct buffer_head **bhs)
 937{
 938	int i, err;
 939
 940	for (i = 0; i < bh_count; i++) {
 941		bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
 942		if (IS_ERR(bhs[i])) {
 943			err = PTR_ERR(bhs[i]);
 944			bh_count = i;
 945			goto out_brelse;
 946		}
 947	}
 948
 949	for (i = 0; i < bh_count; i++)
 950		/* Note that NULL bhs[i] is valid because of holes. */
 951		if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
 952			ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
 953
 954	if (!wait)
 955		return 0;
 956
 957	for (i = 0; i < bh_count; i++)
 958		if (bhs[i])
 959			wait_on_buffer(bhs[i]);
 960
 961	for (i = 0; i < bh_count; i++) {
 962		if (bhs[i] && !buffer_uptodate(bhs[i])) {
 963			err = -EIO;
 964			goto out_brelse;
 965		}
 966	}
 967	return 0;
 968
 969out_brelse:
 970	for (i = 0; i < bh_count; i++) {
 971		brelse(bhs[i]);
 972		bhs[i] = NULL;
 973	}
 974	return err;
 975}
 976
 977int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
 978			   struct buffer_head *head,
 979			   unsigned from,
 980			   unsigned to,
 981			   int *partial,
 982			   int (*fn)(handle_t *handle, struct inode *inode,
 983				     struct buffer_head *bh))
 984{
 985	struct buffer_head *bh;
 986	unsigned block_start, block_end;
 987	unsigned blocksize = head->b_size;
 988	int err, ret = 0;
 989	struct buffer_head *next;
 990
 991	for (bh = head, block_start = 0;
 992	     ret == 0 && (bh != head || !block_start);
 993	     block_start = block_end, bh = next) {
 994		next = bh->b_this_page;
 995		block_end = block_start + blocksize;
 996		if (block_end <= from || block_start >= to) {
 997			if (partial && !buffer_uptodate(bh))
 998				*partial = 1;
 999			continue;
1000		}
1001		err = (*fn)(handle, inode, bh);
1002		if (!ret)
1003			ret = err;
1004	}
1005	return ret;
1006}
1007
1008/*
1009 * To preserve ordering, it is essential that the hole instantiation and
1010 * the data write be encapsulated in a single transaction.  We cannot
1011 * close off a transaction and start a new one between the ext4_get_block()
1012 * and the commit_write().  So doing the jbd2_journal_start at the start of
1013 * prepare_write() is the right place.
1014 *
1015 * Also, this function can nest inside ext4_writepage().  In that case, we
1016 * *know* that ext4_writepage() has generated enough buffer credits to do the
1017 * whole page.  So we won't block on the journal in that case, which is good,
1018 * because the caller may be PF_MEMALLOC.
 
1019 *
1020 * By accident, ext4 can be reentered when a transaction is open via
1021 * quota file writes.  If we were to commit the transaction while thus
1022 * reentered, there can be a deadlock - we would be holding a quota
1023 * lock, and the commit would never complete if another thread had a
1024 * transaction open and was blocking on the quota lock - a ranking
1025 * violation.
1026 *
1027 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1028 * will _not_ run commit under these circumstances because handle->h_ref
1029 * is elevated.  We'll still have enough credits for the tiny quotafile
1030 * write.
1031 */
1032int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1033				struct buffer_head *bh)
1034{
1035	int dirty = buffer_dirty(bh);
1036	int ret;
1037
1038	if (!buffer_mapped(bh) || buffer_freed(bh))
1039		return 0;
1040	/*
1041	 * __block_write_begin() could have dirtied some buffers. Clean
1042	 * the dirty bit as jbd2_journal_get_write_access() could complain
1043	 * otherwise about fs integrity issues. Setting of the dirty bit
1044	 * by __block_write_begin() isn't a real problem here as we clear
1045	 * the bit before releasing a page lock and thus writeback cannot
1046	 * ever write the buffer.
1047	 */
1048	if (dirty)
1049		clear_buffer_dirty(bh);
1050	BUFFER_TRACE(bh, "get write access");
1051	ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1052					    EXT4_JTR_NONE);
1053	if (!ret && dirty)
1054		ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1055	return ret;
1056}
1057
1058#ifdef CONFIG_FS_ENCRYPTION
1059static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1060				  get_block_t *get_block)
1061{
1062	unsigned from = pos & (PAGE_SIZE - 1);
1063	unsigned to = from + len;
1064	struct inode *inode = page->mapping->host;
1065	unsigned block_start, block_end;
1066	sector_t block;
1067	int err = 0;
1068	unsigned blocksize = inode->i_sb->s_blocksize;
1069	unsigned bbits;
1070	struct buffer_head *bh, *head, *wait[2];
1071	int nr_wait = 0;
1072	int i;
1073
1074	BUG_ON(!PageLocked(page));
1075	BUG_ON(from > PAGE_SIZE);
1076	BUG_ON(to > PAGE_SIZE);
1077	BUG_ON(from > to);
1078
1079	if (!page_has_buffers(page))
1080		create_empty_buffers(page, blocksize, 0);
1081	head = page_buffers(page);
1082	bbits = ilog2(blocksize);
1083	block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1084
1085	for (bh = head, block_start = 0; bh != head || !block_start;
1086	    block++, block_start = block_end, bh = bh->b_this_page) {
1087		block_end = block_start + blocksize;
1088		if (block_end <= from || block_start >= to) {
1089			if (PageUptodate(page)) {
1090				set_buffer_uptodate(bh);
1091			}
1092			continue;
1093		}
1094		if (buffer_new(bh))
1095			clear_buffer_new(bh);
1096		if (!buffer_mapped(bh)) {
1097			WARN_ON(bh->b_size != blocksize);
1098			err = get_block(inode, block, bh, 1);
1099			if (err)
1100				break;
1101			if (buffer_new(bh)) {
1102				if (PageUptodate(page)) {
1103					clear_buffer_new(bh);
1104					set_buffer_uptodate(bh);
1105					mark_buffer_dirty(bh);
1106					continue;
1107				}
1108				if (block_end > to || block_start < from)
1109					zero_user_segments(page, to, block_end,
1110							   block_start, from);
1111				continue;
1112			}
1113		}
1114		if (PageUptodate(page)) {
1115			set_buffer_uptodate(bh);
1116			continue;
1117		}
1118		if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1119		    !buffer_unwritten(bh) &&
1120		    (block_start < from || block_end > to)) {
1121			ext4_read_bh_lock(bh, 0, false);
1122			wait[nr_wait++] = bh;
1123		}
1124	}
1125	/*
1126	 * If we issued read requests, let them complete.
1127	 */
1128	for (i = 0; i < nr_wait; i++) {
1129		wait_on_buffer(wait[i]);
1130		if (!buffer_uptodate(wait[i]))
1131			err = -EIO;
1132	}
1133	if (unlikely(err)) {
1134		page_zero_new_buffers(page, from, to);
1135	} else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1136		for (i = 0; i < nr_wait; i++) {
1137			int err2;
1138
1139			err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1140								bh_offset(wait[i]));
1141			if (err2) {
1142				clear_buffer_uptodate(wait[i]);
1143				err = err2;
1144			}
1145		}
1146	}
1147
1148	return err;
1149}
1150#endif
1151
1152static int ext4_write_begin(struct file *file, struct address_space *mapping,
1153			    loff_t pos, unsigned len,
1154			    struct page **pagep, void **fsdata)
1155{
1156	struct inode *inode = mapping->host;
1157	int ret, needed_blocks;
1158	handle_t *handle;
1159	int retries = 0;
1160	struct page *page;
1161	pgoff_t index;
1162	unsigned from, to;
1163
1164	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1165		return -EIO;
1166
1167	trace_ext4_write_begin(inode, pos, len);
1168	/*
1169	 * Reserve one block more for addition to orphan list in case
1170	 * we allocate blocks but write fails for some reason
1171	 */
1172	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1173	index = pos >> PAGE_SHIFT;
1174	from = pos & (PAGE_SIZE - 1);
1175	to = from + len;
1176
1177	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1178		ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1179						    pagep);
1180		if (ret < 0)
1181			return ret;
1182		if (ret == 1)
1183			return 0;
1184	}
1185
1186	/*
1187	 * grab_cache_page_write_begin() can take a long time if the
1188	 * system is thrashing due to memory pressure, or if the page
1189	 * is being written back.  So grab it first before we start
1190	 * the transaction handle.  This also allows us to allocate
1191	 * the page (if needed) without using GFP_NOFS.
1192	 */
1193retry_grab:
1194	page = grab_cache_page_write_begin(mapping, index);
1195	if (!page)
1196		return -ENOMEM;
1197	/*
1198	 * The same as page allocation, we prealloc buffer heads before
1199	 * starting the handle.
1200	 */
1201	if (!page_has_buffers(page))
1202		create_empty_buffers(page, inode->i_sb->s_blocksize, 0);
1203
1204	unlock_page(page);
1205
1206retry_journal:
1207	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1208	if (IS_ERR(handle)) {
1209		put_page(page);
1210		return PTR_ERR(handle);
1211	}
1212
1213	lock_page(page);
1214	if (page->mapping != mapping) {
1215		/* The page got truncated from under us */
1216		unlock_page(page);
1217		put_page(page);
1218		ext4_journal_stop(handle);
1219		goto retry_grab;
 
1220	}
1221	/* In case writeback began while the page was unlocked */
1222	wait_for_stable_page(page);
1223
1224#ifdef CONFIG_FS_ENCRYPTION
1225	if (ext4_should_dioread_nolock(inode))
1226		ret = ext4_block_write_begin(page, pos, len,
1227					     ext4_get_block_unwritten);
1228	else
1229		ret = ext4_block_write_begin(page, pos, len,
1230					     ext4_get_block);
1231#else
1232	if (ext4_should_dioread_nolock(inode))
1233		ret = __block_write_begin(page, pos, len,
1234					  ext4_get_block_unwritten);
1235	else
1236		ret = __block_write_begin(page, pos, len, ext4_get_block);
1237#endif
1238	if (!ret && ext4_should_journal_data(inode)) {
1239		ret = ext4_walk_page_buffers(handle, inode,
1240					     page_buffers(page), from, to, NULL,
1241					     do_journal_get_write_access);
1242	}
1243
1244	if (ret) {
1245		bool extended = (pos + len > inode->i_size) &&
1246				!ext4_verity_in_progress(inode);
1247
1248		unlock_page(page);
 
1249		/*
1250		 * __block_write_begin may have instantiated a few blocks
1251		 * outside i_size.  Trim these off again. Don't need
1252		 * i_size_read because we hold i_rwsem.
1253		 *
1254		 * Add inode to orphan list in case we crash before
1255		 * truncate finishes
1256		 */
1257		if (extended && ext4_can_truncate(inode))
1258			ext4_orphan_add(handle, inode);
1259
1260		ext4_journal_stop(handle);
1261		if (extended) {
1262			ext4_truncate_failed_write(inode);
1263			/*
1264			 * If truncate failed early the inode might
1265			 * still be on the orphan list; we need to
1266			 * make sure the inode is removed from the
1267			 * orphan list in that case.
1268			 */
1269			if (inode->i_nlink)
1270				ext4_orphan_del(NULL, inode);
1271		}
 
1272
1273		if (ret == -ENOSPC &&
1274		    ext4_should_retry_alloc(inode->i_sb, &retries))
1275			goto retry_journal;
1276		put_page(page);
1277		return ret;
1278	}
1279	*pagep = page;
1280	return ret;
1281}
1282
1283/* For write_end() in data=journal mode */
1284static int write_end_fn(handle_t *handle, struct inode *inode,
1285			struct buffer_head *bh)
1286{
1287	int ret;
1288	if (!buffer_mapped(bh) || buffer_freed(bh))
1289		return 0;
1290	set_buffer_uptodate(bh);
1291	ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1292	clear_buffer_meta(bh);
1293	clear_buffer_prio(bh);
1294	return ret;
1295}
1296
1297/*
1298 * We need to pick up the new inode size which generic_commit_write gave us
1299 * `file' can be NULL - eg, when called from page_symlink().
1300 *
1301 * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1302 * buffers are managed internally.
1303 */
1304static int ext4_write_end(struct file *file,
1305			  struct address_space *mapping,
1306			  loff_t pos, unsigned len, unsigned copied,
1307			  struct page *page, void *fsdata)
1308{
 
 
1309	handle_t *handle = ext4_journal_current_handle();
1310	struct inode *inode = mapping->host;
1311	loff_t old_size = inode->i_size;
1312	int ret = 0, ret2;
1313	int i_size_changed = 0;
1314	bool verity = ext4_verity_in_progress(inode);
1315
1316	trace_ext4_write_end(inode, pos, len, copied);
1317
1318	if (ext4_has_inline_data(inode) &&
1319	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1320		return ext4_write_inline_data_end(inode, pos, len, copied, page);
1321
1322	copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1323	/*
1324	 * it's important to update i_size while still holding page lock:
 
 
 
1325	 * page writeout could otherwise come in and zero beyond i_size.
1326	 *
1327	 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1328	 * blocks are being written past EOF, so skip the i_size update.
1329	 */
1330	if (!verity)
1331		i_size_changed = ext4_update_inode_size(inode, pos + copied);
 
 
 
 
 
 
 
 
 
 
 
1332	unlock_page(page);
1333	put_page(page);
1334
1335	if (old_size < pos && !verity)
1336		pagecache_isize_extended(inode, old_size, pos);
1337	/*
1338	 * Don't mark the inode dirty under page lock. First, it unnecessarily
1339	 * makes the holding time of page lock longer. Second, it forces lock
1340	 * ordering of page lock and transaction start for journaling
1341	 * filesystems.
1342	 */
1343	if (i_size_changed)
1344		ret = ext4_mark_inode_dirty(handle, inode);
1345
1346	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1347		/* if we have allocated more blocks and copied
1348		 * less. We will have blocks allocated outside
1349		 * inode->i_size. So truncate them
1350		 */
1351		ext4_orphan_add(handle, inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1352
 
 
 
 
 
 
 
 
 
 
 
 
 
1353	ret2 = ext4_journal_stop(handle);
1354	if (!ret)
1355		ret = ret2;
1356
1357	if (pos + len > inode->i_size && !verity) {
1358		ext4_truncate_failed_write(inode);
1359		/*
1360		 * If truncate failed early the inode might still be
1361		 * on the orphan list; we need to make sure the inode
1362		 * is removed from the orphan list in that case.
1363		 */
1364		if (inode->i_nlink)
1365			ext4_orphan_del(NULL, inode);
1366	}
1367
 
1368	return ret ? ret : copied;
1369}
1370
1371/*
1372 * This is a private version of page_zero_new_buffers() which doesn't
1373 * set the buffer to be dirty, since in data=journalled mode we need
1374 * to call ext4_handle_dirty_metadata() instead.
1375 */
1376static void ext4_journalled_zero_new_buffers(handle_t *handle,
1377					    struct inode *inode,
1378					    struct page *page,
1379					    unsigned from, unsigned to)
1380{
1381	unsigned int block_start = 0, block_end;
1382	struct buffer_head *head, *bh;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1383
1384	bh = head = page_buffers(page);
1385	do {
1386		block_end = block_start + bh->b_size;
1387		if (buffer_new(bh)) {
1388			if (block_end > from && block_start < to) {
1389				if (!PageUptodate(page)) {
1390					unsigned start, size;
1391
1392					start = max(from, block_start);
1393					size = min(to, block_end) - start;
 
 
 
 
 
 
 
 
1394
1395					zero_user(page, start, size);
1396					write_end_fn(handle, inode, bh);
1397				}
1398				clear_buffer_new(bh);
1399			}
1400		}
1401		block_start = block_end;
1402		bh = bh->b_this_page;
1403	} while (bh != head);
1404}
1405
1406static int ext4_journalled_write_end(struct file *file,
1407				     struct address_space *mapping,
1408				     loff_t pos, unsigned len, unsigned copied,
1409				     struct page *page, void *fsdata)
1410{
1411	handle_t *handle = ext4_journal_current_handle();
1412	struct inode *inode = mapping->host;
1413	loff_t old_size = inode->i_size;
1414	int ret = 0, ret2;
1415	int partial = 0;
1416	unsigned from, to;
1417	int size_changed = 0;
1418	bool verity = ext4_verity_in_progress(inode);
1419
1420	trace_ext4_journalled_write_end(inode, pos, len, copied);
1421	from = pos & (PAGE_SIZE - 1);
1422	to = from + len;
1423
1424	BUG_ON(!ext4_handle_valid(handle));
1425
1426	if (ext4_has_inline_data(inode))
1427		return ext4_write_inline_data_end(inode, pos, len, copied, page);
 
 
 
1428
1429	if (unlikely(copied < len) && !PageUptodate(page)) {
1430		copied = 0;
1431		ext4_journalled_zero_new_buffers(handle, inode, page, from, to);
1432	} else {
1433		if (unlikely(copied < len))
1434			ext4_journalled_zero_new_buffers(handle, inode, page,
1435							 from + copied, to);
1436		ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
1437					     from, from + copied, &partial,
1438					     write_end_fn);
1439		if (!partial)
1440			SetPageUptodate(page);
1441	}
1442	if (!verity)
1443		size_changed = ext4_update_inode_size(inode, pos + copied);
1444	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1445	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1446	unlock_page(page);
1447	put_page(page);
1448
1449	if (old_size < pos && !verity)
1450		pagecache_isize_extended(inode, old_size, pos);
1451
1452	if (size_changed) {
1453		ret2 = ext4_mark_inode_dirty(handle, inode);
1454		if (!ret)
1455			ret = ret2;
1456	}
1457
1458	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
 
 
1459		/* if we have allocated more blocks and copied
1460		 * less. We will have blocks allocated outside
1461		 * inode->i_size. So truncate them
1462		 */
1463		ext4_orphan_add(handle, inode);
1464
1465	ret2 = ext4_journal_stop(handle);
1466	if (!ret)
1467		ret = ret2;
1468	if (pos + len > inode->i_size && !verity) {
1469		ext4_truncate_failed_write(inode);
1470		/*
1471		 * If truncate failed early the inode might still be
1472		 * on the orphan list; we need to make sure the inode
1473		 * is removed from the orphan list in that case.
1474		 */
1475		if (inode->i_nlink)
1476			ext4_orphan_del(NULL, inode);
1477	}
1478
1479	return ret ? ret : copied;
1480}
1481
1482/*
1483 * Reserve space for a single cluster
1484 */
1485static int ext4_da_reserve_space(struct inode *inode)
1486{
 
1487	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1488	struct ext4_inode_info *ei = EXT4_I(inode);
 
1489	int ret;
1490
1491	/*
 
 
 
 
 
 
 
 
 
 
 
1492	 * We will charge metadata quota at writeout time; this saves
1493	 * us from metadata over-estimation, though we may go over by
1494	 * a small amount in the end.  Here we just reserve for data.
1495	 */
1496	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1497	if (ret)
1498		return ret;
1499
1500	spin_lock(&ei->i_block_reservation_lock);
1501	if (ext4_claim_free_clusters(sbi, 1, 0)) {
1502		spin_unlock(&ei->i_block_reservation_lock);
1503		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
 
 
 
 
 
1504		return -ENOSPC;
1505	}
 
1506	ei->i_reserved_data_blocks++;
1507	trace_ext4_da_reserve_space(inode);
1508	spin_unlock(&ei->i_block_reservation_lock);
1509
1510	return 0;       /* success */
1511}
1512
1513void ext4_da_release_space(struct inode *inode, int to_free)
1514{
1515	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1516	struct ext4_inode_info *ei = EXT4_I(inode);
1517
1518	if (!to_free)
1519		return;		/* Nothing to release, exit */
1520
1521	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1522
1523	trace_ext4_da_release_space(inode, to_free);
1524	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1525		/*
1526		 * if there aren't enough reserved blocks, then the
1527		 * counter is messed up somewhere.  Since this
1528		 * function is called from invalidate page, it's
1529		 * harmless to return without any action.
1530		 */
1531		ext4_warning(inode->i_sb, "ext4_da_release_space: "
1532			 "ino %lu, to_free %d with only %d reserved "
1533			 "data blocks", inode->i_ino, to_free,
1534			 ei->i_reserved_data_blocks);
1535		WARN_ON(1);
1536		to_free = ei->i_reserved_data_blocks;
1537	}
1538	ei->i_reserved_data_blocks -= to_free;
1539
 
 
 
 
 
 
 
 
 
 
 
 
1540	/* update fs dirty data blocks counter */
1541	percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1542
1543	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1544
1545	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1546}
1547
1548/*
1549 * Delayed allocation stuff
1550 */
1551
1552struct mpage_da_data {
1553	/* These are input fields for ext4_do_writepages() */
1554	struct inode *inode;
1555	struct writeback_control *wbc;
1556	unsigned int can_map:1;	/* Can writepages call map blocks? */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1557
1558	/* These are internal state of ext4_do_writepages() */
1559	pgoff_t first_page;	/* The first page to write */
1560	pgoff_t next_page;	/* Current page to examine */
1561	pgoff_t last_page;	/* Last page to examine */
1562	/*
1563	 * Extent to map - this can be after first_page because that can be
1564	 * fully mapped. We somewhat abuse m_flags to store whether the extent
1565	 * is delalloc or unwritten.
1566	 */
1567	struct ext4_map_blocks map;
1568	struct ext4_io_submit io_submit;	/* IO submission data */
1569	unsigned int do_map:1;
1570	unsigned int scanned_until_end:1;
1571};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1572
1573static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1574				       bool invalidate)
1575{
1576	unsigned nr, i;
1577	pgoff_t index, end;
1578	struct folio_batch fbatch;
1579	struct inode *inode = mpd->inode;
1580	struct address_space *mapping = inode->i_mapping;
1581
1582	/* This is necessary when next_page == 0. */
1583	if (mpd->first_page >= mpd->next_page)
1584		return;
1585
1586	mpd->scanned_until_end = 0;
1587	index = mpd->first_page;
1588	end   = mpd->next_page - 1;
1589	if (invalidate) {
1590		ext4_lblk_t start, last;
1591		start = index << (PAGE_SHIFT - inode->i_blkbits);
1592		last = end << (PAGE_SHIFT - inode->i_blkbits);
1593
1594		/*
1595		 * avoid racing with extent status tree scans made by
1596		 * ext4_insert_delayed_block()
1597		 */
1598		down_write(&EXT4_I(inode)->i_data_sem);
1599		ext4_es_remove_extent(inode, start, last - start + 1);
1600		up_write(&EXT4_I(inode)->i_data_sem);
1601	}
1602
1603	folio_batch_init(&fbatch);
1604	while (index <= end) {
1605		nr = filemap_get_folios(mapping, &index, end, &fbatch);
1606		if (nr == 0)
1607			break;
1608		for (i = 0; i < nr; i++) {
1609			struct folio *folio = fbatch.folios[i];
1610
1611			if (folio->index < mpd->first_page)
1612				continue;
1613			if (folio->index + folio_nr_pages(folio) - 1 > end)
1614				continue;
1615			BUG_ON(!folio_test_locked(folio));
1616			BUG_ON(folio_test_writeback(folio));
1617			if (invalidate) {
1618				if (folio_mapped(folio))
1619					folio_clear_dirty_for_io(folio);
1620				block_invalidate_folio(folio, 0,
1621						folio_size(folio));
1622				folio_clear_uptodate(folio);
1623			}
1624			folio_unlock(folio);
1625		}
1626		folio_batch_release(&fbatch);
 
1627	}
 
1628}
1629
1630static void ext4_print_free_blocks(struct inode *inode)
1631{
1632	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1633	struct super_block *sb = inode->i_sb;
1634	struct ext4_inode_info *ei = EXT4_I(inode);
1635
1636	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1637	       EXT4_C2B(EXT4_SB(inode->i_sb),
1638			ext4_count_free_clusters(sb)));
1639	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1640	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1641	       (long long) EXT4_C2B(EXT4_SB(sb),
1642		percpu_counter_sum(&sbi->s_freeclusters_counter)));
1643	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1644	       (long long) EXT4_C2B(EXT4_SB(sb),
1645		percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1646	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1647	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1648		 ei->i_reserved_data_blocks);
1649	return;
1650}
1651
1652static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode,
1653				      struct buffer_head *bh)
1654{
1655	return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1656}
1657
1658/*
1659 * ext4_insert_delayed_block - adds a delayed block to the extents status
1660 *                             tree, incrementing the reserved cluster/block
1661 *                             count or making a pending reservation
1662 *                             where needed
1663 *
1664 * @inode - file containing the newly added block
1665 * @lblk - logical block to be added
 
1666 *
1667 * Returns 0 on success, negative error code on failure.
1668 */
1669static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1670{
1671	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1672	int ret;
1673	bool allocated = false;
1674	bool reserved = false;
 
 
1675
1676	/*
1677	 * If the cluster containing lblk is shared with a delayed,
1678	 * written, or unwritten extent in a bigalloc file system, it's
1679	 * already been accounted for and does not need to be reserved.
1680	 * A pending reservation must be made for the cluster if it's
1681	 * shared with a written or unwritten extent and doesn't already
1682	 * have one.  Written and unwritten extents can be purged from the
1683	 * extents status tree if the system is under memory pressure, so
1684	 * it's necessary to examine the extent tree if a search of the
1685	 * extents status tree doesn't get a match.
1686	 */
1687	if (sbi->s_cluster_ratio == 1) {
1688		ret = ext4_da_reserve_space(inode);
1689		if (ret != 0)   /* ENOSPC */
1690			goto errout;
1691		reserved = true;
1692	} else {   /* bigalloc */
1693		if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1694			if (!ext4_es_scan_clu(inode,
1695					      &ext4_es_is_mapped, lblk)) {
1696				ret = ext4_clu_mapped(inode,
1697						      EXT4_B2C(sbi, lblk));
1698				if (ret < 0)
1699					goto errout;
1700				if (ret == 0) {
1701					ret = ext4_da_reserve_space(inode);
1702					if (ret != 0)   /* ENOSPC */
1703						goto errout;
1704					reserved = true;
1705				} else {
1706					allocated = true;
1707				}
1708			} else {
1709				allocated = true;
1710			}
1711		}
1712	}
 
1713
1714	ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1715	if (ret && reserved)
1716		ext4_da_release_space(inode, 1);
1717
1718errout:
1719	return ret;
1720}
1721
1722/*
1723 * This function is grabs code from the very beginning of
1724 * ext4_map_blocks, but assumes that the caller is from delayed write
1725 * time. This function looks up the requested blocks and sets the
1726 * buffer delay bit under the protection of i_data_sem.
1727 */
1728static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1729			      struct ext4_map_blocks *map,
1730			      struct buffer_head *bh)
1731{
1732	struct extent_status es;
1733	int retval;
1734	sector_t invalid_block = ~((sector_t) 0xffff);
1735#ifdef ES_AGGRESSIVE_TEST
1736	struct ext4_map_blocks orig_map;
1737
1738	memcpy(&orig_map, map, sizeof(*map));
1739#endif
1740
1741	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1742		invalid_block = ~0;
1743
1744	map->m_flags = 0;
1745	ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1746		  (unsigned long) map->m_lblk);
1747
1748	/* Lookup extent status tree firstly */
1749	if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1750		if (ext4_es_is_hole(&es)) {
1751			retval = 0;
1752			down_read(&EXT4_I(inode)->i_data_sem);
1753			goto add_delayed;
1754		}
1755
1756		/*
1757		 * Delayed extent could be allocated by fallocate.
1758		 * So we need to check it.
 
 
 
1759		 */
1760		if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1761			map_bh(bh, inode->i_sb, invalid_block);
1762			set_buffer_new(bh);
1763			set_buffer_delay(bh);
1764			return 0;
 
 
 
 
 
 
1765		}
 
 
 
 
 
 
 
 
1766
1767		map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1768		retval = es.es_len - (iblock - es.es_lblk);
1769		if (retval > map->m_len)
1770			retval = map->m_len;
1771		map->m_len = retval;
1772		if (ext4_es_is_written(&es))
1773			map->m_flags |= EXT4_MAP_MAPPED;
1774		else if (ext4_es_is_unwritten(&es))
1775			map->m_flags |= EXT4_MAP_UNWRITTEN;
1776		else
1777			BUG();
1778
1779#ifdef ES_AGGRESSIVE_TEST
1780		ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1781#endif
1782		return retval;
 
1783	}
1784
1785	/*
1786	 * Try to see if we can get the block without requesting a new
1787	 * file system block.
1788	 */
1789	down_read(&EXT4_I(inode)->i_data_sem);
1790	if (ext4_has_inline_data(inode))
1791		retval = 0;
1792	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1793		retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1794	else
1795		retval = ext4_ind_map_blocks(NULL, inode, map, 0);
 
 
 
 
1796
1797add_delayed:
1798	if (retval == 0) {
1799		int ret;
 
1800
1801		/*
1802		 * XXX: __block_prepare_write() unmaps passed block,
1803		 * is it OK?
1804		 */
1805
1806		ret = ext4_insert_delayed_block(inode, map->m_lblk);
1807		if (ret != 0) {
1808			retval = ret;
1809			goto out_unlock;
1810		}
 
 
 
 
 
 
 
 
 
 
1811
1812		map_bh(bh, inode->i_sb, invalid_block);
1813		set_buffer_new(bh);
1814		set_buffer_delay(bh);
1815	} else if (retval > 0) {
1816		int ret;
1817		unsigned int status;
 
 
1818
1819		if (unlikely(retval != map->m_len)) {
1820			ext4_warning(inode->i_sb,
1821				     "ES len assertion failed for inode "
1822				     "%lu: retval %d != map->m_len %d",
1823				     inode->i_ino, retval, map->m_len);
1824			WARN_ON(1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1825		}
 
 
 
 
 
 
 
 
 
 
1826
1827		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1828				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1829		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1830					    map->m_pblk, status);
1831		if (ret != 0)
1832			retval = ret;
 
1833	}
1834
1835out_unlock:
1836	up_read((&EXT4_I(inode)->i_data_sem));
 
 
 
 
 
 
1837
1838	return retval;
 
 
1839}
1840
1841/*
1842 * This is a special get_block_t callback which is used by
1843 * ext4_da_write_begin().  It will either return mapped block or
1844 * reserve space for a single block.
1845 *
1846 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1847 * We also have b_blocknr = -1 and b_bdev initialized properly
1848 *
1849 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1850 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1851 * initialized properly.
1852 */
1853int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1854			   struct buffer_head *bh, int create)
1855{
1856	struct ext4_map_blocks map;
1857	int ret = 0;
 
 
 
 
1858
1859	BUG_ON(create == 0);
1860	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1861
1862	map.m_lblk = iblock;
1863	map.m_len = 1;
1864
1865	/*
1866	 * first, we need to know whether the block is allocated already
1867	 * preallocated blocks are unmapped but should treated
1868	 * the same as allocated blocks.
1869	 */
1870	ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1871	if (ret <= 0)
1872		return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1873
1874	map_bh(bh, inode->i_sb, map.m_pblk);
1875	ext4_update_bh_state(bh, map.m_flags);
1876
1877	if (buffer_unwritten(bh)) {
1878		/* A delayed write to unwritten bh should be marked
1879		 * new and mapped.  Mapped ensures that we don't do
1880		 * get_block multiple times when we write to the same
1881		 * offset and new ensures that we do proper zero out
1882		 * for partial write.
1883		 */
1884		set_buffer_new(bh);
1885		set_buffer_mapped(bh);
1886	}
1887	return 0;
1888}
1889
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1890static int __ext4_journalled_writepage(struct page *page,
1891				       unsigned int len)
1892{
1893	struct address_space *mapping = page->mapping;
1894	struct inode *inode = mapping->host;
 
1895	handle_t *handle = NULL;
1896	int ret = 0, err = 0;
1897	int inline_data = ext4_has_inline_data(inode);
1898	struct buffer_head *inode_bh = NULL;
1899	loff_t size;
1900
1901	ClearPageChecked(page);
1902
1903	if (inline_data) {
1904		BUG_ON(page->index != 0);
1905		BUG_ON(len > ext4_get_max_inline_size(inode));
1906		inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1907		if (inode_bh == NULL)
1908			goto out;
1909	}
1910	/*
1911	 * We need to release the page lock before we start the
1912	 * journal, so grab a reference so the page won't disappear
1913	 * out from under us.
1914	 */
1915	get_page(page);
1916	unlock_page(page);
1917
1918	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1919				    ext4_writepage_trans_blocks(inode));
1920	if (IS_ERR(handle)) {
1921		ret = PTR_ERR(handle);
1922		put_page(page);
1923		goto out_no_pagelock;
1924	}
1925	BUG_ON(!ext4_handle_valid(handle));
1926
1927	lock_page(page);
1928	put_page(page);
1929	size = i_size_read(inode);
1930	if (page->mapping != mapping || page_offset(page) > size) {
1931		/* The page got truncated from under us */
1932		ext4_journal_stop(handle);
1933		ret = 0;
1934		goto out;
1935	}
1936
1937	if (inline_data) {
1938		ret = ext4_mark_inode_dirty(handle, inode);
1939	} else {
1940		struct buffer_head *page_bufs = page_buffers(page);
1941
1942		if (page->index == size >> PAGE_SHIFT)
1943			len = size & ~PAGE_MASK;
1944		else
1945			len = PAGE_SIZE;
1946
1947		ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1948					     NULL, do_journal_get_write_access);
1949
1950		err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1951					     NULL, write_end_fn);
1952	}
1953	if (ret == 0)
1954		ret = err;
1955	err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1956	if (ret == 0)
1957		ret = err;
1958	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1959	err = ext4_journal_stop(handle);
1960	if (!ret)
1961		ret = err;
1962
 
1963	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1964out:
1965	unlock_page(page);
1966out_no_pagelock:
1967	brelse(inode_bh);
1968	return ret;
1969}
1970
 
 
 
1971/*
1972 * Note that we don't need to start a transaction unless we're journaling data
1973 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1974 * need to file the inode to the transaction's list in ordered mode because if
1975 * we are writing back data added by write(), the inode is already there and if
1976 * we are writing back data modified via mmap(), no one guarantees in which
1977 * transaction the data will hit the disk. In case we are journaling data, we
1978 * cannot start transaction directly because transaction start ranks above page
1979 * lock so we have to do some magic.
1980 *
1981 * This function can get called via...
1982 *   - ext4_writepages after taking page lock (have journal handle)
1983 *   - journal_submit_inode_data_buffers (no journal handle)
1984 *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1985 *   - grab_page_cache when doing write_begin (have journal handle)
1986 *
1987 * We don't do any block allocation in this function. If we have page with
1988 * multiple blocks we need to write those buffer_heads that are mapped. This
1989 * is important for mmaped based write. So if we do with blocksize 1K
1990 * truncate(f, 1024);
1991 * a = mmap(f, 0, 4096);
1992 * a[0] = 'a';
1993 * truncate(f, 4096);
1994 * we have in the page first buffer_head mapped via page_mkwrite call back
1995 * but other buffer_heads would be unmapped but dirty (dirty done via the
1996 * do_wp_page). So writepage should write the first block. If we modify
1997 * the mmap area beyond 1024 we will again get a page_fault and the
1998 * page_mkwrite callback will do the block allocation and mark the
1999 * buffer_heads mapped.
2000 *
2001 * We redirty the page if we have any buffer_heads that is either delay or
2002 * unwritten in the page.
2003 *
2004 * We can get recursively called as show below.
2005 *
2006 *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2007 *		ext4_writepage()
2008 *
2009 * But since we don't do any block allocation we should not deadlock.
2010 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2011 */
2012static int ext4_writepage(struct page *page,
2013			  struct writeback_control *wbc)
2014{
2015	struct folio *folio = page_folio(page);
2016	int ret = 0;
2017	loff_t size;
2018	unsigned int len;
2019	struct buffer_head *page_bufs = NULL;
2020	struct inode *inode = page->mapping->host;
2021	struct ext4_io_submit io_submit;
2022
2023	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2024		folio_invalidate(folio, 0, folio_size(folio));
2025		folio_unlock(folio);
2026		return -EIO;
2027	}
2028
2029	trace_ext4_writepage(page);
2030	size = i_size_read(inode);
2031	if (page->index == size >> PAGE_SHIFT &&
2032	    !ext4_verity_in_progress(inode))
2033		len = size & ~PAGE_MASK;
2034	else
2035		len = PAGE_SIZE;
2036
2037	/* Should never happen but for bugs in other kernel subsystems */
 
 
 
 
2038	if (!page_has_buffers(page)) {
2039		ext4_warning_inode(inode,
2040		   "page %lu does not have buffers attached", page->index);
2041		ClearPageDirty(page);
2042		unlock_page(page);
2043		return 0;
2044	}
2045
2046	page_bufs = page_buffers(page);
2047	/*
2048	 * We cannot do block allocation or other extent handling in this
2049	 * function. If there are buffers needing that, we have to redirty
2050	 * the page. But we may reach here when we do a journal commit via
2051	 * journal_submit_inode_data_buffers() and in that case we must write
2052	 * allocated buffers to achieve data=ordered mode guarantees.
2053	 *
2054	 * Also, if there is only one buffer per page (the fs block
2055	 * size == the page size), if one buffer needs block
2056	 * allocation or needs to modify the extent tree to clear the
2057	 * unwritten flag, we know that the page can't be written at
2058	 * all, so we might as well refuse the write immediately.
2059	 * Unfortunately if the block size != page size, we can't as
2060	 * easily detect this case using ext4_walk_page_buffers(), but
2061	 * for the extremely common case, this is an optimization that
2062	 * skips a useless round trip through ext4_bio_write_page().
2063	 */
2064	if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL,
2065				   ext4_bh_delay_or_unwritten)) {
2066		redirty_page_for_writepage(wbc, page);
2067		if ((current->flags & PF_MEMALLOC) ||
2068		    (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2069			/*
2070			 * For memory cleaning there's no point in writing only
2071			 * some buffers. So just bail out. Warn if we came here
2072			 * from direct reclaim.
2073			 */
2074			WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2075							== PF_MEMALLOC);
2076			unlock_page(page);
2077			return 0;
2078		}
 
2079	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2080
2081	if (PageChecked(page) && ext4_should_journal_data(inode))
2082		/*
2083		 * It's mmapped pagecache.  Add buffers and journal it.  There
2084		 * doesn't seem much point in redirtying the page here.
2085		 */
2086		return __ext4_journalled_writepage(page, len);
2087
2088	ext4_io_submit_init(&io_submit, wbc);
2089	io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2090	if (!io_submit.io_end) {
2091		redirty_page_for_writepage(wbc, page);
2092		unlock_page(page);
2093		return -ENOMEM;
2094	}
2095	ret = ext4_bio_write_page(&io_submit, page, len);
2096	ext4_io_submit(&io_submit);
2097	/* Drop io_end reference we got from init */
2098	ext4_put_io_end_defer(io_submit.io_end);
2099	return ret;
2100}
2101
2102static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2103{
2104	int len;
2105	loff_t size;
2106	int err;
2107
2108	BUG_ON(page->index != mpd->first_page);
2109	clear_page_dirty_for_io(page);
2110	/*
2111	 * We have to be very careful here!  Nothing protects writeback path
2112	 * against i_size changes and the page can be writeably mapped into
2113	 * page tables. So an application can be growing i_size and writing
2114	 * data through mmap while writeback runs. clear_page_dirty_for_io()
2115	 * write-protects our page in page tables and the page cannot get
2116	 * written to again until we release page lock. So only after
2117	 * clear_page_dirty_for_io() we are safe to sample i_size for
2118	 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2119	 * on the barrier provided by TestClearPageDirty in
2120	 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2121	 * after page tables are updated.
2122	 */
2123	size = i_size_read(mpd->inode);
2124	if (page->index == size >> PAGE_SHIFT &&
2125	    !ext4_verity_in_progress(mpd->inode))
2126		len = size & ~PAGE_MASK;
2127	else
2128		len = PAGE_SIZE;
2129	err = ext4_bio_write_page(&mpd->io_submit, page, len);
2130	if (!err)
2131		mpd->wbc->nr_to_write--;
2132	mpd->first_page++;
2133
2134	return err;
2135}
2136
2137#define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2138
2139/*
2140 * mballoc gives us at most this number of blocks...
2141 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2142 * The rest of mballoc seems to handle chunks up to full group size.
2143 */
2144#define MAX_WRITEPAGES_EXTENT_LEN 2048
2145
2146/*
2147 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2148 *
2149 * @mpd - extent of blocks
2150 * @lblk - logical number of the block in the file
2151 * @bh - buffer head we want to add to the extent
2152 *
2153 * The function is used to collect contig. blocks in the same state. If the
2154 * buffer doesn't require mapping for writeback and we haven't started the
2155 * extent of buffers to map yet, the function returns 'true' immediately - the
2156 * caller can write the buffer right away. Otherwise the function returns true
2157 * if the block has been added to the extent, false if the block couldn't be
2158 * added.
2159 */
2160static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2161				   struct buffer_head *bh)
2162{
2163	struct ext4_map_blocks *map = &mpd->map;
2164
2165	/* Buffer that doesn't need mapping for writeback? */
2166	if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2167	    (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2168		/* So far no extent to map => we write the buffer right away */
2169		if (map->m_len == 0)
2170			return true;
2171		return false;
2172	}
2173
2174	/* First block in the extent? */
2175	if (map->m_len == 0) {
2176		/* We cannot map unless handle is started... */
2177		if (!mpd->do_map)
2178			return false;
2179		map->m_lblk = lblk;
2180		map->m_len = 1;
2181		map->m_flags = bh->b_state & BH_FLAGS;
2182		return true;
2183	}
2184
2185	/* Don't go larger than mballoc is willing to allocate */
2186	if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2187		return false;
2188
2189	/* Can we merge the block to our big extent? */
2190	if (lblk == map->m_lblk + map->m_len &&
2191	    (bh->b_state & BH_FLAGS) == map->m_flags) {
2192		map->m_len++;
2193		return true;
2194	}
2195	return false;
2196}
2197
2198/*
2199 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2200 *
2201 * @mpd - extent of blocks for mapping
2202 * @head - the first buffer in the page
2203 * @bh - buffer we should start processing from
2204 * @lblk - logical number of the block in the file corresponding to @bh
2205 *
2206 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2207 * the page for IO if all buffers in this page were mapped and there's no
2208 * accumulated extent of buffers to map or add buffers in the page to the
2209 * extent of buffers to map. The function returns 1 if the caller can continue
2210 * by processing the next page, 0 if it should stop adding buffers to the
2211 * extent to map because we cannot extend it anymore. It can also return value
2212 * < 0 in case of error during IO submission.
2213 */
2214static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2215				   struct buffer_head *head,
2216				   struct buffer_head *bh,
2217				   ext4_lblk_t lblk)
2218{
2219	struct inode *inode = mpd->inode;
2220	int err;
2221	ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2222							>> inode->i_blkbits;
2223
2224	if (ext4_verity_in_progress(inode))
2225		blocks = EXT_MAX_BLOCKS;
2226
2227	do {
2228		BUG_ON(buffer_locked(bh));
2229
2230		if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2231			/* Found extent to map? */
2232			if (mpd->map.m_len)
2233				return 0;
2234			/* Buffer needs mapping and handle is not started? */
2235			if (!mpd->do_map)
2236				return 0;
2237			/* Everything mapped so far and we hit EOF */
2238			break;
2239		}
2240	} while (lblk++, (bh = bh->b_this_page) != head);
2241	/* So far everything mapped? Submit the page for IO. */
2242	if (mpd->map.m_len == 0) {
2243		err = mpage_submit_page(mpd, head->b_page);
2244		if (err < 0)
2245			return err;
2246	}
2247	if (lblk >= blocks) {
2248		mpd->scanned_until_end = 1;
2249		return 0;
2250	}
2251	return 1;
2252}
2253
2254/*
2255 * mpage_process_page - update page buffers corresponding to changed extent and
2256 *		       may submit fully mapped page for IO
2257 *
2258 * @mpd		- description of extent to map, on return next extent to map
2259 * @m_lblk	- logical block mapping.
2260 * @m_pblk	- corresponding physical mapping.
2261 * @map_bh	- determines on return whether this page requires any further
2262 *		  mapping or not.
2263 * Scan given page buffers corresponding to changed extent and update buffer
2264 * state according to new extent state.
2265 * We map delalloc buffers to their physical location, clear unwritten bits.
2266 * If the given page is not fully mapped, we update @map to the next extent in
2267 * the given page that needs mapping & return @map_bh as true.
2268 */
2269static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2270			      ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2271			      bool *map_bh)
2272{
2273	struct buffer_head *head, *bh;
2274	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2275	ext4_lblk_t lblk = *m_lblk;
2276	ext4_fsblk_t pblock = *m_pblk;
2277	int err = 0;
2278	int blkbits = mpd->inode->i_blkbits;
2279	ssize_t io_end_size = 0;
2280	struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2281
2282	bh = head = page_buffers(page);
2283	do {
2284		if (lblk < mpd->map.m_lblk)
2285			continue;
2286		if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2287			/*
2288			 * Buffer after end of mapped extent.
2289			 * Find next buffer in the page to map.
2290			 */
2291			mpd->map.m_len = 0;
2292			mpd->map.m_flags = 0;
2293			io_end_vec->size += io_end_size;
2294
2295			err = mpage_process_page_bufs(mpd, head, bh, lblk);
2296			if (err > 0)
2297				err = 0;
2298			if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2299				io_end_vec = ext4_alloc_io_end_vec(io_end);
2300				if (IS_ERR(io_end_vec)) {
2301					err = PTR_ERR(io_end_vec);
2302					goto out;
2303				}
2304				io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2305			}
2306			*map_bh = true;
2307			goto out;
2308		}
2309		if (buffer_delay(bh)) {
2310			clear_buffer_delay(bh);
2311			bh->b_blocknr = pblock++;
2312		}
2313		clear_buffer_unwritten(bh);
2314		io_end_size += (1 << blkbits);
2315	} while (lblk++, (bh = bh->b_this_page) != head);
2316
2317	io_end_vec->size += io_end_size;
2318	*map_bh = false;
2319out:
2320	*m_lblk = lblk;
2321	*m_pblk = pblock;
2322	return err;
2323}
2324
2325/*
2326 * mpage_map_buffers - update buffers corresponding to changed extent and
2327 *		       submit fully mapped pages for IO
2328 *
2329 * @mpd - description of extent to map, on return next extent to map
2330 *
2331 * Scan buffers corresponding to changed extent (we expect corresponding pages
2332 * to be already locked) and update buffer state according to new extent state.
2333 * We map delalloc buffers to their physical location, clear unwritten bits,
2334 * and mark buffers as uninit when we perform writes to unwritten extents
2335 * and do extent conversion after IO is finished. If the last page is not fully
2336 * mapped, we update @map to the next extent in the last page that needs
2337 * mapping. Otherwise we submit the page for IO.
2338 */
2339static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2340{
2341	struct folio_batch fbatch;
2342	unsigned nr, i;
2343	struct inode *inode = mpd->inode;
2344	int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2345	pgoff_t start, end;
2346	ext4_lblk_t lblk;
2347	ext4_fsblk_t pblock;
2348	int err;
2349	bool map_bh = false;
2350
2351	start = mpd->map.m_lblk >> bpp_bits;
2352	end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2353	lblk = start << bpp_bits;
2354	pblock = mpd->map.m_pblk;
2355
2356	folio_batch_init(&fbatch);
2357	while (start <= end) {
2358		nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2359		if (nr == 0)
2360			break;
2361		for (i = 0; i < nr; i++) {
2362			struct page *page = &fbatch.folios[i]->page;
2363
2364			err = mpage_process_page(mpd, page, &lblk, &pblock,
2365						 &map_bh);
2366			/*
2367			 * If map_bh is true, means page may require further bh
2368			 * mapping, or maybe the page was submitted for IO.
2369			 * So we return to call further extent mapping.
2370			 */
2371			if (err < 0 || map_bh)
2372				goto out;
2373			/* Page fully mapped - let IO run! */
2374			err = mpage_submit_page(mpd, page);
2375			if (err < 0)
2376				goto out;
2377		}
2378		folio_batch_release(&fbatch);
2379	}
2380	/* Extent fully mapped and matches with page boundary. We are done. */
2381	mpd->map.m_len = 0;
2382	mpd->map.m_flags = 0;
2383	return 0;
2384out:
2385	folio_batch_release(&fbatch);
2386	return err;
2387}
2388
2389static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2390{
2391	struct inode *inode = mpd->inode;
2392	struct ext4_map_blocks *map = &mpd->map;
2393	int get_blocks_flags;
2394	int err, dioread_nolock;
2395
2396	trace_ext4_da_write_pages_extent(inode, map);
2397	/*
2398	 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2399	 * to convert an unwritten extent to be initialized (in the case
2400	 * where we have written into one or more preallocated blocks).  It is
2401	 * possible that we're going to need more metadata blocks than
2402	 * previously reserved. However we must not fail because we're in
2403	 * writeback and there is nothing we can do about it so it might result
2404	 * in data loss.  So use reserved blocks to allocate metadata if
2405	 * possible.
2406	 *
2407	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2408	 * the blocks in question are delalloc blocks.  This indicates
2409	 * that the blocks and quotas has already been checked when
2410	 * the data was copied into the page cache.
2411	 */
2412	get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2413			   EXT4_GET_BLOCKS_METADATA_NOFAIL |
2414			   EXT4_GET_BLOCKS_IO_SUBMIT;
2415	dioread_nolock = ext4_should_dioread_nolock(inode);
2416	if (dioread_nolock)
2417		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2418	if (map->m_flags & BIT(BH_Delay))
2419		get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2420
2421	err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2422	if (err < 0)
2423		return err;
2424	if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2425		if (!mpd->io_submit.io_end->handle &&
2426		    ext4_handle_valid(handle)) {
2427			mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2428			handle->h_rsv_handle = NULL;
2429		}
2430		ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2431	}
2432
2433	BUG_ON(map->m_len == 0);
2434	return 0;
2435}
2436
2437/*
2438 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2439 *				 mpd->len and submit pages underlying it for IO
2440 *
2441 * @handle - handle for journal operations
2442 * @mpd - extent to map
2443 * @give_up_on_write - we set this to true iff there is a fatal error and there
2444 *                     is no hope of writing the data. The caller should discard
2445 *                     dirty pages to avoid infinite loops.
2446 *
2447 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2448 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2449 * them to initialized or split the described range from larger unwritten
2450 * extent. Note that we need not map all the described range since allocation
2451 * can return less blocks or the range is covered by more unwritten extents. We
2452 * cannot map more because we are limited by reserved transaction credits. On
2453 * the other hand we always make sure that the last touched page is fully
2454 * mapped so that it can be written out (and thus forward progress is
2455 * guaranteed). After mapping we submit all mapped pages for IO.
2456 */
2457static int mpage_map_and_submit_extent(handle_t *handle,
2458				       struct mpage_da_data *mpd,
2459				       bool *give_up_on_write)
2460{
2461	struct inode *inode = mpd->inode;
2462	struct ext4_map_blocks *map = &mpd->map;
2463	int err;
2464	loff_t disksize;
2465	int progress = 0;
2466	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2467	struct ext4_io_end_vec *io_end_vec;
2468
2469	io_end_vec = ext4_alloc_io_end_vec(io_end);
2470	if (IS_ERR(io_end_vec))
2471		return PTR_ERR(io_end_vec);
2472	io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2473	do {
2474		err = mpage_map_one_extent(handle, mpd);
2475		if (err < 0) {
2476			struct super_block *sb = inode->i_sb;
2477
2478			if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2479			    ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2480				goto invalidate_dirty_pages;
2481			/*
2482			 * Let the uper layers retry transient errors.
2483			 * In the case of ENOSPC, if ext4_count_free_blocks()
2484			 * is non-zero, a commit should free up blocks.
2485			 */
2486			if ((err == -ENOMEM) ||
2487			    (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2488				if (progress)
2489					goto update_disksize;
2490				return err;
2491			}
2492			ext4_msg(sb, KERN_CRIT,
2493				 "Delayed block allocation failed for "
2494				 "inode %lu at logical offset %llu with"
2495				 " max blocks %u with error %d",
2496				 inode->i_ino,
2497				 (unsigned long long)map->m_lblk,
2498				 (unsigned)map->m_len, -err);
2499			ext4_msg(sb, KERN_CRIT,
2500				 "This should not happen!! Data will "
2501				 "be lost\n");
2502			if (err == -ENOSPC)
2503				ext4_print_free_blocks(inode);
2504		invalidate_dirty_pages:
2505			*give_up_on_write = true;
2506			return err;
2507		}
2508		progress = 1;
2509		/*
2510		 * Update buffer state, submit mapped pages, and get us new
2511		 * extent to map
2512		 */
2513		err = mpage_map_and_submit_buffers(mpd);
2514		if (err < 0)
2515			goto update_disksize;
2516	} while (map->m_len);
2517
2518update_disksize:
2519	/*
2520	 * Update on-disk size after IO is submitted.  Races with
2521	 * truncate are avoided by checking i_size under i_data_sem.
2522	 */
2523	disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2524	if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2525		int err2;
2526		loff_t i_size;
2527
2528		down_write(&EXT4_I(inode)->i_data_sem);
2529		i_size = i_size_read(inode);
2530		if (disksize > i_size)
2531			disksize = i_size;
2532		if (disksize > EXT4_I(inode)->i_disksize)
2533			EXT4_I(inode)->i_disksize = disksize;
2534		up_write(&EXT4_I(inode)->i_data_sem);
2535		err2 = ext4_mark_inode_dirty(handle, inode);
2536		if (err2) {
2537			ext4_error_err(inode->i_sb, -err2,
2538				       "Failed to mark inode %lu dirty",
2539				       inode->i_ino);
2540		}
2541		if (!err)
2542			err = err2;
2543	}
2544	return err;
2545}
2546
2547/*
2548 * Calculate the total number of credits to reserve for one writepages
2549 * iteration. This is called from ext4_writepages(). We map an extent of
2550 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2551 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2552 * bpp - 1 blocks in bpp different extents.
2553 */
2554static int ext4_da_writepages_trans_blocks(struct inode *inode)
2555{
2556	int bpp = ext4_journal_blocks_per_page(inode);
2557
2558	return ext4_meta_trans_blocks(inode,
2559				MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2560}
2561
2562/* Return true if the page needs to be written as part of transaction commit */
2563static bool ext4_page_nomap_can_writeout(struct page *page)
2564{
2565	struct buffer_head *bh, *head;
2566
2567	bh = head = page_buffers(page);
2568	do {
2569		if (buffer_dirty(bh) && buffer_mapped(bh) && !buffer_delay(bh))
2570			return true;
2571	} while ((bh = bh->b_this_page) != head);
2572	return false;
 
 
 
 
 
2573}
2574
2575/*
2576 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2577 * 				 needing mapping, submit mapped pages
2578 *
2579 * @mpd - where to look for pages
2580 *
2581 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2582 * IO immediately. If we cannot map blocks, we submit just already mapped
2583 * buffers in the page for IO and keep page dirty. When we can map blocks and
2584 * we find a page which isn't mapped we start accumulating extent of buffers
2585 * underlying these pages that needs mapping (formed by either delayed or
2586 * unwritten buffers). We also lock the pages containing these buffers. The
2587 * extent found is returned in @mpd structure (starting at mpd->lblk with
2588 * length mpd->len blocks).
2589 *
2590 * Note that this function can attach bios to one io_end structure which are
2591 * neither logically nor physically contiguous. Although it may seem as an
2592 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2593 * case as we need to track IO to all buffers underlying a page in one io_end.
2594 */
2595static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2596{
2597	struct address_space *mapping = mpd->inode->i_mapping;
2598	struct pagevec pvec;
2599	unsigned int nr_pages;
2600	long left = mpd->wbc->nr_to_write;
2601	pgoff_t index = mpd->first_page;
2602	pgoff_t end = mpd->last_page;
2603	xa_mark_t tag;
2604	int i, err = 0;
2605	int blkbits = mpd->inode->i_blkbits;
2606	ext4_lblk_t lblk;
2607	struct buffer_head *head;
2608
2609	if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2610		tag = PAGECACHE_TAG_TOWRITE;
2611	else
2612		tag = PAGECACHE_TAG_DIRTY;
2613
2614	pagevec_init(&pvec);
2615	mpd->map.m_len = 0;
2616	mpd->next_page = index;
2617	while (index <= end) {
2618		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2619				tag);
2620		if (nr_pages == 0)
2621			break;
2622
2623		for (i = 0; i < nr_pages; i++) {
2624			struct page *page = pvec.pages[i];
2625
2626			/*
2627			 * Accumulated enough dirty pages? This doesn't apply
2628			 * to WB_SYNC_ALL mode. For integrity sync we have to
2629			 * keep going because someone may be concurrently
2630			 * dirtying pages, and we might have synced a lot of
2631			 * newly appeared dirty pages, but have not synced all
2632			 * of the old dirty pages.
2633			 */
2634			if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2635				goto out;
2636
2637			/* If we can't merge this page, we are done. */
2638			if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2639				goto out;
 
 
 
 
 
 
 
 
2640
2641			lock_page(page);
 
2642			/*
2643			 * If the page is no longer dirty, or its mapping no
2644			 * longer corresponds to inode we are writing (which
2645			 * means it has been truncated or invalidated), or the
2646			 * page is already under writeback and we are not doing
2647			 * a data integrity writeback, skip the page
 
2648			 */
2649			if (!PageDirty(page) ||
2650			    (PageWriteback(page) &&
2651			     (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2652			    unlikely(page->mapping != mapping)) {
2653				unlock_page(page);
2654				continue;
2655			}
2656
2657			wait_on_page_writeback(page);
2658			BUG_ON(PageWriteback(page));
2659
2660			/*
2661			 * Should never happen but for buggy code in
2662			 * other subsystems that call
2663			 * set_page_dirty() without properly warning
2664			 * the file system first.  See [1] for more
2665			 * information.
2666			 *
2667			 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2668			 */
2669			if (!page_has_buffers(page)) {
2670				ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2671				ClearPageDirty(page);
2672				unlock_page(page);
2673				continue;
2674			}
2675
2676			if (mpd->map.m_len == 0)
2677				mpd->first_page = page->index;
2678			mpd->next_page = page->index + 1;
2679			/*
2680			 * Writeout for transaction commit where we cannot
2681			 * modify metadata is simple. Just submit the page.
2682			 */
2683			if (!mpd->can_map) {
2684				if (ext4_page_nomap_can_writeout(page)) {
2685					err = mpage_submit_page(mpd, page);
2686					if (err < 0)
2687						goto out;
2688				} else {
2689					unlock_page(page);
2690					mpd->first_page++;
2691				}
2692			} else {
2693				/* Add all dirty buffers to mpd */
2694				lblk = ((ext4_lblk_t)page->index) <<
2695					(PAGE_SHIFT - blkbits);
 
2696				head = page_buffers(page);
2697				err = mpage_process_page_bufs(mpd, head, head,
2698							      lblk);
2699				if (err <= 0)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2700					goto out;
2701				err = 0;
2702			}
2703			left--;
2704		}
2705		pagevec_release(&pvec);
2706		cond_resched();
2707	}
2708	mpd->scanned_until_end = 1;
2709	return 0;
 
 
2710out:
2711	pagevec_release(&pvec);
2712	return err;
 
2713}
2714
2715static int ext4_writepage_cb(struct page *page, struct writeback_control *wbc,
2716			     void *data)
2717{
2718	return ext4_writepage(page, wbc);
2719}
2720
2721static int ext4_do_writepages(struct mpage_da_data *mpd)
 
2722{
2723	struct writeback_control *wbc = mpd->wbc;
2724	pgoff_t	writeback_index = 0;
2725	long nr_to_write = wbc->nr_to_write;
2726	int range_whole = 0;
2727	int cycled = 1;
2728	handle_t *handle = NULL;
2729	struct inode *inode = mpd->inode;
2730	struct address_space *mapping = inode->i_mapping;
2731	int needed_blocks, rsv_blocks = 0, ret = 0;
 
 
 
 
 
2732	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2733	struct blk_plug plug;
2734	bool give_up_on_write = false;
2735
2736	trace_ext4_writepages(inode, wbc);
2737
2738	/*
2739	 * No pages to write? This is mainly a kludge to avoid starting
2740	 * a transaction for special inodes like journal inode on last iput()
2741	 * because that could violate lock ordering on umount
2742	 */
2743	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2744		goto out_writepages;
2745
2746	if (ext4_should_journal_data(inode)) {
2747		blk_start_plug(&plug);
2748		ret = write_cache_pages(mapping, wbc, ext4_writepage_cb, NULL);
2749		blk_finish_plug(&plug);
2750		goto out_writepages;
2751	}
2752
2753	/*
2754	 * If the filesystem has aborted, it is read-only, so return
2755	 * right away instead of dumping stack traces later on that
2756	 * will obscure the real source of the problem.  We test
2757	 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2758	 * the latter could be true if the filesystem is mounted
2759	 * read-only, and in that case, ext4_writepages should
2760	 * *never* be called, so if that ever happens, we would want
2761	 * the stack trace.
2762	 */
2763	if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2764		     ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2765		ret = -EROFS;
2766		goto out_writepages;
2767	}
2768
2769	/*
2770	 * If we have inline data and arrive here, it means that
2771	 * we will soon create the block for the 1st page, so
2772	 * we'd better clear the inline data here.
2773	 */
2774	if (ext4_has_inline_data(inode)) {
2775		/* Just inode will be modified... */
2776		handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2777		if (IS_ERR(handle)) {
2778			ret = PTR_ERR(handle);
2779			goto out_writepages;
2780		}
2781		BUG_ON(ext4_test_inode_state(inode,
2782				EXT4_STATE_MAY_INLINE_DATA));
2783		ext4_destroy_inline_data(handle, inode);
2784		ext4_journal_stop(handle);
2785	}
2786
2787	if (ext4_should_dioread_nolock(inode)) {
2788		/*
2789		 * We may need to convert up to one extent per block in
2790		 * the page and we may dirty the inode.
2791		 */
2792		rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2793						PAGE_SIZE >> inode->i_blkbits);
2794	}
2795
2796	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2797		range_whole = 1;
2798
 
2799	if (wbc->range_cyclic) {
2800		writeback_index = mapping->writeback_index;
2801		if (writeback_index)
2802			cycled = 0;
2803		mpd->first_page = writeback_index;
2804		mpd->last_page = -1;
 
 
2805	} else {
2806		mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2807		mpd->last_page = wbc->range_end >> PAGE_SHIFT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2808	}
2809
2810	ext4_io_submit_init(&mpd->io_submit, wbc);
2811retry:
2812	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2813		tag_pages_for_writeback(mapping, mpd->first_page,
2814					mpd->last_page);
2815	blk_start_plug(&plug);
2816
2817	/*
2818	 * First writeback pages that don't need mapping - we can avoid
2819	 * starting a transaction unnecessarily and also avoid being blocked
2820	 * in the block layer on device congestion while having transaction
2821	 * started.
2822	 */
2823	mpd->do_map = 0;
2824	mpd->scanned_until_end = 0;
2825	mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2826	if (!mpd->io_submit.io_end) {
2827		ret = -ENOMEM;
2828		goto unplug;
2829	}
2830	ret = mpage_prepare_extent_to_map(mpd);
2831	/* Unlock pages we didn't use */
2832	mpage_release_unused_pages(mpd, false);
2833	/* Submit prepared bio */
2834	ext4_io_submit(&mpd->io_submit);
2835	ext4_put_io_end_defer(mpd->io_submit.io_end);
2836	mpd->io_submit.io_end = NULL;
2837	if (ret < 0)
2838		goto unplug;
2839
2840	while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2841		/* For each extent of pages we use new io_end */
2842		mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2843		if (!mpd->io_submit.io_end) {
2844			ret = -ENOMEM;
2845			break;
2846		}
2847
2848		WARN_ON_ONCE(!mpd->can_map);
2849		/*
2850		 * We have two constraints: We find one extent to map and we
2851		 * must always write out whole page (makes a difference when
2852		 * blocksize < pagesize) so that we don't block on IO when we
2853		 * try to write out the rest of the page. Journalled mode is
2854		 * not supported by delalloc.
2855		 */
2856		BUG_ON(ext4_should_journal_data(inode));
2857		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2858
2859		/* start a new transaction */
2860		handle = ext4_journal_start_with_reserve(inode,
2861				EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2862		if (IS_ERR(handle)) {
2863			ret = PTR_ERR(handle);
2864			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2865			       "%ld pages, ino %lu; err %d", __func__,
2866				wbc->nr_to_write, inode->i_ino, ret);
2867			/* Release allocated io_end */
2868			ext4_put_io_end(mpd->io_submit.io_end);
2869			mpd->io_submit.io_end = NULL;
2870			break;
2871		}
2872		mpd->do_map = 1;
2873
2874		trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2875		ret = mpage_prepare_extent_to_map(mpd);
2876		if (!ret && mpd->map.m_len)
2877			ret = mpage_map_and_submit_extent(handle, mpd,
2878					&give_up_on_write);
2879		/*
2880		 * Caution: If the handle is synchronous,
2881		 * ext4_journal_stop() can wait for transaction commit
2882		 * to finish which may depend on writeback of pages to
2883		 * complete or on page lock to be released.  In that
2884		 * case, we have to wait until after we have
2885		 * submitted all the IO, released page locks we hold,
2886		 * and dropped io_end reference (for extent conversion
2887		 * to be able to complete) before stopping the handle.
2888		 */
2889		if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2890			ext4_journal_stop(handle);
2891			handle = NULL;
2892			mpd->do_map = 0;
 
 
 
 
 
2893		}
2894		/* Unlock pages we didn't use */
2895		mpage_release_unused_pages(mpd, give_up_on_write);
2896		/* Submit prepared bio */
2897		ext4_io_submit(&mpd->io_submit);
2898
2899		/*
2900		 * Drop our io_end reference we got from init. We have
2901		 * to be careful and use deferred io_end finishing if
2902		 * we are still holding the transaction as we can
2903		 * release the last reference to io_end which may end
2904		 * up doing unwritten extent conversion.
2905		 */
2906		if (handle) {
2907			ext4_put_io_end_defer(mpd->io_submit.io_end);
2908			ext4_journal_stop(handle);
2909		} else
2910			ext4_put_io_end(mpd->io_submit.io_end);
2911		mpd->io_submit.io_end = NULL;
2912
2913		if (ret == -ENOSPC && sbi->s_journal) {
2914			/*
2915			 * Commit the transaction which would
2916			 * free blocks released in the transaction
2917			 * and try again
2918			 */
2919			jbd2_journal_force_commit_nested(sbi->s_journal);
2920			ret = 0;
2921			continue;
2922		}
2923		/* Fatal error - ENOMEM, EIO... */
2924		if (ret)
 
 
 
 
 
 
 
 
 
 
2925			break;
2926	}
2927unplug:
2928	blk_finish_plug(&plug);
2929	if (!ret && !cycled && wbc->nr_to_write > 0) {
2930		cycled = 1;
2931		mpd->last_page = writeback_index - 1;
2932		mpd->first_page = 0;
 
2933		goto retry;
2934	}
2935
2936	/* Update index */
 
2937	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2938		/*
2939		 * Set the writeback_index so that range_cyclic
2940		 * mode will write it back later
2941		 */
2942		mapping->writeback_index = mpd->first_page;
2943
2944out_writepages:
2945	trace_ext4_writepages_result(inode, wbc, ret,
2946				     nr_to_write - wbc->nr_to_write);
2947	return ret;
2948}
2949
2950static int ext4_writepages(struct address_space *mapping,
2951			   struct writeback_control *wbc)
2952{
2953	struct super_block *sb = mapping->host->i_sb;
2954	struct mpage_da_data mpd = {
2955		.inode = mapping->host,
2956		.wbc = wbc,
2957		.can_map = 1,
2958	};
2959	int ret;
2960
2961	if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
2962		return -EIO;
2963
2964	percpu_down_read(&EXT4_SB(sb)->s_writepages_rwsem);
2965	ret = ext4_do_writepages(&mpd);
2966	percpu_up_read(&EXT4_SB(sb)->s_writepages_rwsem);
2967
2968	return ret;
2969}
2970
2971int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2972{
2973	struct writeback_control wbc = {
2974		.sync_mode = WB_SYNC_ALL,
2975		.nr_to_write = LONG_MAX,
2976		.range_start = jinode->i_dirty_start,
2977		.range_end = jinode->i_dirty_end,
2978	};
2979	struct mpage_da_data mpd = {
2980		.inode = jinode->i_vfs_inode,
2981		.wbc = &wbc,
2982		.can_map = 0,
2983	};
2984	return ext4_do_writepages(&mpd);
2985}
2986
2987static int ext4_dax_writepages(struct address_space *mapping,
2988			       struct writeback_control *wbc)
2989{
2990	int ret;
2991	long nr_to_write = wbc->nr_to_write;
2992	struct inode *inode = mapping->host;
2993	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2994
2995	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2996		return -EIO;
2997
2998	percpu_down_read(&sbi->s_writepages_rwsem);
2999	trace_ext4_writepages(inode, wbc);
3000
3001	ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
3002	trace_ext4_writepages_result(inode, wbc, ret,
3003				     nr_to_write - wbc->nr_to_write);
3004	percpu_up_read(&sbi->s_writepages_rwsem);
3005	return ret;
3006}
3007
 
3008static int ext4_nonda_switch(struct super_block *sb)
3009{
3010	s64 free_clusters, dirty_clusters;
3011	struct ext4_sb_info *sbi = EXT4_SB(sb);
3012
3013	/*
3014	 * switch to non delalloc mode if we are running low
3015	 * on free block. The free block accounting via percpu
3016	 * counters can get slightly wrong with percpu_counter_batch getting
3017	 * accumulated on each CPU without updating global counters
3018	 * Delalloc need an accurate free block accounting. So switch
3019	 * to non delalloc when we are near to error range.
3020	 */
3021	free_clusters =
3022		percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3023	dirty_clusters =
3024		percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3025	/*
3026	 * Start pushing delalloc when 1/2 of free blocks are dirty.
3027	 */
3028	if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3029		try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3030
3031	if (2 * free_clusters < 3 * dirty_clusters ||
3032	    free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3033		/*
3034		 * free block count is less than 150% of dirty blocks
3035		 * or free blocks is less than watermark
3036		 */
3037		return 1;
3038	}
 
 
 
 
 
 
 
3039	return 0;
3040}
3041
3042static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3043			       loff_t pos, unsigned len,
3044			       struct page **pagep, void **fsdata)
3045{
3046	int ret, retries = 0;
3047	struct page *page;
3048	pgoff_t index;
3049	struct inode *inode = mapping->host;
 
3050
3051	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3052		return -EIO;
3053
3054	index = pos >> PAGE_SHIFT;
3055
3056	if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
3057		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3058		return ext4_write_begin(file, mapping, pos,
3059					len, pagep, fsdata);
3060	}
3061	*fsdata = (void *)0;
3062	trace_ext4_da_write_begin(inode, pos, len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3063
3064	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3065		ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
3066						      pagep, fsdata);
3067		if (ret < 0)
3068			return ret;
3069		if (ret == 1)
3070			return 0;
3071	}
 
3072
3073retry:
3074	page = grab_cache_page_write_begin(mapping, index);
3075	if (!page)
3076		return -ENOMEM;
3077
3078	/* In case writeback began while the page was unlocked */
3079	wait_for_stable_page(page);
3080
3081#ifdef CONFIG_FS_ENCRYPTION
3082	ret = ext4_block_write_begin(page, pos, len,
3083				     ext4_da_get_block_prep);
3084#else
3085	ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3086#endif
3087	if (ret < 0) {
3088		unlock_page(page);
3089		put_page(page);
 
3090		/*
3091		 * block_write_begin may have instantiated a few blocks
3092		 * outside i_size.  Trim these off again. Don't need
3093		 * i_size_read because we hold inode lock.
3094		 */
3095		if (pos + len > inode->i_size)
3096			ext4_truncate_failed_write(inode);
3097
3098		if (ret == -ENOSPC &&
3099		    ext4_should_retry_alloc(inode->i_sb, &retries))
3100			goto retry;
3101		return ret;
3102	}
3103
3104	*pagep = page;
 
 
3105	return ret;
3106}
3107
3108/*
3109 * Check if we should update i_disksize
3110 * when write to the end of file but not require block allocation
3111 */
3112static int ext4_da_should_update_i_disksize(struct page *page,
3113					    unsigned long offset)
3114{
3115	struct buffer_head *bh;
3116	struct inode *inode = page->mapping->host;
3117	unsigned int idx;
3118	int i;
3119
3120	bh = page_buffers(page);
3121	idx = offset >> inode->i_blkbits;
3122
3123	for (i = 0; i < idx; i++)
3124		bh = bh->b_this_page;
3125
3126	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3127		return 0;
3128	return 1;
3129}
3130
3131static int ext4_da_write_end(struct file *file,
3132			     struct address_space *mapping,
3133			     loff_t pos, unsigned len, unsigned copied,
3134			     struct page *page, void *fsdata)
3135{
3136	struct inode *inode = mapping->host;
 
 
3137	loff_t new_i_size;
3138	unsigned long start, end;
3139	int write_mode = (int)(unsigned long)fsdata;
3140
3141	if (write_mode == FALL_BACK_TO_NONDELALLOC)
3142		return ext4_write_end(file, mapping, pos,
3143				      len, copied, page, fsdata);
 
 
 
 
 
 
 
 
3144
3145	trace_ext4_da_write_end(inode, pos, len, copied);
 
 
 
 
 
 
 
 
3146
3147	if (write_mode != CONVERT_INLINE_DATA &&
3148	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3149	    ext4_has_inline_data(inode))
3150		return ext4_write_inline_data_end(inode, pos, len, copied, page);
 
 
 
 
 
 
 
 
3151
3152	start = pos & (PAGE_SIZE - 1);
3153	end = start + copied - 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3154
 
 
3155	/*
3156	 * Since we are holding inode lock, we are sure i_disksize <=
3157	 * i_size. We also know that if i_disksize < i_size, there are
3158	 * delalloc writes pending in the range upto i_size. If the end of
3159	 * the current write is <= i_size, there's no need to touch
3160	 * i_disksize since writeback will push i_disksize upto i_size
3161	 * eventually. If the end of the current write is > i_size and
3162	 * inside an allocated block (ext4_da_should_update_i_disksize()
3163	 * check), we need to update i_disksize here as neither
3164	 * ext4_writepage() nor certain ext4_writepages() paths not
3165	 * allocating blocks update i_disksize.
3166	 *
3167	 * Note that we defer inode dirtying to generic_write_end() /
3168	 * ext4_da_write_inline_data_end().
3169	 */
3170	new_i_size = pos + copied;
3171	if (copied && new_i_size > inode->i_size &&
3172	    ext4_da_should_update_i_disksize(page, end))
3173		ext4_update_i_disksize(inode, new_i_size);
 
 
 
 
3174
3175	return generic_write_end(file, mapping, pos, len, copied, page, fsdata);
3176}
3177
3178/*
3179 * Force all delayed allocation blocks to be allocated for a given inode.
3180 */
3181int ext4_alloc_da_blocks(struct inode *inode)
3182{
3183	trace_ext4_alloc_da_blocks(inode);
3184
3185	if (!EXT4_I(inode)->i_reserved_data_blocks)
 
3186		return 0;
3187
3188	/*
3189	 * We do something simple for now.  The filemap_flush() will
3190	 * also start triggering a write of the data blocks, which is
3191	 * not strictly speaking necessary (and for users of
3192	 * laptop_mode, not even desirable).  However, to do otherwise
3193	 * would require replicating code paths in:
3194	 *
3195	 * ext4_writepages() ->
3196	 *    write_cache_pages() ---> (via passed in callback function)
3197	 *        __mpage_da_writepage() -->
3198	 *           mpage_add_bh_to_extent()
3199	 *           mpage_da_map_blocks()
3200	 *
3201	 * The problem is that write_cache_pages(), located in
3202	 * mm/page-writeback.c, marks pages clean in preparation for
3203	 * doing I/O, which is not desirable if we're not planning on
3204	 * doing I/O at all.
3205	 *
3206	 * We could call write_cache_pages(), and then redirty all of
3207	 * the pages by calling redirty_page_for_writepage() but that
3208	 * would be ugly in the extreme.  So instead we would need to
3209	 * replicate parts of the code in the above functions,
3210	 * simplifying them because we wouldn't actually intend to
3211	 * write out the pages, but rather only collect contiguous
3212	 * logical block extents, call the multi-block allocator, and
3213	 * then update the buffer heads with the block allocations.
3214	 *
3215	 * For now, though, we'll cheat by calling filemap_flush(),
3216	 * which will map the blocks, and start the I/O, but not
3217	 * actually wait for the I/O to complete.
3218	 */
3219	return filemap_flush(inode->i_mapping);
3220}
3221
3222/*
3223 * bmap() is special.  It gets used by applications such as lilo and by
3224 * the swapper to find the on-disk block of a specific piece of data.
3225 *
3226 * Naturally, this is dangerous if the block concerned is still in the
3227 * journal.  If somebody makes a swapfile on an ext4 data-journaling
3228 * filesystem and enables swap, then they may get a nasty shock when the
3229 * data getting swapped to that swapfile suddenly gets overwritten by
3230 * the original zero's written out previously to the journal and
3231 * awaiting writeback in the kernel's buffer cache.
3232 *
3233 * So, if we see any bmap calls here on a modified, data-journaled file,
3234 * take extra steps to flush any blocks which might be in the cache.
3235 */
3236static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3237{
3238	struct inode *inode = mapping->host;
3239	journal_t *journal;
3240	sector_t ret = 0;
3241	int err;
3242
3243	inode_lock_shared(inode);
3244	/*
3245	 * We can get here for an inline file via the FIBMAP ioctl
3246	 */
3247	if (ext4_has_inline_data(inode))
3248		goto out;
3249
3250	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3251			test_opt(inode->i_sb, DELALLOC)) {
3252		/*
3253		 * With delalloc we want to sync the file
3254		 * so that we can make sure we allocate
3255		 * blocks for file
3256		 */
3257		filemap_write_and_wait(mapping);
3258	}
3259
3260	if (EXT4_JOURNAL(inode) &&
3261	    ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3262		/*
3263		 * This is a REALLY heavyweight approach, but the use of
3264		 * bmap on dirty files is expected to be extremely rare:
3265		 * only if we run lilo or swapon on a freshly made file
3266		 * do we expect this to happen.
3267		 *
3268		 * (bmap requires CAP_SYS_RAWIO so this does not
3269		 * represent an unprivileged user DOS attack --- we'd be
3270		 * in trouble if mortal users could trigger this path at
3271		 * will.)
3272		 *
3273		 * NB. EXT4_STATE_JDATA is not set on files other than
3274		 * regular files.  If somebody wants to bmap a directory
3275		 * or symlink and gets confused because the buffer
3276		 * hasn't yet been flushed to disk, they deserve
3277		 * everything they get.
3278		 */
3279
3280		ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3281		journal = EXT4_JOURNAL(inode);
3282		jbd2_journal_lock_updates(journal);
3283		err = jbd2_journal_flush(journal, 0);
3284		jbd2_journal_unlock_updates(journal);
3285
3286		if (err)
3287			goto out;
3288	}
3289
3290	ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3291
3292out:
3293	inode_unlock_shared(inode);
3294	return ret;
3295}
3296
3297static int ext4_read_folio(struct file *file, struct folio *folio)
3298{
3299	struct page *page = &folio->page;
3300	int ret = -EAGAIN;
3301	struct inode *inode = page->mapping->host;
3302
3303	trace_ext4_readpage(page);
3304
3305	if (ext4_has_inline_data(inode))
3306		ret = ext4_readpage_inline(inode, page);
3307
3308	if (ret == -EAGAIN)
3309		return ext4_mpage_readpages(inode, NULL, page);
3310
3311	return ret;
3312}
3313
3314static void ext4_readahead(struct readahead_control *rac)
 
 
3315{
3316	struct inode *inode = rac->mapping->host;
3317
3318	/* If the file has inline data, no need to do readahead. */
3319	if (ext4_has_inline_data(inode))
3320		return;
3321
3322	ext4_mpage_readpages(inode, rac, NULL);
3323}
3324
3325static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3326				size_t length)
3327{
3328	trace_ext4_invalidate_folio(folio, offset, length);
 
3329
3330	/* No journalling happens on data buffers when this function is used */
3331	WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3332
3333	block_invalidate_folio(folio, offset, length);
 
 
 
 
 
 
 
 
 
3334}
3335
3336static int __ext4_journalled_invalidate_folio(struct folio *folio,
3337					    size_t offset, size_t length)
3338{
3339	journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3340
3341	trace_ext4_journalled_invalidate_folio(folio, offset, length);
3342
3343	/*
 
 
 
 
 
3344	 * If it's a full truncate we just forget about the pending dirtying
3345	 */
3346	if (offset == 0 && length == folio_size(folio))
3347		folio_clear_checked(folio);
3348
3349	return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3350}
3351
3352/* Wrapper for aops... */
3353static void ext4_journalled_invalidate_folio(struct folio *folio,
3354					   size_t offset,
3355					   size_t length)
3356{
3357	WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3358}
3359
3360static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3361{
3362	journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3363
3364	trace_ext4_releasepage(&folio->page);
3365
3366	/* Page has dirty journalled data -> cannot release */
3367	if (folio_test_checked(folio))
3368		return false;
3369	if (journal)
3370		return jbd2_journal_try_to_free_buffers(journal, folio);
3371	else
3372		return try_to_free_buffers(folio);
3373}
3374
3375static bool ext4_inode_datasync_dirty(struct inode *inode)
 
 
 
 
 
 
3376{
3377	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3378
3379	if (journal) {
3380		if (jbd2_transaction_committed(journal,
3381			EXT4_I(inode)->i_datasync_tid))
3382			return false;
3383		if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3384			return !list_empty(&EXT4_I(inode)->i_fc_list);
3385		return true;
3386	}
3387
3388	/* Any metadata buffers to write? */
3389	if (!list_empty(&inode->i_mapping->private_list))
3390		return true;
3391	return inode->i_state & I_DIRTY_DATASYNC;
3392}
3393
3394static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3395			   struct ext4_map_blocks *map, loff_t offset,
3396			   loff_t length, unsigned int flags)
3397{
3398	u8 blkbits = inode->i_blkbits;
 
 
 
 
3399
3400	/*
3401	 * Writes that span EOF might trigger an I/O size update on completion,
3402	 * so consider them to be dirty for the purpose of O_DSYNC, even if
3403	 * there is no other metadata changes being made or are pending.
3404	 */
3405	iomap->flags = 0;
3406	if (ext4_inode_datasync_dirty(inode) ||
3407	    offset + length > i_size_read(inode))
3408		iomap->flags |= IOMAP_F_DIRTY;
3409
3410	if (map->m_flags & EXT4_MAP_NEW)
3411		iomap->flags |= IOMAP_F_NEW;
 
 
 
 
 
 
 
 
 
 
 
 
 
3412
3413	if (flags & IOMAP_DAX)
3414		iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3415	else
3416		iomap->bdev = inode->i_sb->s_bdev;
3417	iomap->offset = (u64) map->m_lblk << blkbits;
3418	iomap->length = (u64) map->m_len << blkbits;
3419
3420	if ((map->m_flags & EXT4_MAP_MAPPED) &&
3421	    !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3422		iomap->flags |= IOMAP_F_MERGED;
3423
3424	/*
3425	 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3426	 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3427	 * set. In order for any allocated unwritten extents to be converted
3428	 * into written extents correctly within the ->end_io() handler, we
3429	 * need to ensure that the iomap->type is set appropriately. Hence, the
3430	 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3431	 * been set first.
3432	 */
3433	if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3434		iomap->type = IOMAP_UNWRITTEN;
3435		iomap->addr = (u64) map->m_pblk << blkbits;
3436		if (flags & IOMAP_DAX)
3437			iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3438	} else if (map->m_flags & EXT4_MAP_MAPPED) {
3439		iomap->type = IOMAP_MAPPED;
3440		iomap->addr = (u64) map->m_pblk << blkbits;
3441		if (flags & IOMAP_DAX)
3442			iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3443	} else {
3444		iomap->type = IOMAP_HOLE;
3445		iomap->addr = IOMAP_NULL_ADDR;
3446	}
3447}
3448
3449static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3450			    unsigned int flags)
3451{
3452	handle_t *handle;
3453	u8 blkbits = inode->i_blkbits;
3454	int ret, dio_credits, m_flags = 0, retries = 0;
3455
3456	/*
3457	 * Trim the mapping request to the maximum value that we can map at
3458	 * once for direct I/O.
3459	 */
3460	if (map->m_len > DIO_MAX_BLOCKS)
3461		map->m_len = DIO_MAX_BLOCKS;
3462	dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3463
3464retry:
3465	/*
3466	 * Either we allocate blocks and then don't get an unwritten extent, so
3467	 * in that case we have reserved enough credits. Or, the blocks are
3468	 * already allocated and unwritten. In that case, the extent conversion
3469	 * fits into the credits as well.
3470	 */
3471	handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3472	if (IS_ERR(handle))
3473		return PTR_ERR(handle);
3474
3475	/*
3476	 * DAX and direct I/O are the only two operations that are currently
3477	 * supported with IOMAP_WRITE.
3478	 */
3479	WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3480	if (flags & IOMAP_DAX)
3481		m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3482	/*
3483	 * We use i_size instead of i_disksize here because delalloc writeback
3484	 * can complete at any point during the I/O and subsequently push the
3485	 * i_disksize out to i_size. This could be beyond where direct I/O is
3486	 * happening and thus expose allocated blocks to direct I/O reads.
3487	 */
3488	else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3489		m_flags = EXT4_GET_BLOCKS_CREATE;
3490	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3491		m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3492
3493	ret = ext4_map_blocks(handle, inode, map, m_flags);
 
 
3494
3495	/*
3496	 * We cannot fill holes in indirect tree based inodes as that could
3497	 * expose stale data in the case of a crash. Use the magic error code
3498	 * to fallback to buffered I/O.
3499	 */
3500	if (!m_flags && !ret)
3501		ret = -ENOTBLK;
3502
3503	ext4_journal_stop(handle);
3504	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3505		goto retry;
3506
3507	return ret;
3508}
3509
3510
3511static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3512		unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3513{
3514	int ret;
3515	struct ext4_map_blocks map;
3516	u8 blkbits = inode->i_blkbits;
 
3517
3518	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3519		return -EINVAL;
3520
3521	if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3522		return -ERANGE;
 
 
 
 
3523
3524	/*
3525	 * Calculate the first and last logical blocks respectively.
 
3526	 */
3527	map.m_lblk = offset >> blkbits;
3528	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3529			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3530
3531	if (flags & IOMAP_WRITE) {
3532		/*
3533		 * We check here if the blocks are already allocated, then we
3534		 * don't need to start a journal txn and we can directly return
3535		 * the mapping information. This could boost performance
3536		 * especially in multi-threaded overwrite requests.
3537		 */
3538		if (offset + length <= i_size_read(inode)) {
3539			ret = ext4_map_blocks(NULL, inode, &map, 0);
3540			if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3541				goto out;
3542		}
3543		ret = ext4_iomap_alloc(inode, &map, flags);
3544	} else {
3545		ret = ext4_map_blocks(NULL, inode, &map, 0);
3546	}
3547
3548	if (ret < 0)
3549		return ret;
3550out:
3551	/*
3552	 * When inline encryption is enabled, sometimes I/O to an encrypted file
3553	 * has to be broken up to guarantee DUN contiguity.  Handle this by
3554	 * limiting the length of the mapping returned.
3555	 */
3556	map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3557
3558	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3559
3560	return 0;
3561}
3562
3563static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3564		loff_t length, unsigned flags, struct iomap *iomap,
3565		struct iomap *srcmap)
3566{
3567	int ret;
 
3568
 
 
 
 
 
 
 
 
 
3569	/*
3570	 * Even for writes we don't need to allocate blocks, so just pretend
3571	 * we are reading to save overhead of starting a transaction.
 
3572	 */
3573	flags &= ~IOMAP_WRITE;
3574	ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3575	WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3576	return ret;
3577}
3578
3579static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3580			  ssize_t written, unsigned flags, struct iomap *iomap)
3581{
3582	/*
3583	 * Check to see whether an error occurred while writing out the data to
3584	 * the allocated blocks. If so, return the magic error code so that we
3585	 * fallback to buffered I/O and attempt to complete the remainder of
3586	 * the I/O. Any blocks that may have been allocated in preparation for
3587	 * the direct I/O will be reused during buffered I/O.
3588	 */
3589	if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3590		return -ENOTBLK;
3591
 
 
3592	return 0;
3593}
3594
3595const struct iomap_ops ext4_iomap_ops = {
3596	.iomap_begin		= ext4_iomap_begin,
3597	.iomap_end		= ext4_iomap_end,
3598};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3599
3600const struct iomap_ops ext4_iomap_overwrite_ops = {
3601	.iomap_begin		= ext4_iomap_overwrite_begin,
3602	.iomap_end		= ext4_iomap_end,
3603};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3604
3605static bool ext4_iomap_is_delalloc(struct inode *inode,
3606				   struct ext4_map_blocks *map)
3607{
3608	struct extent_status es;
3609	ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3610
3611	ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3612				  map->m_lblk, end, &es);
3613
3614	if (!es.es_len || es.es_lblk > end)
3615		return false;
3616
3617	if (es.es_lblk > map->m_lblk) {
3618		map->m_len = es.es_lblk - map->m_lblk;
3619		return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3620	}
3621
3622	offset = map->m_lblk - es.es_lblk;
3623	map->m_len = es.es_len - offset;
3624
3625	return true;
3626}
3627
3628static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3629				   loff_t length, unsigned int flags,
3630				   struct iomap *iomap, struct iomap *srcmap)
3631{
3632	int ret;
3633	bool delalloc = false;
3634	struct ext4_map_blocks map;
3635	u8 blkbits = inode->i_blkbits;
3636
3637	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3638		return -EINVAL;
3639
3640	if (ext4_has_inline_data(inode)) {
3641		ret = ext4_inline_data_iomap(inode, iomap);
3642		if (ret != -EAGAIN) {
3643			if (ret == 0 && offset >= iomap->length)
3644				ret = -ENOENT;
3645			return ret;
3646		}
3647	}
3648
3649	/*
3650	 * Calculate the first and last logical block respectively.
3651	 */
3652	map.m_lblk = offset >> blkbits;
3653	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3654			  EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3655
3656	/*
3657	 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3658	 * So handle it here itself instead of querying ext4_map_blocks().
3659	 * Since ext4_map_blocks() will warn about it and will return
3660	 * -EIO error.
3661	 */
3662	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3663		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3664
3665		if (offset >= sbi->s_bitmap_maxbytes) {
3666			map.m_flags = 0;
3667			goto set_iomap;
3668		}
3669	}
3670
3671	ret = ext4_map_blocks(NULL, inode, &map, 0);
3672	if (ret < 0)
3673		return ret;
3674	if (ret == 0)
3675		delalloc = ext4_iomap_is_delalloc(inode, &map);
3676
3677set_iomap:
3678	ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3679	if (delalloc && iomap->type == IOMAP_HOLE)
3680		iomap->type = IOMAP_DELALLOC;
3681
3682	return 0;
3683}
3684
3685const struct iomap_ops ext4_iomap_report_ops = {
3686	.iomap_begin = ext4_iomap_begin_report,
3687};
3688
3689/*
3690 * Whenever the folio is being dirtied, corresponding buffers should already
3691 * be attached to the transaction (we take care of this in ext4_page_mkwrite()
3692 * and ext4_write_begin()). However we cannot move buffers to dirty transaction
3693 * lists here because ->dirty_folio is called under VFS locks and the folio
3694 * is not necessarily locked.
3695 *
3696 * We cannot just dirty the folio and leave attached buffers clean, because the
3697 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3698 * or jbddirty because all the journalling code will explode.
3699 *
3700 * So what we do is to mark the folio "pending dirty" and next time writepage
3701 * is called, propagate that into the buffers appropriately.
3702 */
3703static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3704		struct folio *folio)
3705{
3706	WARN_ON_ONCE(!folio_buffers(folio));
3707	folio_set_checked(folio);
3708	return filemap_dirty_folio(mapping, folio);
3709}
3710
3711static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3712{
3713	WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3714	WARN_ON_ONCE(!folio_buffers(folio));
3715	return block_dirty_folio(mapping, folio);
3716}
 
 
 
 
 
 
 
 
3717
3718static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3719				    struct file *file, sector_t *span)
3720{
3721	return iomap_swapfile_activate(sis, file, span,
3722				       &ext4_iomap_report_ops);
3723}
3724
3725static const struct address_space_operations ext4_aops = {
3726	.read_folio		= ext4_read_folio,
3727	.readahead		= ext4_readahead,
3728	.writepages		= ext4_writepages,
3729	.write_begin		= ext4_write_begin,
3730	.write_end		= ext4_write_end,
3731	.dirty_folio		= ext4_dirty_folio,
3732	.bmap			= ext4_bmap,
3733	.invalidate_folio	= ext4_invalidate_folio,
3734	.release_folio		= ext4_release_folio,
3735	.direct_IO		= noop_direct_IO,
3736	.migrate_folio		= buffer_migrate_folio,
3737	.is_partially_uptodate  = block_is_partially_uptodate,
3738	.error_remove_page	= generic_error_remove_page,
3739	.swap_activate		= ext4_iomap_swap_activate,
3740};
3741
3742static const struct address_space_operations ext4_journalled_aops = {
3743	.read_folio		= ext4_read_folio,
3744	.readahead		= ext4_readahead,
3745	.writepages		= ext4_writepages,
3746	.write_begin		= ext4_write_begin,
3747	.write_end		= ext4_journalled_write_end,
3748	.dirty_folio		= ext4_journalled_dirty_folio,
3749	.bmap			= ext4_bmap,
3750	.invalidate_folio	= ext4_journalled_invalidate_folio,
3751	.release_folio		= ext4_release_folio,
3752	.direct_IO		= noop_direct_IO,
3753	.migrate_folio		= buffer_migrate_folio_norefs,
3754	.is_partially_uptodate  = block_is_partially_uptodate,
3755	.error_remove_page	= generic_error_remove_page,
3756	.swap_activate		= ext4_iomap_swap_activate,
3757};
3758
3759static const struct address_space_operations ext4_da_aops = {
3760	.read_folio		= ext4_read_folio,
3761	.readahead		= ext4_readahead,
3762	.writepages		= ext4_writepages,
 
3763	.write_begin		= ext4_da_write_begin,
3764	.write_end		= ext4_da_write_end,
3765	.dirty_folio		= ext4_dirty_folio,
3766	.bmap			= ext4_bmap,
3767	.invalidate_folio	= ext4_invalidate_folio,
3768	.release_folio		= ext4_release_folio,
3769	.direct_IO		= noop_direct_IO,
3770	.migrate_folio		= buffer_migrate_folio,
3771	.is_partially_uptodate  = block_is_partially_uptodate,
3772	.error_remove_page	= generic_error_remove_page,
3773	.swap_activate		= ext4_iomap_swap_activate,
3774};
3775
3776static const struct address_space_operations ext4_dax_aops = {
3777	.writepages		= ext4_dax_writepages,
3778	.direct_IO		= noop_direct_IO,
3779	.dirty_folio		= noop_dirty_folio,
3780	.bmap			= ext4_bmap,
3781	.swap_activate		= ext4_iomap_swap_activate,
3782};
3783
3784void ext4_set_aops(struct inode *inode)
3785{
3786	switch (ext4_inode_journal_mode(inode)) {
3787	case EXT4_INODE_ORDERED_DATA_MODE:
3788	case EXT4_INODE_WRITEBACK_DATA_MODE:
3789		break;
3790	case EXT4_INODE_JOURNAL_DATA_MODE:
3791		inode->i_mapping->a_ops = &ext4_journalled_aops;
3792		return;
3793	default:
3794		BUG();
3795	}
3796	if (IS_DAX(inode))
3797		inode->i_mapping->a_ops = &ext4_dax_aops;
3798	else if (test_opt(inode->i_sb, DELALLOC))
3799		inode->i_mapping->a_ops = &ext4_da_aops;
 
 
3800	else
3801		inode->i_mapping->a_ops = &ext4_aops;
3802}
3803
3804static int __ext4_block_zero_page_range(handle_t *handle,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3805		struct address_space *mapping, loff_t from, loff_t length)
3806{
3807	ext4_fsblk_t index = from >> PAGE_SHIFT;
3808	unsigned offset = from & (PAGE_SIZE-1);
3809	unsigned blocksize, pos;
3810	ext4_lblk_t iblock;
3811	struct inode *inode = mapping->host;
3812	struct buffer_head *bh;
3813	struct page *page;
3814	int err = 0;
3815
3816	page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3817				   mapping_gfp_constraint(mapping, ~__GFP_FS));
3818	if (!page)
3819		return -ENOMEM;
3820
3821	blocksize = inode->i_sb->s_blocksize;
 
 
 
 
 
 
 
 
3822
3823	iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3824
3825	if (!page_has_buffers(page))
3826		create_empty_buffers(page, blocksize, 0);
3827
3828	/* Find the buffer that contains "offset" */
3829	bh = page_buffers(page);
3830	pos = blocksize;
3831	while (offset >= pos) {
3832		bh = bh->b_this_page;
3833		iblock++;
3834		pos += blocksize;
3835	}
 
 
3836	if (buffer_freed(bh)) {
3837		BUFFER_TRACE(bh, "freed: skip");
3838		goto unlock;
3839	}
 
3840	if (!buffer_mapped(bh)) {
3841		BUFFER_TRACE(bh, "unmapped");
3842		ext4_get_block(inode, iblock, bh, 0);
3843		/* unmapped? It's a hole - nothing to do */
3844		if (!buffer_mapped(bh)) {
3845			BUFFER_TRACE(bh, "still unmapped");
3846			goto unlock;
3847		}
3848	}
3849
3850	/* Ok, it's mapped. Make sure it's up-to-date */
3851	if (PageUptodate(page))
3852		set_buffer_uptodate(bh);
3853
3854	if (!buffer_uptodate(bh)) {
3855		err = ext4_read_bh_lock(bh, 0, true);
3856		if (err)
 
 
 
3857			goto unlock;
3858		if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3859			/* We expect the key to be set. */
3860			BUG_ON(!fscrypt_has_encryption_key(inode));
3861			err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3862							       bh_offset(bh));
3863			if (err) {
3864				clear_buffer_uptodate(bh);
3865				goto unlock;
3866			}
3867		}
3868	}
 
3869	if (ext4_should_journal_data(inode)) {
3870		BUFFER_TRACE(bh, "get write access");
3871		err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3872						    EXT4_JTR_NONE);
3873		if (err)
3874			goto unlock;
3875	}
 
3876	zero_user(page, offset, length);
 
3877	BUFFER_TRACE(bh, "zeroed end of block");
3878
 
3879	if (ext4_should_journal_data(inode)) {
3880		err = ext4_handle_dirty_metadata(handle, inode, bh);
3881	} else {
3882		err = 0;
 
3883		mark_buffer_dirty(bh);
3884		if (ext4_should_order_data(inode))
3885			err = ext4_jbd2_inode_add_write(handle, inode, from,
3886					length);
3887	}
3888
3889unlock:
3890	unlock_page(page);
3891	put_page(page);
3892	return err;
3893}
3894
3895/*
3896 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3897 * starting from file offset 'from'.  The range to be zero'd must
3898 * be contained with in one block.  If the specified range exceeds
3899 * the end of the block it will be shortened to end of the block
3900 * that corresponds to 'from'
3901 */
3902static int ext4_block_zero_page_range(handle_t *handle,
3903		struct address_space *mapping, loff_t from, loff_t length)
3904{
3905	struct inode *inode = mapping->host;
3906	unsigned offset = from & (PAGE_SIZE-1);
3907	unsigned blocksize = inode->i_sb->s_blocksize;
3908	unsigned max = blocksize - (offset & (blocksize - 1));
3909
3910	/*
3911	 * correct length if it does not fall between
3912	 * 'from' and the end of the block
3913	 */
3914	if (length > max || length < 0)
3915		length = max;
3916
3917	if (IS_DAX(inode)) {
3918		return dax_zero_range(inode, from, length, NULL,
3919				      &ext4_iomap_ops);
3920	}
3921	return __ext4_block_zero_page_range(handle, mapping, from, length);
3922}
3923
3924/*
3925 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3926 * up to the end of the block which corresponds to `from'.
3927 * This required during truncate. We need to physically zero the tail end
3928 * of that block so it doesn't yield old data if the file is later grown.
3929 */
3930static int ext4_block_truncate_page(handle_t *handle,
3931		struct address_space *mapping, loff_t from)
3932{
3933	unsigned offset = from & (PAGE_SIZE-1);
3934	unsigned length;
3935	unsigned blocksize;
3936	struct inode *inode = mapping->host;
3937
3938	/* If we are processing an encrypted inode during orphan list handling */
3939	if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3940		return 0;
3941
3942	blocksize = inode->i_sb->s_blocksize;
3943	length = blocksize - (offset & (blocksize - 1));
3944
3945	return ext4_block_zero_page_range(handle, mapping, from, length);
3946}
3947
3948int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3949			     loff_t lstart, loff_t length)
3950{
3951	struct super_block *sb = inode->i_sb;
3952	struct address_space *mapping = inode->i_mapping;
3953	unsigned partial_start, partial_end;
3954	ext4_fsblk_t start, end;
3955	loff_t byte_end = (lstart + length - 1);
3956	int err = 0;
3957
3958	partial_start = lstart & (sb->s_blocksize - 1);
3959	partial_end = byte_end & (sb->s_blocksize - 1);
3960
3961	start = lstart >> sb->s_blocksize_bits;
3962	end = byte_end >> sb->s_blocksize_bits;
3963
3964	/* Handle partial zero within the single block */
3965	if (start == end &&
3966	    (partial_start || (partial_end != sb->s_blocksize - 1))) {
3967		err = ext4_block_zero_page_range(handle, mapping,
3968						 lstart, length);
3969		return err;
3970	}
3971	/* Handle partial zero out on the start of the range */
3972	if (partial_start) {
3973		err = ext4_block_zero_page_range(handle, mapping,
3974						 lstart, sb->s_blocksize);
3975		if (err)
3976			return err;
3977	}
3978	/* Handle partial zero out on the end of the range */
3979	if (partial_end != sb->s_blocksize - 1)
3980		err = ext4_block_zero_page_range(handle, mapping,
3981						 byte_end - partial_end,
3982						 partial_end + 1);
3983	return err;
3984}
3985
3986int ext4_can_truncate(struct inode *inode)
3987{
3988	if (S_ISREG(inode->i_mode))
3989		return 1;
3990	if (S_ISDIR(inode->i_mode))
3991		return 1;
3992	if (S_ISLNK(inode->i_mode))
3993		return !ext4_inode_is_fast_symlink(inode);
3994	return 0;
3995}
3996
3997/*
3998 * We have to make sure i_disksize gets properly updated before we truncate
3999 * page cache due to hole punching or zero range. Otherwise i_disksize update
4000 * can get lost as it may have been postponed to submission of writeback but
4001 * that will never happen after we truncate page cache.
4002 */
4003int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4004				      loff_t len)
4005{
4006	handle_t *handle;
4007	int ret;
4008
4009	loff_t size = i_size_read(inode);
4010
4011	WARN_ON(!inode_is_locked(inode));
4012	if (offset > size || offset + len < size)
4013		return 0;
4014
4015	if (EXT4_I(inode)->i_disksize >= size)
4016		return 0;
4017
4018	handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4019	if (IS_ERR(handle))
4020		return PTR_ERR(handle);
4021	ext4_update_i_disksize(inode, size);
4022	ret = ext4_mark_inode_dirty(handle, inode);
4023	ext4_journal_stop(handle);
4024
4025	return ret;
4026}
4027
4028static void ext4_wait_dax_page(struct inode *inode)
4029{
4030	filemap_invalidate_unlock(inode->i_mapping);
4031	schedule();
4032	filemap_invalidate_lock(inode->i_mapping);
4033}
4034
4035int ext4_break_layouts(struct inode *inode)
4036{
4037	struct page *page;
4038	int error;
4039
4040	if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
4041		return -EINVAL;
4042
4043	do {
4044		page = dax_layout_busy_page(inode->i_mapping);
4045		if (!page)
4046			return 0;
4047
4048		error = ___wait_var_event(&page->_refcount,
4049				atomic_read(&page->_refcount) == 1,
4050				TASK_INTERRUPTIBLE, 0, 0,
4051				ext4_wait_dax_page(inode));
4052	} while (error == 0);
4053
4054	return error;
4055}
4056
4057/*
4058 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4059 * associated with the given offset and length
4060 *
4061 * @inode:  File inode
4062 * @offset: The offset where the hole will begin
4063 * @len:    The length of the hole
4064 *
4065 * Returns: 0 on success or negative on failure
4066 */
4067
4068int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4069{
4070	struct inode *inode = file_inode(file);
4071	struct super_block *sb = inode->i_sb;
4072	ext4_lblk_t first_block, stop_block;
4073	struct address_space *mapping = inode->i_mapping;
4074	loff_t first_block_offset, last_block_offset, max_length;
4075	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4076	handle_t *handle;
4077	unsigned int credits;
4078	int ret = 0, ret2 = 0;
4079
4080	trace_ext4_punch_hole(inode, offset, length, 0);
4081
4082	/*
4083	 * Write out all dirty pages to avoid race conditions
4084	 * Then release them.
4085	 */
4086	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4087		ret = filemap_write_and_wait_range(mapping, offset,
4088						   offset + length - 1);
4089		if (ret)
4090			return ret;
4091	}
4092
4093	inode_lock(inode);
4094
4095	/* No need to punch hole beyond i_size */
4096	if (offset >= inode->i_size)
4097		goto out_mutex;
4098
4099	/*
4100	 * If the hole extends beyond i_size, set the hole
4101	 * to end after the page that contains i_size
4102	 */
4103	if (offset + length > inode->i_size) {
4104		length = inode->i_size +
4105		   PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4106		   offset;
4107	}
4108
4109	/*
4110	 * For punch hole the length + offset needs to be within one block
4111	 * before last range. Adjust the length if it goes beyond that limit.
4112	 */
4113	max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4114	if (offset + length > max_length)
4115		length = max_length - offset;
4116
4117	if (offset & (sb->s_blocksize - 1) ||
4118	    (offset + length) & (sb->s_blocksize - 1)) {
4119		/*
4120		 * Attach jinode to inode for jbd2 if we do any zeroing of
4121		 * partial block
4122		 */
4123		ret = ext4_inode_attach_jinode(inode);
4124		if (ret < 0)
4125			goto out_mutex;
4126
4127	}
4128
4129	/* Wait all existing dio workers, newcomers will block on i_rwsem */
4130	inode_dio_wait(inode);
4131
4132	ret = file_modified(file);
4133	if (ret)
4134		goto out_mutex;
4135
4136	/*
4137	 * Prevent page faults from reinstantiating pages we have released from
4138	 * page cache.
4139	 */
4140	filemap_invalidate_lock(mapping);
4141
4142	ret = ext4_break_layouts(inode);
4143	if (ret)
4144		goto out_dio;
4145
4146	first_block_offset = round_up(offset, sb->s_blocksize);
4147	last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4148
4149	/* Now release the pages and zero block aligned part of pages*/
4150	if (last_block_offset > first_block_offset) {
4151		ret = ext4_update_disksize_before_punch(inode, offset, length);
4152		if (ret)
4153			goto out_dio;
4154		truncate_pagecache_range(inode, first_block_offset,
4155					 last_block_offset);
4156	}
4157
4158	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4159		credits = ext4_writepage_trans_blocks(inode);
4160	else
4161		credits = ext4_blocks_for_truncate(inode);
4162	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4163	if (IS_ERR(handle)) {
4164		ret = PTR_ERR(handle);
4165		ext4_std_error(sb, ret);
4166		goto out_dio;
4167	}
4168
4169	ret = ext4_zero_partial_blocks(handle, inode, offset,
4170				       length);
4171	if (ret)
4172		goto out_stop;
4173
4174	first_block = (offset + sb->s_blocksize - 1) >>
4175		EXT4_BLOCK_SIZE_BITS(sb);
4176	stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4177
4178	/* If there are blocks to remove, do it */
4179	if (stop_block > first_block) {
4180
4181		down_write(&EXT4_I(inode)->i_data_sem);
4182		ext4_discard_preallocations(inode, 0);
4183
4184		ret = ext4_es_remove_extent(inode, first_block,
4185					    stop_block - first_block);
4186		if (ret) {
4187			up_write(&EXT4_I(inode)->i_data_sem);
4188			goto out_stop;
4189		}
4190
4191		if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4192			ret = ext4_ext_remove_space(inode, first_block,
4193						    stop_block - 1);
4194		else
4195			ret = ext4_ind_remove_space(handle, inode, first_block,
4196						    stop_block);
4197
4198		up_write(&EXT4_I(inode)->i_data_sem);
4199	}
4200	ext4_fc_track_range(handle, inode, first_block, stop_block);
4201	if (IS_SYNC(inode))
4202		ext4_handle_sync(handle);
4203
4204	inode->i_mtime = inode->i_ctime = current_time(inode);
4205	ret2 = ext4_mark_inode_dirty(handle, inode);
4206	if (unlikely(ret2))
4207		ret = ret2;
4208	if (ret >= 0)
4209		ext4_update_inode_fsync_trans(handle, inode, 1);
4210out_stop:
4211	ext4_journal_stop(handle);
4212out_dio:
4213	filemap_invalidate_unlock(mapping);
4214out_mutex:
4215	inode_unlock(inode);
4216	return ret;
4217}
4218
4219int ext4_inode_attach_jinode(struct inode *inode)
4220{
4221	struct ext4_inode_info *ei = EXT4_I(inode);
4222	struct jbd2_inode *jinode;
4223
4224	if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4225		return 0;
4226
4227	jinode = jbd2_alloc_inode(GFP_KERNEL);
4228	spin_lock(&inode->i_lock);
4229	if (!ei->jinode) {
4230		if (!jinode) {
4231			spin_unlock(&inode->i_lock);
4232			return -ENOMEM;
4233		}
4234		ei->jinode = jinode;
4235		jbd2_journal_init_jbd_inode(ei->jinode, inode);
4236		jinode = NULL;
4237	}
4238	spin_unlock(&inode->i_lock);
4239	if (unlikely(jinode != NULL))
4240		jbd2_free_inode(jinode);
4241	return 0;
4242}
4243
4244/*
4245 * ext4_truncate()
4246 *
4247 * We block out ext4_get_block() block instantiations across the entire
4248 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4249 * simultaneously on behalf of the same inode.
4250 *
4251 * As we work through the truncate and commit bits of it to the journal there
4252 * is one core, guiding principle: the file's tree must always be consistent on
4253 * disk.  We must be able to restart the truncate after a crash.
4254 *
4255 * The file's tree may be transiently inconsistent in memory (although it
4256 * probably isn't), but whenever we close off and commit a journal transaction,
4257 * the contents of (the filesystem + the journal) must be consistent and
4258 * restartable.  It's pretty simple, really: bottom up, right to left (although
4259 * left-to-right works OK too).
4260 *
4261 * Note that at recovery time, journal replay occurs *before* the restart of
4262 * truncate against the orphan inode list.
4263 *
4264 * The committed inode has the new, desired i_size (which is the same as
4265 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4266 * that this inode's truncate did not complete and it will again call
4267 * ext4_truncate() to have another go.  So there will be instantiated blocks
4268 * to the right of the truncation point in a crashed ext4 filesystem.  But
4269 * that's fine - as long as they are linked from the inode, the post-crash
4270 * ext4_truncate() run will find them and release them.
4271 */
4272int ext4_truncate(struct inode *inode)
4273{
4274	struct ext4_inode_info *ei = EXT4_I(inode);
4275	unsigned int credits;
4276	int err = 0, err2;
4277	handle_t *handle;
4278	struct address_space *mapping = inode->i_mapping;
4279
4280	/*
4281	 * There is a possibility that we're either freeing the inode
4282	 * or it's a completely new inode. In those cases we might not
4283	 * have i_rwsem locked because it's not necessary.
4284	 */
4285	if (!(inode->i_state & (I_NEW|I_FREEING)))
4286		WARN_ON(!inode_is_locked(inode));
4287	trace_ext4_truncate_enter(inode);
4288
4289	if (!ext4_can_truncate(inode))
4290		goto out_trace;
 
 
4291
4292	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4293		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4294
4295	if (ext4_has_inline_data(inode)) {
4296		int has_inline = 1;
4297
4298		err = ext4_inline_data_truncate(inode, &has_inline);
4299		if (err || has_inline)
4300			goto out_trace;
4301	}
4302
4303	/* If we zero-out tail of the page, we have to create jinode for jbd2 */
4304	if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4305		err = ext4_inode_attach_jinode(inode);
4306		if (err)
4307			goto out_trace;
4308	}
4309
4310	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4311		credits = ext4_writepage_trans_blocks(inode);
4312	else
4313		credits = ext4_blocks_for_truncate(inode);
4314
4315	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4316	if (IS_ERR(handle)) {
4317		err = PTR_ERR(handle);
4318		goto out_trace;
4319	}
4320
4321	if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4322		ext4_block_truncate_page(handle, mapping, inode->i_size);
4323
4324	/*
4325	 * We add the inode to the orphan list, so that if this
4326	 * truncate spans multiple transactions, and we crash, we will
4327	 * resume the truncate when the filesystem recovers.  It also
4328	 * marks the inode dirty, to catch the new size.
4329	 *
4330	 * Implication: the file must always be in a sane, consistent
4331	 * truncatable state while each transaction commits.
4332	 */
4333	err = ext4_orphan_add(handle, inode);
4334	if (err)
4335		goto out_stop;
4336
4337	down_write(&EXT4_I(inode)->i_data_sem);
4338
4339	ext4_discard_preallocations(inode, 0);
4340
4341	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4342		err = ext4_ext_truncate(handle, inode);
4343	else
4344		ext4_ind_truncate(handle, inode);
4345
4346	up_write(&ei->i_data_sem);
4347	if (err)
4348		goto out_stop;
4349
4350	if (IS_SYNC(inode))
4351		ext4_handle_sync(handle);
4352
4353out_stop:
4354	/*
4355	 * If this was a simple ftruncate() and the file will remain alive,
4356	 * then we need to clear up the orphan record which we created above.
4357	 * However, if this was a real unlink then we were called by
4358	 * ext4_evict_inode(), and we allow that function to clean up the
4359	 * orphan info for us.
4360	 */
4361	if (inode->i_nlink)
4362		ext4_orphan_del(handle, inode);
4363
4364	inode->i_mtime = inode->i_ctime = current_time(inode);
4365	err2 = ext4_mark_inode_dirty(handle, inode);
4366	if (unlikely(err2 && !err))
4367		err = err2;
4368	ext4_journal_stop(handle);
4369
4370out_trace:
4371	trace_ext4_truncate_exit(inode);
4372	return err;
4373}
4374
4375static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4376{
4377	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4378		return inode_peek_iversion_raw(inode);
4379	else
4380		return inode_peek_iversion(inode);
4381}
4382
4383static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4384				 struct ext4_inode_info *ei)
4385{
4386	struct inode *inode = &(ei->vfs_inode);
4387	u64 i_blocks = READ_ONCE(inode->i_blocks);
4388	struct super_block *sb = inode->i_sb;
4389
4390	if (i_blocks <= ~0U) {
4391		/*
4392		 * i_blocks can be represented in a 32 bit variable
4393		 * as multiple of 512 bytes
4394		 */
4395		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4396		raw_inode->i_blocks_high = 0;
4397		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4398		return 0;
4399	}
4400
4401	/*
4402	 * This should never happen since sb->s_maxbytes should not have
4403	 * allowed this, sb->s_maxbytes was set according to the huge_file
4404	 * feature in ext4_fill_super().
4405	 */
4406	if (!ext4_has_feature_huge_file(sb))
4407		return -EFSCORRUPTED;
4408
4409	if (i_blocks <= 0xffffffffffffULL) {
4410		/*
4411		 * i_blocks can be represented in a 48 bit variable
4412		 * as multiple of 512 bytes
4413		 */
4414		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4415		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4416		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4417	} else {
4418		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4419		/* i_block is stored in file system block size */
4420		i_blocks = i_blocks >> (inode->i_blkbits - 9);
4421		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4422		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4423	}
4424	return 0;
4425}
4426
4427static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4428{
4429	struct ext4_inode_info *ei = EXT4_I(inode);
4430	uid_t i_uid;
4431	gid_t i_gid;
4432	projid_t i_projid;
4433	int block;
4434	int err;
4435
4436	err = ext4_inode_blocks_set(raw_inode, ei);
4437
4438	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4439	i_uid = i_uid_read(inode);
4440	i_gid = i_gid_read(inode);
4441	i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4442	if (!(test_opt(inode->i_sb, NO_UID32))) {
4443		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4444		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4445		/*
4446		 * Fix up interoperability with old kernels. Otherwise,
4447		 * old inodes get re-used with the upper 16 bits of the
4448		 * uid/gid intact.
4449		 */
4450		if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4451			raw_inode->i_uid_high = 0;
4452			raw_inode->i_gid_high = 0;
4453		} else {
4454			raw_inode->i_uid_high =
4455				cpu_to_le16(high_16_bits(i_uid));
4456			raw_inode->i_gid_high =
4457				cpu_to_le16(high_16_bits(i_gid));
4458		}
4459	} else {
4460		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4461		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4462		raw_inode->i_uid_high = 0;
4463		raw_inode->i_gid_high = 0;
4464	}
4465	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4466
4467	EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4468	EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4469	EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4470	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4471
4472	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4473	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4474	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4475		raw_inode->i_file_acl_high =
4476			cpu_to_le16(ei->i_file_acl >> 32);
4477	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4478	ext4_isize_set(raw_inode, ei->i_disksize);
4479
4480	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4481	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4482		if (old_valid_dev(inode->i_rdev)) {
4483			raw_inode->i_block[0] =
4484				cpu_to_le32(old_encode_dev(inode->i_rdev));
4485			raw_inode->i_block[1] = 0;
4486		} else {
4487			raw_inode->i_block[0] = 0;
4488			raw_inode->i_block[1] =
4489				cpu_to_le32(new_encode_dev(inode->i_rdev));
4490			raw_inode->i_block[2] = 0;
4491		}
4492	} else if (!ext4_has_inline_data(inode)) {
4493		for (block = 0; block < EXT4_N_BLOCKS; block++)
4494			raw_inode->i_block[block] = ei->i_data[block];
4495	}
4496
4497	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4498		u64 ivers = ext4_inode_peek_iversion(inode);
4499
4500		raw_inode->i_disk_version = cpu_to_le32(ivers);
4501		if (ei->i_extra_isize) {
4502			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4503				raw_inode->i_version_hi =
4504					cpu_to_le32(ivers >> 32);
4505			raw_inode->i_extra_isize =
4506				cpu_to_le16(ei->i_extra_isize);
4507		}
4508	}
4509
4510	if (i_projid != EXT4_DEF_PROJID &&
4511	    !ext4_has_feature_project(inode->i_sb))
4512		err = err ?: -EFSCORRUPTED;
4513
4514	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4515	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4516		raw_inode->i_projid = cpu_to_le32(i_projid);
4517
4518	ext4_inode_csum_set(inode, raw_inode, ei);
4519	return err;
4520}
4521
4522/*
4523 * ext4_get_inode_loc returns with an extra refcount against the inode's
4524 * underlying buffer_head on success. If we pass 'inode' and it does not
4525 * have in-inode xattr, we have all inode data in memory that is needed
4526 * to recreate the on-disk version of this inode.
4527 */
4528static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4529				struct inode *inode, struct ext4_iloc *iloc,
4530				ext4_fsblk_t *ret_block)
4531{
4532	struct ext4_group_desc	*gdp;
4533	struct buffer_head	*bh;
 
4534	ext4_fsblk_t		block;
4535	struct blk_plug		plug;
4536	int			inodes_per_block, inode_offset;
4537
4538	iloc->bh = NULL;
4539	if (ino < EXT4_ROOT_INO ||
4540	    ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4541		return -EFSCORRUPTED;
4542
4543	iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4544	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4545	if (!gdp)
4546		return -EIO;
4547
4548	/*
4549	 * Figure out the offset within the block group inode table
4550	 */
4551	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4552	inode_offset = ((ino - 1) %
4553			EXT4_INODES_PER_GROUP(sb));
 
4554	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4555
4556	block = ext4_inode_table(sb, gdp);
4557	if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4558	    (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4559		ext4_error(sb, "Invalid inode table block %llu in "
4560			   "block_group %u", block, iloc->block_group);
4561		return -EFSCORRUPTED;
4562	}
4563	block += (inode_offset / inodes_per_block);
4564
4565	bh = sb_getblk(sb, block);
4566	if (unlikely(!bh))
4567		return -ENOMEM;
4568	if (ext4_buffer_uptodate(bh))
4569		goto has_buffer;
4570
4571	lock_buffer(bh);
4572	if (ext4_buffer_uptodate(bh)) {
4573		/* Someone brought it uptodate while we waited */
4574		unlock_buffer(bh);
4575		goto has_buffer;
4576	}
4577
4578	/*
4579	 * If we have all information of the inode in memory and this
4580	 * is the only valid inode in the block, we need not read the
4581	 * block.
4582	 */
4583	if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4584		struct buffer_head *bitmap_bh;
4585		int i, start;
4586
4587		start = inode_offset & ~(inodes_per_block - 1);
4588
4589		/* Is the inode bitmap in cache? */
4590		bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4591		if (unlikely(!bitmap_bh))
4592			goto make_io;
4593
4594		/*
4595		 * If the inode bitmap isn't in cache then the
4596		 * optimisation may end up performing two reads instead
4597		 * of one, so skip it.
 
4598		 */
4599		if (!buffer_uptodate(bitmap_bh)) {
4600			brelse(bitmap_bh);
4601			goto make_io;
4602		}
4603		for (i = start; i < start + inodes_per_block; i++) {
4604			if (i == inode_offset)
4605				continue;
4606			if (ext4_test_bit(i, bitmap_bh->b_data))
4607				break;
4608		}
4609		brelse(bitmap_bh);
4610		if (i == start + inodes_per_block) {
4611			struct ext4_inode *raw_inode =
4612				(struct ext4_inode *) (bh->b_data + iloc->offset);
4613
4614			/* all other inodes are free, so skip I/O */
4615			memset(bh->b_data, 0, bh->b_size);
4616			if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4617				ext4_fill_raw_inode(inode, raw_inode);
4618			set_buffer_uptodate(bh);
 
 
 
4619			unlock_buffer(bh);
4620			goto has_buffer;
4621		}
4622	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4623
4624make_io:
4625	/*
4626	 * If we need to do any I/O, try to pre-readahead extra
4627	 * blocks from the inode table.
4628	 */
4629	blk_start_plug(&plug);
4630	if (EXT4_SB(sb)->s_inode_readahead_blks) {
4631		ext4_fsblk_t b, end, table;
4632		unsigned num;
4633		__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4634
4635		table = ext4_inode_table(sb, gdp);
4636		/* s_inode_readahead_blks is always a power of 2 */
4637		b = block & ~((ext4_fsblk_t) ra_blks - 1);
4638		if (table > b)
4639			b = table;
4640		end = b + ra_blks;
4641		num = EXT4_INODES_PER_GROUP(sb);
4642		if (ext4_has_group_desc_csum(sb))
4643			num -= ext4_itable_unused_count(sb, gdp);
4644		table += num / inodes_per_block;
4645		if (end > table)
4646			end = table;
4647		while (b <= end)
4648			ext4_sb_breadahead_unmovable(sb, b++);
4649	}
4650
4651	/*
4652	 * There are other valid inodes in the buffer, this inode
4653	 * has in-inode xattrs, or we don't have this inode in memory.
4654	 * Read the block from disk.
4655	 */
4656	trace_ext4_load_inode(sb, ino);
4657	ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4658	blk_finish_plug(&plug);
4659	wait_on_buffer(bh);
4660	ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4661	if (!buffer_uptodate(bh)) {
4662		if (ret_block)
4663			*ret_block = block;
4664		brelse(bh);
4665		return -EIO;
 
4666	}
4667has_buffer:
4668	iloc->bh = bh;
4669	return 0;
4670}
4671
4672static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4673					struct ext4_iloc *iloc)
4674{
4675	ext4_fsblk_t err_blk = 0;
4676	int ret;
4677
4678	ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4679					&err_blk);
4680
4681	if (ret == -EIO)
4682		ext4_error_inode_block(inode, err_blk, EIO,
4683					"unable to read itable block");
4684
4685	return ret;
4686}
4687
4688int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4689{
4690	ext4_fsblk_t err_blk = 0;
4691	int ret;
4692
4693	ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4694					&err_blk);
4695
4696	if (ret == -EIO)
4697		ext4_error_inode_block(inode, err_blk, EIO,
4698					"unable to read itable block");
4699
4700	return ret;
4701}
4702
4703
4704int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4705			  struct ext4_iloc *iloc)
4706{
4707	return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4708}
4709
4710static bool ext4_should_enable_dax(struct inode *inode)
4711{
4712	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4713
4714	if (test_opt2(inode->i_sb, DAX_NEVER))
4715		return false;
4716	if (!S_ISREG(inode->i_mode))
4717		return false;
4718	if (ext4_should_journal_data(inode))
4719		return false;
4720	if (ext4_has_inline_data(inode))
4721		return false;
4722	if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4723		return false;
4724	if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4725		return false;
4726	if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4727		return false;
4728	if (test_opt(inode->i_sb, DAX_ALWAYS))
4729		return true;
4730
4731	return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4732}
4733
4734void ext4_set_inode_flags(struct inode *inode, bool init)
4735{
4736	unsigned int flags = EXT4_I(inode)->i_flags;
4737	unsigned int new_fl = 0;
4738
4739	WARN_ON_ONCE(IS_DAX(inode) && init);
4740
 
4741	if (flags & EXT4_SYNC_FL)
4742		new_fl |= S_SYNC;
4743	if (flags & EXT4_APPEND_FL)
4744		new_fl |= S_APPEND;
4745	if (flags & EXT4_IMMUTABLE_FL)
4746		new_fl |= S_IMMUTABLE;
4747	if (flags & EXT4_NOATIME_FL)
4748		new_fl |= S_NOATIME;
4749	if (flags & EXT4_DIRSYNC_FL)
4750		new_fl |= S_DIRSYNC;
 
 
 
 
 
 
 
4751
4752	/* Because of the way inode_set_flags() works we must preserve S_DAX
4753	 * here if already set. */
4754	new_fl |= (inode->i_flags & S_DAX);
4755	if (init && ext4_should_enable_dax(inode))
4756		new_fl |= S_DAX;
4757
4758	if (flags & EXT4_ENCRYPT_FL)
4759		new_fl |= S_ENCRYPTED;
4760	if (flags & EXT4_CASEFOLD_FL)
4761		new_fl |= S_CASEFOLD;
4762	if (flags & EXT4_VERITY_FL)
4763		new_fl |= S_VERITY;
4764	inode_set_flags(inode, new_fl,
4765			S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4766			S_ENCRYPTED|S_CASEFOLD|S_VERITY);
 
 
4767}
4768
4769static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4770				  struct ext4_inode_info *ei)
4771{
4772	blkcnt_t i_blocks ;
4773	struct inode *inode = &(ei->vfs_inode);
4774	struct super_block *sb = inode->i_sb;
4775
4776	if (ext4_has_feature_huge_file(sb)) {
 
4777		/* we are using combined 48 bit field */
4778		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4779					le32_to_cpu(raw_inode->i_blocks_lo);
4780		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4781			/* i_blocks represent file system block size */
4782			return i_blocks  << (inode->i_blkbits - 9);
4783		} else {
4784			return i_blocks;
4785		}
4786	} else {
4787		return le32_to_cpu(raw_inode->i_blocks_lo);
4788	}
4789}
4790
4791static inline int ext4_iget_extra_inode(struct inode *inode,
4792					 struct ext4_inode *raw_inode,
4793					 struct ext4_inode_info *ei)
4794{
4795	__le32 *magic = (void *)raw_inode +
4796			EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4797
4798	if (EXT4_INODE_HAS_XATTR_SPACE(inode)  &&
4799	    *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4800		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4801		return ext4_find_inline_data_nolock(inode);
4802	} else
4803		EXT4_I(inode)->i_inline_off = 0;
4804	return 0;
4805}
4806
4807int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4808{
4809	if (!ext4_has_feature_project(inode->i_sb))
4810		return -EOPNOTSUPP;
4811	*projid = EXT4_I(inode)->i_projid;
4812	return 0;
4813}
4814
4815/*
4816 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4817 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4818 * set.
4819 */
4820static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4821{
4822	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4823		inode_set_iversion_raw(inode, val);
4824	else
4825		inode_set_iversion_queried(inode, val);
4826}
4827
4828struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4829			  ext4_iget_flags flags, const char *function,
4830			  unsigned int line)
4831{
4832	struct ext4_iloc iloc;
4833	struct ext4_inode *raw_inode;
4834	struct ext4_inode_info *ei;
4835	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4836	struct inode *inode;
4837	journal_t *journal = EXT4_SB(sb)->s_journal;
4838	long ret;
4839	loff_t size;
4840	int block;
4841	uid_t i_uid;
4842	gid_t i_gid;
4843	projid_t i_projid;
4844
4845	if ((!(flags & EXT4_IGET_SPECIAL) &&
4846	     ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4847	      ino == le32_to_cpu(es->s_usr_quota_inum) ||
4848	      ino == le32_to_cpu(es->s_grp_quota_inum) ||
4849	      ino == le32_to_cpu(es->s_prj_quota_inum) ||
4850	      ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4851	    (ino < EXT4_ROOT_INO) ||
4852	    (ino > le32_to_cpu(es->s_inodes_count))) {
4853		if (flags & EXT4_IGET_HANDLE)
4854			return ERR_PTR(-ESTALE);
4855		__ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4856			     "inode #%lu: comm %s: iget: illegal inode #",
4857			     ino, current->comm);
4858		return ERR_PTR(-EFSCORRUPTED);
4859	}
4860
4861	inode = iget_locked(sb, ino);
4862	if (!inode)
4863		return ERR_PTR(-ENOMEM);
4864	if (!(inode->i_state & I_NEW))
4865		return inode;
4866
4867	ei = EXT4_I(inode);
4868	iloc.bh = NULL;
4869
4870	ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4871	if (ret < 0)
4872		goto bad_inode;
4873	raw_inode = ext4_raw_inode(&iloc);
4874
4875	if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4876		ext4_error_inode(inode, function, line, 0,
4877				 "iget: root inode unallocated");
4878		ret = -EFSCORRUPTED;
4879		goto bad_inode;
4880	}
4881
4882	if ((flags & EXT4_IGET_HANDLE) &&
4883	    (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4884		ret = -ESTALE;
4885		goto bad_inode;
4886	}
4887
4888	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4889		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4890		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4891			EXT4_INODE_SIZE(inode->i_sb) ||
4892		    (ei->i_extra_isize & 3)) {
4893			ext4_error_inode(inode, function, line, 0,
4894					 "iget: bad extra_isize %u "
4895					 "(inode size %u)",
4896					 ei->i_extra_isize,
4897					 EXT4_INODE_SIZE(inode->i_sb));
4898			ret = -EFSCORRUPTED;
4899			goto bad_inode;
4900		}
4901	} else
4902		ei->i_extra_isize = 0;
4903
4904	/* Precompute checksum seed for inode metadata */
4905	if (ext4_has_metadata_csum(sb)) {
4906		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4907		__u32 csum;
4908		__le32 inum = cpu_to_le32(inode->i_ino);
4909		__le32 gen = raw_inode->i_generation;
4910		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4911				   sizeof(inum));
4912		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4913					      sizeof(gen));
4914	}
4915
4916	if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4917	    ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4918	     (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4919		ext4_error_inode_err(inode, function, line, 0,
4920				EFSBADCRC, "iget: checksum invalid");
4921		ret = -EFSBADCRC;
4922		goto bad_inode;
4923	}
4924
4925	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4926	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4927	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4928	if (ext4_has_feature_project(sb) &&
4929	    EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4930	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4931		i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4932	else
4933		i_projid = EXT4_DEF_PROJID;
4934
4935	if (!(test_opt(inode->i_sb, NO_UID32))) {
4936		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4937		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4938	}
4939	i_uid_write(inode, i_uid);
4940	i_gid_write(inode, i_gid);
4941	ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4942	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4943
4944	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
4945	ei->i_inline_off = 0;
4946	ei->i_dir_start_lookup = 0;
4947	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4948	/* We now have enough fields to check if the inode was active or not.
4949	 * This is needed because nfsd might try to access dead inodes
4950	 * the test is that same one that e2fsck uses
4951	 * NeilBrown 1999oct15
4952	 */
4953	if (inode->i_nlink == 0) {
4954		if ((inode->i_mode == 0 ||
4955		     !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4956		    ino != EXT4_BOOT_LOADER_INO) {
4957			/* this inode is deleted */
4958			ret = -ESTALE;
4959			goto bad_inode;
4960		}
4961		/* The only unlinked inodes we let through here have
4962		 * valid i_mode and are being read by the orphan
4963		 * recovery code: that's fine, we're about to complete
4964		 * the process of deleting those.
4965		 * OR it is the EXT4_BOOT_LOADER_INO which is
4966		 * not initialized on a new filesystem. */
4967	}
4968	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4969	ext4_set_inode_flags(inode, true);
4970	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4971	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4972	if (ext4_has_feature_64bit(sb))
4973		ei->i_file_acl |=
4974			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4975	inode->i_size = ext4_isize(sb, raw_inode);
4976	if ((size = i_size_read(inode)) < 0) {
4977		ext4_error_inode(inode, function, line, 0,
4978				 "iget: bad i_size value: %lld", size);
4979		ret = -EFSCORRUPTED;
4980		goto bad_inode;
4981	}
4982	/*
4983	 * If dir_index is not enabled but there's dir with INDEX flag set,
4984	 * we'd normally treat htree data as empty space. But with metadata
4985	 * checksumming that corrupts checksums so forbid that.
4986	 */
4987	if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4988	    ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4989		ext4_error_inode(inode, function, line, 0,
4990			 "iget: Dir with htree data on filesystem without dir_index feature.");
4991		ret = -EFSCORRUPTED;
4992		goto bad_inode;
4993	}
4994	ei->i_disksize = inode->i_size;
4995#ifdef CONFIG_QUOTA
4996	ei->i_reserved_quota = 0;
4997#endif
4998	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4999	ei->i_block_group = iloc.block_group;
5000	ei->i_last_alloc_group = ~0;
5001	/*
5002	 * NOTE! The in-memory inode i_data array is in little-endian order
5003	 * even on big-endian machines: we do NOT byteswap the block numbers!
5004	 */
5005	for (block = 0; block < EXT4_N_BLOCKS; block++)
5006		ei->i_data[block] = raw_inode->i_block[block];
5007	INIT_LIST_HEAD(&ei->i_orphan);
5008	ext4_fc_init_inode(&ei->vfs_inode);
5009
5010	/*
5011	 * Set transaction id's of transactions that have to be committed
5012	 * to finish f[data]sync. We set them to currently running transaction
5013	 * as we cannot be sure that the inode or some of its metadata isn't
5014	 * part of the transaction - the inode could have been reclaimed and
5015	 * now it is reread from disk.
5016	 */
5017	if (journal) {
5018		transaction_t *transaction;
5019		tid_t tid;
5020
5021		read_lock(&journal->j_state_lock);
5022		if (journal->j_running_transaction)
5023			transaction = journal->j_running_transaction;
5024		else
5025			transaction = journal->j_committing_transaction;
5026		if (transaction)
5027			tid = transaction->t_tid;
5028		else
5029			tid = journal->j_commit_sequence;
5030		read_unlock(&journal->j_state_lock);
5031		ei->i_sync_tid = tid;
5032		ei->i_datasync_tid = tid;
5033	}
5034
5035	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
 
 
 
 
 
 
5036		if (ei->i_extra_isize == 0) {
5037			/* The extra space is currently unused. Use it. */
5038			BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5039			ei->i_extra_isize = sizeof(struct ext4_inode) -
5040					    EXT4_GOOD_OLD_INODE_SIZE;
5041		} else {
5042			ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5043			if (ret)
5044				goto bad_inode;
 
 
5045		}
5046	}
 
5047
5048	EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5049	EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5050	EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5051	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5052
5053	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5054		u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5055
5056		if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5057			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5058				ivers |=
5059		    (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5060		}
5061		ext4_inode_set_iversion_queried(inode, ivers);
5062	}
5063
5064	ret = 0;
5065	if (ei->i_file_acl &&
5066	    !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
5067		ext4_error_inode(inode, function, line, 0,
5068				 "iget: bad extended attribute block %llu",
5069				 ei->i_file_acl);
5070		ret = -EFSCORRUPTED;
5071		goto bad_inode;
5072	} else if (!ext4_has_inline_data(inode)) {
5073		/* validate the block references in the inode */
5074		if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
5075			(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5076			(S_ISLNK(inode->i_mode) &&
5077			!ext4_inode_is_fast_symlink(inode)))) {
5078			if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5079				ret = ext4_ext_check_inode(inode);
5080			else
5081				ret = ext4_ind_check_inode(inode);
5082		}
5083	}
5084	if (ret)
5085		goto bad_inode;
5086
5087	if (S_ISREG(inode->i_mode)) {
5088		inode->i_op = &ext4_file_inode_operations;
5089		inode->i_fop = &ext4_file_operations;
5090		ext4_set_aops(inode);
5091	} else if (S_ISDIR(inode->i_mode)) {
5092		inode->i_op = &ext4_dir_inode_operations;
5093		inode->i_fop = &ext4_dir_operations;
5094	} else if (S_ISLNK(inode->i_mode)) {
5095		/* VFS does not allow setting these so must be corruption */
5096		if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5097			ext4_error_inode(inode, function, line, 0,
5098					 "iget: immutable or append flags "
5099					 "not allowed on symlinks");
5100			ret = -EFSCORRUPTED;
5101			goto bad_inode;
5102		}
5103		if (IS_ENCRYPTED(inode)) {
5104			inode->i_op = &ext4_encrypted_symlink_inode_operations;
5105		} else if (ext4_inode_is_fast_symlink(inode)) {
5106			inode->i_link = (char *)ei->i_data;
5107			inode->i_op = &ext4_fast_symlink_inode_operations;
5108			nd_terminate_link(ei->i_data, inode->i_size,
5109				sizeof(ei->i_data) - 1);
5110		} else {
5111			inode->i_op = &ext4_symlink_inode_operations;
 
5112		}
5113	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5114	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5115		inode->i_op = &ext4_special_inode_operations;
5116		if (raw_inode->i_block[0])
5117			init_special_inode(inode, inode->i_mode,
5118			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5119		else
5120			init_special_inode(inode, inode->i_mode,
5121			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5122	} else if (ino == EXT4_BOOT_LOADER_INO) {
5123		make_bad_inode(inode);
5124	} else {
5125		ret = -EFSCORRUPTED;
5126		ext4_error_inode(inode, function, line, 0,
5127				 "iget: bogus i_mode (%o)", inode->i_mode);
5128		goto bad_inode;
5129	}
5130	if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
5131		ext4_error_inode(inode, function, line, 0,
5132				 "casefold flag without casefold feature");
5133	if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) {
5134		ext4_error_inode(inode, function, line, 0,
5135				 "bad inode without EXT4_IGET_BAD flag");
5136		ret = -EUCLEAN;
5137		goto bad_inode;
5138	}
5139
5140	brelse(iloc.bh);
 
5141	unlock_new_inode(inode);
5142	return inode;
5143
5144bad_inode:
5145	brelse(iloc.bh);
5146	iget_failed(inode);
5147	return ERR_PTR(ret);
5148}
5149
5150static void __ext4_update_other_inode_time(struct super_block *sb,
5151					   unsigned long orig_ino,
5152					   unsigned long ino,
5153					   struct ext4_inode *raw_inode)
5154{
5155	struct inode *inode;
 
 
5156
5157	inode = find_inode_by_ino_rcu(sb, ino);
5158	if (!inode)
5159		return;
5160
5161	if (!inode_is_dirtytime_only(inode))
5162		return;
5163
5164	spin_lock(&inode->i_lock);
5165	if (inode_is_dirtytime_only(inode)) {
5166		struct ext4_inode_info	*ei = EXT4_I(inode);
5167
5168		inode->i_state &= ~I_DIRTY_TIME;
5169		spin_unlock(&inode->i_lock);
5170
5171		spin_lock(&ei->i_raw_lock);
5172		EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5173		EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5174		EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5175		ext4_inode_csum_set(inode, raw_inode, ei);
5176		spin_unlock(&ei->i_raw_lock);
5177		trace_ext4_other_inode_update_time(inode, orig_ino);
5178		return;
5179	}
5180	spin_unlock(&inode->i_lock);
5181}
5182
5183/*
5184 * Opportunistically update the other time fields for other inodes in
5185 * the same inode table block.
5186 */
5187static void ext4_update_other_inodes_time(struct super_block *sb,
5188					  unsigned long orig_ino, char *buf)
5189{
5190	unsigned long ino;
5191	int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5192	int inode_size = EXT4_INODE_SIZE(sb);
5193
5194	/*
5195	 * Calculate the first inode in the inode table block.  Inode
5196	 * numbers are one-based.  That is, the first inode in a block
5197	 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5198	 */
5199	ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5200	rcu_read_lock();
5201	for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5202		if (ino == orig_ino)
5203			continue;
5204		__ext4_update_other_inode_time(sb, orig_ino, ino,
5205					       (struct ext4_inode *)buf);
5206	}
5207	rcu_read_unlock();
5208}
5209
5210/*
5211 * Post the struct inode info into an on-disk inode location in the
5212 * buffer-cache.  This gobbles the caller's reference to the
5213 * buffer_head in the inode location struct.
5214 *
5215 * The caller must have write access to iloc->bh.
5216 */
5217static int ext4_do_update_inode(handle_t *handle,
5218				struct inode *inode,
5219				struct ext4_iloc *iloc)
5220{
5221	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5222	struct ext4_inode_info *ei = EXT4_I(inode);
5223	struct buffer_head *bh = iloc->bh;
5224	struct super_block *sb = inode->i_sb;
5225	int err;
5226	int need_datasync = 0, set_large_file = 0;
5227
5228	spin_lock(&ei->i_raw_lock);
5229
5230	/*
5231	 * For fields not tracked in the in-memory inode, initialise them
5232	 * to zero for new inodes.
5233	 */
5234	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5235		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5236
5237	if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5238		need_datasync = 1;
5239	if (ei->i_disksize > 0x7fffffffULL) {
5240		if (!ext4_has_feature_large_file(sb) ||
5241		    EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5242			set_large_file = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5243	}
 
 
 
 
 
 
5244
5245	err = ext4_fill_raw_inode(inode, raw_inode);
5246	spin_unlock(&ei->i_raw_lock);
5247	if (err) {
5248		EXT4_ERROR_INODE(inode, "corrupted inode contents");
5249		goto out_brelse;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5250	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5251
5252	if (inode->i_sb->s_flags & SB_LAZYTIME)
5253		ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5254					      bh->b_data);
 
 
 
 
5255
5256	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5257	err = ext4_handle_dirty_metadata(handle, NULL, bh);
5258	if (err)
5259		goto out_error;
5260	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5261	if (set_large_file) {
5262		BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5263		err = ext4_journal_get_write_access(handle, sb,
5264						    EXT4_SB(sb)->s_sbh,
5265						    EXT4_JTR_NONE);
5266		if (err)
5267			goto out_error;
5268		lock_buffer(EXT4_SB(sb)->s_sbh);
5269		ext4_set_feature_large_file(sb);
5270		ext4_superblock_csum_set(sb);
5271		unlock_buffer(EXT4_SB(sb)->s_sbh);
5272		ext4_handle_sync(handle);
5273		err = ext4_handle_dirty_metadata(handle, NULL,
5274						 EXT4_SB(sb)->s_sbh);
5275	}
5276	ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5277out_error:
5278	ext4_std_error(inode->i_sb, err);
5279out_brelse:
5280	brelse(bh);
 
5281	return err;
5282}
5283
5284/*
5285 * ext4_write_inode()
5286 *
5287 * We are called from a few places:
5288 *
5289 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5290 *   Here, there will be no transaction running. We wait for any running
5291 *   transaction to commit.
5292 *
5293 * - Within flush work (sys_sync(), kupdate and such).
5294 *   We wait on commit, if told to.
5295 *
5296 * - Within iput_final() -> write_inode_now()
5297 *   We wait on commit, if told to.
 
5298 *
5299 * In all cases it is actually safe for us to return without doing anything,
5300 * because the inode has been copied into a raw inode buffer in
5301 * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
5302 * writeback.
5303 *
5304 * Note that we are absolutely dependent upon all inode dirtiers doing the
5305 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5306 * which we are interested.
5307 *
5308 * It would be a bug for them to not do this.  The code:
5309 *
5310 *	mark_inode_dirty(inode)
5311 *	stuff();
5312 *	inode->i_size = expr;
5313 *
5314 * is in error because write_inode() could occur while `stuff()' is running,
5315 * and the new i_size will be lost.  Plus the inode will no longer be on the
5316 * superblock's dirty inode list.
5317 */
5318int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5319{
5320	int err;
5321
5322	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5323	    sb_rdonly(inode->i_sb))
5324		return 0;
5325
5326	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5327		return -EIO;
5328
5329	if (EXT4_SB(inode->i_sb)->s_journal) {
5330		if (ext4_journal_current_handle()) {
5331			ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5332			dump_stack();
5333			return -EIO;
5334		}
5335
5336		/*
5337		 * No need to force transaction in WB_SYNC_NONE mode. Also
5338		 * ext4_sync_fs() will force the commit after everything is
5339		 * written.
5340		 */
5341		if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5342			return 0;
5343
5344		err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5345						EXT4_I(inode)->i_sync_tid);
5346	} else {
5347		struct ext4_iloc iloc;
5348
5349		err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5350		if (err)
5351			return err;
5352		/*
5353		 * sync(2) will flush the whole buffer cache. No need to do
5354		 * it here separately for each inode.
5355		 */
5356		if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5357			sync_dirty_buffer(iloc.bh);
5358		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5359			ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5360					       "IO error syncing inode");
5361			err = -EIO;
5362		}
5363		brelse(iloc.bh);
5364	}
5365	return err;
5366}
5367
5368/*
5369 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5370 * buffers that are attached to a folio straddling i_size and are undergoing
5371 * commit. In that case we have to wait for commit to finish and try again.
5372 */
5373static void ext4_wait_for_tail_page_commit(struct inode *inode)
5374{
5375	unsigned offset;
5376	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5377	tid_t commit_tid = 0;
5378	int ret;
5379
5380	offset = inode->i_size & (PAGE_SIZE - 1);
5381	/*
5382	 * If the folio is fully truncated, we don't need to wait for any commit
5383	 * (and we even should not as __ext4_journalled_invalidate_folio() may
5384	 * strip all buffers from the folio but keep the folio dirty which can then
5385	 * confuse e.g. concurrent ext4_writepage() seeing dirty folio without
5386	 * buffers). Also we don't need to wait for any commit if all buffers in
5387	 * the folio remain valid. This is most beneficial for the common case of
5388	 * blocksize == PAGESIZE.
5389	 */
5390	if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5391		return;
5392	while (1) {
5393		struct folio *folio = filemap_lock_folio(inode->i_mapping,
5394				      inode->i_size >> PAGE_SHIFT);
5395		if (!folio)
5396			return;
5397		ret = __ext4_journalled_invalidate_folio(folio, offset,
5398						folio_size(folio) - offset);
5399		folio_unlock(folio);
5400		folio_put(folio);
5401		if (ret != -EBUSY)
5402			return;
5403		commit_tid = 0;
5404		read_lock(&journal->j_state_lock);
5405		if (journal->j_committing_transaction)
5406			commit_tid = journal->j_committing_transaction->t_tid;
5407		read_unlock(&journal->j_state_lock);
5408		if (commit_tid)
5409			jbd2_log_wait_commit(journal, commit_tid);
5410	}
5411}
5412
5413/*
5414 * ext4_setattr()
5415 *
5416 * Called from notify_change.
5417 *
5418 * We want to trap VFS attempts to truncate the file as soon as
5419 * possible.  In particular, we want to make sure that when the VFS
5420 * shrinks i_size, we put the inode on the orphan list and modify
5421 * i_disksize immediately, so that during the subsequent flushing of
5422 * dirty pages and freeing of disk blocks, we can guarantee that any
5423 * commit will leave the blocks being flushed in an unused state on
5424 * disk.  (On recovery, the inode will get truncated and the blocks will
5425 * be freed, so we have a strong guarantee that no future commit will
5426 * leave these blocks visible to the user.)
5427 *
5428 * Another thing we have to assure is that if we are in ordered mode
5429 * and inode is still attached to the committing transaction, we must
5430 * we start writeout of all the dirty pages which are being truncated.
5431 * This way we are sure that all the data written in the previous
5432 * transaction are already on disk (truncate waits for pages under
5433 * writeback).
5434 *
5435 * Called with inode->i_rwsem down.
5436 */
5437int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5438		 struct iattr *attr)
5439{
5440	struct inode *inode = d_inode(dentry);
5441	int error, rc = 0;
5442	int orphan = 0;
5443	const unsigned int ia_valid = attr->ia_valid;
5444	bool inc_ivers = true;
5445
5446	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5447		return -EIO;
5448
5449	if (unlikely(IS_IMMUTABLE(inode)))
5450		return -EPERM;
5451
5452	if (unlikely(IS_APPEND(inode) &&
5453		     (ia_valid & (ATTR_MODE | ATTR_UID |
5454				  ATTR_GID | ATTR_TIMES_SET))))
5455		return -EPERM;
5456
5457	error = setattr_prepare(mnt_userns, dentry, attr);
5458	if (error)
5459		return error;
5460
5461	error = fscrypt_prepare_setattr(dentry, attr);
5462	if (error)
5463		return error;
5464
5465	error = fsverity_prepare_setattr(dentry, attr);
5466	if (error)
5467		return error;
5468
5469	if (is_quota_modification(mnt_userns, inode, attr)) {
5470		error = dquot_initialize(inode);
5471		if (error)
5472			return error;
5473	}
5474
5475	if (i_uid_needs_update(mnt_userns, attr, inode) ||
5476	    i_gid_needs_update(mnt_userns, attr, inode)) {
5477		handle_t *handle;
5478
5479		/* (user+group)*(old+new) structure, inode write (sb,
5480		 * inode block, ? - but truncate inode update has it) */
5481		handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5482			(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5483			 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5484		if (IS_ERR(handle)) {
5485			error = PTR_ERR(handle);
5486			goto err_out;
5487		}
5488
5489		/* dquot_transfer() calls back ext4_get_inode_usage() which
5490		 * counts xattr inode references.
5491		 */
5492		down_read(&EXT4_I(inode)->xattr_sem);
5493		error = dquot_transfer(mnt_userns, inode, attr);
5494		up_read(&EXT4_I(inode)->xattr_sem);
5495
5496		if (error) {
5497			ext4_journal_stop(handle);
5498			return error;
5499		}
5500		/* Update corresponding info in inode so that everything is in
5501		 * one transaction */
5502		i_uid_update(mnt_userns, attr, inode);
5503		i_gid_update(mnt_userns, attr, inode);
 
 
5504		error = ext4_mark_inode_dirty(handle, inode);
5505		ext4_journal_stop(handle);
5506		if (unlikely(error)) {
5507			return error;
5508		}
5509	}
5510
5511	if (attr->ia_valid & ATTR_SIZE) {
5512		handle_t *handle;
5513		loff_t oldsize = inode->i_size;
5514		loff_t old_disksize;
5515		int shrink = (attr->ia_size < inode->i_size);
5516
5517		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5518			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5519
5520			if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5521				return -EFBIG;
5522			}
5523		}
5524		if (!S_ISREG(inode->i_mode)) {
5525			return -EINVAL;
5526		}
 
5527
5528		if (attr->ia_size == inode->i_size)
5529			inc_ivers = false;
 
 
5530
5531		if (shrink) {
5532			if (ext4_should_order_data(inode)) {
5533				error = ext4_begin_ordered_truncate(inode,
5534							    attr->ia_size);
5535				if (error)
5536					goto err_out;
5537			}
5538			/*
5539			 * Blocks are going to be removed from the inode. Wait
5540			 * for dio in flight.
5541			 */
5542			inode_dio_wait(inode);
5543		}
5544
5545		filemap_invalidate_lock(inode->i_mapping);
5546
5547		rc = ext4_break_layouts(inode);
5548		if (rc) {
5549			filemap_invalidate_unlock(inode->i_mapping);
5550			goto err_out;
5551		}
 
 
 
 
 
 
 
 
 
5552
5553		if (attr->ia_size != inode->i_size) {
5554			handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5555			if (IS_ERR(handle)) {
5556				error = PTR_ERR(handle);
5557				goto out_mmap_sem;
5558			}
5559			if (ext4_handle_valid(handle) && shrink) {
5560				error = ext4_orphan_add(handle, inode);
5561				orphan = 1;
5562			}
5563			/*
5564			 * Update c/mtime on truncate up, ext4_truncate() will
5565			 * update c/mtime in shrink case below
5566			 */
5567			if (!shrink) {
5568				inode->i_mtime = current_time(inode);
5569				inode->i_ctime = inode->i_mtime;
5570			}
5571
5572			if (shrink)
5573				ext4_fc_track_range(handle, inode,
5574					(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5575					inode->i_sb->s_blocksize_bits,
5576					EXT_MAX_BLOCKS - 1);
5577			else
5578				ext4_fc_track_range(
5579					handle, inode,
5580					(oldsize > 0 ? oldsize - 1 : oldsize) >>
5581					inode->i_sb->s_blocksize_bits,
5582					(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5583					inode->i_sb->s_blocksize_bits);
5584
5585			down_write(&EXT4_I(inode)->i_data_sem);
5586			old_disksize = EXT4_I(inode)->i_disksize;
5587			EXT4_I(inode)->i_disksize = attr->ia_size;
5588			rc = ext4_mark_inode_dirty(handle, inode);
5589			if (!error)
5590				error = rc;
5591			/*
5592			 * We have to update i_size under i_data_sem together
5593			 * with i_disksize to avoid races with writeback code
5594			 * running ext4_wb_update_i_disksize().
5595			 */
5596			if (!error)
5597				i_size_write(inode, attr->ia_size);
5598			else
5599				EXT4_I(inode)->i_disksize = old_disksize;
5600			up_write(&EXT4_I(inode)->i_data_sem);
5601			ext4_journal_stop(handle);
5602			if (error)
5603				goto out_mmap_sem;
5604			if (!shrink) {
5605				pagecache_isize_extended(inode, oldsize,
5606							 inode->i_size);
5607			} else if (ext4_should_journal_data(inode)) {
5608				ext4_wait_for_tail_page_commit(inode);
5609			}
5610		}
 
5611
5612		/*
5613		 * Truncate pagecache after we've waited for commit
5614		 * in data=journal mode to make pages freeable.
5615		 */
5616		truncate_pagecache(inode, inode->i_size);
5617		/*
5618		 * Call ext4_truncate() even if i_size didn't change to
5619		 * truncate possible preallocated blocks.
5620		 */
5621		if (attr->ia_size <= oldsize) {
5622			rc = ext4_truncate(inode);
5623			if (rc)
5624				error = rc;
5625		}
5626out_mmap_sem:
5627		filemap_invalidate_unlock(inode->i_mapping);
5628	}
5629
5630	if (!error) {
5631		if (inc_ivers)
5632			inode_inc_iversion(inode);
5633		setattr_copy(mnt_userns, inode, attr);
5634		mark_inode_dirty(inode);
5635	}
5636
5637	/*
5638	 * If the call to ext4_truncate failed to get a transaction handle at
5639	 * all, we need to clean up the in-core orphan list manually.
5640	 */
5641	if (orphan && inode->i_nlink)
5642		ext4_orphan_del(NULL, inode);
5643
5644	if (!error && (ia_valid & ATTR_MODE))
5645		rc = posix_acl_chmod(mnt_userns, dentry, inode->i_mode);
5646
5647err_out:
5648	if  (error)
5649		ext4_std_error(inode->i_sb, error);
5650	if (!error)
5651		error = rc;
5652	return error;
5653}
5654
5655u32 ext4_dio_alignment(struct inode *inode)
 
5656{
5657	if (fsverity_active(inode))
5658		return 0;
5659	if (ext4_should_journal_data(inode))
5660		return 0;
5661	if (ext4_has_inline_data(inode))
5662		return 0;
5663	if (IS_ENCRYPTED(inode)) {
5664		if (!fscrypt_dio_supported(inode))
5665			return 0;
5666		return i_blocksize(inode);
5667	}
5668	return 1; /* use the iomap defaults */
5669}
5670
5671int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5672		 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5673{
5674	struct inode *inode = d_inode(path->dentry);
5675	struct ext4_inode *raw_inode;
5676	struct ext4_inode_info *ei = EXT4_I(inode);
5677	unsigned int flags;
5678
5679	if ((request_mask & STATX_BTIME) &&
5680	    EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5681		stat->result_mask |= STATX_BTIME;
5682		stat->btime.tv_sec = ei->i_crtime.tv_sec;
5683		stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5684	}
5685
5686	/*
5687	 * Return the DIO alignment restrictions if requested.  We only return
5688	 * this information when requested, since on encrypted files it might
5689	 * take a fair bit of work to get if the file wasn't opened recently.
5690	 */
5691	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5692		u32 dio_align = ext4_dio_alignment(inode);
5693
5694		stat->result_mask |= STATX_DIOALIGN;
5695		if (dio_align == 1) {
5696			struct block_device *bdev = inode->i_sb->s_bdev;
5697
5698			/* iomap defaults */
5699			stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5700			stat->dio_offset_align = bdev_logical_block_size(bdev);
5701		} else {
5702			stat->dio_mem_align = dio_align;
5703			stat->dio_offset_align = dio_align;
5704		}
5705	}
5706
5707	flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5708	if (flags & EXT4_APPEND_FL)
5709		stat->attributes |= STATX_ATTR_APPEND;
5710	if (flags & EXT4_COMPR_FL)
5711		stat->attributes |= STATX_ATTR_COMPRESSED;
5712	if (flags & EXT4_ENCRYPT_FL)
5713		stat->attributes |= STATX_ATTR_ENCRYPTED;
5714	if (flags & EXT4_IMMUTABLE_FL)
5715		stat->attributes |= STATX_ATTR_IMMUTABLE;
5716	if (flags & EXT4_NODUMP_FL)
5717		stat->attributes |= STATX_ATTR_NODUMP;
5718	if (flags & EXT4_VERITY_FL)
5719		stat->attributes |= STATX_ATTR_VERITY;
5720
5721	stat->attributes_mask |= (STATX_ATTR_APPEND |
5722				  STATX_ATTR_COMPRESSED |
5723				  STATX_ATTR_ENCRYPTED |
5724				  STATX_ATTR_IMMUTABLE |
5725				  STATX_ATTR_NODUMP |
5726				  STATX_ATTR_VERITY);
5727
5728	generic_fillattr(mnt_userns, inode, stat);
5729	return 0;
5730}
5731
5732int ext4_file_getattr(struct user_namespace *mnt_userns,
5733		      const struct path *path, struct kstat *stat,
5734		      u32 request_mask, unsigned int query_flags)
5735{
5736	struct inode *inode = d_inode(path->dentry);
5737	u64 delalloc_blocks;
5738
5739	ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5740
5741	/*
5742	 * If there is inline data in the inode, the inode will normally not
5743	 * have data blocks allocated (it may have an external xattr block).
5744	 * Report at least one sector for such files, so tools like tar, rsync,
5745	 * others don't incorrectly think the file is completely sparse.
5746	 */
5747	if (unlikely(ext4_has_inline_data(inode)))
5748		stat->blocks += (stat->size + 511) >> 9;
5749
5750	/*
5751	 * We can't update i_blocks if the block allocation is delayed
5752	 * otherwise in the case of system crash before the real block
5753	 * allocation is done, we will have i_blocks inconsistent with
5754	 * on-disk file blocks.
5755	 * We always keep i_blocks updated together with real
5756	 * allocation. But to not confuse with user, stat
5757	 * will return the blocks that include the delayed allocation
5758	 * blocks for this file.
5759	 */
5760	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5761				   EXT4_I(inode)->i_reserved_data_blocks);
5762	stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5763	return 0;
5764}
5765
5766static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5767				   int pextents)
5768{
5769	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5770		return ext4_ind_trans_blocks(inode, lblocks);
5771	return ext4_ext_index_trans_blocks(inode, pextents);
5772}
5773
5774/*
5775 * Account for index blocks, block groups bitmaps and block group
5776 * descriptor blocks if modify datablocks and index blocks
5777 * worse case, the indexs blocks spread over different block groups
5778 *
5779 * If datablocks are discontiguous, they are possible to spread over
5780 * different block groups too. If they are contiguous, with flexbg,
5781 * they could still across block group boundary.
5782 *
5783 * Also account for superblock, inode, quota and xattr blocks
5784 */
5785static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5786				  int pextents)
5787{
5788	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5789	int gdpblocks;
5790	int idxblocks;
5791	int ret = 0;
5792
5793	/*
5794	 * How many index blocks need to touch to map @lblocks logical blocks
5795	 * to @pextents physical extents?
 
 
 
 
5796	 */
5797	idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5798
5799	ret = idxblocks;
5800
5801	/*
5802	 * Now let's see how many group bitmaps and group descriptors need
5803	 * to account
5804	 */
5805	groups = idxblocks + pextents;
 
 
 
 
 
5806	gdpblocks = groups;
5807	if (groups > ngroups)
5808		groups = ngroups;
5809	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5810		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5811
5812	/* bitmaps and block group descriptor blocks */
5813	ret += groups + gdpblocks;
5814
5815	/* Blocks for super block, inode, quota and xattr blocks */
5816	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5817
5818	return ret;
5819}
5820
5821/*
5822 * Calculate the total number of credits to reserve to fit
5823 * the modification of a single pages into a single transaction,
5824 * which may include multiple chunks of block allocations.
5825 *
5826 * This could be called via ext4_write_begin()
5827 *
5828 * We need to consider the worse case, when
5829 * one new block per extent.
5830 */
5831int ext4_writepage_trans_blocks(struct inode *inode)
5832{
5833	int bpp = ext4_journal_blocks_per_page(inode);
5834	int ret;
5835
5836	ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5837
5838	/* Account for data blocks for journalled mode */
5839	if (ext4_should_journal_data(inode))
5840		ret += bpp;
5841	return ret;
5842}
5843
5844/*
5845 * Calculate the journal credits for a chunk of data modification.
5846 *
5847 * This is called from DIO, fallocate or whoever calling
5848 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5849 *
5850 * journal buffers for data blocks are not included here, as DIO
5851 * and fallocate do no need to journal data buffers.
5852 */
5853int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5854{
5855	return ext4_meta_trans_blocks(inode, nrblocks, 1);
5856}
5857
5858/*
5859 * The caller must have previously called ext4_reserve_inode_write().
5860 * Give this, we know that the caller already has write access to iloc->bh.
5861 */
5862int ext4_mark_iloc_dirty(handle_t *handle,
5863			 struct inode *inode, struct ext4_iloc *iloc)
5864{
5865	int err = 0;
5866
5867	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5868		put_bh(iloc->bh);
5869		return -EIO;
5870	}
5871	ext4_fc_track_inode(handle, inode);
5872
5873	/* the do_update_inode consumes one bh->b_count */
5874	get_bh(iloc->bh);
5875
5876	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5877	err = ext4_do_update_inode(handle, inode, iloc);
5878	put_bh(iloc->bh);
5879	return err;
5880}
5881
5882/*
5883 * On success, We end up with an outstanding reference count against
5884 * iloc->bh.  This _must_ be cleaned up later.
5885 */
5886
5887int
5888ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5889			 struct ext4_iloc *iloc)
5890{
5891	int err;
5892
5893	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5894		return -EIO;
5895
5896	err = ext4_get_inode_loc(inode, iloc);
5897	if (!err) {
5898		BUFFER_TRACE(iloc->bh, "get_write_access");
5899		err = ext4_journal_get_write_access(handle, inode->i_sb,
5900						    iloc->bh, EXT4_JTR_NONE);
5901		if (err) {
5902			brelse(iloc->bh);
5903			iloc->bh = NULL;
5904		}
5905	}
5906	ext4_std_error(inode->i_sb, err);
5907	return err;
5908}
5909
5910static int __ext4_expand_extra_isize(struct inode *inode,
5911				     unsigned int new_extra_isize,
5912				     struct ext4_iloc *iloc,
5913				     handle_t *handle, int *no_expand)
 
 
 
 
5914{
5915	struct ext4_inode *raw_inode;
5916	struct ext4_xattr_ibody_header *header;
5917	unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5918	struct ext4_inode_info *ei = EXT4_I(inode);
5919	int error;
5920
5921	/* this was checked at iget time, but double check for good measure */
5922	if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5923	    (ei->i_extra_isize & 3)) {
5924		EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5925				 ei->i_extra_isize,
5926				 EXT4_INODE_SIZE(inode->i_sb));
5927		return -EFSCORRUPTED;
5928	}
5929	if ((new_extra_isize < ei->i_extra_isize) ||
5930	    (new_extra_isize < 4) ||
5931	    (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5932		return -EINVAL;	/* Should never happen */
5933
5934	raw_inode = ext4_raw_inode(iloc);
5935
5936	header = IHDR(inode, raw_inode);
5937
5938	/* No extended attributes present */
5939	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5940	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5941		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5942		       EXT4_I(inode)->i_extra_isize, 0,
5943		       new_extra_isize - EXT4_I(inode)->i_extra_isize);
5944		EXT4_I(inode)->i_extra_isize = new_extra_isize;
5945		return 0;
5946	}
5947
5948	/*
5949	 * We may need to allocate external xattr block so we need quotas
5950	 * initialized. Here we can be called with various locks held so we
5951	 * cannot affort to initialize quotas ourselves. So just bail.
5952	 */
5953	if (dquot_initialize_needed(inode))
5954		return -EAGAIN;
5955
5956	/* try to expand with EAs present */
5957	error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5958					   raw_inode, handle);
5959	if (error) {
5960		/*
5961		 * Inode size expansion failed; don't try again
5962		 */
5963		*no_expand = 1;
5964	}
5965
5966	return error;
5967}
5968
5969/*
5970 * Expand an inode by new_extra_isize bytes.
5971 * Returns 0 on success or negative error number on failure.
5972 */
5973static int ext4_try_to_expand_extra_isize(struct inode *inode,
5974					  unsigned int new_extra_isize,
5975					  struct ext4_iloc iloc,
5976					  handle_t *handle)
5977{
5978	int no_expand;
5979	int error;
5980
5981	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5982		return -EOVERFLOW;
5983
5984	/*
5985	 * In nojournal mode, we can immediately attempt to expand
5986	 * the inode.  When journaled, we first need to obtain extra
5987	 * buffer credits since we may write into the EA block
5988	 * with this same handle. If journal_extend fails, then it will
5989	 * only result in a minor loss of functionality for that inode.
5990	 * If this is felt to be critical, then e2fsck should be run to
5991	 * force a large enough s_min_extra_isize.
5992	 */
5993	if (ext4_journal_extend(handle,
5994				EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5995		return -ENOSPC;
5996
5997	if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5998		return -EBUSY;
5999
6000	error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
6001					  handle, &no_expand);
6002	ext4_write_unlock_xattr(inode, &no_expand);
6003
6004	return error;
6005}
6006
6007int ext4_expand_extra_isize(struct inode *inode,
6008			    unsigned int new_extra_isize,
6009			    struct ext4_iloc *iloc)
6010{
6011	handle_t *handle;
6012	int no_expand;
6013	int error, rc;
6014
6015	if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
6016		brelse(iloc->bh);
6017		return -EOVERFLOW;
6018	}
6019
6020	handle = ext4_journal_start(inode, EXT4_HT_INODE,
6021				    EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6022	if (IS_ERR(handle)) {
6023		error = PTR_ERR(handle);
6024		brelse(iloc->bh);
6025		return error;
6026	}
6027
6028	ext4_write_lock_xattr(inode, &no_expand);
6029
6030	BUFFER_TRACE(iloc->bh, "get_write_access");
6031	error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
6032					      EXT4_JTR_NONE);
6033	if (error) {
6034		brelse(iloc->bh);
6035		goto out_unlock;
6036	}
6037
6038	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6039					  handle, &no_expand);
6040
6041	rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6042	if (!error)
6043		error = rc;
6044
6045out_unlock:
6046	ext4_write_unlock_xattr(inode, &no_expand);
6047	ext4_journal_stop(handle);
6048	return error;
6049}
6050
6051/*
6052 * What we do here is to mark the in-core inode as clean with respect to inode
6053 * dirtiness (it may still be data-dirty).
6054 * This means that the in-core inode may be reaped by prune_icache
6055 * without having to perform any I/O.  This is a very good thing,
6056 * because *any* task may call prune_icache - even ones which
6057 * have a transaction open against a different journal.
6058 *
6059 * Is this cheating?  Not really.  Sure, we haven't written the
6060 * inode out, but prune_icache isn't a user-visible syncing function.
6061 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6062 * we start and wait on commits.
 
 
 
 
 
 
 
 
6063 */
6064int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
6065				const char *func, unsigned int line)
6066{
6067	struct ext4_iloc iloc;
6068	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6069	int err;
 
6070
6071	might_sleep();
6072	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6073	err = ext4_reserve_inode_write(handle, inode, &iloc);
6074	if (err)
6075		goto out;
6076
6077	if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6078		ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6079					       iloc, handle);
6080
6081	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
6082out:
6083	if (unlikely(err))
6084		ext4_error_inode_err(inode, func, line, 0, err,
6085					"mark_inode_dirty error");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6086	return err;
6087}
6088
6089/*
6090 * ext4_dirty_inode() is called from __mark_inode_dirty()
6091 *
6092 * We're really interested in the case where a file is being extended.
6093 * i_size has been changed by generic_commit_write() and we thus need
6094 * to include the updated inode in the current transaction.
6095 *
6096 * Also, dquot_alloc_block() will always dirty the inode when blocks
6097 * are allocated to the file.
6098 *
6099 * If the inode is marked synchronous, we don't honour that here - doing
6100 * so would cause a commit on atime updates, which we don't bother doing.
6101 * We handle synchronous inodes at the highest possible level.
6102 */
6103void ext4_dirty_inode(struct inode *inode, int flags)
6104{
6105	handle_t *handle;
6106
6107	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6108	if (IS_ERR(handle))
6109		return;
 
6110	ext4_mark_inode_dirty(handle, inode);
 
6111	ext4_journal_stop(handle);
 
 
6112}
6113
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6114int ext4_change_inode_journal_flag(struct inode *inode, int val)
6115{
6116	journal_t *journal;
6117	handle_t *handle;
6118	int err;
6119	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6120
6121	/*
6122	 * We have to be very careful here: changing a data block's
6123	 * journaling status dynamically is dangerous.  If we write a
6124	 * data block to the journal, change the status and then delete
6125	 * that block, we risk forgetting to revoke the old log record
6126	 * from the journal and so a subsequent replay can corrupt data.
6127	 * So, first we make sure that the journal is empty and that
6128	 * nobody is changing anything.
6129	 */
6130
6131	journal = EXT4_JOURNAL(inode);
6132	if (!journal)
6133		return 0;
6134	if (is_journal_aborted(journal))
6135		return -EROFS;
6136
6137	/* Wait for all existing dio workers */
6138	inode_dio_wait(inode);
6139
6140	/*
6141	 * Before flushing the journal and switching inode's aops, we have
6142	 * to flush all dirty data the inode has. There can be outstanding
6143	 * delayed allocations, there can be unwritten extents created by
6144	 * fallocate or buffered writes in dioread_nolock mode covered by
6145	 * dirty data which can be converted only after flushing the dirty
6146	 * data (and journalled aops don't know how to handle these cases).
6147	 */
6148	if (val) {
6149		filemap_invalidate_lock(inode->i_mapping);
6150		err = filemap_write_and_wait(inode->i_mapping);
6151		if (err < 0) {
6152			filemap_invalidate_unlock(inode->i_mapping);
6153			return err;
6154		}
6155	}
6156
6157	percpu_down_write(&sbi->s_writepages_rwsem);
6158	jbd2_journal_lock_updates(journal);
 
6159
6160	/*
6161	 * OK, there are no updates running now, and all cached data is
6162	 * synced to disk.  We are now in a completely consistent state
6163	 * which doesn't have anything in the journal, and we know that
6164	 * no filesystem updates are running, so it is safe to modify
6165	 * the inode's in-core data-journaling state flag now.
6166	 */
6167
6168	if (val)
6169		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6170	else {
6171		err = jbd2_journal_flush(journal, 0);
6172		if (err < 0) {
6173			jbd2_journal_unlock_updates(journal);
6174			percpu_up_write(&sbi->s_writepages_rwsem);
6175			return err;
6176		}
6177		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6178	}
6179	ext4_set_aops(inode);
6180
6181	jbd2_journal_unlock_updates(journal);
6182	percpu_up_write(&sbi->s_writepages_rwsem);
6183
6184	if (val)
6185		filemap_invalidate_unlock(inode->i_mapping);
6186
6187	/* Finally we can mark the inode as dirty. */
6188
6189	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6190	if (IS_ERR(handle))
6191		return PTR_ERR(handle);
6192
6193	ext4_fc_mark_ineligible(inode->i_sb,
6194		EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6195	err = ext4_mark_inode_dirty(handle, inode);
6196	ext4_handle_sync(handle);
6197	ext4_journal_stop(handle);
6198	ext4_std_error(inode->i_sb, err);
6199
6200	return err;
6201}
6202
6203static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6204			    struct buffer_head *bh)
6205{
6206	return !buffer_mapped(bh);
6207}
6208
6209vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6210{
6211	struct vm_area_struct *vma = vmf->vma;
6212	struct page *page = vmf->page;
6213	loff_t size;
6214	unsigned long len;
6215	int err;
6216	vm_fault_t ret;
6217	struct file *file = vma->vm_file;
6218	struct inode *inode = file_inode(file);
6219	struct address_space *mapping = inode->i_mapping;
6220	handle_t *handle;
6221	get_block_t *get_block;
6222	int retries = 0;
6223
6224	if (unlikely(IS_IMMUTABLE(inode)))
6225		return VM_FAULT_SIGBUS;
6226
6227	sb_start_pagefault(inode->i_sb);
6228	file_update_time(vma->vm_file);
6229
6230	filemap_invalidate_lock_shared(mapping);
6231
6232	err = ext4_convert_inline_data(inode);
6233	if (err)
6234		goto out_ret;
6235
6236	/*
6237	 * On data journalling we skip straight to the transaction handle:
6238	 * there's no delalloc; page truncated will be checked later; the
6239	 * early return w/ all buffers mapped (calculates size/len) can't
6240	 * be used; and there's no dioread_nolock, so only ext4_get_block.
6241	 */
6242	if (ext4_should_journal_data(inode))
6243		goto retry_alloc;
6244
6245	/* Delalloc case is easy... */
6246	if (test_opt(inode->i_sb, DELALLOC) &&
 
6247	    !ext4_nonda_switch(inode->i_sb)) {
6248		do {
6249			err = block_page_mkwrite(vma, vmf,
6250						   ext4_da_get_block_prep);
6251		} while (err == -ENOSPC &&
6252		       ext4_should_retry_alloc(inode->i_sb, &retries));
6253		goto out_ret;
6254	}
6255
6256	lock_page(page);
6257	size = i_size_read(inode);
6258	/* Page got truncated from under us? */
6259	if (page->mapping != mapping || page_offset(page) > size) {
6260		unlock_page(page);
6261		ret = VM_FAULT_NOPAGE;
6262		goto out;
6263	}
6264
6265	if (page->index == size >> PAGE_SHIFT)
6266		len = size & ~PAGE_MASK;
6267	else
6268		len = PAGE_SIZE;
6269	/*
6270	 * Return if we have all the buffers mapped. This avoids the need to do
6271	 * journal_start/journal_stop which can block and take a long time
6272	 *
6273	 * This cannot be done for data journalling, as we have to add the
6274	 * inode to the transaction's list to writeprotect pages on commit.
6275	 */
6276	if (page_has_buffers(page)) {
6277		if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page),
6278					    0, len, NULL,
6279					    ext4_bh_unmapped)) {
6280			/* Wait so that we don't change page under IO */
6281			wait_for_stable_page(page);
6282			ret = VM_FAULT_LOCKED;
6283			goto out;
6284		}
6285	}
6286	unlock_page(page);
6287	/* OK, we need to fill the hole... */
6288	if (ext4_should_dioread_nolock(inode))
6289		get_block = ext4_get_block_unwritten;
6290	else
6291		get_block = ext4_get_block;
6292retry_alloc:
6293	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6294				    ext4_writepage_trans_blocks(inode));
6295	if (IS_ERR(handle)) {
6296		ret = VM_FAULT_SIGBUS;
6297		goto out;
6298	}
6299	/*
6300	 * Data journalling can't use block_page_mkwrite() because it
6301	 * will set_buffer_dirty() before do_journal_get_write_access()
6302	 * thus might hit warning messages for dirty metadata buffers.
6303	 */
6304	if (!ext4_should_journal_data(inode)) {
6305		err = block_page_mkwrite(vma, vmf, get_block);
6306	} else {
6307		lock_page(page);
6308		size = i_size_read(inode);
6309		/* Page got truncated from under us? */
6310		if (page->mapping != mapping || page_offset(page) > size) {
6311			ret = VM_FAULT_NOPAGE;
6312			goto out_error;
6313		}
6314
6315		if (page->index == size >> PAGE_SHIFT)
6316			len = size & ~PAGE_MASK;
6317		else
6318			len = PAGE_SIZE;
6319
6320		err = __block_write_begin(page, 0, len, ext4_get_block);
6321		if (!err) {
6322			ret = VM_FAULT_SIGBUS;
6323			if (ext4_walk_page_buffers(handle, inode,
6324					page_buffers(page), 0, len, NULL,
6325					do_journal_get_write_access))
6326				goto out_error;
6327			if (ext4_walk_page_buffers(handle, inode,
6328					page_buffers(page), 0, len, NULL,
6329					write_end_fn))
6330				goto out_error;
6331			if (ext4_jbd2_inode_add_write(handle, inode,
6332						      page_offset(page), len))
6333				goto out_error;
6334			ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6335		} else {
6336			unlock_page(page);
6337		}
 
6338	}
6339	ext4_journal_stop(handle);
6340	if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6341		goto retry_alloc;
6342out_ret:
6343	ret = block_page_mkwrite_return(err);
6344out:
6345	filemap_invalidate_unlock_shared(mapping);
6346	sb_end_pagefault(inode->i_sb);
6347	return ret;
6348out_error:
6349	unlock_page(page);
6350	ext4_journal_stop(handle);
6351	goto out;
6352}