2 * linux/fs/ext4/inode.c
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)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/buffer_head.h>
28 #include <linux/writeback.h>
29 #include <linux/pagevec.h>
30 #include <linux/mpage.h>
31 #include <linux/namei.h>
32 #include <linux/uio.h>
33 #include <linux/bio.h>
34 #include <linux/workqueue.h>
35 #include <linux/kernel.h>
36 #include <linux/printk.h>
37 #include <linux/slab.h>
38 #include <linux/bitops.h>
40 #include "ext4_jbd2.h"
45 #include <trace/events/ext4.h>
47 #define MPAGE_DA_EXTENT_TAIL 0x01
49 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
50 struct ext4_inode_info *ei)
52 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 csum_lo = le16_to_cpu(raw->i_checksum_lo);
58 raw->i_checksum_lo = 0;
59 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
60 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
61 csum_hi = le16_to_cpu(raw->i_checksum_hi);
62 raw->i_checksum_hi = 0;
65 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
66 EXT4_INODE_SIZE(inode->i_sb));
68 raw->i_checksum_lo = cpu_to_le16(csum_lo);
69 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
70 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
71 raw->i_checksum_hi = cpu_to_le16(csum_hi);
76 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
77 struct ext4_inode_info *ei)
79 __u32 provided, calculated;
81 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
82 cpu_to_le32(EXT4_OS_LINUX) ||
83 !ext4_has_metadata_csum(inode->i_sb))
86 provided = le16_to_cpu(raw->i_checksum_lo);
87 calculated = ext4_inode_csum(inode, raw, ei);
88 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
89 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
90 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
94 return provided == calculated;
97 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
98 struct ext4_inode_info *ei)
102 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
103 cpu_to_le32(EXT4_OS_LINUX) ||
104 !ext4_has_metadata_csum(inode->i_sb))
107 csum = ext4_inode_csum(inode, raw, ei);
108 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
109 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
110 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
111 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
114 static inline int ext4_begin_ordered_truncate(struct inode *inode,
117 trace_ext4_begin_ordered_truncate(inode, new_size);
119 * If jinode is zero, then we never opened the file for
120 * writing, so there's no need to call
121 * jbd2_journal_begin_ordered_truncate() since there's no
122 * outstanding writes we need to flush.
124 if (!EXT4_I(inode)->jinode)
126 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
127 EXT4_I(inode)->jinode,
131 static void ext4_invalidatepage(struct page *page, unsigned int offset,
132 unsigned int length);
133 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
134 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
135 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
139 * Test whether an inode is a fast symlink.
141 int ext4_inode_is_fast_symlink(struct inode *inode)
143 int ea_blocks = EXT4_I(inode)->i_file_acl ?
144 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
146 if (ext4_has_inline_data(inode))
149 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
157 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
168 BUG_ON(EXT4_JOURNAL(inode) == NULL);
169 jbd_debug(2, "restarting handle %p\n", handle);
170 up_write(&EXT4_I(inode)->i_data_sem);
171 ret = ext4_journal_restart(handle, nblocks);
172 down_write(&EXT4_I(inode)->i_data_sem);
173 ext4_discard_preallocations(inode);
179 * Called at the last iput() if i_nlink is zero.
181 void ext4_evict_inode(struct inode *inode)
186 trace_ext4_evict_inode(inode);
188 if (inode->i_nlink) {
190 * When journalling data dirty buffers are tracked only in the
191 * journal. So although mm thinks everything is clean and
192 * ready for reaping the inode might still have some pages to
193 * write in the running transaction or waiting to be
194 * checkpointed. Thus calling jbd2_journal_invalidatepage()
195 * (via truncate_inode_pages()) to discard these buffers can
196 * cause data loss. Also even if we did not discard these
197 * buffers, we would have no way to find them after the inode
198 * is reaped and thus user could see stale data if he tries to
199 * read them before the transaction is checkpointed. So be
200 * careful and force everything to disk here... We use
201 * ei->i_datasync_tid to store the newest transaction
202 * containing inode's data.
204 * Note that directories do not have this problem because they
205 * don't use page cache.
207 if (ext4_should_journal_data(inode) &&
208 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
209 inode->i_ino != EXT4_JOURNAL_INO) {
210 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
211 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
213 jbd2_complete_transaction(journal, commit_tid);
214 filemap_write_and_wait(&inode->i_data);
216 truncate_inode_pages_final(&inode->i_data);
218 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
222 if (is_bad_inode(inode))
224 dquot_initialize(inode);
226 if (ext4_should_order_data(inode))
227 ext4_begin_ordered_truncate(inode, 0);
228 truncate_inode_pages_final(&inode->i_data);
230 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
236 sb_start_intwrite(inode->i_sb);
237 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
238 ext4_blocks_for_truncate(inode)+3);
239 if (IS_ERR(handle)) {
240 ext4_std_error(inode->i_sb, PTR_ERR(handle));
242 * If we're going to skip the normal cleanup, we still need to
243 * make sure that the in-core orphan linked list is properly
246 ext4_orphan_del(NULL, inode);
247 sb_end_intwrite(inode->i_sb);
252 ext4_handle_sync(handle);
254 err = ext4_mark_inode_dirty(handle, inode);
256 ext4_warning(inode->i_sb,
257 "couldn't mark inode dirty (err %d)", err);
261 ext4_truncate(inode);
264 * ext4_ext_truncate() doesn't reserve any slop when it
265 * restarts journal transactions; therefore there may not be
266 * enough credits left in the handle to remove the inode from
267 * the orphan list and set the dtime field.
269 if (!ext4_handle_has_enough_credits(handle, 3)) {
270 err = ext4_journal_extend(handle, 3);
272 err = ext4_journal_restart(handle, 3);
274 ext4_warning(inode->i_sb,
275 "couldn't extend journal (err %d)", err);
277 ext4_journal_stop(handle);
278 ext4_orphan_del(NULL, inode);
279 sb_end_intwrite(inode->i_sb);
285 * Kill off the orphan record which ext4_truncate created.
286 * AKPM: I think this can be inside the above `if'.
287 * Note that ext4_orphan_del() has to be able to cope with the
288 * deletion of a non-existent orphan - this is because we don't
289 * know if ext4_truncate() actually created an orphan record.
290 * (Well, we could do this if we need to, but heck - it works)
292 ext4_orphan_del(handle, inode);
293 EXT4_I(inode)->i_dtime = get_seconds();
296 * One subtle ordering requirement: if anything has gone wrong
297 * (transaction abort, IO errors, whatever), then we can still
298 * do these next steps (the fs will already have been marked as
299 * having errors), but we can't free the inode if the mark_dirty
302 if (ext4_mark_inode_dirty(handle, inode))
303 /* If that failed, just do the required in-core inode clear. */
304 ext4_clear_inode(inode);
306 ext4_free_inode(handle, inode);
307 ext4_journal_stop(handle);
308 sb_end_intwrite(inode->i_sb);
311 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
315 qsize_t *ext4_get_reserved_space(struct inode *inode)
317 return &EXT4_I(inode)->i_reserved_quota;
322 * Called with i_data_sem down, which is important since we can call
323 * ext4_discard_preallocations() from here.
325 void ext4_da_update_reserve_space(struct inode *inode,
326 int used, int quota_claim)
328 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
329 struct ext4_inode_info *ei = EXT4_I(inode);
331 spin_lock(&ei->i_block_reservation_lock);
332 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
333 if (unlikely(used > ei->i_reserved_data_blocks)) {
334 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
335 "with only %d reserved data blocks",
336 __func__, inode->i_ino, used,
337 ei->i_reserved_data_blocks);
339 used = ei->i_reserved_data_blocks;
342 /* Update per-inode reservations */
343 ei->i_reserved_data_blocks -= used;
344 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
346 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
348 /* Update quota subsystem for data blocks */
350 dquot_claim_block(inode, EXT4_C2B(sbi, used));
353 * We did fallocate with an offset that is already delayed
354 * allocated. So on delayed allocated writeback we should
355 * not re-claim the quota for fallocated blocks.
357 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
361 * If we have done all the pending block allocations and if
362 * there aren't any writers on the inode, we can discard the
363 * inode's preallocations.
365 if ((ei->i_reserved_data_blocks == 0) &&
366 (atomic_read(&inode->i_writecount) == 0))
367 ext4_discard_preallocations(inode);
370 static int __check_block_validity(struct inode *inode, const char *func,
372 struct ext4_map_blocks *map)
374 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
376 ext4_error_inode(inode, func, line, map->m_pblk,
377 "lblock %lu mapped to illegal pblock "
378 "(length %d)", (unsigned long) map->m_lblk,
385 #define check_block_validity(inode, map) \
386 __check_block_validity((inode), __func__, __LINE__, (map))
388 #ifdef ES_AGGRESSIVE_TEST
389 static void ext4_map_blocks_es_recheck(handle_t *handle,
391 struct ext4_map_blocks *es_map,
392 struct ext4_map_blocks *map,
399 * There is a race window that the result is not the same.
400 * e.g. xfstests #223 when dioread_nolock enables. The reason
401 * is that we lookup a block mapping in extent status tree with
402 * out taking i_data_sem. So at the time the unwritten extent
403 * could be converted.
405 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
406 down_read(&EXT4_I(inode)->i_data_sem);
407 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
408 retval = ext4_ext_map_blocks(handle, inode, map, flags &
409 EXT4_GET_BLOCKS_KEEP_SIZE);
411 retval = ext4_ind_map_blocks(handle, inode, map, flags &
412 EXT4_GET_BLOCKS_KEEP_SIZE);
414 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
415 up_read((&EXT4_I(inode)->i_data_sem));
418 * We don't check m_len because extent will be collpased in status
419 * tree. So the m_len might not equal.
421 if (es_map->m_lblk != map->m_lblk ||
422 es_map->m_flags != map->m_flags ||
423 es_map->m_pblk != map->m_pblk) {
424 printk("ES cache assertion failed for inode: %lu "
425 "es_cached ex [%d/%d/%llu/%x] != "
426 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
427 inode->i_ino, es_map->m_lblk, es_map->m_len,
428 es_map->m_pblk, es_map->m_flags, map->m_lblk,
429 map->m_len, map->m_pblk, map->m_flags,
433 #endif /* ES_AGGRESSIVE_TEST */
436 * The ext4_map_blocks() function tries to look up the requested blocks,
437 * and returns if the blocks are already mapped.
439 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
440 * and store the allocated blocks in the result buffer head and mark it
443 * If file type is extents based, it will call ext4_ext_map_blocks(),
444 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
447 * On success, it returns the number of blocks being mapped or allocated.
448 * if create==0 and the blocks are pre-allocated and unwritten block,
449 * the result buffer head is unmapped. If the create ==1, it will make sure
450 * the buffer head is mapped.
452 * It returns 0 if plain look up failed (blocks have not been allocated), in
453 * that case, buffer head is unmapped
455 * It returns the error in case of allocation failure.
457 int ext4_map_blocks(handle_t *handle, struct inode *inode,
458 struct ext4_map_blocks *map, int flags)
460 struct extent_status es;
463 #ifdef ES_AGGRESSIVE_TEST
464 struct ext4_map_blocks orig_map;
466 memcpy(&orig_map, map, sizeof(*map));
470 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
471 "logical block %lu\n", inode->i_ino, flags, map->m_len,
472 (unsigned long) map->m_lblk);
475 * ext4_map_blocks returns an int, and m_len is an unsigned int
477 if (unlikely(map->m_len > INT_MAX))
478 map->m_len = INT_MAX;
480 /* We can handle the block number less than EXT_MAX_BLOCKS */
481 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
484 /* Lookup extent status tree firstly */
485 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
486 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
487 map->m_pblk = ext4_es_pblock(&es) +
488 map->m_lblk - es.es_lblk;
489 map->m_flags |= ext4_es_is_written(&es) ?
490 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
491 retval = es.es_len - (map->m_lblk - es.es_lblk);
492 if (retval > map->m_len)
495 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
500 #ifdef ES_AGGRESSIVE_TEST
501 ext4_map_blocks_es_recheck(handle, inode, map,
508 * Try to see if we can get the block without requesting a new
511 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
512 down_read(&EXT4_I(inode)->i_data_sem);
513 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
514 retval = ext4_ext_map_blocks(handle, inode, map, flags &
515 EXT4_GET_BLOCKS_KEEP_SIZE);
517 retval = ext4_ind_map_blocks(handle, inode, map, flags &
518 EXT4_GET_BLOCKS_KEEP_SIZE);
523 if (unlikely(retval != map->m_len)) {
524 ext4_warning(inode->i_sb,
525 "ES len assertion failed for inode "
526 "%lu: retval %d != map->m_len %d",
527 inode->i_ino, retval, map->m_len);
531 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
532 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
533 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
534 !(status & EXTENT_STATUS_WRITTEN) &&
535 ext4_find_delalloc_range(inode, map->m_lblk,
536 map->m_lblk + map->m_len - 1))
537 status |= EXTENT_STATUS_DELAYED;
538 ret = ext4_es_insert_extent(inode, map->m_lblk,
539 map->m_len, map->m_pblk, status);
543 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
544 up_read((&EXT4_I(inode)->i_data_sem));
547 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
548 ret = check_block_validity(inode, map);
553 /* If it is only a block(s) look up */
554 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
558 * Returns if the blocks have already allocated
560 * Note that if blocks have been preallocated
561 * ext4_ext_get_block() returns the create = 0
562 * with buffer head unmapped.
564 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
566 * If we need to convert extent to unwritten
567 * we continue and do the actual work in
568 * ext4_ext_map_blocks()
570 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
574 * Here we clear m_flags because after allocating an new extent,
575 * it will be set again.
577 map->m_flags &= ~EXT4_MAP_FLAGS;
580 * New blocks allocate and/or writing to unwritten extent
581 * will possibly result in updating i_data, so we take
582 * the write lock of i_data_sem, and call get_block()
583 * with create == 1 flag.
585 down_write(&EXT4_I(inode)->i_data_sem);
588 * We need to check for EXT4 here because migrate
589 * could have changed the inode type in between
591 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
592 retval = ext4_ext_map_blocks(handle, inode, map, flags);
594 retval = ext4_ind_map_blocks(handle, inode, map, flags);
596 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
598 * We allocated new blocks which will result in
599 * i_data's format changing. Force the migrate
600 * to fail by clearing migrate flags
602 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
606 * Update reserved blocks/metadata blocks after successful
607 * block allocation which had been deferred till now. We don't
608 * support fallocate for non extent files. So we can update
609 * reserve space here.
612 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
613 ext4_da_update_reserve_space(inode, retval, 1);
619 if (unlikely(retval != map->m_len)) {
620 ext4_warning(inode->i_sb,
621 "ES len assertion failed for inode "
622 "%lu: retval %d != map->m_len %d",
623 inode->i_ino, retval, map->m_len);
628 * If the extent has been zeroed out, we don't need to update
629 * extent status tree.
631 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
632 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
633 if (ext4_es_is_written(&es))
636 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
637 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
638 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
639 !(status & EXTENT_STATUS_WRITTEN) &&
640 ext4_find_delalloc_range(inode, map->m_lblk,
641 map->m_lblk + map->m_len - 1))
642 status |= EXTENT_STATUS_DELAYED;
643 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
644 map->m_pblk, status);
650 up_write((&EXT4_I(inode)->i_data_sem));
651 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
652 ret = check_block_validity(inode, map);
659 static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate)
661 struct inode *inode = bh->b_assoc_map->host;
662 /* XXX: breaks on 32-bit > 16GB. Is that even supported? */
663 loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits;
667 WARN_ON(!buffer_unwritten(bh));
668 err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size);
671 /* Maximum number of blocks we map for direct IO at once. */
672 #define DIO_MAX_BLOCKS 4096
674 static int _ext4_get_block(struct inode *inode, sector_t iblock,
675 struct buffer_head *bh, int flags)
677 handle_t *handle = ext4_journal_current_handle();
678 struct ext4_map_blocks map;
679 int ret = 0, started = 0;
682 if (ext4_has_inline_data(inode))
686 map.m_len = bh->b_size >> inode->i_blkbits;
688 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
689 /* Direct IO write... */
690 if (map.m_len > DIO_MAX_BLOCKS)
691 map.m_len = DIO_MAX_BLOCKS;
692 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
693 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
695 if (IS_ERR(handle)) {
696 ret = PTR_ERR(handle);
702 ret = ext4_map_blocks(handle, inode, &map, flags);
704 ext4_io_end_t *io_end = ext4_inode_aio(inode);
706 map_bh(bh, inode->i_sb, map.m_pblk);
707 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
708 if (IS_DAX(inode) && buffer_unwritten(bh) && !io_end) {
709 bh->b_assoc_map = inode->i_mapping;
710 bh->b_private = (void *)(unsigned long)iblock;
711 bh->b_end_io = ext4_end_io_unwritten;
713 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
714 set_buffer_defer_completion(bh);
715 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
719 ext4_journal_stop(handle);
723 int ext4_get_block(struct inode *inode, sector_t iblock,
724 struct buffer_head *bh, int create)
726 return _ext4_get_block(inode, iblock, bh,
727 create ? EXT4_GET_BLOCKS_CREATE : 0);
731 * `handle' can be NULL if create is zero
733 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
734 ext4_lblk_t block, int create)
736 struct ext4_map_blocks map;
737 struct buffer_head *bh;
740 J_ASSERT(handle != NULL || create == 0);
744 err = ext4_map_blocks(handle, inode, &map,
745 create ? EXT4_GET_BLOCKS_CREATE : 0);
748 return create ? ERR_PTR(-ENOSPC) : NULL;
752 bh = sb_getblk(inode->i_sb, map.m_pblk);
754 return ERR_PTR(-ENOMEM);
755 if (map.m_flags & EXT4_MAP_NEW) {
756 J_ASSERT(create != 0);
757 J_ASSERT(handle != NULL);
760 * Now that we do not always journal data, we should
761 * keep in mind whether this should always journal the
762 * new buffer as metadata. For now, regular file
763 * writes use ext4_get_block instead, so it's not a
767 BUFFER_TRACE(bh, "call get_create_access");
768 err = ext4_journal_get_create_access(handle, bh);
773 if (!buffer_uptodate(bh)) {
774 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
775 set_buffer_uptodate(bh);
778 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
779 err = ext4_handle_dirty_metadata(handle, inode, bh);
783 BUFFER_TRACE(bh, "not a new buffer");
790 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
791 ext4_lblk_t block, int create)
793 struct buffer_head *bh;
795 bh = ext4_getblk(handle, inode, block, create);
798 if (!bh || buffer_uptodate(bh))
800 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
802 if (buffer_uptodate(bh))
805 return ERR_PTR(-EIO);
808 int ext4_walk_page_buffers(handle_t *handle,
809 struct buffer_head *head,
813 int (*fn)(handle_t *handle,
814 struct buffer_head *bh))
816 struct buffer_head *bh;
817 unsigned block_start, block_end;
818 unsigned blocksize = head->b_size;
820 struct buffer_head *next;
822 for (bh = head, block_start = 0;
823 ret == 0 && (bh != head || !block_start);
824 block_start = block_end, bh = next) {
825 next = bh->b_this_page;
826 block_end = block_start + blocksize;
827 if (block_end <= from || block_start >= to) {
828 if (partial && !buffer_uptodate(bh))
832 err = (*fn)(handle, bh);
840 * To preserve ordering, it is essential that the hole instantiation and
841 * the data write be encapsulated in a single transaction. We cannot
842 * close off a transaction and start a new one between the ext4_get_block()
843 * and the commit_write(). So doing the jbd2_journal_start at the start of
844 * prepare_write() is the right place.
846 * Also, this function can nest inside ext4_writepage(). In that case, we
847 * *know* that ext4_writepage() has generated enough buffer credits to do the
848 * whole page. So we won't block on the journal in that case, which is good,
849 * because the caller may be PF_MEMALLOC.
851 * By accident, ext4 can be reentered when a transaction is open via
852 * quota file writes. If we were to commit the transaction while thus
853 * reentered, there can be a deadlock - we would be holding a quota
854 * lock, and the commit would never complete if another thread had a
855 * transaction open and was blocking on the quota lock - a ranking
858 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
859 * will _not_ run commit under these circumstances because handle->h_ref
860 * is elevated. We'll still have enough credits for the tiny quotafile
863 int do_journal_get_write_access(handle_t *handle,
864 struct buffer_head *bh)
866 int dirty = buffer_dirty(bh);
869 if (!buffer_mapped(bh) || buffer_freed(bh))
872 * __block_write_begin() could have dirtied some buffers. Clean
873 * the dirty bit as jbd2_journal_get_write_access() could complain
874 * otherwise about fs integrity issues. Setting of the dirty bit
875 * by __block_write_begin() isn't a real problem here as we clear
876 * the bit before releasing a page lock and thus writeback cannot
877 * ever write the buffer.
880 clear_buffer_dirty(bh);
881 BUFFER_TRACE(bh, "get write access");
882 ret = ext4_journal_get_write_access(handle, bh);
884 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
888 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
889 struct buffer_head *bh_result, int create);
891 #ifdef CONFIG_EXT4_FS_ENCRYPTION
892 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
893 get_block_t *get_block)
895 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
896 unsigned to = from + len;
897 struct inode *inode = page->mapping->host;
898 unsigned block_start, block_end;
901 unsigned blocksize = inode->i_sb->s_blocksize;
903 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
904 bool decrypt = false;
906 BUG_ON(!PageLocked(page));
907 BUG_ON(from > PAGE_CACHE_SIZE);
908 BUG_ON(to > PAGE_CACHE_SIZE);
911 if (!page_has_buffers(page))
912 create_empty_buffers(page, blocksize, 0);
913 head = page_buffers(page);
914 bbits = ilog2(blocksize);
915 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
917 for (bh = head, block_start = 0; bh != head || !block_start;
918 block++, block_start = block_end, bh = bh->b_this_page) {
919 block_end = block_start + blocksize;
920 if (block_end <= from || block_start >= to) {
921 if (PageUptodate(page)) {
922 if (!buffer_uptodate(bh))
923 set_buffer_uptodate(bh);
928 clear_buffer_new(bh);
929 if (!buffer_mapped(bh)) {
930 WARN_ON(bh->b_size != blocksize);
931 err = get_block(inode, block, bh, 1);
934 if (buffer_new(bh)) {
935 unmap_underlying_metadata(bh->b_bdev,
937 if (PageUptodate(page)) {
938 clear_buffer_new(bh);
939 set_buffer_uptodate(bh);
940 mark_buffer_dirty(bh);
943 if (block_end > to || block_start < from)
944 zero_user_segments(page, to, block_end,
949 if (PageUptodate(page)) {
950 if (!buffer_uptodate(bh))
951 set_buffer_uptodate(bh);
954 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
955 !buffer_unwritten(bh) &&
956 (block_start < from || block_end > to)) {
957 ll_rw_block(READ, 1, &bh);
959 decrypt = ext4_encrypted_inode(inode) &&
960 S_ISREG(inode->i_mode);
964 * If we issued read requests, let them complete.
966 while (wait_bh > wait) {
967 wait_on_buffer(*--wait_bh);
968 if (!buffer_uptodate(*wait_bh))
972 page_zero_new_buffers(page, from, to);
974 err = ext4_decrypt_one(inode, page);
979 static int ext4_write_begin(struct file *file, struct address_space *mapping,
980 loff_t pos, unsigned len, unsigned flags,
981 struct page **pagep, void **fsdata)
983 struct inode *inode = mapping->host;
984 int ret, needed_blocks;
991 trace_ext4_write_begin(inode, pos, len, flags);
993 * Reserve one block more for addition to orphan list in case
994 * we allocate blocks but write fails for some reason
996 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
997 index = pos >> PAGE_CACHE_SHIFT;
998 from = pos & (PAGE_CACHE_SIZE - 1);
1001 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1002 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1011 * grab_cache_page_write_begin() can take a long time if the
1012 * system is thrashing due to memory pressure, or if the page
1013 * is being written back. So grab it first before we start
1014 * the transaction handle. This also allows us to allocate
1015 * the page (if needed) without using GFP_NOFS.
1018 page = grab_cache_page_write_begin(mapping, index, flags);
1024 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1025 if (IS_ERR(handle)) {
1026 page_cache_release(page);
1027 return PTR_ERR(handle);
1031 if (page->mapping != mapping) {
1032 /* The page got truncated from under us */
1034 page_cache_release(page);
1035 ext4_journal_stop(handle);
1038 /* In case writeback began while the page was unlocked */
1039 wait_for_stable_page(page);
1041 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1042 if (ext4_should_dioread_nolock(inode))
1043 ret = ext4_block_write_begin(page, pos, len,
1044 ext4_get_block_write);
1046 ret = ext4_block_write_begin(page, pos, len,
1049 if (ext4_should_dioread_nolock(inode))
1050 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1052 ret = __block_write_begin(page, pos, len, ext4_get_block);
1054 if (!ret && ext4_should_journal_data(inode)) {
1055 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1057 do_journal_get_write_access);
1063 * __block_write_begin may have instantiated a few blocks
1064 * outside i_size. Trim these off again. Don't need
1065 * i_size_read because we hold i_mutex.
1067 * Add inode to orphan list in case we crash before
1070 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1071 ext4_orphan_add(handle, inode);
1073 ext4_journal_stop(handle);
1074 if (pos + len > inode->i_size) {
1075 ext4_truncate_failed_write(inode);
1077 * If truncate failed early the inode might
1078 * still be on the orphan list; we need to
1079 * make sure the inode is removed from the
1080 * orphan list in that case.
1083 ext4_orphan_del(NULL, inode);
1086 if (ret == -ENOSPC &&
1087 ext4_should_retry_alloc(inode->i_sb, &retries))
1089 page_cache_release(page);
1096 /* For write_end() in data=journal mode */
1097 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1100 if (!buffer_mapped(bh) || buffer_freed(bh))
1102 set_buffer_uptodate(bh);
1103 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1104 clear_buffer_meta(bh);
1105 clear_buffer_prio(bh);
1110 * We need to pick up the new inode size which generic_commit_write gave us
1111 * `file' can be NULL - eg, when called from page_symlink().
1113 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1114 * buffers are managed internally.
1116 static int ext4_write_end(struct file *file,
1117 struct address_space *mapping,
1118 loff_t pos, unsigned len, unsigned copied,
1119 struct page *page, void *fsdata)
1121 handle_t *handle = ext4_journal_current_handle();
1122 struct inode *inode = mapping->host;
1123 loff_t old_size = inode->i_size;
1125 int i_size_changed = 0;
1127 trace_ext4_write_end(inode, pos, len, copied);
1128 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1129 ret = ext4_jbd2_file_inode(handle, inode);
1132 page_cache_release(page);
1137 if (ext4_has_inline_data(inode)) {
1138 ret = ext4_write_inline_data_end(inode, pos, len,
1144 copied = block_write_end(file, mapping, pos,
1145 len, copied, page, fsdata);
1147 * it's important to update i_size while still holding page lock:
1148 * page writeout could otherwise come in and zero beyond i_size.
1150 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1152 page_cache_release(page);
1155 pagecache_isize_extended(inode, old_size, pos);
1157 * Don't mark the inode dirty under page lock. First, it unnecessarily
1158 * makes the holding time of page lock longer. Second, it forces lock
1159 * ordering of page lock and transaction start for journaling
1163 ext4_mark_inode_dirty(handle, inode);
1165 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1166 /* if we have allocated more blocks and copied
1167 * less. We will have blocks allocated outside
1168 * inode->i_size. So truncate them
1170 ext4_orphan_add(handle, inode);
1172 ret2 = ext4_journal_stop(handle);
1176 if (pos + len > inode->i_size) {
1177 ext4_truncate_failed_write(inode);
1179 * If truncate failed early the inode might still be
1180 * on the orphan list; we need to make sure the inode
1181 * is removed from the orphan list in that case.
1184 ext4_orphan_del(NULL, inode);
1187 return ret ? ret : copied;
1190 static int ext4_journalled_write_end(struct file *file,
1191 struct address_space *mapping,
1192 loff_t pos, unsigned len, unsigned copied,
1193 struct page *page, void *fsdata)
1195 handle_t *handle = ext4_journal_current_handle();
1196 struct inode *inode = mapping->host;
1197 loff_t old_size = inode->i_size;
1201 int size_changed = 0;
1203 trace_ext4_journalled_write_end(inode, pos, len, copied);
1204 from = pos & (PAGE_CACHE_SIZE - 1);
1207 BUG_ON(!ext4_handle_valid(handle));
1209 if (ext4_has_inline_data(inode))
1210 copied = ext4_write_inline_data_end(inode, pos, len,
1214 if (!PageUptodate(page))
1216 page_zero_new_buffers(page, from+copied, to);
1219 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1220 to, &partial, write_end_fn);
1222 SetPageUptodate(page);
1224 size_changed = ext4_update_inode_size(inode, pos + copied);
1225 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1226 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1228 page_cache_release(page);
1231 pagecache_isize_extended(inode, old_size, pos);
1234 ret2 = ext4_mark_inode_dirty(handle, inode);
1239 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1240 /* if we have allocated more blocks and copied
1241 * less. We will have blocks allocated outside
1242 * inode->i_size. So truncate them
1244 ext4_orphan_add(handle, inode);
1246 ret2 = ext4_journal_stop(handle);
1249 if (pos + len > inode->i_size) {
1250 ext4_truncate_failed_write(inode);
1252 * If truncate failed early the inode might still be
1253 * on the orphan list; we need to make sure the inode
1254 * is removed from the orphan list in that case.
1257 ext4_orphan_del(NULL, inode);
1260 return ret ? ret : copied;
1264 * Reserve a single cluster located at lblock
1266 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1268 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1269 struct ext4_inode_info *ei = EXT4_I(inode);
1270 unsigned int md_needed;
1274 * We will charge metadata quota at writeout time; this saves
1275 * us from metadata over-estimation, though we may go over by
1276 * a small amount in the end. Here we just reserve for data.
1278 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1283 * recalculate the amount of metadata blocks to reserve
1284 * in order to allocate nrblocks
1285 * worse case is one extent per block
1287 spin_lock(&ei->i_block_reservation_lock);
1289 * ext4_calc_metadata_amount() has side effects, which we have
1290 * to be prepared undo if we fail to claim space.
1293 trace_ext4_da_reserve_space(inode, 0);
1295 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1296 spin_unlock(&ei->i_block_reservation_lock);
1297 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1300 ei->i_reserved_data_blocks++;
1301 spin_unlock(&ei->i_block_reservation_lock);
1303 return 0; /* success */
1306 static void ext4_da_release_space(struct inode *inode, int to_free)
1308 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1309 struct ext4_inode_info *ei = EXT4_I(inode);
1312 return; /* Nothing to release, exit */
1314 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1316 trace_ext4_da_release_space(inode, to_free);
1317 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1319 * if there aren't enough reserved blocks, then the
1320 * counter is messed up somewhere. Since this
1321 * function is called from invalidate page, it's
1322 * harmless to return without any action.
1324 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1325 "ino %lu, to_free %d with only %d reserved "
1326 "data blocks", inode->i_ino, to_free,
1327 ei->i_reserved_data_blocks);
1329 to_free = ei->i_reserved_data_blocks;
1331 ei->i_reserved_data_blocks -= to_free;
1333 /* update fs dirty data blocks counter */
1334 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1336 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1338 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1341 static void ext4_da_page_release_reservation(struct page *page,
1342 unsigned int offset,
1343 unsigned int length)
1345 int to_release = 0, contiguous_blks = 0;
1346 struct buffer_head *head, *bh;
1347 unsigned int curr_off = 0;
1348 struct inode *inode = page->mapping->host;
1349 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1350 unsigned int stop = offset + length;
1354 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1356 head = page_buffers(page);
1359 unsigned int next_off = curr_off + bh->b_size;
1361 if (next_off > stop)
1364 if ((offset <= curr_off) && (buffer_delay(bh))) {
1367 clear_buffer_delay(bh);
1368 } else if (contiguous_blks) {
1369 lblk = page->index <<
1370 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1371 lblk += (curr_off >> inode->i_blkbits) -
1373 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1374 contiguous_blks = 0;
1376 curr_off = next_off;
1377 } while ((bh = bh->b_this_page) != head);
1379 if (contiguous_blks) {
1380 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1381 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1382 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1385 /* If we have released all the blocks belonging to a cluster, then we
1386 * need to release the reserved space for that cluster. */
1387 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1388 while (num_clusters > 0) {
1389 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1390 ((num_clusters - 1) << sbi->s_cluster_bits);
1391 if (sbi->s_cluster_ratio == 1 ||
1392 !ext4_find_delalloc_cluster(inode, lblk))
1393 ext4_da_release_space(inode, 1);
1400 * Delayed allocation stuff
1403 struct mpage_da_data {
1404 struct inode *inode;
1405 struct writeback_control *wbc;
1407 pgoff_t first_page; /* The first page to write */
1408 pgoff_t next_page; /* Current page to examine */
1409 pgoff_t last_page; /* Last page to examine */
1411 * Extent to map - this can be after first_page because that can be
1412 * fully mapped. We somewhat abuse m_flags to store whether the extent
1413 * is delalloc or unwritten.
1415 struct ext4_map_blocks map;
1416 struct ext4_io_submit io_submit; /* IO submission data */
1419 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1424 struct pagevec pvec;
1425 struct inode *inode = mpd->inode;
1426 struct address_space *mapping = inode->i_mapping;
1428 /* This is necessary when next_page == 0. */
1429 if (mpd->first_page >= mpd->next_page)
1432 index = mpd->first_page;
1433 end = mpd->next_page - 1;
1435 ext4_lblk_t start, last;
1436 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1437 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1438 ext4_es_remove_extent(inode, start, last - start + 1);
1441 pagevec_init(&pvec, 0);
1442 while (index <= end) {
1443 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1446 for (i = 0; i < nr_pages; i++) {
1447 struct page *page = pvec.pages[i];
1448 if (page->index > end)
1450 BUG_ON(!PageLocked(page));
1451 BUG_ON(PageWriteback(page));
1453 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1454 ClearPageUptodate(page);
1458 index = pvec.pages[nr_pages - 1]->index + 1;
1459 pagevec_release(&pvec);
1463 static void ext4_print_free_blocks(struct inode *inode)
1465 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1466 struct super_block *sb = inode->i_sb;
1467 struct ext4_inode_info *ei = EXT4_I(inode);
1469 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1470 EXT4_C2B(EXT4_SB(inode->i_sb),
1471 ext4_count_free_clusters(sb)));
1472 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1473 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1474 (long long) EXT4_C2B(EXT4_SB(sb),
1475 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1476 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1477 (long long) EXT4_C2B(EXT4_SB(sb),
1478 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1479 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1480 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1481 ei->i_reserved_data_blocks);
1485 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1487 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1491 * This function is grabs code from the very beginning of
1492 * ext4_map_blocks, but assumes that the caller is from delayed write
1493 * time. This function looks up the requested blocks and sets the
1494 * buffer delay bit under the protection of i_data_sem.
1496 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1497 struct ext4_map_blocks *map,
1498 struct buffer_head *bh)
1500 struct extent_status es;
1502 sector_t invalid_block = ~((sector_t) 0xffff);
1503 #ifdef ES_AGGRESSIVE_TEST
1504 struct ext4_map_blocks orig_map;
1506 memcpy(&orig_map, map, sizeof(*map));
1509 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1513 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1514 "logical block %lu\n", inode->i_ino, map->m_len,
1515 (unsigned long) map->m_lblk);
1517 /* Lookup extent status tree firstly */
1518 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1519 if (ext4_es_is_hole(&es)) {
1521 down_read(&EXT4_I(inode)->i_data_sem);
1526 * Delayed extent could be allocated by fallocate.
1527 * So we need to check it.
1529 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1530 map_bh(bh, inode->i_sb, invalid_block);
1532 set_buffer_delay(bh);
1536 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1537 retval = es.es_len - (iblock - es.es_lblk);
1538 if (retval > map->m_len)
1539 retval = map->m_len;
1540 map->m_len = retval;
1541 if (ext4_es_is_written(&es))
1542 map->m_flags |= EXT4_MAP_MAPPED;
1543 else if (ext4_es_is_unwritten(&es))
1544 map->m_flags |= EXT4_MAP_UNWRITTEN;
1548 #ifdef ES_AGGRESSIVE_TEST
1549 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1555 * Try to see if we can get the block without requesting a new
1556 * file system block.
1558 down_read(&EXT4_I(inode)->i_data_sem);
1559 if (ext4_has_inline_data(inode))
1561 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1562 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1564 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1570 * XXX: __block_prepare_write() unmaps passed block,
1574 * If the block was allocated from previously allocated cluster,
1575 * then we don't need to reserve it again. However we still need
1576 * to reserve metadata for every block we're going to write.
1578 if (EXT4_SB(inode->i_sb)->s_cluster_ratio <= 1 ||
1579 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1580 ret = ext4_da_reserve_space(inode, iblock);
1582 /* not enough space to reserve */
1588 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1589 ~0, EXTENT_STATUS_DELAYED);
1595 map_bh(bh, inode->i_sb, invalid_block);
1597 set_buffer_delay(bh);
1598 } else if (retval > 0) {
1600 unsigned int status;
1602 if (unlikely(retval != map->m_len)) {
1603 ext4_warning(inode->i_sb,
1604 "ES len assertion failed for inode "
1605 "%lu: retval %d != map->m_len %d",
1606 inode->i_ino, retval, map->m_len);
1610 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1611 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1612 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1613 map->m_pblk, status);
1619 up_read((&EXT4_I(inode)->i_data_sem));
1625 * This is a special get_block_t callback which is used by
1626 * ext4_da_write_begin(). It will either return mapped block or
1627 * reserve space for a single block.
1629 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1630 * We also have b_blocknr = -1 and b_bdev initialized properly
1632 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1633 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1634 * initialized properly.
1636 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1637 struct buffer_head *bh, int create)
1639 struct ext4_map_blocks map;
1642 BUG_ON(create == 0);
1643 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1645 map.m_lblk = iblock;
1649 * first, we need to know whether the block is allocated already
1650 * preallocated blocks are unmapped but should treated
1651 * the same as allocated blocks.
1653 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1657 map_bh(bh, inode->i_sb, map.m_pblk);
1658 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1660 if (buffer_unwritten(bh)) {
1661 /* A delayed write to unwritten bh should be marked
1662 * new and mapped. Mapped ensures that we don't do
1663 * get_block multiple times when we write to the same
1664 * offset and new ensures that we do proper zero out
1665 * for partial write.
1668 set_buffer_mapped(bh);
1673 static int bget_one(handle_t *handle, struct buffer_head *bh)
1679 static int bput_one(handle_t *handle, struct buffer_head *bh)
1685 static int __ext4_journalled_writepage(struct page *page,
1688 struct address_space *mapping = page->mapping;
1689 struct inode *inode = mapping->host;
1690 struct buffer_head *page_bufs = NULL;
1691 handle_t *handle = NULL;
1692 int ret = 0, err = 0;
1693 int inline_data = ext4_has_inline_data(inode);
1694 struct buffer_head *inode_bh = NULL;
1696 ClearPageChecked(page);
1699 BUG_ON(page->index != 0);
1700 BUG_ON(len > ext4_get_max_inline_size(inode));
1701 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1702 if (inode_bh == NULL)
1705 page_bufs = page_buffers(page);
1710 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1714 * We need to release the page lock before we start the
1715 * journal, so grab a reference so the page won't disappear
1716 * out from under us.
1721 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1722 ext4_writepage_trans_blocks(inode));
1723 if (IS_ERR(handle)) {
1724 ret = PTR_ERR(handle);
1726 goto out_no_pagelock;
1728 BUG_ON(!ext4_handle_valid(handle));
1732 if (page->mapping != mapping) {
1733 /* The page got truncated from under us */
1734 ext4_journal_stop(handle);
1740 BUFFER_TRACE(inode_bh, "get write access");
1741 ret = ext4_journal_get_write_access(handle, inode_bh);
1743 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1746 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1747 do_journal_get_write_access);
1749 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1754 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1755 err = ext4_journal_stop(handle);
1759 if (!ext4_has_inline_data(inode))
1760 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1762 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1771 * Note that we don't need to start a transaction unless we're journaling data
1772 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1773 * need to file the inode to the transaction's list in ordered mode because if
1774 * we are writing back data added by write(), the inode is already there and if
1775 * we are writing back data modified via mmap(), no one guarantees in which
1776 * transaction the data will hit the disk. In case we are journaling data, we
1777 * cannot start transaction directly because transaction start ranks above page
1778 * lock so we have to do some magic.
1780 * This function can get called via...
1781 * - ext4_writepages after taking page lock (have journal handle)
1782 * - journal_submit_inode_data_buffers (no journal handle)
1783 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1784 * - grab_page_cache when doing write_begin (have journal handle)
1786 * We don't do any block allocation in this function. If we have page with
1787 * multiple blocks we need to write those buffer_heads that are mapped. This
1788 * is important for mmaped based write. So if we do with blocksize 1K
1789 * truncate(f, 1024);
1790 * a = mmap(f, 0, 4096);
1792 * truncate(f, 4096);
1793 * we have in the page first buffer_head mapped via page_mkwrite call back
1794 * but other buffer_heads would be unmapped but dirty (dirty done via the
1795 * do_wp_page). So writepage should write the first block. If we modify
1796 * the mmap area beyond 1024 we will again get a page_fault and the
1797 * page_mkwrite callback will do the block allocation and mark the
1798 * buffer_heads mapped.
1800 * We redirty the page if we have any buffer_heads that is either delay or
1801 * unwritten in the page.
1803 * We can get recursively called as show below.
1805 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1808 * But since we don't do any block allocation we should not deadlock.
1809 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1811 static int ext4_writepage(struct page *page,
1812 struct writeback_control *wbc)
1817 struct buffer_head *page_bufs = NULL;
1818 struct inode *inode = page->mapping->host;
1819 struct ext4_io_submit io_submit;
1820 bool keep_towrite = false;
1822 trace_ext4_writepage(page);
1823 size = i_size_read(inode);
1824 if (page->index == size >> PAGE_CACHE_SHIFT)
1825 len = size & ~PAGE_CACHE_MASK;
1827 len = PAGE_CACHE_SIZE;
1829 page_bufs = page_buffers(page);
1831 * We cannot do block allocation or other extent handling in this
1832 * function. If there are buffers needing that, we have to redirty
1833 * the page. But we may reach here when we do a journal commit via
1834 * journal_submit_inode_data_buffers() and in that case we must write
1835 * allocated buffers to achieve data=ordered mode guarantees.
1837 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1838 ext4_bh_delay_or_unwritten)) {
1839 redirty_page_for_writepage(wbc, page);
1840 if (current->flags & PF_MEMALLOC) {
1842 * For memory cleaning there's no point in writing only
1843 * some buffers. So just bail out. Warn if we came here
1844 * from direct reclaim.
1846 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1851 keep_towrite = true;
1854 if (PageChecked(page) && ext4_should_journal_data(inode))
1856 * It's mmapped pagecache. Add buffers and journal it. There
1857 * doesn't seem much point in redirtying the page here.
1859 return __ext4_journalled_writepage(page, len);
1861 ext4_io_submit_init(&io_submit, wbc);
1862 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1863 if (!io_submit.io_end) {
1864 redirty_page_for_writepage(wbc, page);
1868 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1869 ext4_io_submit(&io_submit);
1870 /* Drop io_end reference we got from init */
1871 ext4_put_io_end_defer(io_submit.io_end);
1875 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1878 loff_t size = i_size_read(mpd->inode);
1881 BUG_ON(page->index != mpd->first_page);
1882 if (page->index == size >> PAGE_CACHE_SHIFT)
1883 len = size & ~PAGE_CACHE_MASK;
1885 len = PAGE_CACHE_SIZE;
1886 clear_page_dirty_for_io(page);
1887 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1889 mpd->wbc->nr_to_write--;
1895 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1898 * mballoc gives us at most this number of blocks...
1899 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1900 * The rest of mballoc seems to handle chunks up to full group size.
1902 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1905 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1907 * @mpd - extent of blocks
1908 * @lblk - logical number of the block in the file
1909 * @bh - buffer head we want to add to the extent
1911 * The function is used to collect contig. blocks in the same state. If the
1912 * buffer doesn't require mapping for writeback and we haven't started the
1913 * extent of buffers to map yet, the function returns 'true' immediately - the
1914 * caller can write the buffer right away. Otherwise the function returns true
1915 * if the block has been added to the extent, false if the block couldn't be
1918 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1919 struct buffer_head *bh)
1921 struct ext4_map_blocks *map = &mpd->map;
1923 /* Buffer that doesn't need mapping for writeback? */
1924 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1925 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1926 /* So far no extent to map => we write the buffer right away */
1927 if (map->m_len == 0)
1932 /* First block in the extent? */
1933 if (map->m_len == 0) {
1936 map->m_flags = bh->b_state & BH_FLAGS;
1940 /* Don't go larger than mballoc is willing to allocate */
1941 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1944 /* Can we merge the block to our big extent? */
1945 if (lblk == map->m_lblk + map->m_len &&
1946 (bh->b_state & BH_FLAGS) == map->m_flags) {
1954 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1956 * @mpd - extent of blocks for mapping
1957 * @head - the first buffer in the page
1958 * @bh - buffer we should start processing from
1959 * @lblk - logical number of the block in the file corresponding to @bh
1961 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1962 * the page for IO if all buffers in this page were mapped and there's no
1963 * accumulated extent of buffers to map or add buffers in the page to the
1964 * extent of buffers to map. The function returns 1 if the caller can continue
1965 * by processing the next page, 0 if it should stop adding buffers to the
1966 * extent to map because we cannot extend it anymore. It can also return value
1967 * < 0 in case of error during IO submission.
1969 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1970 struct buffer_head *head,
1971 struct buffer_head *bh,
1974 struct inode *inode = mpd->inode;
1976 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1977 >> inode->i_blkbits;
1980 BUG_ON(buffer_locked(bh));
1982 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1983 /* Found extent to map? */
1986 /* Everything mapped so far and we hit EOF */
1989 } while (lblk++, (bh = bh->b_this_page) != head);
1990 /* So far everything mapped? Submit the page for IO. */
1991 if (mpd->map.m_len == 0) {
1992 err = mpage_submit_page(mpd, head->b_page);
1996 return lblk < blocks;
2000 * mpage_map_buffers - update buffers corresponding to changed extent and
2001 * submit fully mapped pages for IO
2003 * @mpd - description of extent to map, on return next extent to map
2005 * Scan buffers corresponding to changed extent (we expect corresponding pages
2006 * to be already locked) and update buffer state according to new extent state.
2007 * We map delalloc buffers to their physical location, clear unwritten bits,
2008 * and mark buffers as uninit when we perform writes to unwritten extents
2009 * and do extent conversion after IO is finished. If the last page is not fully
2010 * mapped, we update @map to the next extent in the last page that needs
2011 * mapping. Otherwise we submit the page for IO.
2013 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2015 struct pagevec pvec;
2017 struct inode *inode = mpd->inode;
2018 struct buffer_head *head, *bh;
2019 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2025 start = mpd->map.m_lblk >> bpp_bits;
2026 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2027 lblk = start << bpp_bits;
2028 pblock = mpd->map.m_pblk;
2030 pagevec_init(&pvec, 0);
2031 while (start <= end) {
2032 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2036 for (i = 0; i < nr_pages; i++) {
2037 struct page *page = pvec.pages[i];
2039 if (page->index > end)
2041 /* Up to 'end' pages must be contiguous */
2042 BUG_ON(page->index != start);
2043 bh = head = page_buffers(page);
2045 if (lblk < mpd->map.m_lblk)
2047 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2049 * Buffer after end of mapped extent.
2050 * Find next buffer in the page to map.
2053 mpd->map.m_flags = 0;
2055 * FIXME: If dioread_nolock supports
2056 * blocksize < pagesize, we need to make
2057 * sure we add size mapped so far to
2058 * io_end->size as the following call
2059 * can submit the page for IO.
2061 err = mpage_process_page_bufs(mpd, head,
2063 pagevec_release(&pvec);
2068 if (buffer_delay(bh)) {
2069 clear_buffer_delay(bh);
2070 bh->b_blocknr = pblock++;
2072 clear_buffer_unwritten(bh);
2073 } while (lblk++, (bh = bh->b_this_page) != head);
2076 * FIXME: This is going to break if dioread_nolock
2077 * supports blocksize < pagesize as we will try to
2078 * convert potentially unmapped parts of inode.
2080 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2081 /* Page fully mapped - let IO run! */
2082 err = mpage_submit_page(mpd, page);
2084 pagevec_release(&pvec);
2089 pagevec_release(&pvec);
2091 /* Extent fully mapped and matches with page boundary. We are done. */
2093 mpd->map.m_flags = 0;
2097 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2099 struct inode *inode = mpd->inode;
2100 struct ext4_map_blocks *map = &mpd->map;
2101 int get_blocks_flags;
2102 int err, dioread_nolock;
2104 trace_ext4_da_write_pages_extent(inode, map);
2106 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2107 * to convert an unwritten extent to be initialized (in the case
2108 * where we have written into one or more preallocated blocks). It is
2109 * possible that we're going to need more metadata blocks than
2110 * previously reserved. However we must not fail because we're in
2111 * writeback and there is nothing we can do about it so it might result
2112 * in data loss. So use reserved blocks to allocate metadata if
2115 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2116 * the blocks in question are delalloc blocks. This indicates
2117 * that the blocks and quotas has already been checked when
2118 * the data was copied into the page cache.
2120 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2121 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2122 dioread_nolock = ext4_should_dioread_nolock(inode);
2124 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2125 if (map->m_flags & (1 << BH_Delay))
2126 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2128 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2131 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2132 if (!mpd->io_submit.io_end->handle &&
2133 ext4_handle_valid(handle)) {
2134 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2135 handle->h_rsv_handle = NULL;
2137 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2140 BUG_ON(map->m_len == 0);
2141 if (map->m_flags & EXT4_MAP_NEW) {
2142 struct block_device *bdev = inode->i_sb->s_bdev;
2145 for (i = 0; i < map->m_len; i++)
2146 unmap_underlying_metadata(bdev, map->m_pblk + i);
2152 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2153 * mpd->len and submit pages underlying it for IO
2155 * @handle - handle for journal operations
2156 * @mpd - extent to map
2157 * @give_up_on_write - we set this to true iff there is a fatal error and there
2158 * is no hope of writing the data. The caller should discard
2159 * dirty pages to avoid infinite loops.
2161 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2162 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2163 * them to initialized or split the described range from larger unwritten
2164 * extent. Note that we need not map all the described range since allocation
2165 * can return less blocks or the range is covered by more unwritten extents. We
2166 * cannot map more because we are limited by reserved transaction credits. On
2167 * the other hand we always make sure that the last touched page is fully
2168 * mapped so that it can be written out (and thus forward progress is
2169 * guaranteed). After mapping we submit all mapped pages for IO.
2171 static int mpage_map_and_submit_extent(handle_t *handle,
2172 struct mpage_da_data *mpd,
2173 bool *give_up_on_write)
2175 struct inode *inode = mpd->inode;
2176 struct ext4_map_blocks *map = &mpd->map;
2181 mpd->io_submit.io_end->offset =
2182 ((loff_t)map->m_lblk) << inode->i_blkbits;
2184 err = mpage_map_one_extent(handle, mpd);
2186 struct super_block *sb = inode->i_sb;
2188 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2189 goto invalidate_dirty_pages;
2191 * Let the uper layers retry transient errors.
2192 * In the case of ENOSPC, if ext4_count_free_blocks()
2193 * is non-zero, a commit should free up blocks.
2195 if ((err == -ENOMEM) ||
2196 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2198 goto update_disksize;
2201 ext4_msg(sb, KERN_CRIT,
2202 "Delayed block allocation failed for "
2203 "inode %lu at logical offset %llu with"
2204 " max blocks %u with error %d",
2206 (unsigned long long)map->m_lblk,
2207 (unsigned)map->m_len, -err);
2208 ext4_msg(sb, KERN_CRIT,
2209 "This should not happen!! Data will "
2212 ext4_print_free_blocks(inode);
2213 invalidate_dirty_pages:
2214 *give_up_on_write = true;
2219 * Update buffer state, submit mapped pages, and get us new
2222 err = mpage_map_and_submit_buffers(mpd);
2224 goto update_disksize;
2225 } while (map->m_len);
2229 * Update on-disk size after IO is submitted. Races with
2230 * truncate are avoided by checking i_size under i_data_sem.
2232 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2233 if (disksize > EXT4_I(inode)->i_disksize) {
2237 down_write(&EXT4_I(inode)->i_data_sem);
2238 i_size = i_size_read(inode);
2239 if (disksize > i_size)
2241 if (disksize > EXT4_I(inode)->i_disksize)
2242 EXT4_I(inode)->i_disksize = disksize;
2243 err2 = ext4_mark_inode_dirty(handle, inode);
2244 up_write(&EXT4_I(inode)->i_data_sem);
2246 ext4_error(inode->i_sb,
2247 "Failed to mark inode %lu dirty",
2256 * Calculate the total number of credits to reserve for one writepages
2257 * iteration. This is called from ext4_writepages(). We map an extent of
2258 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2259 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2260 * bpp - 1 blocks in bpp different extents.
2262 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2264 int bpp = ext4_journal_blocks_per_page(inode);
2266 return ext4_meta_trans_blocks(inode,
2267 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2271 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2272 * and underlying extent to map
2274 * @mpd - where to look for pages
2276 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2277 * IO immediately. When we find a page which isn't mapped we start accumulating
2278 * extent of buffers underlying these pages that needs mapping (formed by
2279 * either delayed or unwritten buffers). We also lock the pages containing
2280 * these buffers. The extent found is returned in @mpd structure (starting at
2281 * mpd->lblk with length mpd->len blocks).
2283 * Note that this function can attach bios to one io_end structure which are
2284 * neither logically nor physically contiguous. Although it may seem as an
2285 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2286 * case as we need to track IO to all buffers underlying a page in one io_end.
2288 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2290 struct address_space *mapping = mpd->inode->i_mapping;
2291 struct pagevec pvec;
2292 unsigned int nr_pages;
2293 long left = mpd->wbc->nr_to_write;
2294 pgoff_t index = mpd->first_page;
2295 pgoff_t end = mpd->last_page;
2298 int blkbits = mpd->inode->i_blkbits;
2300 struct buffer_head *head;
2302 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2303 tag = PAGECACHE_TAG_TOWRITE;
2305 tag = PAGECACHE_TAG_DIRTY;
2307 pagevec_init(&pvec, 0);
2309 mpd->next_page = index;
2310 while (index <= end) {
2311 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2312 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2316 for (i = 0; i < nr_pages; i++) {
2317 struct page *page = pvec.pages[i];
2320 * At this point, the page may be truncated or
2321 * invalidated (changing page->mapping to NULL), or
2322 * even swizzled back from swapper_space to tmpfs file
2323 * mapping. However, page->index will not change
2324 * because we have a reference on the page.
2326 if (page->index > end)
2330 * Accumulated enough dirty pages? This doesn't apply
2331 * to WB_SYNC_ALL mode. For integrity sync we have to
2332 * keep going because someone may be concurrently
2333 * dirtying pages, and we might have synced a lot of
2334 * newly appeared dirty pages, but have not synced all
2335 * of the old dirty pages.
2337 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2340 /* If we can't merge this page, we are done. */
2341 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2346 * If the page is no longer dirty, or its mapping no
2347 * longer corresponds to inode we are writing (which
2348 * means it has been truncated or invalidated), or the
2349 * page is already under writeback and we are not doing
2350 * a data integrity writeback, skip the page
2352 if (!PageDirty(page) ||
2353 (PageWriteback(page) &&
2354 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2355 unlikely(page->mapping != mapping)) {
2360 wait_on_page_writeback(page);
2361 BUG_ON(PageWriteback(page));
2363 if (mpd->map.m_len == 0)
2364 mpd->first_page = page->index;
2365 mpd->next_page = page->index + 1;
2366 /* Add all dirty buffers to mpd */
2367 lblk = ((ext4_lblk_t)page->index) <<
2368 (PAGE_CACHE_SHIFT - blkbits);
2369 head = page_buffers(page);
2370 err = mpage_process_page_bufs(mpd, head, head, lblk);
2376 pagevec_release(&pvec);
2381 pagevec_release(&pvec);
2385 static int __writepage(struct page *page, struct writeback_control *wbc,
2388 struct address_space *mapping = data;
2389 int ret = ext4_writepage(page, wbc);
2390 mapping_set_error(mapping, ret);
2394 static int ext4_writepages(struct address_space *mapping,
2395 struct writeback_control *wbc)
2397 pgoff_t writeback_index = 0;
2398 long nr_to_write = wbc->nr_to_write;
2399 int range_whole = 0;
2401 handle_t *handle = NULL;
2402 struct mpage_da_data mpd;
2403 struct inode *inode = mapping->host;
2404 int needed_blocks, rsv_blocks = 0, ret = 0;
2405 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2407 struct blk_plug plug;
2408 bool give_up_on_write = false;
2410 trace_ext4_writepages(inode, wbc);
2413 * No pages to write? This is mainly a kludge to avoid starting
2414 * a transaction for special inodes like journal inode on last iput()
2415 * because that could violate lock ordering on umount
2417 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2418 goto out_writepages;
2420 if (ext4_should_journal_data(inode)) {
2421 struct blk_plug plug;
2423 blk_start_plug(&plug);
2424 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2425 blk_finish_plug(&plug);
2426 goto out_writepages;
2430 * If the filesystem has aborted, it is read-only, so return
2431 * right away instead of dumping stack traces later on that
2432 * will obscure the real source of the problem. We test
2433 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2434 * the latter could be true if the filesystem is mounted
2435 * read-only, and in that case, ext4_writepages should
2436 * *never* be called, so if that ever happens, we would want
2439 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2441 goto out_writepages;
2444 if (ext4_should_dioread_nolock(inode)) {
2446 * We may need to convert up to one extent per block in
2447 * the page and we may dirty the inode.
2449 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2453 * If we have inline data and arrive here, it means that
2454 * we will soon create the block for the 1st page, so
2455 * we'd better clear the inline data here.
2457 if (ext4_has_inline_data(inode)) {
2458 /* Just inode will be modified... */
2459 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2460 if (IS_ERR(handle)) {
2461 ret = PTR_ERR(handle);
2462 goto out_writepages;
2464 BUG_ON(ext4_test_inode_state(inode,
2465 EXT4_STATE_MAY_INLINE_DATA));
2466 ext4_destroy_inline_data(handle, inode);
2467 ext4_journal_stop(handle);
2470 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2473 if (wbc->range_cyclic) {
2474 writeback_index = mapping->writeback_index;
2475 if (writeback_index)
2477 mpd.first_page = writeback_index;
2480 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2481 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2486 ext4_io_submit_init(&mpd.io_submit, wbc);
2488 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2489 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2491 blk_start_plug(&plug);
2492 while (!done && mpd.first_page <= mpd.last_page) {
2493 /* For each extent of pages we use new io_end */
2494 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2495 if (!mpd.io_submit.io_end) {
2501 * We have two constraints: We find one extent to map and we
2502 * must always write out whole page (makes a difference when
2503 * blocksize < pagesize) so that we don't block on IO when we
2504 * try to write out the rest of the page. Journalled mode is
2505 * not supported by delalloc.
2507 BUG_ON(ext4_should_journal_data(inode));
2508 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2510 /* start a new transaction */
2511 handle = ext4_journal_start_with_reserve(inode,
2512 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2513 if (IS_ERR(handle)) {
2514 ret = PTR_ERR(handle);
2515 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2516 "%ld pages, ino %lu; err %d", __func__,
2517 wbc->nr_to_write, inode->i_ino, ret);
2518 /* Release allocated io_end */
2519 ext4_put_io_end(mpd.io_submit.io_end);
2523 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2524 ret = mpage_prepare_extent_to_map(&mpd);
2527 ret = mpage_map_and_submit_extent(handle, &mpd,
2531 * We scanned the whole range (or exhausted
2532 * nr_to_write), submitted what was mapped and
2533 * didn't find anything needing mapping. We are
2539 ext4_journal_stop(handle);
2540 /* Submit prepared bio */
2541 ext4_io_submit(&mpd.io_submit);
2542 /* Unlock pages we didn't use */
2543 mpage_release_unused_pages(&mpd, give_up_on_write);
2544 /* Drop our io_end reference we got from init */
2545 ext4_put_io_end(mpd.io_submit.io_end);
2547 if (ret == -ENOSPC && sbi->s_journal) {
2549 * Commit the transaction which would
2550 * free blocks released in the transaction
2553 jbd2_journal_force_commit_nested(sbi->s_journal);
2557 /* Fatal error - ENOMEM, EIO... */
2561 blk_finish_plug(&plug);
2562 if (!ret && !cycled && wbc->nr_to_write > 0) {
2564 mpd.last_page = writeback_index - 1;
2570 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2572 * Set the writeback_index so that range_cyclic
2573 * mode will write it back later
2575 mapping->writeback_index = mpd.first_page;
2578 trace_ext4_writepages_result(inode, wbc, ret,
2579 nr_to_write - wbc->nr_to_write);
2583 static int ext4_nonda_switch(struct super_block *sb)
2585 s64 free_clusters, dirty_clusters;
2586 struct ext4_sb_info *sbi = EXT4_SB(sb);
2589 * switch to non delalloc mode if we are running low
2590 * on free block. The free block accounting via percpu
2591 * counters can get slightly wrong with percpu_counter_batch getting
2592 * accumulated on each CPU without updating global counters
2593 * Delalloc need an accurate free block accounting. So switch
2594 * to non delalloc when we are near to error range.
2597 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2599 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2601 * Start pushing delalloc when 1/2 of free blocks are dirty.
2603 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2604 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2606 if (2 * free_clusters < 3 * dirty_clusters ||
2607 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2609 * free block count is less than 150% of dirty blocks
2610 * or free blocks is less than watermark
2617 /* We always reserve for an inode update; the superblock could be there too */
2618 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2620 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2621 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2624 if (pos + len <= 0x7fffffffULL)
2627 /* We might need to update the superblock to set LARGE_FILE */
2631 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2632 loff_t pos, unsigned len, unsigned flags,
2633 struct page **pagep, void **fsdata)
2635 int ret, retries = 0;
2638 struct inode *inode = mapping->host;
2641 index = pos >> PAGE_CACHE_SHIFT;
2643 if (ext4_nonda_switch(inode->i_sb)) {
2644 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2645 return ext4_write_begin(file, mapping, pos,
2646 len, flags, pagep, fsdata);
2648 *fsdata = (void *)0;
2649 trace_ext4_da_write_begin(inode, pos, len, flags);
2651 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2652 ret = ext4_da_write_inline_data_begin(mapping, inode,
2662 * grab_cache_page_write_begin() can take a long time if the
2663 * system is thrashing due to memory pressure, or if the page
2664 * is being written back. So grab it first before we start
2665 * the transaction handle. This also allows us to allocate
2666 * the page (if needed) without using GFP_NOFS.
2669 page = grab_cache_page_write_begin(mapping, index, flags);
2675 * With delayed allocation, we don't log the i_disksize update
2676 * if there is delayed block allocation. But we still need
2677 * to journalling the i_disksize update if writes to the end
2678 * of file which has an already mapped buffer.
2681 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2682 ext4_da_write_credits(inode, pos, len));
2683 if (IS_ERR(handle)) {
2684 page_cache_release(page);
2685 return PTR_ERR(handle);
2689 if (page->mapping != mapping) {
2690 /* The page got truncated from under us */
2692 page_cache_release(page);
2693 ext4_journal_stop(handle);
2696 /* In case writeback began while the page was unlocked */
2697 wait_for_stable_page(page);
2699 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2700 ret = ext4_block_write_begin(page, pos, len,
2701 ext4_da_get_block_prep);
2703 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2707 ext4_journal_stop(handle);
2709 * block_write_begin may have instantiated a few blocks
2710 * outside i_size. Trim these off again. Don't need
2711 * i_size_read because we hold i_mutex.
2713 if (pos + len > inode->i_size)
2714 ext4_truncate_failed_write(inode);
2716 if (ret == -ENOSPC &&
2717 ext4_should_retry_alloc(inode->i_sb, &retries))
2720 page_cache_release(page);
2729 * Check if we should update i_disksize
2730 * when write to the end of file but not require block allocation
2732 static int ext4_da_should_update_i_disksize(struct page *page,
2733 unsigned long offset)
2735 struct buffer_head *bh;
2736 struct inode *inode = page->mapping->host;
2740 bh = page_buffers(page);
2741 idx = offset >> inode->i_blkbits;
2743 for (i = 0; i < idx; i++)
2744 bh = bh->b_this_page;
2746 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2751 static int ext4_da_write_end(struct file *file,
2752 struct address_space *mapping,
2753 loff_t pos, unsigned len, unsigned copied,
2754 struct page *page, void *fsdata)
2756 struct inode *inode = mapping->host;
2758 handle_t *handle = ext4_journal_current_handle();
2760 unsigned long start, end;
2761 int write_mode = (int)(unsigned long)fsdata;
2763 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2764 return ext4_write_end(file, mapping, pos,
2765 len, copied, page, fsdata);
2767 trace_ext4_da_write_end(inode, pos, len, copied);
2768 start = pos & (PAGE_CACHE_SIZE - 1);
2769 end = start + copied - 1;
2772 * generic_write_end() will run mark_inode_dirty() if i_size
2773 * changes. So let's piggyback the i_disksize mark_inode_dirty
2776 new_i_size = pos + copied;
2777 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2778 if (ext4_has_inline_data(inode) ||
2779 ext4_da_should_update_i_disksize(page, end)) {
2780 ext4_update_i_disksize(inode, new_i_size);
2781 /* We need to mark inode dirty even if
2782 * new_i_size is less that inode->i_size
2783 * bu greater than i_disksize.(hint delalloc)
2785 ext4_mark_inode_dirty(handle, inode);
2789 if (write_mode != CONVERT_INLINE_DATA &&
2790 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2791 ext4_has_inline_data(inode))
2792 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2795 ret2 = generic_write_end(file, mapping, pos, len, copied,
2801 ret2 = ext4_journal_stop(handle);
2805 return ret ? ret : copied;
2808 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2809 unsigned int length)
2812 * Drop reserved blocks
2814 BUG_ON(!PageLocked(page));
2815 if (!page_has_buffers(page))
2818 ext4_da_page_release_reservation(page, offset, length);
2821 ext4_invalidatepage(page, offset, length);
2827 * Force all delayed allocation blocks to be allocated for a given inode.
2829 int ext4_alloc_da_blocks(struct inode *inode)
2831 trace_ext4_alloc_da_blocks(inode);
2833 if (!EXT4_I(inode)->i_reserved_data_blocks)
2837 * We do something simple for now. The filemap_flush() will
2838 * also start triggering a write of the data blocks, which is
2839 * not strictly speaking necessary (and for users of
2840 * laptop_mode, not even desirable). However, to do otherwise
2841 * would require replicating code paths in:
2843 * ext4_writepages() ->
2844 * write_cache_pages() ---> (via passed in callback function)
2845 * __mpage_da_writepage() -->
2846 * mpage_add_bh_to_extent()
2847 * mpage_da_map_blocks()
2849 * The problem is that write_cache_pages(), located in
2850 * mm/page-writeback.c, marks pages clean in preparation for
2851 * doing I/O, which is not desirable if we're not planning on
2854 * We could call write_cache_pages(), and then redirty all of
2855 * the pages by calling redirty_page_for_writepage() but that
2856 * would be ugly in the extreme. So instead we would need to
2857 * replicate parts of the code in the above functions,
2858 * simplifying them because we wouldn't actually intend to
2859 * write out the pages, but rather only collect contiguous
2860 * logical block extents, call the multi-block allocator, and
2861 * then update the buffer heads with the block allocations.
2863 * For now, though, we'll cheat by calling filemap_flush(),
2864 * which will map the blocks, and start the I/O, but not
2865 * actually wait for the I/O to complete.
2867 return filemap_flush(inode->i_mapping);
2871 * bmap() is special. It gets used by applications such as lilo and by
2872 * the swapper to find the on-disk block of a specific piece of data.
2874 * Naturally, this is dangerous if the block concerned is still in the
2875 * journal. If somebody makes a swapfile on an ext4 data-journaling
2876 * filesystem and enables swap, then they may get a nasty shock when the
2877 * data getting swapped to that swapfile suddenly gets overwritten by
2878 * the original zero's written out previously to the journal and
2879 * awaiting writeback in the kernel's buffer cache.
2881 * So, if we see any bmap calls here on a modified, data-journaled file,
2882 * take extra steps to flush any blocks which might be in the cache.
2884 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2886 struct inode *inode = mapping->host;
2891 * We can get here for an inline file via the FIBMAP ioctl
2893 if (ext4_has_inline_data(inode))
2896 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2897 test_opt(inode->i_sb, DELALLOC)) {
2899 * With delalloc we want to sync the file
2900 * so that we can make sure we allocate
2903 filemap_write_and_wait(mapping);
2906 if (EXT4_JOURNAL(inode) &&
2907 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2909 * This is a REALLY heavyweight approach, but the use of
2910 * bmap on dirty files is expected to be extremely rare:
2911 * only if we run lilo or swapon on a freshly made file
2912 * do we expect this to happen.
2914 * (bmap requires CAP_SYS_RAWIO so this does not
2915 * represent an unprivileged user DOS attack --- we'd be
2916 * in trouble if mortal users could trigger this path at
2919 * NB. EXT4_STATE_JDATA is not set on files other than
2920 * regular files. If somebody wants to bmap a directory
2921 * or symlink and gets confused because the buffer
2922 * hasn't yet been flushed to disk, they deserve
2923 * everything they get.
2926 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2927 journal = EXT4_JOURNAL(inode);
2928 jbd2_journal_lock_updates(journal);
2929 err = jbd2_journal_flush(journal);
2930 jbd2_journal_unlock_updates(journal);
2936 return generic_block_bmap(mapping, block, ext4_get_block);
2939 static int ext4_readpage(struct file *file, struct page *page)
2942 struct inode *inode = page->mapping->host;
2944 trace_ext4_readpage(page);
2946 if (ext4_has_inline_data(inode))
2947 ret = ext4_readpage_inline(inode, page);
2950 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
2956 ext4_readpages(struct file *file, struct address_space *mapping,
2957 struct list_head *pages, unsigned nr_pages)
2959 struct inode *inode = mapping->host;
2961 /* If the file has inline data, no need to do readpages. */
2962 if (ext4_has_inline_data(inode))
2965 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
2968 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2969 unsigned int length)
2971 trace_ext4_invalidatepage(page, offset, length);
2973 /* No journalling happens on data buffers when this function is used */
2974 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2976 block_invalidatepage(page, offset, length);
2979 static int __ext4_journalled_invalidatepage(struct page *page,
2980 unsigned int offset,
2981 unsigned int length)
2983 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2985 trace_ext4_journalled_invalidatepage(page, offset, length);
2988 * If it's a full truncate we just forget about the pending dirtying
2990 if (offset == 0 && length == PAGE_CACHE_SIZE)
2991 ClearPageChecked(page);
2993 return jbd2_journal_invalidatepage(journal, page, offset, length);
2996 /* Wrapper for aops... */
2997 static void ext4_journalled_invalidatepage(struct page *page,
2998 unsigned int offset,
2999 unsigned int length)
3001 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3004 static int ext4_releasepage(struct page *page, gfp_t wait)
3006 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3008 trace_ext4_releasepage(page);
3010 /* Page has dirty journalled data -> cannot release */
3011 if (PageChecked(page))
3014 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3016 return try_to_free_buffers(page);
3020 * ext4_get_block used when preparing for a DIO write or buffer write.
3021 * We allocate an uinitialized extent if blocks haven't been allocated.
3022 * The extent will be converted to initialized after the IO is complete.
3024 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3025 struct buffer_head *bh_result, int create)
3027 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3028 inode->i_ino, create);
3029 return _ext4_get_block(inode, iblock, bh_result,
3030 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3033 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3034 struct buffer_head *bh_result, int create)
3036 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3037 inode->i_ino, create);
3038 return _ext4_get_block(inode, iblock, bh_result,
3039 EXT4_GET_BLOCKS_NO_LOCK);
3042 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3043 ssize_t size, void *private)
3045 ext4_io_end_t *io_end = iocb->private;
3047 /* if not async direct IO just return */
3051 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3052 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3053 iocb->private, io_end->inode->i_ino, iocb, offset,
3056 iocb->private = NULL;
3057 io_end->offset = offset;
3058 io_end->size = size;
3059 ext4_put_io_end(io_end);
3063 * For ext4 extent files, ext4 will do direct-io write to holes,
3064 * preallocated extents, and those write extend the file, no need to
3065 * fall back to buffered IO.
3067 * For holes, we fallocate those blocks, mark them as unwritten
3068 * If those blocks were preallocated, we mark sure they are split, but
3069 * still keep the range to write as unwritten.
3071 * The unwritten extents will be converted to written when DIO is completed.
3072 * For async direct IO, since the IO may still pending when return, we
3073 * set up an end_io call back function, which will do the conversion
3074 * when async direct IO completed.
3076 * If the O_DIRECT write will extend the file then add this inode to the
3077 * orphan list. So recovery will truncate it back to the original size
3078 * if the machine crashes during the write.
3081 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3084 struct file *file = iocb->ki_filp;
3085 struct inode *inode = file->f_mapping->host;
3087 size_t count = iov_iter_count(iter);
3089 get_block_t *get_block_func = NULL;
3091 loff_t final_size = offset + count;
3092 ext4_io_end_t *io_end = NULL;
3094 /* Use the old path for reads and writes beyond i_size. */
3095 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3096 return ext4_ind_direct_IO(iocb, iter, offset);
3098 BUG_ON(iocb->private == NULL);
3101 * Make all waiters for direct IO properly wait also for extent
3102 * conversion. This also disallows race between truncate() and
3103 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3105 if (iov_iter_rw(iter) == WRITE)
3106 inode_dio_begin(inode);
3108 /* If we do a overwrite dio, i_mutex locking can be released */
3109 overwrite = *((int *)iocb->private);
3112 down_read(&EXT4_I(inode)->i_data_sem);
3113 mutex_unlock(&inode->i_mutex);
3117 * We could direct write to holes and fallocate.
3119 * Allocated blocks to fill the hole are marked as
3120 * unwritten to prevent parallel buffered read to expose
3121 * the stale data before DIO complete the data IO.
3123 * As to previously fallocated extents, ext4 get_block will
3124 * just simply mark the buffer mapped but still keep the
3125 * extents unwritten.
3127 * For non AIO case, we will convert those unwritten extents
3128 * to written after return back from blockdev_direct_IO.
3130 * For async DIO, the conversion needs to be deferred when the
3131 * IO is completed. The ext4 end_io callback function will be
3132 * called to take care of the conversion work. Here for async
3133 * case, we allocate an io_end structure to hook to the iocb.
3135 iocb->private = NULL;
3136 ext4_inode_aio_set(inode, NULL);
3137 if (!is_sync_kiocb(iocb)) {
3138 io_end = ext4_init_io_end(inode, GFP_NOFS);
3144 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3146 iocb->private = ext4_get_io_end(io_end);
3148 * we save the io structure for current async direct
3149 * IO, so that later ext4_map_blocks() could flag the
3150 * io structure whether there is a unwritten extents
3151 * needs to be converted when IO is completed.
3153 ext4_inode_aio_set(inode, io_end);
3157 get_block_func = ext4_get_block_write_nolock;
3159 get_block_func = ext4_get_block_write;
3160 dio_flags = DIO_LOCKING;
3162 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3163 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3166 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3167 ext4_end_io_dio, dio_flags);
3169 ret = __blockdev_direct_IO(iocb, inode,
3170 inode->i_sb->s_bdev, iter, offset,
3172 ext4_end_io_dio, NULL, dio_flags);
3175 * Put our reference to io_end. This can free the io_end structure e.g.
3176 * in sync IO case or in case of error. It can even perform extent
3177 * conversion if all bios we submitted finished before we got here.
3178 * Note that in that case iocb->private can be already set to NULL
3182 ext4_inode_aio_set(inode, NULL);
3183 ext4_put_io_end(io_end);
3185 * When no IO was submitted ext4_end_io_dio() was not
3186 * called so we have to put iocb's reference.
3188 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3189 WARN_ON(iocb->private != io_end);
3190 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3191 ext4_put_io_end(io_end);
3192 iocb->private = NULL;
3195 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3196 EXT4_STATE_DIO_UNWRITTEN)) {
3199 * for non AIO case, since the IO is already
3200 * completed, we could do the conversion right here
3202 err = ext4_convert_unwritten_extents(NULL, inode,
3206 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3210 if (iov_iter_rw(iter) == WRITE)
3211 inode_dio_end(inode);
3212 /* take i_mutex locking again if we do a ovewrite dio */
3214 up_read(&EXT4_I(inode)->i_data_sem);
3215 mutex_lock(&inode->i_mutex);
3221 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3224 struct file *file = iocb->ki_filp;
3225 struct inode *inode = file->f_mapping->host;
3226 size_t count = iov_iter_count(iter);
3229 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3230 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3235 * If we are doing data journalling we don't support O_DIRECT
3237 if (ext4_should_journal_data(inode))
3240 /* Let buffer I/O handle the inline data case. */
3241 if (ext4_has_inline_data(inode))
3244 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3245 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3246 ret = ext4_ext_direct_IO(iocb, iter, offset);
3248 ret = ext4_ind_direct_IO(iocb, iter, offset);
3249 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3254 * Pages can be marked dirty completely asynchronously from ext4's journalling
3255 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3256 * much here because ->set_page_dirty is called under VFS locks. The page is
3257 * not necessarily locked.
3259 * We cannot just dirty the page and leave attached buffers clean, because the
3260 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3261 * or jbddirty because all the journalling code will explode.
3263 * So what we do is to mark the page "pending dirty" and next time writepage
3264 * is called, propagate that into the buffers appropriately.
3266 static int ext4_journalled_set_page_dirty(struct page *page)
3268 SetPageChecked(page);
3269 return __set_page_dirty_nobuffers(page);
3272 static const struct address_space_operations ext4_aops = {
3273 .readpage = ext4_readpage,
3274 .readpages = ext4_readpages,
3275 .writepage = ext4_writepage,
3276 .writepages = ext4_writepages,
3277 .write_begin = ext4_write_begin,
3278 .write_end = ext4_write_end,
3280 .invalidatepage = ext4_invalidatepage,
3281 .releasepage = ext4_releasepage,
3282 .direct_IO = ext4_direct_IO,
3283 .migratepage = buffer_migrate_page,
3284 .is_partially_uptodate = block_is_partially_uptodate,
3285 .error_remove_page = generic_error_remove_page,
3288 static const struct address_space_operations ext4_journalled_aops = {
3289 .readpage = ext4_readpage,
3290 .readpages = ext4_readpages,
3291 .writepage = ext4_writepage,
3292 .writepages = ext4_writepages,
3293 .write_begin = ext4_write_begin,
3294 .write_end = ext4_journalled_write_end,
3295 .set_page_dirty = ext4_journalled_set_page_dirty,
3297 .invalidatepage = ext4_journalled_invalidatepage,
3298 .releasepage = ext4_releasepage,
3299 .direct_IO = ext4_direct_IO,
3300 .is_partially_uptodate = block_is_partially_uptodate,
3301 .error_remove_page = generic_error_remove_page,
3304 static const struct address_space_operations ext4_da_aops = {
3305 .readpage = ext4_readpage,
3306 .readpages = ext4_readpages,
3307 .writepage = ext4_writepage,
3308 .writepages = ext4_writepages,
3309 .write_begin = ext4_da_write_begin,
3310 .write_end = ext4_da_write_end,
3312 .invalidatepage = ext4_da_invalidatepage,
3313 .releasepage = ext4_releasepage,
3314 .direct_IO = ext4_direct_IO,
3315 .migratepage = buffer_migrate_page,
3316 .is_partially_uptodate = block_is_partially_uptodate,
3317 .error_remove_page = generic_error_remove_page,
3320 void ext4_set_aops(struct inode *inode)
3322 switch (ext4_inode_journal_mode(inode)) {
3323 case EXT4_INODE_ORDERED_DATA_MODE:
3324 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3326 case EXT4_INODE_WRITEBACK_DATA_MODE:
3327 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3329 case EXT4_INODE_JOURNAL_DATA_MODE:
3330 inode->i_mapping->a_ops = &ext4_journalled_aops;
3335 if (test_opt(inode->i_sb, DELALLOC))
3336 inode->i_mapping->a_ops = &ext4_da_aops;
3338 inode->i_mapping->a_ops = &ext4_aops;
3341 static int __ext4_block_zero_page_range(handle_t *handle,
3342 struct address_space *mapping, loff_t from, loff_t length)
3344 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3345 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3346 unsigned blocksize, pos;
3348 struct inode *inode = mapping->host;
3349 struct buffer_head *bh;
3353 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3354 mapping_gfp_mask(mapping) & ~__GFP_FS);
3358 blocksize = inode->i_sb->s_blocksize;
3360 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3362 if (!page_has_buffers(page))
3363 create_empty_buffers(page, blocksize, 0);
3365 /* Find the buffer that contains "offset" */
3366 bh = page_buffers(page);
3368 while (offset >= pos) {
3369 bh = bh->b_this_page;
3373 if (buffer_freed(bh)) {
3374 BUFFER_TRACE(bh, "freed: skip");
3377 if (!buffer_mapped(bh)) {
3378 BUFFER_TRACE(bh, "unmapped");
3379 ext4_get_block(inode, iblock, bh, 0);
3380 /* unmapped? It's a hole - nothing to do */
3381 if (!buffer_mapped(bh)) {
3382 BUFFER_TRACE(bh, "still unmapped");
3387 /* Ok, it's mapped. Make sure it's up-to-date */
3388 if (PageUptodate(page))
3389 set_buffer_uptodate(bh);
3391 if (!buffer_uptodate(bh)) {
3393 ll_rw_block(READ, 1, &bh);
3395 /* Uhhuh. Read error. Complain and punt. */
3396 if (!buffer_uptodate(bh))
3398 if (S_ISREG(inode->i_mode) &&
3399 ext4_encrypted_inode(inode)) {
3400 /* We expect the key to be set. */
3401 BUG_ON(!ext4_has_encryption_key(inode));
3402 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3403 WARN_ON_ONCE(ext4_decrypt_one(inode, page));
3406 if (ext4_should_journal_data(inode)) {
3407 BUFFER_TRACE(bh, "get write access");
3408 err = ext4_journal_get_write_access(handle, bh);
3412 zero_user(page, offset, length);
3413 BUFFER_TRACE(bh, "zeroed end of block");
3415 if (ext4_should_journal_data(inode)) {
3416 err = ext4_handle_dirty_metadata(handle, inode, bh);
3419 mark_buffer_dirty(bh);
3420 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3421 err = ext4_jbd2_file_inode(handle, inode);
3426 page_cache_release(page);
3431 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3432 * starting from file offset 'from'. The range to be zero'd must
3433 * be contained with in one block. If the specified range exceeds
3434 * the end of the block it will be shortened to end of the block
3435 * that cooresponds to 'from'
3437 static int ext4_block_zero_page_range(handle_t *handle,
3438 struct address_space *mapping, loff_t from, loff_t length)
3440 struct inode *inode = mapping->host;
3441 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3442 unsigned blocksize = inode->i_sb->s_blocksize;
3443 unsigned max = blocksize - (offset & (blocksize - 1));
3446 * correct length if it does not fall between
3447 * 'from' and the end of the block
3449 if (length > max || length < 0)
3453 return dax_zero_page_range(inode, from, length, ext4_get_block);
3454 return __ext4_block_zero_page_range(handle, mapping, from, length);
3458 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3459 * up to the end of the block which corresponds to `from'.
3460 * This required during truncate. We need to physically zero the tail end
3461 * of that block so it doesn't yield old data if the file is later grown.
3463 static int ext4_block_truncate_page(handle_t *handle,
3464 struct address_space *mapping, loff_t from)
3466 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3469 struct inode *inode = mapping->host;
3471 blocksize = inode->i_sb->s_blocksize;
3472 length = blocksize - (offset & (blocksize - 1));
3474 return ext4_block_zero_page_range(handle, mapping, from, length);
3477 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3478 loff_t lstart, loff_t length)
3480 struct super_block *sb = inode->i_sb;
3481 struct address_space *mapping = inode->i_mapping;
3482 unsigned partial_start, partial_end;
3483 ext4_fsblk_t start, end;
3484 loff_t byte_end = (lstart + length - 1);
3487 partial_start = lstart & (sb->s_blocksize - 1);
3488 partial_end = byte_end & (sb->s_blocksize - 1);
3490 start = lstart >> sb->s_blocksize_bits;
3491 end = byte_end >> sb->s_blocksize_bits;
3493 /* Handle partial zero within the single block */
3495 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3496 err = ext4_block_zero_page_range(handle, mapping,
3500 /* Handle partial zero out on the start of the range */
3501 if (partial_start) {
3502 err = ext4_block_zero_page_range(handle, mapping,
3503 lstart, sb->s_blocksize);
3507 /* Handle partial zero out on the end of the range */
3508 if (partial_end != sb->s_blocksize - 1)
3509 err = ext4_block_zero_page_range(handle, mapping,
3510 byte_end - partial_end,
3515 int ext4_can_truncate(struct inode *inode)
3517 if (S_ISREG(inode->i_mode))
3519 if (S_ISDIR(inode->i_mode))
3521 if (S_ISLNK(inode->i_mode))
3522 return !ext4_inode_is_fast_symlink(inode);
3527 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3528 * associated with the given offset and length
3530 * @inode: File inode
3531 * @offset: The offset where the hole will begin
3532 * @len: The length of the hole
3534 * Returns: 0 on success or negative on failure
3537 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3539 struct super_block *sb = inode->i_sb;
3540 ext4_lblk_t first_block, stop_block;
3541 struct address_space *mapping = inode->i_mapping;
3542 loff_t first_block_offset, last_block_offset;
3544 unsigned int credits;
3547 if (!S_ISREG(inode->i_mode))
3550 trace_ext4_punch_hole(inode, offset, length, 0);
3553 * Write out all dirty pages to avoid race conditions
3554 * Then release them.
3556 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3557 ret = filemap_write_and_wait_range(mapping, offset,
3558 offset + length - 1);
3563 mutex_lock(&inode->i_mutex);
3565 /* No need to punch hole beyond i_size */
3566 if (offset >= inode->i_size)
3570 * If the hole extends beyond i_size, set the hole
3571 * to end after the page that contains i_size
3573 if (offset + length > inode->i_size) {
3574 length = inode->i_size +
3575 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3579 if (offset & (sb->s_blocksize - 1) ||
3580 (offset + length) & (sb->s_blocksize - 1)) {
3582 * Attach jinode to inode for jbd2 if we do any zeroing of
3585 ret = ext4_inode_attach_jinode(inode);
3591 first_block_offset = round_up(offset, sb->s_blocksize);
3592 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3594 /* Now release the pages and zero block aligned part of pages*/
3595 if (last_block_offset > first_block_offset)
3596 truncate_pagecache_range(inode, first_block_offset,
3599 /* Wait all existing dio workers, newcomers will block on i_mutex */
3600 ext4_inode_block_unlocked_dio(inode);
3601 inode_dio_wait(inode);
3603 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3604 credits = ext4_writepage_trans_blocks(inode);
3606 credits = ext4_blocks_for_truncate(inode);
3607 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3608 if (IS_ERR(handle)) {
3609 ret = PTR_ERR(handle);
3610 ext4_std_error(sb, ret);
3614 ret = ext4_zero_partial_blocks(handle, inode, offset,
3619 first_block = (offset + sb->s_blocksize - 1) >>
3620 EXT4_BLOCK_SIZE_BITS(sb);
3621 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3623 /* If there are no blocks to remove, return now */
3624 if (first_block >= stop_block)
3627 down_write(&EXT4_I(inode)->i_data_sem);
3628 ext4_discard_preallocations(inode);
3630 ret = ext4_es_remove_extent(inode, first_block,
3631 stop_block - first_block);
3633 up_write(&EXT4_I(inode)->i_data_sem);
3637 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3638 ret = ext4_ext_remove_space(inode, first_block,
3641 ret = ext4_ind_remove_space(handle, inode, first_block,
3644 up_write(&EXT4_I(inode)->i_data_sem);
3646 ext4_handle_sync(handle);
3648 /* Now release the pages again to reduce race window */
3649 if (last_block_offset > first_block_offset)
3650 truncate_pagecache_range(inode, first_block_offset,
3653 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3654 ext4_mark_inode_dirty(handle, inode);
3656 ext4_journal_stop(handle);
3658 ext4_inode_resume_unlocked_dio(inode);
3660 mutex_unlock(&inode->i_mutex);
3664 int ext4_inode_attach_jinode(struct inode *inode)
3666 struct ext4_inode_info *ei = EXT4_I(inode);
3667 struct jbd2_inode *jinode;
3669 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3672 jinode = jbd2_alloc_inode(GFP_KERNEL);
3673 spin_lock(&inode->i_lock);
3676 spin_unlock(&inode->i_lock);
3679 ei->jinode = jinode;
3680 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3683 spin_unlock(&inode->i_lock);
3684 if (unlikely(jinode != NULL))
3685 jbd2_free_inode(jinode);
3692 * We block out ext4_get_block() block instantiations across the entire
3693 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3694 * simultaneously on behalf of the same inode.
3696 * As we work through the truncate and commit bits of it to the journal there
3697 * is one core, guiding principle: the file's tree must always be consistent on
3698 * disk. We must be able to restart the truncate after a crash.
3700 * The file's tree may be transiently inconsistent in memory (although it
3701 * probably isn't), but whenever we close off and commit a journal transaction,
3702 * the contents of (the filesystem + the journal) must be consistent and
3703 * restartable. It's pretty simple, really: bottom up, right to left (although
3704 * left-to-right works OK too).
3706 * Note that at recovery time, journal replay occurs *before* the restart of
3707 * truncate against the orphan inode list.
3709 * The committed inode has the new, desired i_size (which is the same as
3710 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3711 * that this inode's truncate did not complete and it will again call
3712 * ext4_truncate() to have another go. So there will be instantiated blocks
3713 * to the right of the truncation point in a crashed ext4 filesystem. But
3714 * that's fine - as long as they are linked from the inode, the post-crash
3715 * ext4_truncate() run will find them and release them.
3717 void ext4_truncate(struct inode *inode)
3719 struct ext4_inode_info *ei = EXT4_I(inode);
3720 unsigned int credits;
3722 struct address_space *mapping = inode->i_mapping;
3725 * There is a possibility that we're either freeing the inode
3726 * or it's a completely new inode. In those cases we might not
3727 * have i_mutex locked because it's not necessary.
3729 if (!(inode->i_state & (I_NEW|I_FREEING)))
3730 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3731 trace_ext4_truncate_enter(inode);
3733 if (!ext4_can_truncate(inode))
3736 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3738 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3739 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3741 if (ext4_has_inline_data(inode)) {
3744 ext4_inline_data_truncate(inode, &has_inline);
3749 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3750 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3751 if (ext4_inode_attach_jinode(inode) < 0)
3755 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3756 credits = ext4_writepage_trans_blocks(inode);
3758 credits = ext4_blocks_for_truncate(inode);
3760 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3761 if (IS_ERR(handle)) {
3762 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3766 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3767 ext4_block_truncate_page(handle, mapping, inode->i_size);
3770 * We add the inode to the orphan list, so that if this
3771 * truncate spans multiple transactions, and we crash, we will
3772 * resume the truncate when the filesystem recovers. It also
3773 * marks the inode dirty, to catch the new size.
3775 * Implication: the file must always be in a sane, consistent
3776 * truncatable state while each transaction commits.
3778 if (ext4_orphan_add(handle, inode))
3781 down_write(&EXT4_I(inode)->i_data_sem);
3783 ext4_discard_preallocations(inode);
3785 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3786 ext4_ext_truncate(handle, inode);
3788 ext4_ind_truncate(handle, inode);
3790 up_write(&ei->i_data_sem);
3793 ext4_handle_sync(handle);
3797 * If this was a simple ftruncate() and the file will remain alive,
3798 * then we need to clear up the orphan record which we created above.
3799 * However, if this was a real unlink then we were called by
3800 * ext4_evict_inode(), and we allow that function to clean up the
3801 * orphan info for us.
3804 ext4_orphan_del(handle, inode);
3806 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3807 ext4_mark_inode_dirty(handle, inode);
3808 ext4_journal_stop(handle);
3810 trace_ext4_truncate_exit(inode);
3814 * ext4_get_inode_loc returns with an extra refcount against the inode's
3815 * underlying buffer_head on success. If 'in_mem' is true, we have all
3816 * data in memory that is needed to recreate the on-disk version of this
3819 static int __ext4_get_inode_loc(struct inode *inode,
3820 struct ext4_iloc *iloc, int in_mem)
3822 struct ext4_group_desc *gdp;
3823 struct buffer_head *bh;
3824 struct super_block *sb = inode->i_sb;
3826 int inodes_per_block, inode_offset;
3829 if (!ext4_valid_inum(sb, inode->i_ino))
3832 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3833 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3838 * Figure out the offset within the block group inode table
3840 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3841 inode_offset = ((inode->i_ino - 1) %
3842 EXT4_INODES_PER_GROUP(sb));
3843 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3844 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3846 bh = sb_getblk(sb, block);
3849 if (!buffer_uptodate(bh)) {
3853 * If the buffer has the write error flag, we have failed
3854 * to write out another inode in the same block. In this
3855 * case, we don't have to read the block because we may
3856 * read the old inode data successfully.
3858 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3859 set_buffer_uptodate(bh);
3861 if (buffer_uptodate(bh)) {
3862 /* someone brought it uptodate while we waited */
3868 * If we have all information of the inode in memory and this
3869 * is the only valid inode in the block, we need not read the
3873 struct buffer_head *bitmap_bh;
3876 start = inode_offset & ~(inodes_per_block - 1);
3878 /* Is the inode bitmap in cache? */
3879 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3880 if (unlikely(!bitmap_bh))
3884 * If the inode bitmap isn't in cache then the
3885 * optimisation may end up performing two reads instead
3886 * of one, so skip it.
3888 if (!buffer_uptodate(bitmap_bh)) {
3892 for (i = start; i < start + inodes_per_block; i++) {
3893 if (i == inode_offset)
3895 if (ext4_test_bit(i, bitmap_bh->b_data))
3899 if (i == start + inodes_per_block) {
3900 /* all other inodes are free, so skip I/O */
3901 memset(bh->b_data, 0, bh->b_size);
3902 set_buffer_uptodate(bh);
3910 * If we need to do any I/O, try to pre-readahead extra
3911 * blocks from the inode table.
3913 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3914 ext4_fsblk_t b, end, table;
3916 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3918 table = ext4_inode_table(sb, gdp);
3919 /* s_inode_readahead_blks is always a power of 2 */
3920 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3924 num = EXT4_INODES_PER_GROUP(sb);
3925 if (ext4_has_group_desc_csum(sb))
3926 num -= ext4_itable_unused_count(sb, gdp);
3927 table += num / inodes_per_block;
3931 sb_breadahead(sb, b++);
3935 * There are other valid inodes in the buffer, this inode
3936 * has in-inode xattrs, or we don't have this inode in memory.
3937 * Read the block from disk.
3939 trace_ext4_load_inode(inode);
3941 bh->b_end_io = end_buffer_read_sync;
3942 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3944 if (!buffer_uptodate(bh)) {
3945 EXT4_ERROR_INODE_BLOCK(inode, block,
3946 "unable to read itable block");
3956 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3958 /* We have all inode data except xattrs in memory here. */
3959 return __ext4_get_inode_loc(inode, iloc,
3960 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3963 void ext4_set_inode_flags(struct inode *inode)
3965 unsigned int flags = EXT4_I(inode)->i_flags;
3966 unsigned int new_fl = 0;
3968 if (flags & EXT4_SYNC_FL)
3970 if (flags & EXT4_APPEND_FL)
3972 if (flags & EXT4_IMMUTABLE_FL)
3973 new_fl |= S_IMMUTABLE;
3974 if (flags & EXT4_NOATIME_FL)
3975 new_fl |= S_NOATIME;
3976 if (flags & EXT4_DIRSYNC_FL)
3977 new_fl |= S_DIRSYNC;
3978 if (test_opt(inode->i_sb, DAX))
3980 inode_set_flags(inode, new_fl,
3981 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
3984 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3985 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3987 unsigned int vfs_fl;
3988 unsigned long old_fl, new_fl;
3991 vfs_fl = ei->vfs_inode.i_flags;
3992 old_fl = ei->i_flags;
3993 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3994 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3996 if (vfs_fl & S_SYNC)
3997 new_fl |= EXT4_SYNC_FL;
3998 if (vfs_fl & S_APPEND)
3999 new_fl |= EXT4_APPEND_FL;
4000 if (vfs_fl & S_IMMUTABLE)
4001 new_fl |= EXT4_IMMUTABLE_FL;
4002 if (vfs_fl & S_NOATIME)
4003 new_fl |= EXT4_NOATIME_FL;
4004 if (vfs_fl & S_DIRSYNC)
4005 new_fl |= EXT4_DIRSYNC_FL;
4006 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4009 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4010 struct ext4_inode_info *ei)
4013 struct inode *inode = &(ei->vfs_inode);
4014 struct super_block *sb = inode->i_sb;
4016 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4017 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4018 /* we are using combined 48 bit field */
4019 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4020 le32_to_cpu(raw_inode->i_blocks_lo);
4021 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4022 /* i_blocks represent file system block size */
4023 return i_blocks << (inode->i_blkbits - 9);
4028 return le32_to_cpu(raw_inode->i_blocks_lo);
4032 static inline void ext4_iget_extra_inode(struct inode *inode,
4033 struct ext4_inode *raw_inode,
4034 struct ext4_inode_info *ei)
4036 __le32 *magic = (void *)raw_inode +
4037 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4038 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4039 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4040 ext4_find_inline_data_nolock(inode);
4042 EXT4_I(inode)->i_inline_off = 0;
4045 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4047 struct ext4_iloc iloc;
4048 struct ext4_inode *raw_inode;
4049 struct ext4_inode_info *ei;
4050 struct inode *inode;
4051 journal_t *journal = EXT4_SB(sb)->s_journal;
4057 inode = iget_locked(sb, ino);
4059 return ERR_PTR(-ENOMEM);
4060 if (!(inode->i_state & I_NEW))
4066 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4069 raw_inode = ext4_raw_inode(&iloc);
4071 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4072 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4073 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4074 EXT4_INODE_SIZE(inode->i_sb)) {
4075 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4076 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4077 EXT4_INODE_SIZE(inode->i_sb));
4082 ei->i_extra_isize = 0;
4084 /* Precompute checksum seed for inode metadata */
4085 if (ext4_has_metadata_csum(sb)) {
4086 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4088 __le32 inum = cpu_to_le32(inode->i_ino);
4089 __le32 gen = raw_inode->i_generation;
4090 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4092 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4096 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4097 EXT4_ERROR_INODE(inode, "checksum invalid");
4102 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4103 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4104 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4105 if (!(test_opt(inode->i_sb, NO_UID32))) {
4106 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4107 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4109 i_uid_write(inode, i_uid);
4110 i_gid_write(inode, i_gid);
4111 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4113 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4114 ei->i_inline_off = 0;
4115 ei->i_dir_start_lookup = 0;
4116 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4117 /* We now have enough fields to check if the inode was active or not.
4118 * This is needed because nfsd might try to access dead inodes
4119 * the test is that same one that e2fsck uses
4120 * NeilBrown 1999oct15
4122 if (inode->i_nlink == 0) {
4123 if ((inode->i_mode == 0 ||
4124 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4125 ino != EXT4_BOOT_LOADER_INO) {
4126 /* this inode is deleted */
4130 /* The only unlinked inodes we let through here have
4131 * valid i_mode and are being read by the orphan
4132 * recovery code: that's fine, we're about to complete
4133 * the process of deleting those.
4134 * OR it is the EXT4_BOOT_LOADER_INO which is
4135 * not initialized on a new filesystem. */
4137 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4138 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4139 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4140 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4142 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4143 inode->i_size = ext4_isize(raw_inode);
4144 ei->i_disksize = inode->i_size;
4146 ei->i_reserved_quota = 0;
4148 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4149 ei->i_block_group = iloc.block_group;
4150 ei->i_last_alloc_group = ~0;
4152 * NOTE! The in-memory inode i_data array is in little-endian order
4153 * even on big-endian machines: we do NOT byteswap the block numbers!
4155 for (block = 0; block < EXT4_N_BLOCKS; block++)
4156 ei->i_data[block] = raw_inode->i_block[block];
4157 INIT_LIST_HEAD(&ei->i_orphan);
4160 * Set transaction id's of transactions that have to be committed
4161 * to finish f[data]sync. We set them to currently running transaction
4162 * as we cannot be sure that the inode or some of its metadata isn't
4163 * part of the transaction - the inode could have been reclaimed and
4164 * now it is reread from disk.
4167 transaction_t *transaction;
4170 read_lock(&journal->j_state_lock);
4171 if (journal->j_running_transaction)
4172 transaction = journal->j_running_transaction;
4174 transaction = journal->j_committing_transaction;
4176 tid = transaction->t_tid;
4178 tid = journal->j_commit_sequence;
4179 read_unlock(&journal->j_state_lock);
4180 ei->i_sync_tid = tid;
4181 ei->i_datasync_tid = tid;
4184 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4185 if (ei->i_extra_isize == 0) {
4186 /* The extra space is currently unused. Use it. */
4187 ei->i_extra_isize = sizeof(struct ext4_inode) -
4188 EXT4_GOOD_OLD_INODE_SIZE;
4190 ext4_iget_extra_inode(inode, raw_inode, ei);
4194 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4195 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4196 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4197 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4199 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4200 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4201 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4202 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4204 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4209 if (ei->i_file_acl &&
4210 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4211 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4215 } else if (!ext4_has_inline_data(inode)) {
4216 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4217 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4218 (S_ISLNK(inode->i_mode) &&
4219 !ext4_inode_is_fast_symlink(inode))))
4220 /* Validate extent which is part of inode */
4221 ret = ext4_ext_check_inode(inode);
4222 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4223 (S_ISLNK(inode->i_mode) &&
4224 !ext4_inode_is_fast_symlink(inode))) {
4225 /* Validate block references which are part of inode */
4226 ret = ext4_ind_check_inode(inode);
4232 if (S_ISREG(inode->i_mode)) {
4233 inode->i_op = &ext4_file_inode_operations;
4234 inode->i_fop = &ext4_file_operations;
4235 ext4_set_aops(inode);
4236 } else if (S_ISDIR(inode->i_mode)) {
4237 inode->i_op = &ext4_dir_inode_operations;
4238 inode->i_fop = &ext4_dir_operations;
4239 } else if (S_ISLNK(inode->i_mode)) {
4240 if (ext4_inode_is_fast_symlink(inode) &&
4241 !ext4_encrypted_inode(inode)) {
4242 inode->i_op = &ext4_fast_symlink_inode_operations;
4243 nd_terminate_link(ei->i_data, inode->i_size,
4244 sizeof(ei->i_data) - 1);
4246 inode->i_op = &ext4_symlink_inode_operations;
4247 ext4_set_aops(inode);
4249 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4250 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4251 inode->i_op = &ext4_special_inode_operations;
4252 if (raw_inode->i_block[0])
4253 init_special_inode(inode, inode->i_mode,
4254 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4256 init_special_inode(inode, inode->i_mode,
4257 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4258 } else if (ino == EXT4_BOOT_LOADER_INO) {
4259 make_bad_inode(inode);
4262 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4266 ext4_set_inode_flags(inode);
4267 unlock_new_inode(inode);
4273 return ERR_PTR(ret);
4276 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4278 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4279 return ERR_PTR(-EIO);
4280 return ext4_iget(sb, ino);
4283 static int ext4_inode_blocks_set(handle_t *handle,
4284 struct ext4_inode *raw_inode,
4285 struct ext4_inode_info *ei)
4287 struct inode *inode = &(ei->vfs_inode);
4288 u64 i_blocks = inode->i_blocks;
4289 struct super_block *sb = inode->i_sb;
4291 if (i_blocks <= ~0U) {
4293 * i_blocks can be represented in a 32 bit variable
4294 * as multiple of 512 bytes
4296 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4297 raw_inode->i_blocks_high = 0;
4298 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4301 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4304 if (i_blocks <= 0xffffffffffffULL) {
4306 * i_blocks can be represented in a 48 bit variable
4307 * as multiple of 512 bytes
4309 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4310 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4311 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4313 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4314 /* i_block is stored in file system block size */
4315 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4316 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4317 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4322 struct other_inode {
4323 unsigned long orig_ino;
4324 struct ext4_inode *raw_inode;
4327 static int other_inode_match(struct inode * inode, unsigned long ino,
4330 struct other_inode *oi = (struct other_inode *) data;
4332 if ((inode->i_ino != ino) ||
4333 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4334 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4335 ((inode->i_state & I_DIRTY_TIME) == 0))
4337 spin_lock(&inode->i_lock);
4338 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4339 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4340 (inode->i_state & I_DIRTY_TIME)) {
4341 struct ext4_inode_info *ei = EXT4_I(inode);
4343 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4344 spin_unlock(&inode->i_lock);
4346 spin_lock(&ei->i_raw_lock);
4347 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4348 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4349 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4350 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4351 spin_unlock(&ei->i_raw_lock);
4352 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4355 spin_unlock(&inode->i_lock);
4360 * Opportunistically update the other time fields for other inodes in
4361 * the same inode table block.
4363 static void ext4_update_other_inodes_time(struct super_block *sb,
4364 unsigned long orig_ino, char *buf)
4366 struct other_inode oi;
4368 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4369 int inode_size = EXT4_INODE_SIZE(sb);
4371 oi.orig_ino = orig_ino;
4373 * Calculate the first inode in the inode table block. Inode
4374 * numbers are one-based. That is, the first inode in a block
4375 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4377 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4378 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4379 if (ino == orig_ino)
4381 oi.raw_inode = (struct ext4_inode *) buf;
4382 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4387 * Post the struct inode info into an on-disk inode location in the
4388 * buffer-cache. This gobbles the caller's reference to the
4389 * buffer_head in the inode location struct.
4391 * The caller must have write access to iloc->bh.
4393 static int ext4_do_update_inode(handle_t *handle,
4394 struct inode *inode,
4395 struct ext4_iloc *iloc)
4397 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4398 struct ext4_inode_info *ei = EXT4_I(inode);
4399 struct buffer_head *bh = iloc->bh;
4400 struct super_block *sb = inode->i_sb;
4401 int err = 0, rc, block;
4402 int need_datasync = 0, set_large_file = 0;
4406 spin_lock(&ei->i_raw_lock);
4408 /* For fields not tracked in the in-memory inode,
4409 * initialise them to zero for new inodes. */
4410 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4411 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4413 ext4_get_inode_flags(ei);
4414 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4415 i_uid = i_uid_read(inode);
4416 i_gid = i_gid_read(inode);
4417 if (!(test_opt(inode->i_sb, NO_UID32))) {
4418 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4419 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4421 * Fix up interoperability with old kernels. Otherwise, old inodes get
4422 * re-used with the upper 16 bits of the uid/gid intact
4425 raw_inode->i_uid_high =
4426 cpu_to_le16(high_16_bits(i_uid));
4427 raw_inode->i_gid_high =
4428 cpu_to_le16(high_16_bits(i_gid));
4430 raw_inode->i_uid_high = 0;
4431 raw_inode->i_gid_high = 0;
4434 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4435 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4436 raw_inode->i_uid_high = 0;
4437 raw_inode->i_gid_high = 0;
4439 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4441 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4442 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4443 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4444 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4446 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4448 spin_unlock(&ei->i_raw_lock);
4451 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4452 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4453 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4454 raw_inode->i_file_acl_high =
4455 cpu_to_le16(ei->i_file_acl >> 32);
4456 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4457 if (ei->i_disksize != ext4_isize(raw_inode)) {
4458 ext4_isize_set(raw_inode, ei->i_disksize);
4461 if (ei->i_disksize > 0x7fffffffULL) {
4462 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4463 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4464 EXT4_SB(sb)->s_es->s_rev_level ==
4465 cpu_to_le32(EXT4_GOOD_OLD_REV))
4468 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4469 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4470 if (old_valid_dev(inode->i_rdev)) {
4471 raw_inode->i_block[0] =
4472 cpu_to_le32(old_encode_dev(inode->i_rdev));
4473 raw_inode->i_block[1] = 0;
4475 raw_inode->i_block[0] = 0;
4476 raw_inode->i_block[1] =
4477 cpu_to_le32(new_encode_dev(inode->i_rdev));
4478 raw_inode->i_block[2] = 0;
4480 } else if (!ext4_has_inline_data(inode)) {
4481 for (block = 0; block < EXT4_N_BLOCKS; block++)
4482 raw_inode->i_block[block] = ei->i_data[block];
4485 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4486 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4487 if (ei->i_extra_isize) {
4488 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4489 raw_inode->i_version_hi =
4490 cpu_to_le32(inode->i_version >> 32);
4491 raw_inode->i_extra_isize =
4492 cpu_to_le16(ei->i_extra_isize);
4495 ext4_inode_csum_set(inode, raw_inode, ei);
4496 spin_unlock(&ei->i_raw_lock);
4497 if (inode->i_sb->s_flags & MS_LAZYTIME)
4498 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4501 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4502 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4505 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4506 if (set_large_file) {
4507 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4508 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4511 ext4_update_dynamic_rev(sb);
4512 EXT4_SET_RO_COMPAT_FEATURE(sb,
4513 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4514 ext4_handle_sync(handle);
4515 err = ext4_handle_dirty_super(handle, sb);
4517 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4520 ext4_std_error(inode->i_sb, err);
4525 * ext4_write_inode()
4527 * We are called from a few places:
4529 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4530 * Here, there will be no transaction running. We wait for any running
4531 * transaction to commit.
4533 * - Within flush work (sys_sync(), kupdate and such).
4534 * We wait on commit, if told to.
4536 * - Within iput_final() -> write_inode_now()
4537 * We wait on commit, if told to.
4539 * In all cases it is actually safe for us to return without doing anything,
4540 * because the inode has been copied into a raw inode buffer in
4541 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4544 * Note that we are absolutely dependent upon all inode dirtiers doing the
4545 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4546 * which we are interested.
4548 * It would be a bug for them to not do this. The code:
4550 * mark_inode_dirty(inode)
4552 * inode->i_size = expr;
4554 * is in error because write_inode() could occur while `stuff()' is running,
4555 * and the new i_size will be lost. Plus the inode will no longer be on the
4556 * superblock's dirty inode list.
4558 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4562 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4565 if (EXT4_SB(inode->i_sb)->s_journal) {
4566 if (ext4_journal_current_handle()) {
4567 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4573 * No need to force transaction in WB_SYNC_NONE mode. Also
4574 * ext4_sync_fs() will force the commit after everything is
4577 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4580 err = ext4_force_commit(inode->i_sb);
4582 struct ext4_iloc iloc;
4584 err = __ext4_get_inode_loc(inode, &iloc, 0);
4588 * sync(2) will flush the whole buffer cache. No need to do
4589 * it here separately for each inode.
4591 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4592 sync_dirty_buffer(iloc.bh);
4593 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4594 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4595 "IO error syncing inode");
4604 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4605 * buffers that are attached to a page stradding i_size and are undergoing
4606 * commit. In that case we have to wait for commit to finish and try again.
4608 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4612 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4613 tid_t commit_tid = 0;
4616 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4618 * All buffers in the last page remain valid? Then there's nothing to
4619 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4622 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4625 page = find_lock_page(inode->i_mapping,
4626 inode->i_size >> PAGE_CACHE_SHIFT);
4629 ret = __ext4_journalled_invalidatepage(page, offset,
4630 PAGE_CACHE_SIZE - offset);
4632 page_cache_release(page);
4636 read_lock(&journal->j_state_lock);
4637 if (journal->j_committing_transaction)
4638 commit_tid = journal->j_committing_transaction->t_tid;
4639 read_unlock(&journal->j_state_lock);
4641 jbd2_log_wait_commit(journal, commit_tid);
4648 * Called from notify_change.
4650 * We want to trap VFS attempts to truncate the file as soon as
4651 * possible. In particular, we want to make sure that when the VFS
4652 * shrinks i_size, we put the inode on the orphan list and modify
4653 * i_disksize immediately, so that during the subsequent flushing of
4654 * dirty pages and freeing of disk blocks, we can guarantee that any
4655 * commit will leave the blocks being flushed in an unused state on
4656 * disk. (On recovery, the inode will get truncated and the blocks will
4657 * be freed, so we have a strong guarantee that no future commit will
4658 * leave these blocks visible to the user.)
4660 * Another thing we have to assure is that if we are in ordered mode
4661 * and inode is still attached to the committing transaction, we must
4662 * we start writeout of all the dirty pages which are being truncated.
4663 * This way we are sure that all the data written in the previous
4664 * transaction are already on disk (truncate waits for pages under
4667 * Called with inode->i_mutex down.
4669 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4671 struct inode *inode = d_inode(dentry);
4674 const unsigned int ia_valid = attr->ia_valid;
4676 error = inode_change_ok(inode, attr);
4680 if (is_quota_modification(inode, attr))
4681 dquot_initialize(inode);
4682 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4683 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4686 /* (user+group)*(old+new) structure, inode write (sb,
4687 * inode block, ? - but truncate inode update has it) */
4688 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4689 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4690 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4691 if (IS_ERR(handle)) {
4692 error = PTR_ERR(handle);
4695 error = dquot_transfer(inode, attr);
4697 ext4_journal_stop(handle);
4700 /* Update corresponding info in inode so that everything is in
4701 * one transaction */
4702 if (attr->ia_valid & ATTR_UID)
4703 inode->i_uid = attr->ia_uid;
4704 if (attr->ia_valid & ATTR_GID)
4705 inode->i_gid = attr->ia_gid;
4706 error = ext4_mark_inode_dirty(handle, inode);
4707 ext4_journal_stop(handle);
4710 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4713 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4714 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4716 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4720 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4721 inode_inc_iversion(inode);
4723 if (S_ISREG(inode->i_mode) &&
4724 (attr->ia_size < inode->i_size)) {
4725 if (ext4_should_order_data(inode)) {
4726 error = ext4_begin_ordered_truncate(inode,
4731 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4732 if (IS_ERR(handle)) {
4733 error = PTR_ERR(handle);
4736 if (ext4_handle_valid(handle)) {
4737 error = ext4_orphan_add(handle, inode);
4740 down_write(&EXT4_I(inode)->i_data_sem);
4741 EXT4_I(inode)->i_disksize = attr->ia_size;
4742 rc = ext4_mark_inode_dirty(handle, inode);
4746 * We have to update i_size under i_data_sem together
4747 * with i_disksize to avoid races with writeback code
4748 * running ext4_wb_update_i_disksize().
4751 i_size_write(inode, attr->ia_size);
4752 up_write(&EXT4_I(inode)->i_data_sem);
4753 ext4_journal_stop(handle);
4755 ext4_orphan_del(NULL, inode);
4759 loff_t oldsize = inode->i_size;
4761 i_size_write(inode, attr->ia_size);
4762 pagecache_isize_extended(inode, oldsize, inode->i_size);
4766 * Blocks are going to be removed from the inode. Wait
4767 * for dio in flight. Temporarily disable
4768 * dioread_nolock to prevent livelock.
4771 if (!ext4_should_journal_data(inode)) {
4772 ext4_inode_block_unlocked_dio(inode);
4773 inode_dio_wait(inode);
4774 ext4_inode_resume_unlocked_dio(inode);
4776 ext4_wait_for_tail_page_commit(inode);
4779 * Truncate pagecache after we've waited for commit
4780 * in data=journal mode to make pages freeable.
4782 truncate_pagecache(inode, inode->i_size);
4785 * We want to call ext4_truncate() even if attr->ia_size ==
4786 * inode->i_size for cases like truncation of fallocated space
4788 if (attr->ia_valid & ATTR_SIZE)
4789 ext4_truncate(inode);
4792 setattr_copy(inode, attr);
4793 mark_inode_dirty(inode);
4797 * If the call to ext4_truncate failed to get a transaction handle at
4798 * all, we need to clean up the in-core orphan list manually.
4800 if (orphan && inode->i_nlink)
4801 ext4_orphan_del(NULL, inode);
4803 if (!rc && (ia_valid & ATTR_MODE))
4804 rc = posix_acl_chmod(inode, inode->i_mode);
4807 ext4_std_error(inode->i_sb, error);
4813 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4816 struct inode *inode;
4817 unsigned long long delalloc_blocks;
4819 inode = d_inode(dentry);
4820 generic_fillattr(inode, stat);
4823 * If there is inline data in the inode, the inode will normally not
4824 * have data blocks allocated (it may have an external xattr block).
4825 * Report at least one sector for such files, so tools like tar, rsync,
4826 * others doen't incorrectly think the file is completely sparse.
4828 if (unlikely(ext4_has_inline_data(inode)))
4829 stat->blocks += (stat->size + 511) >> 9;
4832 * We can't update i_blocks if the block allocation is delayed
4833 * otherwise in the case of system crash before the real block
4834 * allocation is done, we will have i_blocks inconsistent with
4835 * on-disk file blocks.
4836 * We always keep i_blocks updated together with real
4837 * allocation. But to not confuse with user, stat
4838 * will return the blocks that include the delayed allocation
4839 * blocks for this file.
4841 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4842 EXT4_I(inode)->i_reserved_data_blocks);
4843 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4847 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4850 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4851 return ext4_ind_trans_blocks(inode, lblocks);
4852 return ext4_ext_index_trans_blocks(inode, pextents);
4856 * Account for index blocks, block groups bitmaps and block group
4857 * descriptor blocks if modify datablocks and index blocks
4858 * worse case, the indexs blocks spread over different block groups
4860 * If datablocks are discontiguous, they are possible to spread over
4861 * different block groups too. If they are contiguous, with flexbg,
4862 * they could still across block group boundary.
4864 * Also account for superblock, inode, quota and xattr blocks
4866 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4869 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4875 * How many index blocks need to touch to map @lblocks logical blocks
4876 * to @pextents physical extents?
4878 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4883 * Now let's see how many group bitmaps and group descriptors need
4886 groups = idxblocks + pextents;
4888 if (groups > ngroups)
4890 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4891 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4893 /* bitmaps and block group descriptor blocks */
4894 ret += groups + gdpblocks;
4896 /* Blocks for super block, inode, quota and xattr blocks */
4897 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4903 * Calculate the total number of credits to reserve to fit
4904 * the modification of a single pages into a single transaction,
4905 * which may include multiple chunks of block allocations.
4907 * This could be called via ext4_write_begin()
4909 * We need to consider the worse case, when
4910 * one new block per extent.
4912 int ext4_writepage_trans_blocks(struct inode *inode)
4914 int bpp = ext4_journal_blocks_per_page(inode);
4917 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4919 /* Account for data blocks for journalled mode */
4920 if (ext4_should_journal_data(inode))
4926 * Calculate the journal credits for a chunk of data modification.
4928 * This is called from DIO, fallocate or whoever calling
4929 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4931 * journal buffers for data blocks are not included here, as DIO
4932 * and fallocate do no need to journal data buffers.
4934 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4936 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4940 * The caller must have previously called ext4_reserve_inode_write().
4941 * Give this, we know that the caller already has write access to iloc->bh.
4943 int ext4_mark_iloc_dirty(handle_t *handle,
4944 struct inode *inode, struct ext4_iloc *iloc)
4948 if (IS_I_VERSION(inode))
4949 inode_inc_iversion(inode);
4951 /* the do_update_inode consumes one bh->b_count */
4954 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4955 err = ext4_do_update_inode(handle, inode, iloc);
4961 * On success, We end up with an outstanding reference count against
4962 * iloc->bh. This _must_ be cleaned up later.
4966 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4967 struct ext4_iloc *iloc)
4971 err = ext4_get_inode_loc(inode, iloc);
4973 BUFFER_TRACE(iloc->bh, "get_write_access");
4974 err = ext4_journal_get_write_access(handle, iloc->bh);
4980 ext4_std_error(inode->i_sb, err);
4985 * Expand an inode by new_extra_isize bytes.
4986 * Returns 0 on success or negative error number on failure.
4988 static int ext4_expand_extra_isize(struct inode *inode,
4989 unsigned int new_extra_isize,
4990 struct ext4_iloc iloc,
4993 struct ext4_inode *raw_inode;
4994 struct ext4_xattr_ibody_header *header;
4996 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4999 raw_inode = ext4_raw_inode(&iloc);
5001 header = IHDR(inode, raw_inode);
5003 /* No extended attributes present */
5004 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5005 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5006 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5008 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5012 /* try to expand with EAs present */
5013 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5018 * What we do here is to mark the in-core inode as clean with respect to inode
5019 * dirtiness (it may still be data-dirty).
5020 * This means that the in-core inode may be reaped by prune_icache
5021 * without having to perform any I/O. This is a very good thing,
5022 * because *any* task may call prune_icache - even ones which
5023 * have a transaction open against a different journal.
5025 * Is this cheating? Not really. Sure, we haven't written the
5026 * inode out, but prune_icache isn't a user-visible syncing function.
5027 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5028 * we start and wait on commits.
5030 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5032 struct ext4_iloc iloc;
5033 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5034 static unsigned int mnt_count;
5038 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5039 err = ext4_reserve_inode_write(handle, inode, &iloc);
5040 if (ext4_handle_valid(handle) &&
5041 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5042 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5044 * We need extra buffer credits since we may write into EA block
5045 * with this same handle. If journal_extend fails, then it will
5046 * only result in a minor loss of functionality for that inode.
5047 * If this is felt to be critical, then e2fsck should be run to
5048 * force a large enough s_min_extra_isize.
5050 if ((jbd2_journal_extend(handle,
5051 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5052 ret = ext4_expand_extra_isize(inode,
5053 sbi->s_want_extra_isize,
5056 ext4_set_inode_state(inode,
5057 EXT4_STATE_NO_EXPAND);
5059 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5060 ext4_warning(inode->i_sb,
5061 "Unable to expand inode %lu. Delete"
5062 " some EAs or run e2fsck.",
5065 le16_to_cpu(sbi->s_es->s_mnt_count);
5071 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5076 * ext4_dirty_inode() is called from __mark_inode_dirty()
5078 * We're really interested in the case where a file is being extended.
5079 * i_size has been changed by generic_commit_write() and we thus need
5080 * to include the updated inode in the current transaction.
5082 * Also, dquot_alloc_block() will always dirty the inode when blocks
5083 * are allocated to the file.
5085 * If the inode is marked synchronous, we don't honour that here - doing
5086 * so would cause a commit on atime updates, which we don't bother doing.
5087 * We handle synchronous inodes at the highest possible level.
5089 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5090 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5091 * to copy into the on-disk inode structure are the timestamp files.
5093 void ext4_dirty_inode(struct inode *inode, int flags)
5097 if (flags == I_DIRTY_TIME)
5099 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5103 ext4_mark_inode_dirty(handle, inode);
5105 ext4_journal_stop(handle);
5112 * Bind an inode's backing buffer_head into this transaction, to prevent
5113 * it from being flushed to disk early. Unlike
5114 * ext4_reserve_inode_write, this leaves behind no bh reference and
5115 * returns no iloc structure, so the caller needs to repeat the iloc
5116 * lookup to mark the inode dirty later.
5118 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5120 struct ext4_iloc iloc;
5124 err = ext4_get_inode_loc(inode, &iloc);
5126 BUFFER_TRACE(iloc.bh, "get_write_access");
5127 err = jbd2_journal_get_write_access(handle, iloc.bh);
5129 err = ext4_handle_dirty_metadata(handle,
5135 ext4_std_error(inode->i_sb, err);
5140 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5147 * We have to be very careful here: changing a data block's
5148 * journaling status dynamically is dangerous. If we write a
5149 * data block to the journal, change the status and then delete
5150 * that block, we risk forgetting to revoke the old log record
5151 * from the journal and so a subsequent replay can corrupt data.
5152 * So, first we make sure that the journal is empty and that
5153 * nobody is changing anything.
5156 journal = EXT4_JOURNAL(inode);
5159 if (is_journal_aborted(journal))
5161 /* We have to allocate physical blocks for delalloc blocks
5162 * before flushing journal. otherwise delalloc blocks can not
5163 * be allocated any more. even more truncate on delalloc blocks
5164 * could trigger BUG by flushing delalloc blocks in journal.
5165 * There is no delalloc block in non-journal data mode.
5167 if (val && test_opt(inode->i_sb, DELALLOC)) {
5168 err = ext4_alloc_da_blocks(inode);
5173 /* Wait for all existing dio workers */
5174 ext4_inode_block_unlocked_dio(inode);
5175 inode_dio_wait(inode);
5177 jbd2_journal_lock_updates(journal);
5180 * OK, there are no updates running now, and all cached data is
5181 * synced to disk. We are now in a completely consistent state
5182 * which doesn't have anything in the journal, and we know that
5183 * no filesystem updates are running, so it is safe to modify
5184 * the inode's in-core data-journaling state flag now.
5188 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5190 err = jbd2_journal_flush(journal);
5192 jbd2_journal_unlock_updates(journal);
5193 ext4_inode_resume_unlocked_dio(inode);
5196 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5198 ext4_set_aops(inode);
5200 jbd2_journal_unlock_updates(journal);
5201 ext4_inode_resume_unlocked_dio(inode);
5203 /* Finally we can mark the inode as dirty. */
5205 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5207 return PTR_ERR(handle);
5209 err = ext4_mark_inode_dirty(handle, inode);
5210 ext4_handle_sync(handle);
5211 ext4_journal_stop(handle);
5212 ext4_std_error(inode->i_sb, err);
5217 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5219 return !buffer_mapped(bh);
5222 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5224 struct page *page = vmf->page;
5228 struct file *file = vma->vm_file;
5229 struct inode *inode = file_inode(file);
5230 struct address_space *mapping = inode->i_mapping;
5232 get_block_t *get_block;
5235 sb_start_pagefault(inode->i_sb);
5236 file_update_time(vma->vm_file);
5237 /* Delalloc case is easy... */
5238 if (test_opt(inode->i_sb, DELALLOC) &&
5239 !ext4_should_journal_data(inode) &&
5240 !ext4_nonda_switch(inode->i_sb)) {
5242 ret = __block_page_mkwrite(vma, vmf,
5243 ext4_da_get_block_prep);
5244 } while (ret == -ENOSPC &&
5245 ext4_should_retry_alloc(inode->i_sb, &retries));
5250 size = i_size_read(inode);
5251 /* Page got truncated from under us? */
5252 if (page->mapping != mapping || page_offset(page) > size) {
5254 ret = VM_FAULT_NOPAGE;
5258 if (page->index == size >> PAGE_CACHE_SHIFT)
5259 len = size & ~PAGE_CACHE_MASK;
5261 len = PAGE_CACHE_SIZE;
5263 * Return if we have all the buffers mapped. This avoids the need to do
5264 * journal_start/journal_stop which can block and take a long time
5266 if (page_has_buffers(page)) {
5267 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5269 ext4_bh_unmapped)) {
5270 /* Wait so that we don't change page under IO */
5271 wait_for_stable_page(page);
5272 ret = VM_FAULT_LOCKED;
5277 /* OK, we need to fill the hole... */
5278 if (ext4_should_dioread_nolock(inode))
5279 get_block = ext4_get_block_write;
5281 get_block = ext4_get_block;
5283 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5284 ext4_writepage_trans_blocks(inode));
5285 if (IS_ERR(handle)) {
5286 ret = VM_FAULT_SIGBUS;
5289 ret = __block_page_mkwrite(vma, vmf, get_block);
5290 if (!ret && ext4_should_journal_data(inode)) {
5291 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5292 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5294 ret = VM_FAULT_SIGBUS;
5295 ext4_journal_stop(handle);
5298 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5300 ext4_journal_stop(handle);
5301 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5304 ret = block_page_mkwrite_return(ret);
5306 sb_end_pagefault(inode->i_sb);