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/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/bitops.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
56 int offset = offsetof(struct ext4_inode, i_checksum_lo);
57 unsigned int csum_size = sizeof(dummy_csum);
59 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
60 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
62 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
63 EXT4_GOOD_OLD_INODE_SIZE - offset);
65 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
66 offset = offsetof(struct ext4_inode, i_checksum_hi);
67 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
68 EXT4_GOOD_OLD_INODE_SIZE,
69 offset - EXT4_GOOD_OLD_INODE_SIZE);
70 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
71 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
75 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
76 EXT4_INODE_SIZE(inode->i_sb) - offset);
82 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
83 struct ext4_inode_info *ei)
85 __u32 provided, calculated;
87 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
88 cpu_to_le32(EXT4_OS_LINUX) ||
89 !ext4_has_metadata_csum(inode->i_sb))
92 provided = le16_to_cpu(raw->i_checksum_lo);
93 calculated = ext4_inode_csum(inode, raw, ei);
94 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
95 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
96 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100 return provided == calculated;
103 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
104 struct ext4_inode_info *ei)
108 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
109 cpu_to_le32(EXT4_OS_LINUX) ||
110 !ext4_has_metadata_csum(inode->i_sb))
113 csum = ext4_inode_csum(inode, raw, ei);
114 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
115 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
116 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
117 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
120 static inline int ext4_begin_ordered_truncate(struct inode *inode,
123 trace_ext4_begin_ordered_truncate(inode, new_size);
125 * If jinode is zero, then we never opened the file for
126 * writing, so there's no need to call
127 * jbd2_journal_begin_ordered_truncate() since there's no
128 * outstanding writes we need to flush.
130 if (!EXT4_I(inode)->jinode)
132 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
133 EXT4_I(inode)->jinode,
137 static void ext4_invalidatepage(struct page *page, unsigned int offset,
138 unsigned int length);
139 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
140 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
141 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
145 * Test whether an inode is a fast symlink.
147 int ext4_inode_is_fast_symlink(struct inode *inode)
149 int ea_blocks = EXT4_I(inode)->i_file_acl ?
150 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
152 if (ext4_has_inline_data(inode))
155 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
159 * Restart the transaction associated with *handle. This does a commit,
160 * so before we call here everything must be consistently dirtied against
163 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
169 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
170 * moment, get_block can be called only for blocks inside i_size since
171 * page cache has been already dropped and writes are blocked by
172 * i_mutex. So we can safely drop the i_data_sem here.
174 BUG_ON(EXT4_JOURNAL(inode) == NULL);
175 jbd_debug(2, "restarting handle %p\n", handle);
176 up_write(&EXT4_I(inode)->i_data_sem);
177 ret = ext4_journal_restart(handle, nblocks);
178 down_write(&EXT4_I(inode)->i_data_sem);
179 ext4_discard_preallocations(inode);
185 * Called at the last iput() if i_nlink is zero.
187 void ext4_evict_inode(struct inode *inode)
192 trace_ext4_evict_inode(inode);
194 if (inode->i_nlink) {
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
210 * Note that directories do not have this problem because they
211 * don't use page cache.
213 if (inode->i_ino != EXT4_JOURNAL_INO &&
214 ext4_should_journal_data(inode) &&
215 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
216 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
217 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
219 jbd2_complete_transaction(journal, commit_tid);
220 filemap_write_and_wait(&inode->i_data);
222 truncate_inode_pages_final(&inode->i_data);
227 if (is_bad_inode(inode))
229 dquot_initialize(inode);
231 if (ext4_should_order_data(inode))
232 ext4_begin_ordered_truncate(inode, 0);
233 truncate_inode_pages_final(&inode->i_data);
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it
239 sb_start_intwrite(inode->i_sb);
240 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
241 ext4_blocks_for_truncate(inode)+3);
242 if (IS_ERR(handle)) {
243 ext4_std_error(inode->i_sb, PTR_ERR(handle));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL, inode);
250 sb_end_intwrite(inode->i_sb);
255 ext4_handle_sync(handle);
257 err = ext4_mark_inode_dirty(handle, inode);
259 ext4_warning(inode->i_sb,
260 "couldn't mark inode dirty (err %d)", err);
264 ext4_truncate(inode);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle, 3)) {
273 err = ext4_journal_extend(handle, 3);
275 err = ext4_journal_restart(handle, 3);
277 ext4_warning(inode->i_sb,
278 "couldn't extend journal (err %d)", err);
280 ext4_journal_stop(handle);
281 ext4_orphan_del(NULL, inode);
282 sb_end_intwrite(inode->i_sb);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle, inode);
296 EXT4_I(inode)->i_dtime = get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle, inode))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode);
309 ext4_free_inode(handle, inode);
310 ext4_journal_stop(handle);
311 sb_end_intwrite(inode->i_sb);
314 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
318 qsize_t *ext4_get_reserved_space(struct inode *inode)
320 return &EXT4_I(inode)->i_reserved_quota;
325 * Called with i_data_sem down, which is important since we can call
326 * ext4_discard_preallocations() from here.
328 void ext4_da_update_reserve_space(struct inode *inode,
329 int used, int quota_claim)
331 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
332 struct ext4_inode_info *ei = EXT4_I(inode);
334 spin_lock(&ei->i_block_reservation_lock);
335 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
336 if (unlikely(used > ei->i_reserved_data_blocks)) {
337 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
338 "with only %d reserved data blocks",
339 __func__, inode->i_ino, used,
340 ei->i_reserved_data_blocks);
342 used = ei->i_reserved_data_blocks;
345 /* Update per-inode reservations */
346 ei->i_reserved_data_blocks -= used;
347 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
349 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
351 /* Update quota subsystem for data blocks */
353 dquot_claim_block(inode, EXT4_C2B(sbi, used));
356 * We did fallocate with an offset that is already delayed
357 * allocated. So on delayed allocated writeback we should
358 * not re-claim the quota for fallocated blocks.
360 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
364 * If we have done all the pending block allocations and if
365 * there aren't any writers on the inode, we can discard the
366 * inode's preallocations.
368 if ((ei->i_reserved_data_blocks == 0) &&
369 (atomic_read(&inode->i_writecount) == 0))
370 ext4_discard_preallocations(inode);
373 static int __check_block_validity(struct inode *inode, const char *func,
375 struct ext4_map_blocks *map)
377 if (ext4_has_feature_journal(inode->i_sb) &&
379 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
381 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
382 ext4_error_inode(inode, func, line, map->m_pblk,
383 "lblock %lu mapped to illegal pblock %llu "
384 "(length %d)", (unsigned long) map->m_lblk,
385 map->m_pblk, map->m_len);
386 return -EFSCORRUPTED;
391 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
396 if (ext4_encrypted_inode(inode))
397 return fscrypt_zeroout_range(inode, lblk, pblk, len);
399 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
406 #define check_block_validity(inode, map) \
407 __check_block_validity((inode), __func__, __LINE__, (map))
409 #ifdef ES_AGGRESSIVE_TEST
410 static void ext4_map_blocks_es_recheck(handle_t *handle,
412 struct ext4_map_blocks *es_map,
413 struct ext4_map_blocks *map,
420 * There is a race window that the result is not the same.
421 * e.g. xfstests #223 when dioread_nolock enables. The reason
422 * is that we lookup a block mapping in extent status tree with
423 * out taking i_data_sem. So at the time the unwritten extent
424 * could be converted.
426 down_read(&EXT4_I(inode)->i_data_sem);
427 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
428 retval = ext4_ext_map_blocks(handle, inode, map, flags &
429 EXT4_GET_BLOCKS_KEEP_SIZE);
431 retval = ext4_ind_map_blocks(handle, inode, map, flags &
432 EXT4_GET_BLOCKS_KEEP_SIZE);
434 up_read((&EXT4_I(inode)->i_data_sem));
437 * We don't check m_len because extent will be collpased in status
438 * tree. So the m_len might not equal.
440 if (es_map->m_lblk != map->m_lblk ||
441 es_map->m_flags != map->m_flags ||
442 es_map->m_pblk != map->m_pblk) {
443 printk("ES cache assertion failed for inode: %lu "
444 "es_cached ex [%d/%d/%llu/%x] != "
445 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
446 inode->i_ino, es_map->m_lblk, es_map->m_len,
447 es_map->m_pblk, es_map->m_flags, map->m_lblk,
448 map->m_len, map->m_pblk, map->m_flags,
452 #endif /* ES_AGGRESSIVE_TEST */
455 * The ext4_map_blocks() function tries to look up the requested blocks,
456 * and returns if the blocks are already mapped.
458 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
459 * and store the allocated blocks in the result buffer head and mark it
462 * If file type is extents based, it will call ext4_ext_map_blocks(),
463 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
466 * On success, it returns the number of blocks being mapped or allocated. if
467 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
468 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
470 * It returns 0 if plain look up failed (blocks have not been allocated), in
471 * that case, @map is returned as unmapped but we still do fill map->m_len to
472 * indicate the length of a hole starting at map->m_lblk.
474 * It returns the error in case of allocation failure.
476 int ext4_map_blocks(handle_t *handle, struct inode *inode,
477 struct ext4_map_blocks *map, int flags)
479 struct extent_status es;
482 #ifdef ES_AGGRESSIVE_TEST
483 struct ext4_map_blocks orig_map;
485 memcpy(&orig_map, map, sizeof(*map));
489 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
490 "logical block %lu\n", inode->i_ino, flags, map->m_len,
491 (unsigned long) map->m_lblk);
494 * ext4_map_blocks returns an int, and m_len is an unsigned int
496 if (unlikely(map->m_len > INT_MAX))
497 map->m_len = INT_MAX;
499 /* We can handle the block number less than EXT_MAX_BLOCKS */
500 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
501 return -EFSCORRUPTED;
503 /* Lookup extent status tree firstly */
504 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
505 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
506 map->m_pblk = ext4_es_pblock(&es) +
507 map->m_lblk - es.es_lblk;
508 map->m_flags |= ext4_es_is_written(&es) ?
509 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
510 retval = es.es_len - (map->m_lblk - es.es_lblk);
511 if (retval > map->m_len)
514 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
516 retval = es.es_len - (map->m_lblk - es.es_lblk);
517 if (retval > map->m_len)
524 #ifdef ES_AGGRESSIVE_TEST
525 ext4_map_blocks_es_recheck(handle, inode, map,
532 * Try to see if we can get the block without requesting a new
535 down_read(&EXT4_I(inode)->i_data_sem);
536 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
537 retval = ext4_ext_map_blocks(handle, inode, map, flags &
538 EXT4_GET_BLOCKS_KEEP_SIZE);
540 retval = ext4_ind_map_blocks(handle, inode, map, flags &
541 EXT4_GET_BLOCKS_KEEP_SIZE);
546 if (unlikely(retval != map->m_len)) {
547 ext4_warning(inode->i_sb,
548 "ES len assertion failed for inode "
549 "%lu: retval %d != map->m_len %d",
550 inode->i_ino, retval, map->m_len);
554 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
555 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
556 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
557 !(status & EXTENT_STATUS_WRITTEN) &&
558 ext4_find_delalloc_range(inode, map->m_lblk,
559 map->m_lblk + map->m_len - 1))
560 status |= EXTENT_STATUS_DELAYED;
561 ret = ext4_es_insert_extent(inode, map->m_lblk,
562 map->m_len, map->m_pblk, status);
566 up_read((&EXT4_I(inode)->i_data_sem));
569 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
570 ret = check_block_validity(inode, map);
575 /* If it is only a block(s) look up */
576 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
580 * Returns if the blocks have already allocated
582 * Note that if blocks have been preallocated
583 * ext4_ext_get_block() returns the create = 0
584 * with buffer head unmapped.
586 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
588 * If we need to convert extent to unwritten
589 * we continue and do the actual work in
590 * ext4_ext_map_blocks()
592 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
596 * Here we clear m_flags because after allocating an new extent,
597 * it will be set again.
599 map->m_flags &= ~EXT4_MAP_FLAGS;
602 * New blocks allocate and/or writing to unwritten extent
603 * will possibly result in updating i_data, so we take
604 * the write lock of i_data_sem, and call get_block()
605 * with create == 1 flag.
607 down_write(&EXT4_I(inode)->i_data_sem);
610 * We need to check for EXT4 here because migrate
611 * could have changed the inode type in between
613 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
614 retval = ext4_ext_map_blocks(handle, inode, map, flags);
616 retval = ext4_ind_map_blocks(handle, inode, map, flags);
618 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
620 * We allocated new blocks which will result in
621 * i_data's format changing. Force the migrate
622 * to fail by clearing migrate flags
624 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
628 * Update reserved blocks/metadata blocks after successful
629 * block allocation which had been deferred till now. We don't
630 * support fallocate for non extent files. So we can update
631 * reserve space here.
634 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
635 ext4_da_update_reserve_space(inode, retval, 1);
641 if (unlikely(retval != map->m_len)) {
642 ext4_warning(inode->i_sb,
643 "ES len assertion failed for inode "
644 "%lu: retval %d != map->m_len %d",
645 inode->i_ino, retval, map->m_len);
650 * We have to zeroout blocks before inserting them into extent
651 * status tree. Otherwise someone could look them up there and
652 * use them before they are really zeroed. We also have to
653 * unmap metadata before zeroing as otherwise writeback can
654 * overwrite zeros with stale data from block device.
656 if (flags & EXT4_GET_BLOCKS_ZERO &&
657 map->m_flags & EXT4_MAP_MAPPED &&
658 map->m_flags & EXT4_MAP_NEW) {
661 for (i = 0; i < map->m_len; i++) {
662 unmap_underlying_metadata(inode->i_sb->s_bdev,
665 ret = ext4_issue_zeroout(inode, map->m_lblk,
666 map->m_pblk, map->m_len);
674 * If the extent has been zeroed out, we don't need to update
675 * extent status tree.
677 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
678 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
679 if (ext4_es_is_written(&es))
682 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
683 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
684 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
685 !(status & EXTENT_STATUS_WRITTEN) &&
686 ext4_find_delalloc_range(inode, map->m_lblk,
687 map->m_lblk + map->m_len - 1))
688 status |= EXTENT_STATUS_DELAYED;
689 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
690 map->m_pblk, status);
698 up_write((&EXT4_I(inode)->i_data_sem));
699 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
700 ret = check_block_validity(inode, map);
705 * Inodes with freshly allocated blocks where contents will be
706 * visible after transaction commit must be on transaction's
709 if (map->m_flags & EXT4_MAP_NEW &&
710 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
711 !(flags & EXT4_GET_BLOCKS_ZERO) &&
712 !IS_NOQUOTA(inode) &&
713 ext4_should_order_data(inode)) {
714 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
715 ret = ext4_jbd2_inode_add_wait(handle, inode);
717 ret = ext4_jbd2_inode_add_write(handle, inode);
726 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
727 * we have to be careful as someone else may be manipulating b_state as well.
729 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
731 unsigned long old_state;
732 unsigned long new_state;
734 flags &= EXT4_MAP_FLAGS;
736 /* Dummy buffer_head? Set non-atomically. */
738 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
742 * Someone else may be modifying b_state. Be careful! This is ugly but
743 * once we get rid of using bh as a container for mapping information
744 * to pass to / from get_block functions, this can go away.
747 old_state = READ_ONCE(bh->b_state);
748 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
750 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
753 static int _ext4_get_block(struct inode *inode, sector_t iblock,
754 struct buffer_head *bh, int flags)
756 struct ext4_map_blocks map;
759 if (ext4_has_inline_data(inode))
763 map.m_len = bh->b_size >> inode->i_blkbits;
765 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
768 map_bh(bh, inode->i_sb, map.m_pblk);
769 ext4_update_bh_state(bh, map.m_flags);
770 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
776 int ext4_get_block(struct inode *inode, sector_t iblock,
777 struct buffer_head *bh, int create)
779 return _ext4_get_block(inode, iblock, bh,
780 create ? EXT4_GET_BLOCKS_CREATE : 0);
784 * Get block function used when preparing for buffered write if we require
785 * creating an unwritten extent if blocks haven't been allocated. The extent
786 * will be converted to written after the IO is complete.
788 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
789 struct buffer_head *bh_result, int create)
791 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
792 inode->i_ino, create);
793 return _ext4_get_block(inode, iblock, bh_result,
794 EXT4_GET_BLOCKS_IO_CREATE_EXT);
797 /* Maximum number of blocks we map for direct IO at once. */
798 #define DIO_MAX_BLOCKS 4096
801 * Get blocks function for the cases that need to start a transaction -
802 * generally difference cases of direct IO and DAX IO. It also handles retries
805 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
806 struct buffer_head *bh_result, int flags)
813 /* Trim mapping request to maximum we can map at once for DIO */
814 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
815 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
816 dio_credits = ext4_chunk_trans_blocks(inode,
817 bh_result->b_size >> inode->i_blkbits);
819 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
821 return PTR_ERR(handle);
823 ret = _ext4_get_block(inode, iblock, bh_result, flags);
824 ext4_journal_stop(handle);
826 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
831 /* Get block function for DIO reads and writes to inodes without extents */
832 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
833 struct buffer_head *bh, int create)
835 /* We don't expect handle for direct IO */
836 WARN_ON_ONCE(ext4_journal_current_handle());
839 return _ext4_get_block(inode, iblock, bh, 0);
840 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
844 * Get block function for AIO DIO writes when we create unwritten extent if
845 * blocks are not allocated yet. The extent will be converted to written
846 * after IO is complete.
848 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
849 sector_t iblock, struct buffer_head *bh_result, int create)
853 /* We don't expect handle for direct IO */
854 WARN_ON_ONCE(ext4_journal_current_handle());
856 ret = ext4_get_block_trans(inode, iblock, bh_result,
857 EXT4_GET_BLOCKS_IO_CREATE_EXT);
860 * When doing DIO using unwritten extents, we need io_end to convert
861 * unwritten extents to written on IO completion. We allocate io_end
862 * once we spot unwritten extent and store it in b_private. Generic
863 * DIO code keeps b_private set and furthermore passes the value to
864 * our completion callback in 'private' argument.
866 if (!ret && buffer_unwritten(bh_result)) {
867 if (!bh_result->b_private) {
868 ext4_io_end_t *io_end;
870 io_end = ext4_init_io_end(inode, GFP_KERNEL);
873 bh_result->b_private = io_end;
874 ext4_set_io_unwritten_flag(inode, io_end);
876 set_buffer_defer_completion(bh_result);
883 * Get block function for non-AIO DIO writes when we create unwritten extent if
884 * blocks are not allocated yet. The extent will be converted to written
885 * after IO is complete from ext4_ext_direct_IO() function.
887 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
888 sector_t iblock, struct buffer_head *bh_result, int create)
892 /* We don't expect handle for direct IO */
893 WARN_ON_ONCE(ext4_journal_current_handle());
895 ret = ext4_get_block_trans(inode, iblock, bh_result,
896 EXT4_GET_BLOCKS_IO_CREATE_EXT);
899 * Mark inode as having pending DIO writes to unwritten extents.
900 * ext4_ext_direct_IO() checks this flag and converts extents to
903 if (!ret && buffer_unwritten(bh_result))
904 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
909 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
910 struct buffer_head *bh_result, int create)
914 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
915 inode->i_ino, create);
916 /* We don't expect handle for direct IO */
917 WARN_ON_ONCE(ext4_journal_current_handle());
919 ret = _ext4_get_block(inode, iblock, bh_result, 0);
921 * Blocks should have been preallocated! ext4_file_write_iter() checks
924 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
931 * `handle' can be NULL if create is zero
933 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
934 ext4_lblk_t block, int map_flags)
936 struct ext4_map_blocks map;
937 struct buffer_head *bh;
938 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
941 J_ASSERT(handle != NULL || create == 0);
945 err = ext4_map_blocks(handle, inode, &map, map_flags);
948 return create ? ERR_PTR(-ENOSPC) : NULL;
952 bh = sb_getblk(inode->i_sb, map.m_pblk);
954 return ERR_PTR(-ENOMEM);
955 if (map.m_flags & EXT4_MAP_NEW) {
956 J_ASSERT(create != 0);
957 J_ASSERT(handle != NULL);
960 * Now that we do not always journal data, we should
961 * keep in mind whether this should always journal the
962 * new buffer as metadata. For now, regular file
963 * writes use ext4_get_block instead, so it's not a
967 BUFFER_TRACE(bh, "call get_create_access");
968 err = ext4_journal_get_create_access(handle, bh);
973 if (!buffer_uptodate(bh)) {
974 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
975 set_buffer_uptodate(bh);
978 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
979 err = ext4_handle_dirty_metadata(handle, inode, bh);
983 BUFFER_TRACE(bh, "not a new buffer");
990 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
991 ext4_lblk_t block, int map_flags)
993 struct buffer_head *bh;
995 bh = ext4_getblk(handle, inode, block, map_flags);
998 if (!bh || buffer_uptodate(bh))
1000 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1002 if (buffer_uptodate(bh))
1005 return ERR_PTR(-EIO);
1008 int ext4_walk_page_buffers(handle_t *handle,
1009 struct buffer_head *head,
1013 int (*fn)(handle_t *handle,
1014 struct buffer_head *bh))
1016 struct buffer_head *bh;
1017 unsigned block_start, block_end;
1018 unsigned blocksize = head->b_size;
1020 struct buffer_head *next;
1022 for (bh = head, block_start = 0;
1023 ret == 0 && (bh != head || !block_start);
1024 block_start = block_end, bh = next) {
1025 next = bh->b_this_page;
1026 block_end = block_start + blocksize;
1027 if (block_end <= from || block_start >= to) {
1028 if (partial && !buffer_uptodate(bh))
1032 err = (*fn)(handle, bh);
1040 * To preserve ordering, it is essential that the hole instantiation and
1041 * the data write be encapsulated in a single transaction. We cannot
1042 * close off a transaction and start a new one between the ext4_get_block()
1043 * and the commit_write(). So doing the jbd2_journal_start at the start of
1044 * prepare_write() is the right place.
1046 * Also, this function can nest inside ext4_writepage(). In that case, we
1047 * *know* that ext4_writepage() has generated enough buffer credits to do the
1048 * whole page. So we won't block on the journal in that case, which is good,
1049 * because the caller may be PF_MEMALLOC.
1051 * By accident, ext4 can be reentered when a transaction is open via
1052 * quota file writes. If we were to commit the transaction while thus
1053 * reentered, there can be a deadlock - we would be holding a quota
1054 * lock, and the commit would never complete if another thread had a
1055 * transaction open and was blocking on the quota lock - a ranking
1058 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1059 * will _not_ run commit under these circumstances because handle->h_ref
1060 * is elevated. We'll still have enough credits for the tiny quotafile
1063 int do_journal_get_write_access(handle_t *handle,
1064 struct buffer_head *bh)
1066 int dirty = buffer_dirty(bh);
1069 if (!buffer_mapped(bh) || buffer_freed(bh))
1072 * __block_write_begin() could have dirtied some buffers. Clean
1073 * the dirty bit as jbd2_journal_get_write_access() could complain
1074 * otherwise about fs integrity issues. Setting of the dirty bit
1075 * by __block_write_begin() isn't a real problem here as we clear
1076 * the bit before releasing a page lock and thus writeback cannot
1077 * ever write the buffer.
1080 clear_buffer_dirty(bh);
1081 BUFFER_TRACE(bh, "get write access");
1082 ret = ext4_journal_get_write_access(handle, bh);
1084 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1088 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1089 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1090 get_block_t *get_block)
1092 unsigned from = pos & (PAGE_SIZE - 1);
1093 unsigned to = from + len;
1094 struct inode *inode = page->mapping->host;
1095 unsigned block_start, block_end;
1098 unsigned blocksize = inode->i_sb->s_blocksize;
1100 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1101 bool decrypt = false;
1103 BUG_ON(!PageLocked(page));
1104 BUG_ON(from > PAGE_SIZE);
1105 BUG_ON(to > PAGE_SIZE);
1108 if (!page_has_buffers(page))
1109 create_empty_buffers(page, blocksize, 0);
1110 head = page_buffers(page);
1111 bbits = ilog2(blocksize);
1112 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1114 for (bh = head, block_start = 0; bh != head || !block_start;
1115 block++, block_start = block_end, bh = bh->b_this_page) {
1116 block_end = block_start + blocksize;
1117 if (block_end <= from || block_start >= to) {
1118 if (PageUptodate(page)) {
1119 if (!buffer_uptodate(bh))
1120 set_buffer_uptodate(bh);
1125 clear_buffer_new(bh);
1126 if (!buffer_mapped(bh)) {
1127 WARN_ON(bh->b_size != blocksize);
1128 err = get_block(inode, block, bh, 1);
1131 if (buffer_new(bh)) {
1132 unmap_underlying_metadata(bh->b_bdev,
1134 if (PageUptodate(page)) {
1135 clear_buffer_new(bh);
1136 set_buffer_uptodate(bh);
1137 mark_buffer_dirty(bh);
1140 if (block_end > to || block_start < from)
1141 zero_user_segments(page, to, block_end,
1146 if (PageUptodate(page)) {
1147 if (!buffer_uptodate(bh))
1148 set_buffer_uptodate(bh);
1151 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1152 !buffer_unwritten(bh) &&
1153 (block_start < from || block_end > to)) {
1154 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1156 decrypt = ext4_encrypted_inode(inode) &&
1157 S_ISREG(inode->i_mode);
1161 * If we issued read requests, let them complete.
1163 while (wait_bh > wait) {
1164 wait_on_buffer(*--wait_bh);
1165 if (!buffer_uptodate(*wait_bh))
1169 page_zero_new_buffers(page, from, to);
1171 err = fscrypt_decrypt_page(page);
1176 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1177 loff_t pos, unsigned len, unsigned flags,
1178 struct page **pagep, void **fsdata)
1180 struct inode *inode = mapping->host;
1181 int ret, needed_blocks;
1188 trace_ext4_write_begin(inode, pos, len, flags);
1190 * Reserve one block more for addition to orphan list in case
1191 * we allocate blocks but write fails for some reason
1193 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1194 index = pos >> PAGE_SHIFT;
1195 from = pos & (PAGE_SIZE - 1);
1198 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1199 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1208 * grab_cache_page_write_begin() can take a long time if the
1209 * system is thrashing due to memory pressure, or if the page
1210 * is being written back. So grab it first before we start
1211 * the transaction handle. This also allows us to allocate
1212 * the page (if needed) without using GFP_NOFS.
1215 page = grab_cache_page_write_begin(mapping, index, flags);
1219 * The same as page allocation, we prealloc buffer heads before
1220 * starting the handle.
1222 if (!page_has_buffers(page))
1223 create_empty_buffers(page, inode->i_sb->s_blocksize, 0);
1228 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1229 if (IS_ERR(handle)) {
1231 return PTR_ERR(handle);
1235 if (page->mapping != mapping) {
1236 /* The page got truncated from under us */
1239 ext4_journal_stop(handle);
1242 /* In case writeback began while the page was unlocked */
1243 wait_for_stable_page(page);
1245 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1246 if (ext4_should_dioread_nolock(inode))
1247 ret = ext4_block_write_begin(page, pos, len,
1248 ext4_get_block_unwritten);
1250 ret = ext4_block_write_begin(page, pos, len,
1253 if (ext4_should_dioread_nolock(inode))
1254 ret = __block_write_begin(page, pos, len,
1255 ext4_get_block_unwritten);
1257 ret = __block_write_begin(page, pos, len, ext4_get_block);
1259 if (!ret && ext4_should_journal_data(inode)) {
1260 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1262 do_journal_get_write_access);
1268 * __block_write_begin may have instantiated a few blocks
1269 * outside i_size. Trim these off again. Don't need
1270 * i_size_read because we hold i_mutex.
1272 * Add inode to orphan list in case we crash before
1275 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1276 ext4_orphan_add(handle, inode);
1278 ext4_journal_stop(handle);
1279 if (pos + len > inode->i_size) {
1280 ext4_truncate_failed_write(inode);
1282 * If truncate failed early the inode might
1283 * still be on the orphan list; we need to
1284 * make sure the inode is removed from the
1285 * orphan list in that case.
1288 ext4_orphan_del(NULL, inode);
1291 if (ret == -ENOSPC &&
1292 ext4_should_retry_alloc(inode->i_sb, &retries))
1301 /* For write_end() in data=journal mode */
1302 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1305 if (!buffer_mapped(bh) || buffer_freed(bh))
1307 set_buffer_uptodate(bh);
1308 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1309 clear_buffer_meta(bh);
1310 clear_buffer_prio(bh);
1315 * We need to pick up the new inode size which generic_commit_write gave us
1316 * `file' can be NULL - eg, when called from page_symlink().
1318 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1319 * buffers are managed internally.
1321 static int ext4_write_end(struct file *file,
1322 struct address_space *mapping,
1323 loff_t pos, unsigned len, unsigned copied,
1324 struct page *page, void *fsdata)
1326 handle_t *handle = ext4_journal_current_handle();
1327 struct inode *inode = mapping->host;
1328 loff_t old_size = inode->i_size;
1330 int i_size_changed = 0;
1331 int inline_data = ext4_has_inline_data(inode);
1333 trace_ext4_write_end(inode, pos, len, copied);
1335 ret = ext4_write_inline_data_end(inode, pos, len,
1344 copied = block_write_end(file, mapping, pos,
1345 len, copied, page, fsdata);
1347 * it's important to update i_size while still holding page lock:
1348 * page writeout could otherwise come in and zero beyond i_size.
1350 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1355 pagecache_isize_extended(inode, old_size, pos);
1357 * Don't mark the inode dirty under page lock. First, it unnecessarily
1358 * makes the holding time of page lock longer. Second, it forces lock
1359 * ordering of page lock and transaction start for journaling
1362 if (i_size_changed || inline_data)
1363 ext4_mark_inode_dirty(handle, inode);
1365 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1366 /* if we have allocated more blocks and copied
1367 * less. We will have blocks allocated outside
1368 * inode->i_size. So truncate them
1370 ext4_orphan_add(handle, inode);
1372 ret2 = ext4_journal_stop(handle);
1376 if (pos + len > inode->i_size) {
1377 ext4_truncate_failed_write(inode);
1379 * If truncate failed early the inode might still be
1380 * on the orphan list; we need to make sure the inode
1381 * is removed from the orphan list in that case.
1384 ext4_orphan_del(NULL, inode);
1387 return ret ? ret : copied;
1391 * This is a private version of page_zero_new_buffers() which doesn't
1392 * set the buffer to be dirty, since in data=journalled mode we need
1393 * to call ext4_handle_dirty_metadata() instead.
1395 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1397 unsigned from, unsigned to)
1399 unsigned int block_start = 0, block_end;
1400 struct buffer_head *head, *bh;
1402 bh = head = page_buffers(page);
1404 block_end = block_start + bh->b_size;
1405 if (buffer_new(bh)) {
1406 if (block_end > from && block_start < to) {
1407 if (!PageUptodate(page)) {
1408 unsigned start, size;
1410 start = max(from, block_start);
1411 size = min(to, block_end) - start;
1413 zero_user(page, start, size);
1414 write_end_fn(handle, bh);
1416 clear_buffer_new(bh);
1419 block_start = block_end;
1420 bh = bh->b_this_page;
1421 } while (bh != head);
1424 static int ext4_journalled_write_end(struct file *file,
1425 struct address_space *mapping,
1426 loff_t pos, unsigned len, unsigned copied,
1427 struct page *page, void *fsdata)
1429 handle_t *handle = ext4_journal_current_handle();
1430 struct inode *inode = mapping->host;
1431 loff_t old_size = inode->i_size;
1435 int size_changed = 0;
1436 int inline_data = ext4_has_inline_data(inode);
1438 trace_ext4_journalled_write_end(inode, pos, len, copied);
1439 from = pos & (PAGE_SIZE - 1);
1442 BUG_ON(!ext4_handle_valid(handle));
1445 ret = ext4_write_inline_data_end(inode, pos, len,
1453 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1455 ext4_journalled_zero_new_buffers(handle, page, from, to);
1457 if (unlikely(copied < len))
1458 ext4_journalled_zero_new_buffers(handle, page,
1460 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1461 from + copied, &partial,
1464 SetPageUptodate(page);
1466 size_changed = ext4_update_inode_size(inode, pos + copied);
1467 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1468 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1473 pagecache_isize_extended(inode, old_size, pos);
1475 if (size_changed || inline_data) {
1476 ret2 = ext4_mark_inode_dirty(handle, inode);
1481 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1482 /* if we have allocated more blocks and copied
1483 * less. We will have blocks allocated outside
1484 * inode->i_size. So truncate them
1486 ext4_orphan_add(handle, inode);
1489 ret2 = ext4_journal_stop(handle);
1492 if (pos + len > inode->i_size) {
1493 ext4_truncate_failed_write(inode);
1495 * If truncate failed early the inode might still be
1496 * on the orphan list; we need to make sure the inode
1497 * is removed from the orphan list in that case.
1500 ext4_orphan_del(NULL, inode);
1503 return ret ? ret : copied;
1507 * Reserve space for a single cluster
1509 static int ext4_da_reserve_space(struct inode *inode)
1511 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1512 struct ext4_inode_info *ei = EXT4_I(inode);
1516 * We will charge metadata quota at writeout time; this saves
1517 * us from metadata over-estimation, though we may go over by
1518 * a small amount in the end. Here we just reserve for data.
1520 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1524 spin_lock(&ei->i_block_reservation_lock);
1525 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1526 spin_unlock(&ei->i_block_reservation_lock);
1527 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1530 ei->i_reserved_data_blocks++;
1531 trace_ext4_da_reserve_space(inode);
1532 spin_unlock(&ei->i_block_reservation_lock);
1534 return 0; /* success */
1537 static void ext4_da_release_space(struct inode *inode, int to_free)
1539 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1540 struct ext4_inode_info *ei = EXT4_I(inode);
1543 return; /* Nothing to release, exit */
1545 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1547 trace_ext4_da_release_space(inode, to_free);
1548 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1550 * if there aren't enough reserved blocks, then the
1551 * counter is messed up somewhere. Since this
1552 * function is called from invalidate page, it's
1553 * harmless to return without any action.
1555 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1556 "ino %lu, to_free %d with only %d reserved "
1557 "data blocks", inode->i_ino, to_free,
1558 ei->i_reserved_data_blocks);
1560 to_free = ei->i_reserved_data_blocks;
1562 ei->i_reserved_data_blocks -= to_free;
1564 /* update fs dirty data blocks counter */
1565 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1567 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1569 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1572 static void ext4_da_page_release_reservation(struct page *page,
1573 unsigned int offset,
1574 unsigned int length)
1576 int to_release = 0, contiguous_blks = 0;
1577 struct buffer_head *head, *bh;
1578 unsigned int curr_off = 0;
1579 struct inode *inode = page->mapping->host;
1580 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1581 unsigned int stop = offset + length;
1585 BUG_ON(stop > PAGE_SIZE || stop < length);
1587 head = page_buffers(page);
1590 unsigned int next_off = curr_off + bh->b_size;
1592 if (next_off > stop)
1595 if ((offset <= curr_off) && (buffer_delay(bh))) {
1598 clear_buffer_delay(bh);
1599 } else if (contiguous_blks) {
1600 lblk = page->index <<
1601 (PAGE_SHIFT - inode->i_blkbits);
1602 lblk += (curr_off >> inode->i_blkbits) -
1604 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1605 contiguous_blks = 0;
1607 curr_off = next_off;
1608 } while ((bh = bh->b_this_page) != head);
1610 if (contiguous_blks) {
1611 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1612 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1613 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1616 /* If we have released all the blocks belonging to a cluster, then we
1617 * need to release the reserved space for that cluster. */
1618 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1619 while (num_clusters > 0) {
1620 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1621 ((num_clusters - 1) << sbi->s_cluster_bits);
1622 if (sbi->s_cluster_ratio == 1 ||
1623 !ext4_find_delalloc_cluster(inode, lblk))
1624 ext4_da_release_space(inode, 1);
1631 * Delayed allocation stuff
1634 struct mpage_da_data {
1635 struct inode *inode;
1636 struct writeback_control *wbc;
1638 pgoff_t first_page; /* The first page to write */
1639 pgoff_t next_page; /* Current page to examine */
1640 pgoff_t last_page; /* Last page to examine */
1642 * Extent to map - this can be after first_page because that can be
1643 * fully mapped. We somewhat abuse m_flags to store whether the extent
1644 * is delalloc or unwritten.
1646 struct ext4_map_blocks map;
1647 struct ext4_io_submit io_submit; /* IO submission data */
1650 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1655 struct pagevec pvec;
1656 struct inode *inode = mpd->inode;
1657 struct address_space *mapping = inode->i_mapping;
1659 /* This is necessary when next_page == 0. */
1660 if (mpd->first_page >= mpd->next_page)
1663 index = mpd->first_page;
1664 end = mpd->next_page - 1;
1666 ext4_lblk_t start, last;
1667 start = index << (PAGE_SHIFT - inode->i_blkbits);
1668 last = end << (PAGE_SHIFT - inode->i_blkbits);
1671 * avoid racing with extent status tree scans made by
1672 * ext4_insert_delayed_block()
1674 down_write(&EXT4_I(inode)->i_data_sem);
1675 ext4_es_remove_extent(inode, start, last - start + 1);
1676 up_write(&EXT4_I(inode)->i_data_sem);
1679 pagevec_init(&pvec, 0);
1680 while (index <= end) {
1681 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1684 for (i = 0; i < nr_pages; i++) {
1685 struct page *page = pvec.pages[i];
1686 if (page->index > end)
1688 BUG_ON(!PageLocked(page));
1689 BUG_ON(PageWriteback(page));
1691 if (page_mapped(page))
1692 clear_page_dirty_for_io(page);
1693 block_invalidatepage(page, 0, PAGE_SIZE);
1694 ClearPageUptodate(page);
1698 index = pvec.pages[nr_pages - 1]->index + 1;
1699 pagevec_release(&pvec);
1703 static void ext4_print_free_blocks(struct inode *inode)
1705 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1706 struct super_block *sb = inode->i_sb;
1707 struct ext4_inode_info *ei = EXT4_I(inode);
1709 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1710 EXT4_C2B(EXT4_SB(inode->i_sb),
1711 ext4_count_free_clusters(sb)));
1712 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1713 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1714 (long long) EXT4_C2B(EXT4_SB(sb),
1715 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1716 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1717 (long long) EXT4_C2B(EXT4_SB(sb),
1718 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1719 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1720 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1721 ei->i_reserved_data_blocks);
1725 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1727 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1731 * This function is grabs code from the very beginning of
1732 * ext4_map_blocks, but assumes that the caller is from delayed write
1733 * time. This function looks up the requested blocks and sets the
1734 * buffer delay bit under the protection of i_data_sem.
1736 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1737 struct ext4_map_blocks *map,
1738 struct buffer_head *bh)
1740 struct extent_status es;
1742 sector_t invalid_block = ~((sector_t) 0xffff);
1743 #ifdef ES_AGGRESSIVE_TEST
1744 struct ext4_map_blocks orig_map;
1746 memcpy(&orig_map, map, sizeof(*map));
1749 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1753 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1754 "logical block %lu\n", inode->i_ino, map->m_len,
1755 (unsigned long) map->m_lblk);
1757 /* Lookup extent status tree firstly */
1758 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1759 if (ext4_es_is_hole(&es)) {
1761 down_read(&EXT4_I(inode)->i_data_sem);
1766 * Delayed extent could be allocated by fallocate.
1767 * So we need to check it.
1769 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1770 map_bh(bh, inode->i_sb, invalid_block);
1772 set_buffer_delay(bh);
1776 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1777 retval = es.es_len - (iblock - es.es_lblk);
1778 if (retval > map->m_len)
1779 retval = map->m_len;
1780 map->m_len = retval;
1781 if (ext4_es_is_written(&es))
1782 map->m_flags |= EXT4_MAP_MAPPED;
1783 else if (ext4_es_is_unwritten(&es))
1784 map->m_flags |= EXT4_MAP_UNWRITTEN;
1788 #ifdef ES_AGGRESSIVE_TEST
1789 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1795 * Try to see if we can get the block without requesting a new
1796 * file system block.
1798 down_read(&EXT4_I(inode)->i_data_sem);
1799 if (ext4_has_inline_data(inode))
1801 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1802 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1804 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1810 * XXX: __block_prepare_write() unmaps passed block,
1814 * If the block was allocated from previously allocated cluster,
1815 * then we don't need to reserve it again. However we still need
1816 * to reserve metadata for every block we're going to write.
1818 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1819 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1820 ret = ext4_da_reserve_space(inode);
1822 /* not enough space to reserve */
1828 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1829 ~0, EXTENT_STATUS_DELAYED);
1835 map_bh(bh, inode->i_sb, invalid_block);
1837 set_buffer_delay(bh);
1838 } else if (retval > 0) {
1840 unsigned int status;
1842 if (unlikely(retval != map->m_len)) {
1843 ext4_warning(inode->i_sb,
1844 "ES len assertion failed for inode "
1845 "%lu: retval %d != map->m_len %d",
1846 inode->i_ino, retval, map->m_len);
1850 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1851 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1852 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1853 map->m_pblk, status);
1859 up_read((&EXT4_I(inode)->i_data_sem));
1865 * This is a special get_block_t callback which is used by
1866 * ext4_da_write_begin(). It will either return mapped block or
1867 * reserve space for a single block.
1869 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1870 * We also have b_blocknr = -1 and b_bdev initialized properly
1872 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1873 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1874 * initialized properly.
1876 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1877 struct buffer_head *bh, int create)
1879 struct ext4_map_blocks map;
1882 BUG_ON(create == 0);
1883 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1885 map.m_lblk = iblock;
1889 * first, we need to know whether the block is allocated already
1890 * preallocated blocks are unmapped but should treated
1891 * the same as allocated blocks.
1893 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1897 map_bh(bh, inode->i_sb, map.m_pblk);
1898 ext4_update_bh_state(bh, map.m_flags);
1900 if (buffer_unwritten(bh)) {
1901 /* A delayed write to unwritten bh should be marked
1902 * new and mapped. Mapped ensures that we don't do
1903 * get_block multiple times when we write to the same
1904 * offset and new ensures that we do proper zero out
1905 * for partial write.
1908 set_buffer_mapped(bh);
1913 static int bget_one(handle_t *handle, struct buffer_head *bh)
1919 static int bput_one(handle_t *handle, struct buffer_head *bh)
1925 static int __ext4_journalled_writepage(struct page *page,
1928 struct address_space *mapping = page->mapping;
1929 struct inode *inode = mapping->host;
1930 struct buffer_head *page_bufs = NULL;
1931 handle_t *handle = NULL;
1932 int ret = 0, err = 0;
1933 int inline_data = ext4_has_inline_data(inode);
1934 struct buffer_head *inode_bh = NULL;
1936 ClearPageChecked(page);
1939 BUG_ON(page->index != 0);
1940 BUG_ON(len > ext4_get_max_inline_size(inode));
1941 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1942 if (inode_bh == NULL)
1945 page_bufs = page_buffers(page);
1950 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1954 * We need to release the page lock before we start the
1955 * journal, so grab a reference so the page won't disappear
1956 * out from under us.
1961 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1962 ext4_writepage_trans_blocks(inode));
1963 if (IS_ERR(handle)) {
1964 ret = PTR_ERR(handle);
1966 goto out_no_pagelock;
1968 BUG_ON(!ext4_handle_valid(handle));
1972 if (page->mapping != mapping) {
1973 /* The page got truncated from under us */
1974 ext4_journal_stop(handle);
1980 ret = ext4_mark_inode_dirty(handle, inode);
1982 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1983 do_journal_get_write_access);
1985 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1990 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1991 err = ext4_journal_stop(handle);
1995 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1999 if (!inline_data && page_bufs)
2000 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2007 * Note that we don't need to start a transaction unless we're journaling data
2008 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2009 * need to file the inode to the transaction's list in ordered mode because if
2010 * we are writing back data added by write(), the inode is already there and if
2011 * we are writing back data modified via mmap(), no one guarantees in which
2012 * transaction the data will hit the disk. In case we are journaling data, we
2013 * cannot start transaction directly because transaction start ranks above page
2014 * lock so we have to do some magic.
2016 * This function can get called via...
2017 * - ext4_writepages after taking page lock (have journal handle)
2018 * - journal_submit_inode_data_buffers (no journal handle)
2019 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2020 * - grab_page_cache when doing write_begin (have journal handle)
2022 * We don't do any block allocation in this function. If we have page with
2023 * multiple blocks we need to write those buffer_heads that are mapped. This
2024 * is important for mmaped based write. So if we do with blocksize 1K
2025 * truncate(f, 1024);
2026 * a = mmap(f, 0, 4096);
2028 * truncate(f, 4096);
2029 * we have in the page first buffer_head mapped via page_mkwrite call back
2030 * but other buffer_heads would be unmapped but dirty (dirty done via the
2031 * do_wp_page). So writepage should write the first block. If we modify
2032 * the mmap area beyond 1024 we will again get a page_fault and the
2033 * page_mkwrite callback will do the block allocation and mark the
2034 * buffer_heads mapped.
2036 * We redirty the page if we have any buffer_heads that is either delay or
2037 * unwritten in the page.
2039 * We can get recursively called as show below.
2041 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2044 * But since we don't do any block allocation we should not deadlock.
2045 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2047 static int ext4_writepage(struct page *page,
2048 struct writeback_control *wbc)
2053 struct buffer_head *page_bufs = NULL;
2054 struct inode *inode = page->mapping->host;
2055 struct ext4_io_submit io_submit;
2056 bool keep_towrite = false;
2058 trace_ext4_writepage(page);
2059 size = i_size_read(inode);
2060 if (page->index == size >> PAGE_SHIFT)
2061 len = size & ~PAGE_MASK;
2065 /* Should never happen but for bugs in other kernel subsystems */
2066 if (!page_has_buffers(page)) {
2067 ext4_warning_inode(inode,
2068 "page %lu does not have buffers attached", page->index);
2069 ClearPageDirty(page);
2074 page_bufs = page_buffers(page);
2076 * We cannot do block allocation or other extent handling in this
2077 * function. If there are buffers needing that, we have to redirty
2078 * the page. But we may reach here when we do a journal commit via
2079 * journal_submit_inode_data_buffers() and in that case we must write
2080 * allocated buffers to achieve data=ordered mode guarantees.
2082 * Also, if there is only one buffer per page (the fs block
2083 * size == the page size), if one buffer needs block
2084 * allocation or needs to modify the extent tree to clear the
2085 * unwritten flag, we know that the page can't be written at
2086 * all, so we might as well refuse the write immediately.
2087 * Unfortunately if the block size != page size, we can't as
2088 * easily detect this case using ext4_walk_page_buffers(), but
2089 * for the extremely common case, this is an optimization that
2090 * skips a useless round trip through ext4_bio_write_page().
2092 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2093 ext4_bh_delay_or_unwritten)) {
2094 redirty_page_for_writepage(wbc, page);
2095 if ((current->flags & PF_MEMALLOC) ||
2096 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2098 * For memory cleaning there's no point in writing only
2099 * some buffers. So just bail out. Warn if we came here
2100 * from direct reclaim.
2102 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2107 keep_towrite = true;
2110 if (PageChecked(page) && ext4_should_journal_data(inode))
2112 * It's mmapped pagecache. Add buffers and journal it. There
2113 * doesn't seem much point in redirtying the page here.
2115 return __ext4_journalled_writepage(page, len);
2117 ext4_io_submit_init(&io_submit, wbc);
2118 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2119 if (!io_submit.io_end) {
2120 redirty_page_for_writepage(wbc, page);
2124 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2125 ext4_io_submit(&io_submit);
2126 /* Drop io_end reference we got from init */
2127 ext4_put_io_end_defer(io_submit.io_end);
2131 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2137 BUG_ON(page->index != mpd->first_page);
2138 clear_page_dirty_for_io(page);
2140 * We have to be very careful here! Nothing protects writeback path
2141 * against i_size changes and the page can be writeably mapped into
2142 * page tables. So an application can be growing i_size and writing
2143 * data through mmap while writeback runs. clear_page_dirty_for_io()
2144 * write-protects our page in page tables and the page cannot get
2145 * written to again until we release page lock. So only after
2146 * clear_page_dirty_for_io() we are safe to sample i_size for
2147 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2148 * on the barrier provided by TestClearPageDirty in
2149 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2150 * after page tables are updated.
2152 size = i_size_read(mpd->inode);
2153 if (page->index == size >> PAGE_SHIFT)
2154 len = size & ~PAGE_MASK;
2157 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2159 mpd->wbc->nr_to_write--;
2165 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2168 * mballoc gives us at most this number of blocks...
2169 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2170 * The rest of mballoc seems to handle chunks up to full group size.
2172 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2175 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2177 * @mpd - extent of blocks
2178 * @lblk - logical number of the block in the file
2179 * @bh - buffer head we want to add to the extent
2181 * The function is used to collect contig. blocks in the same state. If the
2182 * buffer doesn't require mapping for writeback and we haven't started the
2183 * extent of buffers to map yet, the function returns 'true' immediately - the
2184 * caller can write the buffer right away. Otherwise the function returns true
2185 * if the block has been added to the extent, false if the block couldn't be
2188 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2189 struct buffer_head *bh)
2191 struct ext4_map_blocks *map = &mpd->map;
2193 /* Buffer that doesn't need mapping for writeback? */
2194 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2195 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2196 /* So far no extent to map => we write the buffer right away */
2197 if (map->m_len == 0)
2202 /* First block in the extent? */
2203 if (map->m_len == 0) {
2206 map->m_flags = bh->b_state & BH_FLAGS;
2210 /* Don't go larger than mballoc is willing to allocate */
2211 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2214 /* Can we merge the block to our big extent? */
2215 if (lblk == map->m_lblk + map->m_len &&
2216 (bh->b_state & BH_FLAGS) == map->m_flags) {
2224 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2226 * @mpd - extent of blocks for mapping
2227 * @head - the first buffer in the page
2228 * @bh - buffer we should start processing from
2229 * @lblk - logical number of the block in the file corresponding to @bh
2231 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2232 * the page for IO if all buffers in this page were mapped and there's no
2233 * accumulated extent of buffers to map or add buffers in the page to the
2234 * extent of buffers to map. The function returns 1 if the caller can continue
2235 * by processing the next page, 0 if it should stop adding buffers to the
2236 * extent to map because we cannot extend it anymore. It can also return value
2237 * < 0 in case of error during IO submission.
2239 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2240 struct buffer_head *head,
2241 struct buffer_head *bh,
2244 struct inode *inode = mpd->inode;
2246 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2247 >> inode->i_blkbits;
2250 BUG_ON(buffer_locked(bh));
2252 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2253 /* Found extent to map? */
2256 /* Everything mapped so far and we hit EOF */
2259 } while (lblk++, (bh = bh->b_this_page) != head);
2260 /* So far everything mapped? Submit the page for IO. */
2261 if (mpd->map.m_len == 0) {
2262 err = mpage_submit_page(mpd, head->b_page);
2266 return lblk < blocks;
2270 * mpage_map_buffers - update buffers corresponding to changed extent and
2271 * submit fully mapped pages for IO
2273 * @mpd - description of extent to map, on return next extent to map
2275 * Scan buffers corresponding to changed extent (we expect corresponding pages
2276 * to be already locked) and update buffer state according to new extent state.
2277 * We map delalloc buffers to their physical location, clear unwritten bits,
2278 * and mark buffers as uninit when we perform writes to unwritten extents
2279 * and do extent conversion after IO is finished. If the last page is not fully
2280 * mapped, we update @map to the next extent in the last page that needs
2281 * mapping. Otherwise we submit the page for IO.
2283 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2285 struct pagevec pvec;
2287 struct inode *inode = mpd->inode;
2288 struct buffer_head *head, *bh;
2289 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2295 start = mpd->map.m_lblk >> bpp_bits;
2296 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2297 lblk = start << bpp_bits;
2298 pblock = mpd->map.m_pblk;
2300 pagevec_init(&pvec, 0);
2301 while (start <= end) {
2302 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2306 for (i = 0; i < nr_pages; i++) {
2307 struct page *page = pvec.pages[i];
2309 if (page->index > end)
2311 /* Up to 'end' pages must be contiguous */
2312 BUG_ON(page->index != start);
2313 bh = head = page_buffers(page);
2315 if (lblk < mpd->map.m_lblk)
2317 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2319 * Buffer after end of mapped extent.
2320 * Find next buffer in the page to map.
2323 mpd->map.m_flags = 0;
2325 * FIXME: If dioread_nolock supports
2326 * blocksize < pagesize, we need to make
2327 * sure we add size mapped so far to
2328 * io_end->size as the following call
2329 * can submit the page for IO.
2331 err = mpage_process_page_bufs(mpd, head,
2333 pagevec_release(&pvec);
2338 if (buffer_delay(bh)) {
2339 clear_buffer_delay(bh);
2340 bh->b_blocknr = pblock++;
2342 clear_buffer_unwritten(bh);
2343 } while (lblk++, (bh = bh->b_this_page) != head);
2346 * FIXME: This is going to break if dioread_nolock
2347 * supports blocksize < pagesize as we will try to
2348 * convert potentially unmapped parts of inode.
2350 mpd->io_submit.io_end->size += PAGE_SIZE;
2351 /* Page fully mapped - let IO run! */
2352 err = mpage_submit_page(mpd, page);
2354 pagevec_release(&pvec);
2359 pagevec_release(&pvec);
2361 /* Extent fully mapped and matches with page boundary. We are done. */
2363 mpd->map.m_flags = 0;
2367 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2369 struct inode *inode = mpd->inode;
2370 struct ext4_map_blocks *map = &mpd->map;
2371 int get_blocks_flags;
2372 int err, dioread_nolock;
2374 trace_ext4_da_write_pages_extent(inode, map);
2376 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2377 * to convert an unwritten extent to be initialized (in the case
2378 * where we have written into one or more preallocated blocks). It is
2379 * possible that we're going to need more metadata blocks than
2380 * previously reserved. However we must not fail because we're in
2381 * writeback and there is nothing we can do about it so it might result
2382 * in data loss. So use reserved blocks to allocate metadata if
2385 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2386 * the blocks in question are delalloc blocks. This indicates
2387 * that the blocks and quotas has already been checked when
2388 * the data was copied into the page cache.
2390 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2391 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2392 EXT4_GET_BLOCKS_IO_SUBMIT;
2393 dioread_nolock = ext4_should_dioread_nolock(inode);
2395 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2396 if (map->m_flags & (1 << BH_Delay))
2397 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2399 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2402 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2403 if (!mpd->io_submit.io_end->handle &&
2404 ext4_handle_valid(handle)) {
2405 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2406 handle->h_rsv_handle = NULL;
2408 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2411 BUG_ON(map->m_len == 0);
2412 if (map->m_flags & EXT4_MAP_NEW) {
2413 struct block_device *bdev = inode->i_sb->s_bdev;
2416 for (i = 0; i < map->m_len; i++)
2417 unmap_underlying_metadata(bdev, map->m_pblk + i);
2423 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2424 * mpd->len and submit pages underlying it for IO
2426 * @handle - handle for journal operations
2427 * @mpd - extent to map
2428 * @give_up_on_write - we set this to true iff there is a fatal error and there
2429 * is no hope of writing the data. The caller should discard
2430 * dirty pages to avoid infinite loops.
2432 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2433 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2434 * them to initialized or split the described range from larger unwritten
2435 * extent. Note that we need not map all the described range since allocation
2436 * can return less blocks or the range is covered by more unwritten extents. We
2437 * cannot map more because we are limited by reserved transaction credits. On
2438 * the other hand we always make sure that the last touched page is fully
2439 * mapped so that it can be written out (and thus forward progress is
2440 * guaranteed). After mapping we submit all mapped pages for IO.
2442 static int mpage_map_and_submit_extent(handle_t *handle,
2443 struct mpage_da_data *mpd,
2444 bool *give_up_on_write)
2446 struct inode *inode = mpd->inode;
2447 struct ext4_map_blocks *map = &mpd->map;
2452 mpd->io_submit.io_end->offset =
2453 ((loff_t)map->m_lblk) << inode->i_blkbits;
2455 err = mpage_map_one_extent(handle, mpd);
2457 struct super_block *sb = inode->i_sb;
2459 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2460 goto invalidate_dirty_pages;
2462 * Let the uper layers retry transient errors.
2463 * In the case of ENOSPC, if ext4_count_free_blocks()
2464 * is non-zero, a commit should free up blocks.
2466 if ((err == -ENOMEM) ||
2467 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2469 goto update_disksize;
2472 ext4_msg(sb, KERN_CRIT,
2473 "Delayed block allocation failed for "
2474 "inode %lu at logical offset %llu with"
2475 " max blocks %u with error %d",
2477 (unsigned long long)map->m_lblk,
2478 (unsigned)map->m_len, -err);
2479 ext4_msg(sb, KERN_CRIT,
2480 "This should not happen!! Data will "
2483 ext4_print_free_blocks(inode);
2484 invalidate_dirty_pages:
2485 *give_up_on_write = true;
2490 * Update buffer state, submit mapped pages, and get us new
2493 err = mpage_map_and_submit_buffers(mpd);
2495 goto update_disksize;
2496 } while (map->m_len);
2500 * Update on-disk size after IO is submitted. Races with
2501 * truncate are avoided by checking i_size under i_data_sem.
2503 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2504 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2508 down_write(&EXT4_I(inode)->i_data_sem);
2509 i_size = i_size_read(inode);
2510 if (disksize > i_size)
2512 if (disksize > EXT4_I(inode)->i_disksize)
2513 EXT4_I(inode)->i_disksize = disksize;
2514 err2 = ext4_mark_inode_dirty(handle, inode);
2515 up_write(&EXT4_I(inode)->i_data_sem);
2517 ext4_error(inode->i_sb,
2518 "Failed to mark inode %lu dirty",
2527 * Calculate the total number of credits to reserve for one writepages
2528 * iteration. This is called from ext4_writepages(). We map an extent of
2529 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2530 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2531 * bpp - 1 blocks in bpp different extents.
2533 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2535 int bpp = ext4_journal_blocks_per_page(inode);
2537 return ext4_meta_trans_blocks(inode,
2538 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2542 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2543 * and underlying extent to map
2545 * @mpd - where to look for pages
2547 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2548 * IO immediately. When we find a page which isn't mapped we start accumulating
2549 * extent of buffers underlying these pages that needs mapping (formed by
2550 * either delayed or unwritten buffers). We also lock the pages containing
2551 * these buffers. The extent found is returned in @mpd structure (starting at
2552 * mpd->lblk with length mpd->len blocks).
2554 * Note that this function can attach bios to one io_end structure which are
2555 * neither logically nor physically contiguous. Although it may seem as an
2556 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2557 * case as we need to track IO to all buffers underlying a page in one io_end.
2559 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2561 struct address_space *mapping = mpd->inode->i_mapping;
2562 struct pagevec pvec;
2563 unsigned int nr_pages;
2564 long left = mpd->wbc->nr_to_write;
2565 pgoff_t index = mpd->first_page;
2566 pgoff_t end = mpd->last_page;
2569 int blkbits = mpd->inode->i_blkbits;
2571 struct buffer_head *head;
2573 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2574 tag = PAGECACHE_TAG_TOWRITE;
2576 tag = PAGECACHE_TAG_DIRTY;
2578 pagevec_init(&pvec, 0);
2580 mpd->next_page = index;
2581 while (index <= end) {
2582 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2583 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2587 for (i = 0; i < nr_pages; i++) {
2588 struct page *page = pvec.pages[i];
2591 * At this point, the page may be truncated or
2592 * invalidated (changing page->mapping to NULL), or
2593 * even swizzled back from swapper_space to tmpfs file
2594 * mapping. However, page->index will not change
2595 * because we have a reference on the page.
2597 if (page->index > end)
2601 * Accumulated enough dirty pages? This doesn't apply
2602 * to WB_SYNC_ALL mode. For integrity sync we have to
2603 * keep going because someone may be concurrently
2604 * dirtying pages, and we might have synced a lot of
2605 * newly appeared dirty pages, but have not synced all
2606 * of the old dirty pages.
2608 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2611 /* If we can't merge this page, we are done. */
2612 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2617 * If the page is no longer dirty, or its mapping no
2618 * longer corresponds to inode we are writing (which
2619 * means it has been truncated or invalidated), or the
2620 * page is already under writeback and we are not doing
2621 * a data integrity writeback, skip the page
2623 if (!PageDirty(page) ||
2624 (PageWriteback(page) &&
2625 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2626 unlikely(page->mapping != mapping)) {
2631 wait_on_page_writeback(page);
2632 BUG_ON(PageWriteback(page));
2635 * Should never happen but for buggy code in
2636 * other subsystems that call
2637 * set_page_dirty() without properly warning
2638 * the file system first. See [1] for more
2641 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2643 if (!page_has_buffers(page)) {
2644 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2645 ClearPageDirty(page);
2650 if (mpd->map.m_len == 0)
2651 mpd->first_page = page->index;
2652 mpd->next_page = page->index + 1;
2653 /* Add all dirty buffers to mpd */
2654 lblk = ((ext4_lblk_t)page->index) <<
2655 (PAGE_SHIFT - blkbits);
2656 head = page_buffers(page);
2657 err = mpage_process_page_bufs(mpd, head, head, lblk);
2663 pagevec_release(&pvec);
2668 pagevec_release(&pvec);
2672 static int __writepage(struct page *page, struct writeback_control *wbc,
2675 struct address_space *mapping = data;
2676 int ret = ext4_writepage(page, wbc);
2677 mapping_set_error(mapping, ret);
2681 static int ext4_writepages(struct address_space *mapping,
2682 struct writeback_control *wbc)
2684 pgoff_t writeback_index = 0;
2685 long nr_to_write = wbc->nr_to_write;
2686 int range_whole = 0;
2688 handle_t *handle = NULL;
2689 struct mpage_da_data mpd;
2690 struct inode *inode = mapping->host;
2691 int needed_blocks, rsv_blocks = 0, ret = 0;
2692 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2694 struct blk_plug plug;
2695 bool give_up_on_write = false;
2697 percpu_down_read(&sbi->s_writepages_rwsem);
2698 trace_ext4_writepages(inode, wbc);
2700 if (dax_mapping(mapping)) {
2701 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2703 goto out_writepages;
2707 * No pages to write? This is mainly a kludge to avoid starting
2708 * a transaction for special inodes like journal inode on last iput()
2709 * because that could violate lock ordering on umount
2711 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2712 goto out_writepages;
2714 if (ext4_should_journal_data(inode)) {
2715 struct blk_plug plug;
2717 blk_start_plug(&plug);
2718 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2719 blk_finish_plug(&plug);
2720 goto out_writepages;
2724 * If the filesystem has aborted, it is read-only, so return
2725 * right away instead of dumping stack traces later on that
2726 * will obscure the real source of the problem. We test
2727 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2728 * the latter could be true if the filesystem is mounted
2729 * read-only, and in that case, ext4_writepages should
2730 * *never* be called, so if that ever happens, we would want
2733 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2735 goto out_writepages;
2738 if (ext4_should_dioread_nolock(inode)) {
2740 * We may need to convert up to one extent per block in
2741 * the page and we may dirty the inode.
2743 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2744 PAGE_SIZE >> inode->i_blkbits);
2748 * If we have inline data and arrive here, it means that
2749 * we will soon create the block for the 1st page, so
2750 * we'd better clear the inline data here.
2752 if (ext4_has_inline_data(inode)) {
2753 /* Just inode will be modified... */
2754 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2755 if (IS_ERR(handle)) {
2756 ret = PTR_ERR(handle);
2757 goto out_writepages;
2759 BUG_ON(ext4_test_inode_state(inode,
2760 EXT4_STATE_MAY_INLINE_DATA));
2761 ext4_destroy_inline_data(handle, inode);
2762 ext4_journal_stop(handle);
2765 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2768 if (wbc->range_cyclic) {
2769 writeback_index = mapping->writeback_index;
2770 if (writeback_index)
2772 mpd.first_page = writeback_index;
2775 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2776 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2781 ext4_io_submit_init(&mpd.io_submit, wbc);
2783 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2784 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2786 blk_start_plug(&plug);
2787 while (!done && mpd.first_page <= mpd.last_page) {
2788 /* For each extent of pages we use new io_end */
2789 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2790 if (!mpd.io_submit.io_end) {
2796 * We have two constraints: We find one extent to map and we
2797 * must always write out whole page (makes a difference when
2798 * blocksize < pagesize) so that we don't block on IO when we
2799 * try to write out the rest of the page. Journalled mode is
2800 * not supported by delalloc.
2802 BUG_ON(ext4_should_journal_data(inode));
2803 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2805 /* start a new transaction */
2806 handle = ext4_journal_start_with_reserve(inode,
2807 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2808 if (IS_ERR(handle)) {
2809 ret = PTR_ERR(handle);
2810 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2811 "%ld pages, ino %lu; err %d", __func__,
2812 wbc->nr_to_write, inode->i_ino, ret);
2813 /* Release allocated io_end */
2814 ext4_put_io_end(mpd.io_submit.io_end);
2818 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2819 ret = mpage_prepare_extent_to_map(&mpd);
2822 ret = mpage_map_and_submit_extent(handle, &mpd,
2826 * We scanned the whole range (or exhausted
2827 * nr_to_write), submitted what was mapped and
2828 * didn't find anything needing mapping. We are
2835 * Caution: If the handle is synchronous,
2836 * ext4_journal_stop() can wait for transaction commit
2837 * to finish which may depend on writeback of pages to
2838 * complete or on page lock to be released. In that
2839 * case, we have to wait until after after we have
2840 * submitted all the IO, released page locks we hold,
2841 * and dropped io_end reference (for extent conversion
2842 * to be able to complete) before stopping the handle.
2844 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2845 ext4_journal_stop(handle);
2848 /* Submit prepared bio */
2849 ext4_io_submit(&mpd.io_submit);
2850 /* Unlock pages we didn't use */
2851 mpage_release_unused_pages(&mpd, give_up_on_write);
2853 * Drop our io_end reference we got from init. We have
2854 * to be careful and use deferred io_end finishing if
2855 * we are still holding the transaction as we can
2856 * release the last reference to io_end which may end
2857 * up doing unwritten extent conversion.
2860 ext4_put_io_end_defer(mpd.io_submit.io_end);
2861 ext4_journal_stop(handle);
2863 ext4_put_io_end(mpd.io_submit.io_end);
2865 if (ret == -ENOSPC && sbi->s_journal) {
2867 * Commit the transaction which would
2868 * free blocks released in the transaction
2871 jbd2_journal_force_commit_nested(sbi->s_journal);
2875 /* Fatal error - ENOMEM, EIO... */
2879 blk_finish_plug(&plug);
2880 if (!ret && !cycled && wbc->nr_to_write > 0) {
2882 mpd.last_page = writeback_index - 1;
2888 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2890 * Set the writeback_index so that range_cyclic
2891 * mode will write it back later
2893 mapping->writeback_index = mpd.first_page;
2896 trace_ext4_writepages_result(inode, wbc, ret,
2897 nr_to_write - wbc->nr_to_write);
2898 percpu_up_read(&sbi->s_writepages_rwsem);
2902 static int ext4_nonda_switch(struct super_block *sb)
2904 s64 free_clusters, dirty_clusters;
2905 struct ext4_sb_info *sbi = EXT4_SB(sb);
2908 * switch to non delalloc mode if we are running low
2909 * on free block. The free block accounting via percpu
2910 * counters can get slightly wrong with percpu_counter_batch getting
2911 * accumulated on each CPU without updating global counters
2912 * Delalloc need an accurate free block accounting. So switch
2913 * to non delalloc when we are near to error range.
2916 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2918 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2920 * Start pushing delalloc when 1/2 of free blocks are dirty.
2922 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2923 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2925 if (2 * free_clusters < 3 * dirty_clusters ||
2926 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2928 * free block count is less than 150% of dirty blocks
2929 * or free blocks is less than watermark
2936 /* We always reserve for an inode update; the superblock could be there too */
2937 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2939 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2942 if (pos + len <= 0x7fffffffULL)
2945 /* We might need to update the superblock to set LARGE_FILE */
2949 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2950 loff_t pos, unsigned len, unsigned flags,
2951 struct page **pagep, void **fsdata)
2953 int ret, retries = 0;
2956 struct inode *inode = mapping->host;
2959 index = pos >> PAGE_SHIFT;
2961 if (ext4_nonda_switch(inode->i_sb)) {
2962 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2963 return ext4_write_begin(file, mapping, pos,
2964 len, flags, pagep, fsdata);
2966 *fsdata = (void *)0;
2967 trace_ext4_da_write_begin(inode, pos, len, flags);
2969 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2970 ret = ext4_da_write_inline_data_begin(mapping, inode,
2980 * grab_cache_page_write_begin() can take a long time if the
2981 * system is thrashing due to memory pressure, or if the page
2982 * is being written back. So grab it first before we start
2983 * the transaction handle. This also allows us to allocate
2984 * the page (if needed) without using GFP_NOFS.
2987 page = grab_cache_page_write_begin(mapping, index, flags);
2993 * With delayed allocation, we don't log the i_disksize update
2994 * if there is delayed block allocation. But we still need
2995 * to journalling the i_disksize update if writes to the end
2996 * of file which has an already mapped buffer.
2999 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3000 ext4_da_write_credits(inode, pos, len));
3001 if (IS_ERR(handle)) {
3003 return PTR_ERR(handle);
3007 if (page->mapping != mapping) {
3008 /* The page got truncated from under us */
3011 ext4_journal_stop(handle);
3014 /* In case writeback began while the page was unlocked */
3015 wait_for_stable_page(page);
3017 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3018 ret = ext4_block_write_begin(page, pos, len,
3019 ext4_da_get_block_prep);
3021 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3025 ext4_journal_stop(handle);
3027 * block_write_begin may have instantiated a few blocks
3028 * outside i_size. Trim these off again. Don't need
3029 * i_size_read because we hold i_mutex.
3031 if (pos + len > inode->i_size)
3032 ext4_truncate_failed_write(inode);
3034 if (ret == -ENOSPC &&
3035 ext4_should_retry_alloc(inode->i_sb, &retries))
3047 * Check if we should update i_disksize
3048 * when write to the end of file but not require block allocation
3050 static int ext4_da_should_update_i_disksize(struct page *page,
3051 unsigned long offset)
3053 struct buffer_head *bh;
3054 struct inode *inode = page->mapping->host;
3058 bh = page_buffers(page);
3059 idx = offset >> inode->i_blkbits;
3061 for (i = 0; i < idx; i++)
3062 bh = bh->b_this_page;
3064 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3069 static int ext4_da_write_end(struct file *file,
3070 struct address_space *mapping,
3071 loff_t pos, unsigned len, unsigned copied,
3072 struct page *page, void *fsdata)
3074 struct inode *inode = mapping->host;
3076 handle_t *handle = ext4_journal_current_handle();
3078 unsigned long start, end;
3079 int write_mode = (int)(unsigned long)fsdata;
3081 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3082 return ext4_write_end(file, mapping, pos,
3083 len, copied, page, fsdata);
3085 trace_ext4_da_write_end(inode, pos, len, copied);
3086 start = pos & (PAGE_SIZE - 1);
3087 end = start + copied - 1;
3090 * generic_write_end() will run mark_inode_dirty() if i_size
3091 * changes. So let's piggyback the i_disksize mark_inode_dirty
3094 new_i_size = pos + copied;
3095 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3096 if (ext4_has_inline_data(inode) ||
3097 ext4_da_should_update_i_disksize(page, end)) {
3098 ext4_update_i_disksize(inode, new_i_size);
3099 /* We need to mark inode dirty even if
3100 * new_i_size is less that inode->i_size
3101 * bu greater than i_disksize.(hint delalloc)
3103 ext4_mark_inode_dirty(handle, inode);
3107 if (write_mode != CONVERT_INLINE_DATA &&
3108 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3109 ext4_has_inline_data(inode))
3110 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3113 ret2 = generic_write_end(file, mapping, pos, len, copied,
3119 ret2 = ext4_journal_stop(handle);
3123 return ret ? ret : copied;
3126 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3127 unsigned int length)
3130 * Drop reserved blocks
3132 BUG_ON(!PageLocked(page));
3133 if (!page_has_buffers(page))
3136 ext4_da_page_release_reservation(page, offset, length);
3139 ext4_invalidatepage(page, offset, length);
3145 * Force all delayed allocation blocks to be allocated for a given inode.
3147 int ext4_alloc_da_blocks(struct inode *inode)
3149 trace_ext4_alloc_da_blocks(inode);
3151 if (!EXT4_I(inode)->i_reserved_data_blocks)
3155 * We do something simple for now. The filemap_flush() will
3156 * also start triggering a write of the data blocks, which is
3157 * not strictly speaking necessary (and for users of
3158 * laptop_mode, not even desirable). However, to do otherwise
3159 * would require replicating code paths in:
3161 * ext4_writepages() ->
3162 * write_cache_pages() ---> (via passed in callback function)
3163 * __mpage_da_writepage() -->
3164 * mpage_add_bh_to_extent()
3165 * mpage_da_map_blocks()
3167 * The problem is that write_cache_pages(), located in
3168 * mm/page-writeback.c, marks pages clean in preparation for
3169 * doing I/O, which is not desirable if we're not planning on
3172 * We could call write_cache_pages(), and then redirty all of
3173 * the pages by calling redirty_page_for_writepage() but that
3174 * would be ugly in the extreme. So instead we would need to
3175 * replicate parts of the code in the above functions,
3176 * simplifying them because we wouldn't actually intend to
3177 * write out the pages, but rather only collect contiguous
3178 * logical block extents, call the multi-block allocator, and
3179 * then update the buffer heads with the block allocations.
3181 * For now, though, we'll cheat by calling filemap_flush(),
3182 * which will map the blocks, and start the I/O, but not
3183 * actually wait for the I/O to complete.
3185 return filemap_flush(inode->i_mapping);
3189 * bmap() is special. It gets used by applications such as lilo and by
3190 * the swapper to find the on-disk block of a specific piece of data.
3192 * Naturally, this is dangerous if the block concerned is still in the
3193 * journal. If somebody makes a swapfile on an ext4 data-journaling
3194 * filesystem and enables swap, then they may get a nasty shock when the
3195 * data getting swapped to that swapfile suddenly gets overwritten by
3196 * the original zero's written out previously to the journal and
3197 * awaiting writeback in the kernel's buffer cache.
3199 * So, if we see any bmap calls here on a modified, data-journaled file,
3200 * take extra steps to flush any blocks which might be in the cache.
3202 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3204 struct inode *inode = mapping->host;
3209 * We can get here for an inline file via the FIBMAP ioctl
3211 if (ext4_has_inline_data(inode))
3214 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3215 test_opt(inode->i_sb, DELALLOC)) {
3217 * With delalloc we want to sync the file
3218 * so that we can make sure we allocate
3221 filemap_write_and_wait(mapping);
3224 if (EXT4_JOURNAL(inode) &&
3225 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3227 * This is a REALLY heavyweight approach, but the use of
3228 * bmap on dirty files is expected to be extremely rare:
3229 * only if we run lilo or swapon on a freshly made file
3230 * do we expect this to happen.
3232 * (bmap requires CAP_SYS_RAWIO so this does not
3233 * represent an unprivileged user DOS attack --- we'd be
3234 * in trouble if mortal users could trigger this path at
3237 * NB. EXT4_STATE_JDATA is not set on files other than
3238 * regular files. If somebody wants to bmap a directory
3239 * or symlink and gets confused because the buffer
3240 * hasn't yet been flushed to disk, they deserve
3241 * everything they get.
3244 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3245 journal = EXT4_JOURNAL(inode);
3246 jbd2_journal_lock_updates(journal);
3247 err = jbd2_journal_flush(journal);
3248 jbd2_journal_unlock_updates(journal);
3254 return generic_block_bmap(mapping, block, ext4_get_block);
3257 static int ext4_readpage(struct file *file, struct page *page)
3260 struct inode *inode = page->mapping->host;
3262 trace_ext4_readpage(page);
3264 if (ext4_has_inline_data(inode))
3265 ret = ext4_readpage_inline(inode, page);
3268 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3274 ext4_readpages(struct file *file, struct address_space *mapping,
3275 struct list_head *pages, unsigned nr_pages)
3277 struct inode *inode = mapping->host;
3279 /* If the file has inline data, no need to do readpages. */
3280 if (ext4_has_inline_data(inode))
3283 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3286 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3287 unsigned int length)
3289 trace_ext4_invalidatepage(page, offset, length);
3291 /* No journalling happens on data buffers when this function is used */
3292 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3294 block_invalidatepage(page, offset, length);
3297 static int __ext4_journalled_invalidatepage(struct page *page,
3298 unsigned int offset,
3299 unsigned int length)
3301 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3303 trace_ext4_journalled_invalidatepage(page, offset, length);
3306 * If it's a full truncate we just forget about the pending dirtying
3308 if (offset == 0 && length == PAGE_SIZE)
3309 ClearPageChecked(page);
3311 return jbd2_journal_invalidatepage(journal, page, offset, length);
3314 /* Wrapper for aops... */
3315 static void ext4_journalled_invalidatepage(struct page *page,
3316 unsigned int offset,
3317 unsigned int length)
3319 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3322 static int ext4_releasepage(struct page *page, gfp_t wait)
3324 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3326 trace_ext4_releasepage(page);
3328 /* Page has dirty journalled data -> cannot release */
3329 if (PageChecked(page))
3332 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3334 return try_to_free_buffers(page);
3337 #ifdef CONFIG_FS_DAX
3339 * Get block function for DAX IO and mmap faults. It takes care of converting
3340 * unwritten extents to written ones and initializes new / converted blocks
3343 int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3344 struct buffer_head *bh_result, int create)
3348 ext4_debug("inode %lu, create flag %d\n", inode->i_ino, create);
3350 return _ext4_get_block(inode, iblock, bh_result, 0);
3352 ret = ext4_get_block_trans(inode, iblock, bh_result,
3353 EXT4_GET_BLOCKS_PRE_IO |
3354 EXT4_GET_BLOCKS_CREATE_ZERO);
3358 if (buffer_unwritten(bh_result)) {
3360 * We are protected by i_mmap_sem or i_mutex so we know block
3361 * cannot go away from under us even though we dropped
3362 * i_data_sem. Convert extent to written and write zeros there.
3364 ret = ext4_get_block_trans(inode, iblock, bh_result,
3365 EXT4_GET_BLOCKS_CONVERT |
3366 EXT4_GET_BLOCKS_CREATE_ZERO);
3371 * At least for now we have to clear BH_New so that DAX code
3372 * doesn't attempt to zero blocks again in a racy way.
3374 clear_buffer_new(bh_result);
3378 /* Just define empty function, it will never get called. */
3379 int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3380 struct buffer_head *bh_result, int create)
3387 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3388 ssize_t size, void *private)
3390 ext4_io_end_t *io_end = private;
3392 /* if not async direct IO just return */
3396 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3397 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3398 io_end, io_end->inode->i_ino, iocb, offset, size);
3401 * Error during AIO DIO. We cannot convert unwritten extents as the
3402 * data was not written. Just clear the unwritten flag and drop io_end.
3405 ext4_clear_io_unwritten_flag(io_end);
3408 io_end->offset = offset;
3409 io_end->size = size;
3410 ext4_put_io_end(io_end);
3416 * Handling of direct IO writes.
3418 * For ext4 extent files, ext4 will do direct-io write even to holes,
3419 * preallocated extents, and those write extend the file, no need to
3420 * fall back to buffered IO.
3422 * For holes, we fallocate those blocks, mark them as unwritten
3423 * If those blocks were preallocated, we mark sure they are split, but
3424 * still keep the range to write as unwritten.
3426 * The unwritten extents will be converted to written when DIO is completed.
3427 * For async direct IO, since the IO may still pending when return, we
3428 * set up an end_io call back function, which will do the conversion
3429 * when async direct IO completed.
3431 * If the O_DIRECT write will extend the file then add this inode to the
3432 * orphan list. So recovery will truncate it back to the original size
3433 * if the machine crashes during the write.
3436 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3438 struct file *file = iocb->ki_filp;
3439 struct inode *inode = file->f_mapping->host;
3441 loff_t offset = iocb->ki_pos;
3442 size_t count = iov_iter_count(iter);
3444 get_block_t *get_block_func = NULL;
3446 loff_t final_size = offset + count;
3450 if (final_size > inode->i_size) {
3451 /* Credits for sb + inode write */
3452 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3453 if (IS_ERR(handle)) {
3454 ret = PTR_ERR(handle);
3457 ret = ext4_orphan_add(handle, inode);
3459 ext4_journal_stop(handle);
3463 ext4_update_i_disksize(inode, inode->i_size);
3464 ext4_journal_stop(handle);
3467 BUG_ON(iocb->private == NULL);
3470 * Make all waiters for direct IO properly wait also for extent
3471 * conversion. This also disallows race between truncate() and
3472 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3474 inode_dio_begin(inode);
3476 /* If we do a overwrite dio, i_mutex locking can be released */
3477 overwrite = *((int *)iocb->private);
3480 inode_unlock(inode);
3483 * For extent mapped files we could direct write to holes and fallocate.
3485 * Allocated blocks to fill the hole are marked as unwritten to prevent
3486 * parallel buffered read to expose the stale data before DIO complete
3489 * As to previously fallocated extents, ext4 get_block will just simply
3490 * mark the buffer mapped but still keep the extents unwritten.
3492 * For non AIO case, we will convert those unwritten extents to written
3493 * after return back from blockdev_direct_IO. That way we save us from
3494 * allocating io_end structure and also the overhead of offloading
3495 * the extent convertion to a workqueue.
3497 * For async DIO, the conversion needs to be deferred when the
3498 * IO is completed. The ext4 end_io callback function will be
3499 * called to take care of the conversion work. Here for async
3500 * case, we allocate an io_end structure to hook to the iocb.
3502 iocb->private = NULL;
3504 get_block_func = ext4_dio_get_block_overwrite;
3505 else if (IS_DAX(inode)) {
3507 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3508 * writes need zeroing either because they can race with page
3509 * faults or because they use partial blocks.
3511 if (round_down(offset, i_blocksize(inode)) >= inode->i_size &&
3512 ext4_aligned_io(inode, offset, count))
3513 get_block_func = ext4_dio_get_block;
3515 get_block_func = ext4_dax_get_block;
3516 dio_flags = DIO_LOCKING;
3517 } else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3518 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3519 get_block_func = ext4_dio_get_block;
3520 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3521 } else if (is_sync_kiocb(iocb)) {
3522 get_block_func = ext4_dio_get_block_unwritten_sync;
3523 dio_flags = DIO_LOCKING;
3525 get_block_func = ext4_dio_get_block_unwritten_async;
3526 dio_flags = DIO_LOCKING;
3528 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3529 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3531 if (IS_DAX(inode)) {
3532 ret = dax_do_io(iocb, inode, iter, get_block_func,
3533 ext4_end_io_dio, dio_flags);
3535 ret = __blockdev_direct_IO(iocb, inode,
3536 inode->i_sb->s_bdev, iter,
3538 ext4_end_io_dio, NULL, dio_flags);
3540 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3541 EXT4_STATE_DIO_UNWRITTEN)) {
3544 * for non AIO case, since the IO is already
3545 * completed, we could do the conversion right here
3547 err = ext4_convert_unwritten_extents(NULL, inode,
3551 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3554 inode_dio_end(inode);
3555 /* take i_mutex locking again if we do a ovewrite dio */
3559 if (ret < 0 && final_size > inode->i_size)
3560 ext4_truncate_failed_write(inode);
3562 /* Handle extending of i_size after direct IO write */
3566 /* Credits for sb + inode write */
3567 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3568 if (IS_ERR(handle)) {
3570 * We wrote the data but cannot extend
3571 * i_size. Bail out. In async io case, we do
3572 * not return error here because we have
3573 * already submmitted the corresponding
3574 * bio. Returning error here makes the caller
3575 * think that this IO is done and failed
3576 * resulting in race with bio's completion
3580 ret = PTR_ERR(handle);
3582 ext4_orphan_del(NULL, inode);
3587 ext4_orphan_del(handle, inode);
3589 loff_t end = offset + ret;
3590 if (end > inode->i_size) {
3591 ext4_update_i_disksize(inode, end);
3592 i_size_write(inode, end);
3594 * We're going to return a positive `ret'
3595 * here due to non-zero-length I/O, so there's
3596 * no way of reporting error returns from
3597 * ext4_mark_inode_dirty() to userspace. So
3600 ext4_mark_inode_dirty(handle, inode);
3603 err = ext4_journal_stop(handle);
3611 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3613 struct address_space *mapping = iocb->ki_filp->f_mapping;
3614 struct inode *inode = mapping->host;
3616 loff_t offset = iocb->ki_pos;
3617 loff_t size = i_size_read(inode);
3623 * Shared inode_lock is enough for us - it protects against concurrent
3624 * writes & truncates and since we take care of writing back page cache,
3625 * we are protected against page writeback as well.
3627 inode_lock_shared(inode);
3628 if (IS_DAX(inode)) {
3629 ret = dax_do_io(iocb, inode, iter, ext4_dio_get_block, NULL, 0);
3631 size_t count = iov_iter_count(iter);
3633 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3634 iocb->ki_pos + count);
3637 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3638 iter, ext4_dio_get_block,
3642 inode_unlock_shared(inode);
3646 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3648 struct file *file = iocb->ki_filp;
3649 struct inode *inode = file->f_mapping->host;
3650 size_t count = iov_iter_count(iter);
3651 loff_t offset = iocb->ki_pos;
3654 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3655 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3660 * If we are doing data journalling we don't support O_DIRECT
3662 if (ext4_should_journal_data(inode))
3665 /* Let buffer I/O handle the inline data case. */
3666 if (ext4_has_inline_data(inode))
3669 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3670 if (iov_iter_rw(iter) == READ)
3671 ret = ext4_direct_IO_read(iocb, iter);
3673 ret = ext4_direct_IO_write(iocb, iter);
3674 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3679 * Pages can be marked dirty completely asynchronously from ext4's journalling
3680 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3681 * much here because ->set_page_dirty is called under VFS locks. The page is
3682 * not necessarily locked.
3684 * We cannot just dirty the page and leave attached buffers clean, because the
3685 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3686 * or jbddirty because all the journalling code will explode.
3688 * So what we do is to mark the page "pending dirty" and next time writepage
3689 * is called, propagate that into the buffers appropriately.
3691 static int ext4_journalled_set_page_dirty(struct page *page)
3693 SetPageChecked(page);
3694 return __set_page_dirty_nobuffers(page);
3697 static const struct address_space_operations ext4_aops = {
3698 .readpage = ext4_readpage,
3699 .readpages = ext4_readpages,
3700 .writepage = ext4_writepage,
3701 .writepages = ext4_writepages,
3702 .write_begin = ext4_write_begin,
3703 .write_end = ext4_write_end,
3705 .invalidatepage = ext4_invalidatepage,
3706 .releasepage = ext4_releasepage,
3707 .direct_IO = ext4_direct_IO,
3708 .migratepage = buffer_migrate_page,
3709 .is_partially_uptodate = block_is_partially_uptodate,
3710 .error_remove_page = generic_error_remove_page,
3713 static const struct address_space_operations ext4_journalled_aops = {
3714 .readpage = ext4_readpage,
3715 .readpages = ext4_readpages,
3716 .writepage = ext4_writepage,
3717 .writepages = ext4_writepages,
3718 .write_begin = ext4_write_begin,
3719 .write_end = ext4_journalled_write_end,
3720 .set_page_dirty = ext4_journalled_set_page_dirty,
3722 .invalidatepage = ext4_journalled_invalidatepage,
3723 .releasepage = ext4_releasepage,
3724 .direct_IO = ext4_direct_IO,
3725 .is_partially_uptodate = block_is_partially_uptodate,
3726 .error_remove_page = generic_error_remove_page,
3729 static const struct address_space_operations ext4_da_aops = {
3730 .readpage = ext4_readpage,
3731 .readpages = ext4_readpages,
3732 .writepage = ext4_writepage,
3733 .writepages = ext4_writepages,
3734 .write_begin = ext4_da_write_begin,
3735 .write_end = ext4_da_write_end,
3737 .invalidatepage = ext4_da_invalidatepage,
3738 .releasepage = ext4_releasepage,
3739 .direct_IO = ext4_direct_IO,
3740 .migratepage = buffer_migrate_page,
3741 .is_partially_uptodate = block_is_partially_uptodate,
3742 .error_remove_page = generic_error_remove_page,
3745 void ext4_set_aops(struct inode *inode)
3747 switch (ext4_inode_journal_mode(inode)) {
3748 case EXT4_INODE_ORDERED_DATA_MODE:
3749 case EXT4_INODE_WRITEBACK_DATA_MODE:
3751 case EXT4_INODE_JOURNAL_DATA_MODE:
3752 inode->i_mapping->a_ops = &ext4_journalled_aops;
3757 if (test_opt(inode->i_sb, DELALLOC))
3758 inode->i_mapping->a_ops = &ext4_da_aops;
3760 inode->i_mapping->a_ops = &ext4_aops;
3763 static int __ext4_block_zero_page_range(handle_t *handle,
3764 struct address_space *mapping, loff_t from, loff_t length)
3766 ext4_fsblk_t index = from >> PAGE_SHIFT;
3767 unsigned offset = from & (PAGE_SIZE-1);
3768 unsigned blocksize, pos;
3770 struct inode *inode = mapping->host;
3771 struct buffer_head *bh;
3775 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3776 mapping_gfp_constraint(mapping, ~__GFP_FS));
3780 blocksize = inode->i_sb->s_blocksize;
3782 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3784 if (!page_has_buffers(page))
3785 create_empty_buffers(page, blocksize, 0);
3787 /* Find the buffer that contains "offset" */
3788 bh = page_buffers(page);
3790 while (offset >= pos) {
3791 bh = bh->b_this_page;
3795 if (buffer_freed(bh)) {
3796 BUFFER_TRACE(bh, "freed: skip");
3799 if (!buffer_mapped(bh)) {
3800 BUFFER_TRACE(bh, "unmapped");
3801 ext4_get_block(inode, iblock, bh, 0);
3802 /* unmapped? It's a hole - nothing to do */
3803 if (!buffer_mapped(bh)) {
3804 BUFFER_TRACE(bh, "still unmapped");
3809 /* Ok, it's mapped. Make sure it's up-to-date */
3810 if (PageUptodate(page))
3811 set_buffer_uptodate(bh);
3813 if (!buffer_uptodate(bh)) {
3815 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3817 /* Uhhuh. Read error. Complain and punt. */
3818 if (!buffer_uptodate(bh))
3820 if (S_ISREG(inode->i_mode) &&
3821 ext4_encrypted_inode(inode)) {
3822 /* We expect the key to be set. */
3823 BUG_ON(!fscrypt_has_encryption_key(inode));
3824 BUG_ON(blocksize != PAGE_SIZE);
3825 WARN_ON_ONCE(fscrypt_decrypt_page(page));
3828 if (ext4_should_journal_data(inode)) {
3829 BUFFER_TRACE(bh, "get write access");
3830 err = ext4_journal_get_write_access(handle, bh);
3834 zero_user(page, offset, length);
3835 BUFFER_TRACE(bh, "zeroed end of block");
3837 if (ext4_should_journal_data(inode)) {
3838 err = ext4_handle_dirty_metadata(handle, inode, bh);
3841 mark_buffer_dirty(bh);
3842 if (ext4_should_order_data(inode))
3843 err = ext4_jbd2_inode_add_write(handle, inode);
3853 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3854 * starting from file offset 'from'. The range to be zero'd must
3855 * be contained with in one block. If the specified range exceeds
3856 * the end of the block it will be shortened to end of the block
3857 * that cooresponds to 'from'
3859 static int ext4_block_zero_page_range(handle_t *handle,
3860 struct address_space *mapping, loff_t from, loff_t length)
3862 struct inode *inode = mapping->host;
3863 unsigned offset = from & (PAGE_SIZE-1);
3864 unsigned blocksize = inode->i_sb->s_blocksize;
3865 unsigned max = blocksize - (offset & (blocksize - 1));
3868 * correct length if it does not fall between
3869 * 'from' and the end of the block
3871 if (length > max || length < 0)
3875 return dax_zero_page_range(inode, from, length, ext4_get_block);
3876 return __ext4_block_zero_page_range(handle, mapping, from, length);
3880 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3881 * up to the end of the block which corresponds to `from'.
3882 * This required during truncate. We need to physically zero the tail end
3883 * of that block so it doesn't yield old data if the file is later grown.
3885 static int ext4_block_truncate_page(handle_t *handle,
3886 struct address_space *mapping, loff_t from)
3888 unsigned offset = from & (PAGE_SIZE-1);
3891 struct inode *inode = mapping->host;
3893 /* If we are processing an encrypted inode during orphan list handling */
3894 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
3897 blocksize = inode->i_sb->s_blocksize;
3898 length = blocksize - (offset & (blocksize - 1));
3900 return ext4_block_zero_page_range(handle, mapping, from, length);
3903 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3904 loff_t lstart, loff_t length)
3906 struct super_block *sb = inode->i_sb;
3907 struct address_space *mapping = inode->i_mapping;
3908 unsigned partial_start, partial_end;
3909 ext4_fsblk_t start, end;
3910 loff_t byte_end = (lstart + length - 1);
3913 partial_start = lstart & (sb->s_blocksize - 1);
3914 partial_end = byte_end & (sb->s_blocksize - 1);
3916 start = lstart >> sb->s_blocksize_bits;
3917 end = byte_end >> sb->s_blocksize_bits;
3919 /* Handle partial zero within the single block */
3921 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3922 err = ext4_block_zero_page_range(handle, mapping,
3926 /* Handle partial zero out on the start of the range */
3927 if (partial_start) {
3928 err = ext4_block_zero_page_range(handle, mapping,
3929 lstart, sb->s_blocksize);
3933 /* Handle partial zero out on the end of the range */
3934 if (partial_end != sb->s_blocksize - 1)
3935 err = ext4_block_zero_page_range(handle, mapping,
3936 byte_end - partial_end,
3941 int ext4_can_truncate(struct inode *inode)
3943 if (S_ISREG(inode->i_mode))
3945 if (S_ISDIR(inode->i_mode))
3947 if (S_ISLNK(inode->i_mode))
3948 return !ext4_inode_is_fast_symlink(inode);
3953 * We have to make sure i_disksize gets properly updated before we truncate
3954 * page cache due to hole punching or zero range. Otherwise i_disksize update
3955 * can get lost as it may have been postponed to submission of writeback but
3956 * that will never happen after we truncate page cache.
3958 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3962 loff_t size = i_size_read(inode);
3964 WARN_ON(!inode_is_locked(inode));
3965 if (offset > size || offset + len < size)
3968 if (EXT4_I(inode)->i_disksize >= size)
3971 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3973 return PTR_ERR(handle);
3974 ext4_update_i_disksize(inode, size);
3975 ext4_mark_inode_dirty(handle, inode);
3976 ext4_journal_stop(handle);
3982 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3983 * associated with the given offset and length
3985 * @inode: File inode
3986 * @offset: The offset where the hole will begin
3987 * @len: The length of the hole
3989 * Returns: 0 on success or negative on failure
3992 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3994 struct super_block *sb = inode->i_sb;
3995 ext4_lblk_t first_block, stop_block;
3996 struct address_space *mapping = inode->i_mapping;
3997 loff_t first_block_offset, last_block_offset, max_length;
3998 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4000 unsigned int credits;
4003 if (!S_ISREG(inode->i_mode))
4006 trace_ext4_punch_hole(inode, offset, length, 0);
4008 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4009 if (ext4_has_inline_data(inode)) {
4010 down_write(&EXT4_I(inode)->i_mmap_sem);
4011 ret = ext4_convert_inline_data(inode);
4012 up_write(&EXT4_I(inode)->i_mmap_sem);
4018 * Write out all dirty pages to avoid race conditions
4019 * Then release them.
4021 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4022 ret = filemap_write_and_wait_range(mapping, offset,
4023 offset + length - 1);
4030 /* No need to punch hole beyond i_size */
4031 if (offset >= inode->i_size)
4035 * If the hole extends beyond i_size, set the hole
4036 * to end after the page that contains i_size
4038 if (offset + length > inode->i_size) {
4039 length = inode->i_size +
4040 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4045 * For punch hole the length + offset needs to be within one block
4046 * before last range. Adjust the length if it goes beyond that limit.
4048 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4049 if (offset + length > max_length)
4050 length = max_length - offset;
4052 if (offset & (sb->s_blocksize - 1) ||
4053 (offset + length) & (sb->s_blocksize - 1)) {
4055 * Attach jinode to inode for jbd2 if we do any zeroing of
4058 ret = ext4_inode_attach_jinode(inode);
4064 /* Wait all existing dio workers, newcomers will block on i_mutex */
4065 ext4_inode_block_unlocked_dio(inode);
4066 inode_dio_wait(inode);
4069 * Prevent page faults from reinstantiating pages we have released from
4072 down_write(&EXT4_I(inode)->i_mmap_sem);
4073 first_block_offset = round_up(offset, sb->s_blocksize);
4074 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4076 /* Now release the pages and zero block aligned part of pages*/
4077 if (last_block_offset > first_block_offset) {
4078 ret = ext4_update_disksize_before_punch(inode, offset, length);
4081 truncate_pagecache_range(inode, first_block_offset,
4085 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4086 credits = ext4_writepage_trans_blocks(inode);
4088 credits = ext4_blocks_for_truncate(inode);
4089 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4090 if (IS_ERR(handle)) {
4091 ret = PTR_ERR(handle);
4092 ext4_std_error(sb, ret);
4096 ret = ext4_zero_partial_blocks(handle, inode, offset,
4101 first_block = (offset + sb->s_blocksize - 1) >>
4102 EXT4_BLOCK_SIZE_BITS(sb);
4103 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4105 /* If there are blocks to remove, do it */
4106 if (stop_block > first_block) {
4108 down_write(&EXT4_I(inode)->i_data_sem);
4109 ext4_discard_preallocations(inode);
4111 ret = ext4_es_remove_extent(inode, first_block,
4112 stop_block - first_block);
4114 up_write(&EXT4_I(inode)->i_data_sem);
4118 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4119 ret = ext4_ext_remove_space(inode, first_block,
4122 ret = ext4_ind_remove_space(handle, inode, first_block,
4125 up_write(&EXT4_I(inode)->i_data_sem);
4128 ext4_handle_sync(handle);
4130 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4131 ext4_mark_inode_dirty(handle, inode);
4133 ext4_update_inode_fsync_trans(handle, inode, 1);
4135 ext4_journal_stop(handle);
4137 up_write(&EXT4_I(inode)->i_mmap_sem);
4138 ext4_inode_resume_unlocked_dio(inode);
4140 inode_unlock(inode);
4144 int ext4_inode_attach_jinode(struct inode *inode)
4146 struct ext4_inode_info *ei = EXT4_I(inode);
4147 struct jbd2_inode *jinode;
4149 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4152 jinode = jbd2_alloc_inode(GFP_KERNEL);
4153 spin_lock(&inode->i_lock);
4156 spin_unlock(&inode->i_lock);
4159 ei->jinode = jinode;
4160 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4163 spin_unlock(&inode->i_lock);
4164 if (unlikely(jinode != NULL))
4165 jbd2_free_inode(jinode);
4172 * We block out ext4_get_block() block instantiations across the entire
4173 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4174 * simultaneously on behalf of the same inode.
4176 * As we work through the truncate and commit bits of it to the journal there
4177 * is one core, guiding principle: the file's tree must always be consistent on
4178 * disk. We must be able to restart the truncate after a crash.
4180 * The file's tree may be transiently inconsistent in memory (although it
4181 * probably isn't), but whenever we close off and commit a journal transaction,
4182 * the contents of (the filesystem + the journal) must be consistent and
4183 * restartable. It's pretty simple, really: bottom up, right to left (although
4184 * left-to-right works OK too).
4186 * Note that at recovery time, journal replay occurs *before* the restart of
4187 * truncate against the orphan inode list.
4189 * The committed inode has the new, desired i_size (which is the same as
4190 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4191 * that this inode's truncate did not complete and it will again call
4192 * ext4_truncate() to have another go. So there will be instantiated blocks
4193 * to the right of the truncation point in a crashed ext4 filesystem. But
4194 * that's fine - as long as they are linked from the inode, the post-crash
4195 * ext4_truncate() run will find them and release them.
4197 void ext4_truncate(struct inode *inode)
4199 struct ext4_inode_info *ei = EXT4_I(inode);
4200 unsigned int credits;
4202 struct address_space *mapping = inode->i_mapping;
4205 * There is a possibility that we're either freeing the inode
4206 * or it's a completely new inode. In those cases we might not
4207 * have i_mutex locked because it's not necessary.
4209 if (!(inode->i_state & (I_NEW|I_FREEING)))
4210 WARN_ON(!inode_is_locked(inode));
4211 trace_ext4_truncate_enter(inode);
4213 if (!ext4_can_truncate(inode))
4216 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4218 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4219 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4221 if (ext4_has_inline_data(inode)) {
4224 ext4_inline_data_truncate(inode, &has_inline);
4229 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4230 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4231 if (ext4_inode_attach_jinode(inode) < 0)
4235 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4236 credits = ext4_writepage_trans_blocks(inode);
4238 credits = ext4_blocks_for_truncate(inode);
4240 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4241 if (IS_ERR(handle)) {
4242 ext4_std_error(inode->i_sb, PTR_ERR(handle));
4246 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4247 ext4_block_truncate_page(handle, mapping, inode->i_size);
4250 * We add the inode to the orphan list, so that if this
4251 * truncate spans multiple transactions, and we crash, we will
4252 * resume the truncate when the filesystem recovers. It also
4253 * marks the inode dirty, to catch the new size.
4255 * Implication: the file must always be in a sane, consistent
4256 * truncatable state while each transaction commits.
4258 if (ext4_orphan_add(handle, inode))
4261 down_write(&EXT4_I(inode)->i_data_sem);
4263 ext4_discard_preallocations(inode);
4265 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4266 ext4_ext_truncate(handle, inode);
4268 ext4_ind_truncate(handle, inode);
4270 up_write(&ei->i_data_sem);
4273 ext4_handle_sync(handle);
4277 * If this was a simple ftruncate() and the file will remain alive,
4278 * then we need to clear up the orphan record which we created above.
4279 * However, if this was a real unlink then we were called by
4280 * ext4_evict_inode(), and we allow that function to clean up the
4281 * orphan info for us.
4284 ext4_orphan_del(handle, inode);
4286 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4287 ext4_mark_inode_dirty(handle, inode);
4288 ext4_journal_stop(handle);
4290 trace_ext4_truncate_exit(inode);
4294 * ext4_get_inode_loc returns with an extra refcount against the inode's
4295 * underlying buffer_head on success. If 'in_mem' is true, we have all
4296 * data in memory that is needed to recreate the on-disk version of this
4299 static int __ext4_get_inode_loc(struct inode *inode,
4300 struct ext4_iloc *iloc, int in_mem)
4302 struct ext4_group_desc *gdp;
4303 struct buffer_head *bh;
4304 struct super_block *sb = inode->i_sb;
4306 int inodes_per_block, inode_offset;
4309 if (inode->i_ino < EXT4_ROOT_INO ||
4310 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4311 return -EFSCORRUPTED;
4313 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4314 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4319 * Figure out the offset within the block group inode table
4321 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4322 inode_offset = ((inode->i_ino - 1) %
4323 EXT4_INODES_PER_GROUP(sb));
4324 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4325 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4327 bh = sb_getblk(sb, block);
4330 if (!buffer_uptodate(bh)) {
4334 * If the buffer has the write error flag, we have failed
4335 * to write out another inode in the same block. In this
4336 * case, we don't have to read the block because we may
4337 * read the old inode data successfully.
4339 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4340 set_buffer_uptodate(bh);
4342 if (buffer_uptodate(bh)) {
4343 /* someone brought it uptodate while we waited */
4349 * If we have all information of the inode in memory and this
4350 * is the only valid inode in the block, we need not read the
4354 struct buffer_head *bitmap_bh;
4357 start = inode_offset & ~(inodes_per_block - 1);
4359 /* Is the inode bitmap in cache? */
4360 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4361 if (unlikely(!bitmap_bh))
4365 * If the inode bitmap isn't in cache then the
4366 * optimisation may end up performing two reads instead
4367 * of one, so skip it.
4369 if (!buffer_uptodate(bitmap_bh)) {
4373 for (i = start; i < start + inodes_per_block; i++) {
4374 if (i == inode_offset)
4376 if (ext4_test_bit(i, bitmap_bh->b_data))
4380 if (i == start + inodes_per_block) {
4381 /* all other inodes are free, so skip I/O */
4382 memset(bh->b_data, 0, bh->b_size);
4383 set_buffer_uptodate(bh);
4391 * If we need to do any I/O, try to pre-readahead extra
4392 * blocks from the inode table.
4394 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4395 ext4_fsblk_t b, end, table;
4397 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4399 table = ext4_inode_table(sb, gdp);
4400 /* s_inode_readahead_blks is always a power of 2 */
4401 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4405 num = EXT4_INODES_PER_GROUP(sb);
4406 if (ext4_has_group_desc_csum(sb))
4407 num -= ext4_itable_unused_count(sb, gdp);
4408 table += num / inodes_per_block;
4412 sb_breadahead(sb, b++);
4416 * There are other valid inodes in the buffer, this inode
4417 * has in-inode xattrs, or we don't have this inode in memory.
4418 * Read the block from disk.
4420 trace_ext4_load_inode(inode);
4422 bh->b_end_io = end_buffer_read_sync;
4423 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4425 if (!buffer_uptodate(bh)) {
4426 EXT4_ERROR_INODE_BLOCK(inode, block,
4427 "unable to read itable block");
4437 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4439 /* We have all inode data except xattrs in memory here. */
4440 return __ext4_get_inode_loc(inode, iloc,
4441 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4444 void ext4_set_inode_flags(struct inode *inode)
4446 unsigned int flags = EXT4_I(inode)->i_flags;
4447 unsigned int new_fl = 0;
4449 if (flags & EXT4_SYNC_FL)
4451 if (flags & EXT4_APPEND_FL)
4453 if (flags & EXT4_IMMUTABLE_FL)
4454 new_fl |= S_IMMUTABLE;
4455 if (flags & EXT4_NOATIME_FL)
4456 new_fl |= S_NOATIME;
4457 if (flags & EXT4_DIRSYNC_FL)
4458 new_fl |= S_DIRSYNC;
4459 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
4461 inode_set_flags(inode, new_fl,
4462 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4465 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4466 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4468 unsigned int vfs_fl;
4469 unsigned long old_fl, new_fl;
4472 vfs_fl = ei->vfs_inode.i_flags;
4473 old_fl = ei->i_flags;
4474 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4475 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4477 if (vfs_fl & S_SYNC)
4478 new_fl |= EXT4_SYNC_FL;
4479 if (vfs_fl & S_APPEND)
4480 new_fl |= EXT4_APPEND_FL;
4481 if (vfs_fl & S_IMMUTABLE)
4482 new_fl |= EXT4_IMMUTABLE_FL;
4483 if (vfs_fl & S_NOATIME)
4484 new_fl |= EXT4_NOATIME_FL;
4485 if (vfs_fl & S_DIRSYNC)
4486 new_fl |= EXT4_DIRSYNC_FL;
4487 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4490 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4491 struct ext4_inode_info *ei)
4494 struct inode *inode = &(ei->vfs_inode);
4495 struct super_block *sb = inode->i_sb;
4497 if (ext4_has_feature_huge_file(sb)) {
4498 /* we are using combined 48 bit field */
4499 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4500 le32_to_cpu(raw_inode->i_blocks_lo);
4501 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4502 /* i_blocks represent file system block size */
4503 return i_blocks << (inode->i_blkbits - 9);
4508 return le32_to_cpu(raw_inode->i_blocks_lo);
4512 static inline void ext4_iget_extra_inode(struct inode *inode,
4513 struct ext4_inode *raw_inode,
4514 struct ext4_inode_info *ei)
4516 __le32 *magic = (void *)raw_inode +
4517 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4518 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4519 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4520 ext4_find_inline_data_nolock(inode);
4522 EXT4_I(inode)->i_inline_off = 0;
4525 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4527 if (!ext4_has_feature_project(inode->i_sb))
4529 *projid = EXT4_I(inode)->i_projid;
4533 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4534 ext4_iget_flags flags, const char *function,
4537 struct ext4_iloc iloc;
4538 struct ext4_inode *raw_inode;
4539 struct ext4_inode_info *ei;
4540 struct inode *inode;
4541 journal_t *journal = EXT4_SB(sb)->s_journal;
4549 if ((!(flags & EXT4_IGET_SPECIAL) &&
4550 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4551 (ino < EXT4_ROOT_INO) ||
4552 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4553 if (flags & EXT4_IGET_HANDLE)
4554 return ERR_PTR(-ESTALE);
4555 __ext4_error(sb, function, line,
4556 "inode #%lu: comm %s: iget: illegal inode #",
4557 ino, current->comm);
4558 return ERR_PTR(-EFSCORRUPTED);
4561 inode = iget_locked(sb, ino);
4563 return ERR_PTR(-ENOMEM);
4564 if (!(inode->i_state & I_NEW))
4570 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4573 raw_inode = ext4_raw_inode(&iloc);
4575 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4576 ext4_error_inode(inode, function, line, 0,
4577 "iget: root inode unallocated");
4578 ret = -EFSCORRUPTED;
4582 if ((flags & EXT4_IGET_HANDLE) &&
4583 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4588 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4589 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4590 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4591 EXT4_INODE_SIZE(inode->i_sb)) {
4592 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4593 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4594 EXT4_INODE_SIZE(inode->i_sb));
4595 ret = -EFSCORRUPTED;
4599 ei->i_extra_isize = 0;
4601 /* Precompute checksum seed for inode metadata */
4602 if (ext4_has_metadata_csum(sb)) {
4603 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4605 __le32 inum = cpu_to_le32(inode->i_ino);
4606 __le32 gen = raw_inode->i_generation;
4607 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4609 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4613 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4614 ext4_error_inode(inode, function, line, 0,
4615 "iget: checksum invalid");
4620 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4621 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4622 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4623 if (ext4_has_feature_project(sb) &&
4624 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4625 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4626 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4628 i_projid = EXT4_DEF_PROJID;
4630 if (!(test_opt(inode->i_sb, NO_UID32))) {
4631 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4632 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4634 i_uid_write(inode, i_uid);
4635 i_gid_write(inode, i_gid);
4636 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4637 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4639 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4640 ei->i_inline_off = 0;
4641 ei->i_dir_start_lookup = 0;
4642 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4643 /* We now have enough fields to check if the inode was active or not.
4644 * This is needed because nfsd might try to access dead inodes
4645 * the test is that same one that e2fsck uses
4646 * NeilBrown 1999oct15
4648 if (inode->i_nlink == 0) {
4649 if ((inode->i_mode == 0 ||
4650 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4651 ino != EXT4_BOOT_LOADER_INO) {
4652 /* this inode is deleted */
4656 /* The only unlinked inodes we let through here have
4657 * valid i_mode and are being read by the orphan
4658 * recovery code: that's fine, we're about to complete
4659 * the process of deleting those.
4660 * OR it is the EXT4_BOOT_LOADER_INO which is
4661 * not initialized on a new filesystem. */
4663 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4664 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4665 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4666 if (ext4_has_feature_64bit(sb))
4668 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4669 inode->i_size = ext4_isize(raw_inode);
4670 if ((size = i_size_read(inode)) < 0) {
4671 ext4_error_inode(inode, function, line, 0,
4672 "iget: bad i_size value: %lld", size);
4673 ret = -EFSCORRUPTED;
4677 * If dir_index is not enabled but there's dir with INDEX flag set,
4678 * we'd normally treat htree data as empty space. But with metadata
4679 * checksumming that corrupts checksums so forbid that.
4681 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4682 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4683 EXT4_ERROR_INODE(inode,
4684 "iget: Dir with htree data on filesystem without dir_index feature.");
4685 ret = -EFSCORRUPTED;
4688 ei->i_disksize = inode->i_size;
4690 ei->i_reserved_quota = 0;
4692 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4693 ei->i_block_group = iloc.block_group;
4694 ei->i_last_alloc_group = ~0;
4696 * NOTE! The in-memory inode i_data array is in little-endian order
4697 * even on big-endian machines: we do NOT byteswap the block numbers!
4699 for (block = 0; block < EXT4_N_BLOCKS; block++)
4700 ei->i_data[block] = raw_inode->i_block[block];
4701 INIT_LIST_HEAD(&ei->i_orphan);
4704 * Set transaction id's of transactions that have to be committed
4705 * to finish f[data]sync. We set them to currently running transaction
4706 * as we cannot be sure that the inode or some of its metadata isn't
4707 * part of the transaction - the inode could have been reclaimed and
4708 * now it is reread from disk.
4711 transaction_t *transaction;
4714 read_lock(&journal->j_state_lock);
4715 if (journal->j_running_transaction)
4716 transaction = journal->j_running_transaction;
4718 transaction = journal->j_committing_transaction;
4720 tid = transaction->t_tid;
4722 tid = journal->j_commit_sequence;
4723 read_unlock(&journal->j_state_lock);
4724 ei->i_sync_tid = tid;
4725 ei->i_datasync_tid = tid;
4728 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4729 if (ei->i_extra_isize == 0) {
4730 /* The extra space is currently unused. Use it. */
4731 ei->i_extra_isize = sizeof(struct ext4_inode) -
4732 EXT4_GOOD_OLD_INODE_SIZE;
4734 ext4_iget_extra_inode(inode, raw_inode, ei);
4738 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4739 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4740 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4741 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4743 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4744 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4745 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4746 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4748 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4753 if (ei->i_file_acl &&
4754 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4755 ext4_error_inode(inode, function, line, 0,
4756 "iget: bad extended attribute block %llu",
4758 ret = -EFSCORRUPTED;
4760 } else if (!ext4_has_inline_data(inode)) {
4761 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4762 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4763 (S_ISLNK(inode->i_mode) &&
4764 !ext4_inode_is_fast_symlink(inode))))
4765 /* Validate extent which is part of inode */
4766 ret = ext4_ext_check_inode(inode);
4767 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4768 (S_ISLNK(inode->i_mode) &&
4769 !ext4_inode_is_fast_symlink(inode))) {
4770 /* Validate block references which are part of inode */
4771 ret = ext4_ind_check_inode(inode);
4777 if (S_ISREG(inode->i_mode)) {
4778 inode->i_op = &ext4_file_inode_operations;
4779 inode->i_fop = &ext4_file_operations;
4780 ext4_set_aops(inode);
4781 } else if (S_ISDIR(inode->i_mode)) {
4782 inode->i_op = &ext4_dir_inode_operations;
4783 inode->i_fop = &ext4_dir_operations;
4784 } else if (S_ISLNK(inode->i_mode)) {
4785 if (ext4_encrypted_inode(inode)) {
4786 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4787 ext4_set_aops(inode);
4788 } else if (ext4_inode_is_fast_symlink(inode)) {
4789 inode->i_link = (char *)ei->i_data;
4790 inode->i_op = &ext4_fast_symlink_inode_operations;
4791 nd_terminate_link(ei->i_data, inode->i_size,
4792 sizeof(ei->i_data) - 1);
4794 inode->i_op = &ext4_symlink_inode_operations;
4795 ext4_set_aops(inode);
4797 inode_nohighmem(inode);
4798 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4799 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4800 inode->i_op = &ext4_special_inode_operations;
4801 if (raw_inode->i_block[0])
4802 init_special_inode(inode, inode->i_mode,
4803 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4805 init_special_inode(inode, inode->i_mode,
4806 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4807 } else if (ino == EXT4_BOOT_LOADER_INO) {
4808 make_bad_inode(inode);
4810 ret = -EFSCORRUPTED;
4811 ext4_error_inode(inode, function, line, 0,
4812 "iget: bogus i_mode (%o)", inode->i_mode);
4816 ext4_set_inode_flags(inode);
4817 unlock_new_inode(inode);
4823 return ERR_PTR(ret);
4826 static int ext4_inode_blocks_set(handle_t *handle,
4827 struct ext4_inode *raw_inode,
4828 struct ext4_inode_info *ei)
4830 struct inode *inode = &(ei->vfs_inode);
4831 u64 i_blocks = READ_ONCE(inode->i_blocks);
4832 struct super_block *sb = inode->i_sb;
4834 if (i_blocks <= ~0U) {
4836 * i_blocks can be represented in a 32 bit variable
4837 * as multiple of 512 bytes
4839 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4840 raw_inode->i_blocks_high = 0;
4841 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4844 if (!ext4_has_feature_huge_file(sb))
4847 if (i_blocks <= 0xffffffffffffULL) {
4849 * i_blocks can be represented in a 48 bit variable
4850 * as multiple of 512 bytes
4852 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4853 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4854 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4856 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4857 /* i_block is stored in file system block size */
4858 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4859 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4860 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4865 struct other_inode {
4866 unsigned long orig_ino;
4867 struct ext4_inode *raw_inode;
4870 static int other_inode_match(struct inode * inode, unsigned long ino,
4873 struct other_inode *oi = (struct other_inode *) data;
4875 if ((inode->i_ino != ino) ||
4876 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4877 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4878 ((inode->i_state & I_DIRTY_TIME) == 0))
4880 spin_lock(&inode->i_lock);
4881 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4882 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4883 (inode->i_state & I_DIRTY_TIME)) {
4884 struct ext4_inode_info *ei = EXT4_I(inode);
4886 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4887 spin_unlock(&inode->i_lock);
4889 spin_lock(&ei->i_raw_lock);
4890 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4891 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4892 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4893 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4894 spin_unlock(&ei->i_raw_lock);
4895 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4898 spin_unlock(&inode->i_lock);
4903 * Opportunistically update the other time fields for other inodes in
4904 * the same inode table block.
4906 static void ext4_update_other_inodes_time(struct super_block *sb,
4907 unsigned long orig_ino, char *buf)
4909 struct other_inode oi;
4911 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4912 int inode_size = EXT4_INODE_SIZE(sb);
4914 oi.orig_ino = orig_ino;
4916 * Calculate the first inode in the inode table block. Inode
4917 * numbers are one-based. That is, the first inode in a block
4918 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4920 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4921 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4922 if (ino == orig_ino)
4924 oi.raw_inode = (struct ext4_inode *) buf;
4925 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4930 * Post the struct inode info into an on-disk inode location in the
4931 * buffer-cache. This gobbles the caller's reference to the
4932 * buffer_head in the inode location struct.
4934 * The caller must have write access to iloc->bh.
4936 static int ext4_do_update_inode(handle_t *handle,
4937 struct inode *inode,
4938 struct ext4_iloc *iloc)
4940 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4941 struct ext4_inode_info *ei = EXT4_I(inode);
4942 struct buffer_head *bh = iloc->bh;
4943 struct super_block *sb = inode->i_sb;
4945 int need_datasync = 0, set_large_file = 0;
4950 spin_lock(&ei->i_raw_lock);
4952 /* For fields not tracked in the in-memory inode,
4953 * initialise them to zero for new inodes. */
4954 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4955 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4957 ext4_get_inode_flags(ei);
4958 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4959 i_uid = i_uid_read(inode);
4960 i_gid = i_gid_read(inode);
4961 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4962 if (!(test_opt(inode->i_sb, NO_UID32))) {
4963 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4964 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4966 * Fix up interoperability with old kernels. Otherwise, old inodes get
4967 * re-used with the upper 16 bits of the uid/gid intact
4969 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4970 raw_inode->i_uid_high = 0;
4971 raw_inode->i_gid_high = 0;
4973 raw_inode->i_uid_high =
4974 cpu_to_le16(high_16_bits(i_uid));
4975 raw_inode->i_gid_high =
4976 cpu_to_le16(high_16_bits(i_gid));
4979 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4980 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4981 raw_inode->i_uid_high = 0;
4982 raw_inode->i_gid_high = 0;
4984 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4986 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4987 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4988 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4989 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4991 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4993 spin_unlock(&ei->i_raw_lock);
4996 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4997 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4998 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4999 raw_inode->i_file_acl_high =
5000 cpu_to_le16(ei->i_file_acl >> 32);
5001 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5002 if (ei->i_disksize != ext4_isize(raw_inode)) {
5003 ext4_isize_set(raw_inode, ei->i_disksize);
5006 if (ei->i_disksize > 0x7fffffffULL) {
5007 if (!ext4_has_feature_large_file(sb) ||
5008 EXT4_SB(sb)->s_es->s_rev_level ==
5009 cpu_to_le32(EXT4_GOOD_OLD_REV))
5012 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5013 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5014 if (old_valid_dev(inode->i_rdev)) {
5015 raw_inode->i_block[0] =
5016 cpu_to_le32(old_encode_dev(inode->i_rdev));
5017 raw_inode->i_block[1] = 0;
5019 raw_inode->i_block[0] = 0;
5020 raw_inode->i_block[1] =
5021 cpu_to_le32(new_encode_dev(inode->i_rdev));
5022 raw_inode->i_block[2] = 0;
5024 } else if (!ext4_has_inline_data(inode)) {
5025 for (block = 0; block < EXT4_N_BLOCKS; block++)
5026 raw_inode->i_block[block] = ei->i_data[block];
5029 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5030 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5031 if (ei->i_extra_isize) {
5032 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5033 raw_inode->i_version_hi =
5034 cpu_to_le32(inode->i_version >> 32);
5035 raw_inode->i_extra_isize =
5036 cpu_to_le16(ei->i_extra_isize);
5040 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5041 i_projid != EXT4_DEF_PROJID);
5043 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5044 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5045 raw_inode->i_projid = cpu_to_le32(i_projid);
5047 ext4_inode_csum_set(inode, raw_inode, ei);
5048 spin_unlock(&ei->i_raw_lock);
5049 if (inode->i_sb->s_flags & MS_LAZYTIME)
5050 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5053 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5054 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5057 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5058 if (set_large_file) {
5059 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5060 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5063 ext4_update_dynamic_rev(sb);
5064 ext4_set_feature_large_file(sb);
5065 ext4_handle_sync(handle);
5066 err = ext4_handle_dirty_super(handle, sb);
5068 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5071 ext4_std_error(inode->i_sb, err);
5076 * ext4_write_inode()
5078 * We are called from a few places:
5080 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5081 * Here, there will be no transaction running. We wait for any running
5082 * transaction to commit.
5084 * - Within flush work (sys_sync(), kupdate and such).
5085 * We wait on commit, if told to.
5087 * - Within iput_final() -> write_inode_now()
5088 * We wait on commit, if told to.
5090 * In all cases it is actually safe for us to return without doing anything,
5091 * because the inode has been copied into a raw inode buffer in
5092 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5095 * Note that we are absolutely dependent upon all inode dirtiers doing the
5096 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5097 * which we are interested.
5099 * It would be a bug for them to not do this. The code:
5101 * mark_inode_dirty(inode)
5103 * inode->i_size = expr;
5105 * is in error because write_inode() could occur while `stuff()' is running,
5106 * and the new i_size will be lost. Plus the inode will no longer be on the
5107 * superblock's dirty inode list.
5109 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5113 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5116 if (EXT4_SB(inode->i_sb)->s_journal) {
5117 if (ext4_journal_current_handle()) {
5118 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5124 * No need to force transaction in WB_SYNC_NONE mode. Also
5125 * ext4_sync_fs() will force the commit after everything is
5128 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5131 err = ext4_force_commit(inode->i_sb);
5133 struct ext4_iloc iloc;
5135 err = __ext4_get_inode_loc(inode, &iloc, 0);
5139 * sync(2) will flush the whole buffer cache. No need to do
5140 * it here separately for each inode.
5142 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5143 sync_dirty_buffer(iloc.bh);
5144 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5145 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5146 "IO error syncing inode");
5155 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5156 * buffers that are attached to a page stradding i_size and are undergoing
5157 * commit. In that case we have to wait for commit to finish and try again.
5159 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5163 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5164 tid_t commit_tid = 0;
5167 offset = inode->i_size & (PAGE_SIZE - 1);
5169 * If the page is fully truncated, we don't need to wait for any commit
5170 * (and we even should not as __ext4_journalled_invalidatepage() may
5171 * strip all buffers from the page but keep the page dirty which can then
5172 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5173 * buffers). Also we don't need to wait for any commit if all buffers in
5174 * the page remain valid. This is most beneficial for the common case of
5175 * blocksize == PAGESIZE.
5177 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5180 page = find_lock_page(inode->i_mapping,
5181 inode->i_size >> PAGE_SHIFT);
5184 ret = __ext4_journalled_invalidatepage(page, offset,
5185 PAGE_SIZE - offset);
5191 read_lock(&journal->j_state_lock);
5192 if (journal->j_committing_transaction)
5193 commit_tid = journal->j_committing_transaction->t_tid;
5194 read_unlock(&journal->j_state_lock);
5196 jbd2_log_wait_commit(journal, commit_tid);
5203 * Called from notify_change.
5205 * We want to trap VFS attempts to truncate the file as soon as
5206 * possible. In particular, we want to make sure that when the VFS
5207 * shrinks i_size, we put the inode on the orphan list and modify
5208 * i_disksize immediately, so that during the subsequent flushing of
5209 * dirty pages and freeing of disk blocks, we can guarantee that any
5210 * commit will leave the blocks being flushed in an unused state on
5211 * disk. (On recovery, the inode will get truncated and the blocks will
5212 * be freed, so we have a strong guarantee that no future commit will
5213 * leave these blocks visible to the user.)
5215 * Another thing we have to assure is that if we are in ordered mode
5216 * and inode is still attached to the committing transaction, we must
5217 * we start writeout of all the dirty pages which are being truncated.
5218 * This way we are sure that all the data written in the previous
5219 * transaction are already on disk (truncate waits for pages under
5222 * Called with inode->i_mutex down.
5224 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5226 struct inode *inode = d_inode(dentry);
5229 const unsigned int ia_valid = attr->ia_valid;
5231 error = setattr_prepare(dentry, attr);
5235 if (is_quota_modification(inode, attr)) {
5236 error = dquot_initialize(inode);
5240 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5241 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5244 /* (user+group)*(old+new) structure, inode write (sb,
5245 * inode block, ? - but truncate inode update has it) */
5246 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5247 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5248 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5249 if (IS_ERR(handle)) {
5250 error = PTR_ERR(handle);
5253 error = dquot_transfer(inode, attr);
5255 ext4_journal_stop(handle);
5258 /* Update corresponding info in inode so that everything is in
5259 * one transaction */
5260 if (attr->ia_valid & ATTR_UID)
5261 inode->i_uid = attr->ia_uid;
5262 if (attr->ia_valid & ATTR_GID)
5263 inode->i_gid = attr->ia_gid;
5264 error = ext4_mark_inode_dirty(handle, inode);
5265 ext4_journal_stop(handle);
5268 if (attr->ia_valid & ATTR_SIZE) {
5270 loff_t oldsize = inode->i_size;
5271 int shrink = (attr->ia_size <= inode->i_size);
5273 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5274 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5276 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5279 if (!S_ISREG(inode->i_mode))
5282 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5283 inode_inc_iversion(inode);
5285 if (ext4_should_order_data(inode) &&
5286 (attr->ia_size < inode->i_size)) {
5287 error = ext4_begin_ordered_truncate(inode,
5292 if (attr->ia_size != inode->i_size) {
5293 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5294 if (IS_ERR(handle)) {
5295 error = PTR_ERR(handle);
5298 if (ext4_handle_valid(handle) && shrink) {
5299 error = ext4_orphan_add(handle, inode);
5303 * Update c/mtime on truncate up, ext4_truncate() will
5304 * update c/mtime in shrink case below
5307 inode->i_mtime = ext4_current_time(inode);
5308 inode->i_ctime = inode->i_mtime;
5310 down_write(&EXT4_I(inode)->i_data_sem);
5311 EXT4_I(inode)->i_disksize = attr->ia_size;
5312 rc = ext4_mark_inode_dirty(handle, inode);
5316 * We have to update i_size under i_data_sem together
5317 * with i_disksize to avoid races with writeback code
5318 * running ext4_wb_update_i_disksize().
5321 i_size_write(inode, attr->ia_size);
5322 up_write(&EXT4_I(inode)->i_data_sem);
5323 ext4_journal_stop(handle);
5325 if (orphan && inode->i_nlink)
5326 ext4_orphan_del(NULL, inode);
5331 pagecache_isize_extended(inode, oldsize, inode->i_size);
5334 * Blocks are going to be removed from the inode. Wait
5335 * for dio in flight. Temporarily disable
5336 * dioread_nolock to prevent livelock.
5339 if (!ext4_should_journal_data(inode)) {
5340 ext4_inode_block_unlocked_dio(inode);
5341 inode_dio_wait(inode);
5342 ext4_inode_resume_unlocked_dio(inode);
5344 ext4_wait_for_tail_page_commit(inode);
5346 down_write(&EXT4_I(inode)->i_mmap_sem);
5348 * Truncate pagecache after we've waited for commit
5349 * in data=journal mode to make pages freeable.
5351 truncate_pagecache(inode, inode->i_size);
5353 ext4_truncate(inode);
5354 up_write(&EXT4_I(inode)->i_mmap_sem);
5358 setattr_copy(inode, attr);
5359 mark_inode_dirty(inode);
5363 * If the call to ext4_truncate failed to get a transaction handle at
5364 * all, we need to clean up the in-core orphan list manually.
5366 if (orphan && inode->i_nlink)
5367 ext4_orphan_del(NULL, inode);
5369 if (!rc && (ia_valid & ATTR_MODE))
5370 rc = posix_acl_chmod(inode, inode->i_mode);
5373 ext4_std_error(inode->i_sb, error);
5379 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5382 struct inode *inode;
5383 unsigned long long delalloc_blocks;
5385 inode = d_inode(dentry);
5386 generic_fillattr(inode, stat);
5389 * If there is inline data in the inode, the inode will normally not
5390 * have data blocks allocated (it may have an external xattr block).
5391 * Report at least one sector for such files, so tools like tar, rsync,
5392 * others doen't incorrectly think the file is completely sparse.
5394 if (unlikely(ext4_has_inline_data(inode)))
5395 stat->blocks += (stat->size + 511) >> 9;
5398 * We can't update i_blocks if the block allocation is delayed
5399 * otherwise in the case of system crash before the real block
5400 * allocation is done, we will have i_blocks inconsistent with
5401 * on-disk file blocks.
5402 * We always keep i_blocks updated together with real
5403 * allocation. But to not confuse with user, stat
5404 * will return the blocks that include the delayed allocation
5405 * blocks for this file.
5407 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5408 EXT4_I(inode)->i_reserved_data_blocks);
5409 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5413 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5416 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5417 return ext4_ind_trans_blocks(inode, lblocks);
5418 return ext4_ext_index_trans_blocks(inode, pextents);
5422 * Account for index blocks, block groups bitmaps and block group
5423 * descriptor blocks if modify datablocks and index blocks
5424 * worse case, the indexs blocks spread over different block groups
5426 * If datablocks are discontiguous, they are possible to spread over
5427 * different block groups too. If they are contiguous, with flexbg,
5428 * they could still across block group boundary.
5430 * Also account for superblock, inode, quota and xattr blocks
5432 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5435 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5441 * How many index blocks need to touch to map @lblocks logical blocks
5442 * to @pextents physical extents?
5444 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5449 * Now let's see how many group bitmaps and group descriptors need
5452 groups = idxblocks + pextents;
5454 if (groups > ngroups)
5456 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5457 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5459 /* bitmaps and block group descriptor blocks */
5460 ret += groups + gdpblocks;
5462 /* Blocks for super block, inode, quota and xattr blocks */
5463 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5469 * Calculate the total number of credits to reserve to fit
5470 * the modification of a single pages into a single transaction,
5471 * which may include multiple chunks of block allocations.
5473 * This could be called via ext4_write_begin()
5475 * We need to consider the worse case, when
5476 * one new block per extent.
5478 int ext4_writepage_trans_blocks(struct inode *inode)
5480 int bpp = ext4_journal_blocks_per_page(inode);
5483 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5485 /* Account for data blocks for journalled mode */
5486 if (ext4_should_journal_data(inode))
5492 * Calculate the journal credits for a chunk of data modification.
5494 * This is called from DIO, fallocate or whoever calling
5495 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5497 * journal buffers for data blocks are not included here, as DIO
5498 * and fallocate do no need to journal data buffers.
5500 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5502 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5506 * The caller must have previously called ext4_reserve_inode_write().
5507 * Give this, we know that the caller already has write access to iloc->bh.
5509 int ext4_mark_iloc_dirty(handle_t *handle,
5510 struct inode *inode, struct ext4_iloc *iloc)
5514 if (IS_I_VERSION(inode))
5515 inode_inc_iversion(inode);
5517 /* the do_update_inode consumes one bh->b_count */
5520 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5521 err = ext4_do_update_inode(handle, inode, iloc);
5527 * On success, We end up with an outstanding reference count against
5528 * iloc->bh. This _must_ be cleaned up later.
5532 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5533 struct ext4_iloc *iloc)
5537 err = ext4_get_inode_loc(inode, iloc);
5539 BUFFER_TRACE(iloc->bh, "get_write_access");
5540 err = ext4_journal_get_write_access(handle, iloc->bh);
5546 ext4_std_error(inode->i_sb, err);
5551 * Expand an inode by new_extra_isize bytes.
5552 * Returns 0 on success or negative error number on failure.
5554 static int ext4_expand_extra_isize(struct inode *inode,
5555 unsigned int new_extra_isize,
5556 struct ext4_iloc iloc,
5559 struct ext4_inode *raw_inode;
5560 struct ext4_xattr_ibody_header *header;
5561 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5562 struct ext4_inode_info *ei = EXT4_I(inode);
5564 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5567 /* this was checked at iget time, but double check for good measure */
5568 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5569 (ei->i_extra_isize & 3)) {
5570 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5572 EXT4_INODE_SIZE(inode->i_sb));
5573 return -EFSCORRUPTED;
5575 if ((new_extra_isize < ei->i_extra_isize) ||
5576 (new_extra_isize < 4) ||
5577 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5578 return -EINVAL; /* Should never happen */
5580 raw_inode = ext4_raw_inode(&iloc);
5582 header = IHDR(inode, raw_inode);
5584 /* No extended attributes present */
5585 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5586 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5587 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5588 EXT4_I(inode)->i_extra_isize, 0,
5589 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5590 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5594 /* try to expand with EAs present */
5595 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5600 * What we do here is to mark the in-core inode as clean with respect to inode
5601 * dirtiness (it may still be data-dirty).
5602 * This means that the in-core inode may be reaped by prune_icache
5603 * without having to perform any I/O. This is a very good thing,
5604 * because *any* task may call prune_icache - even ones which
5605 * have a transaction open against a different journal.
5607 * Is this cheating? Not really. Sure, we haven't written the
5608 * inode out, but prune_icache isn't a user-visible syncing function.
5609 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5610 * we start and wait on commits.
5612 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5614 struct ext4_iloc iloc;
5615 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5616 static unsigned int mnt_count;
5620 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5621 err = ext4_reserve_inode_write(handle, inode, &iloc);
5624 if (ext4_handle_valid(handle) &&
5625 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5626 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5628 * We need extra buffer credits since we may write into EA block
5629 * with this same handle. If journal_extend fails, then it will
5630 * only result in a minor loss of functionality for that inode.
5631 * If this is felt to be critical, then e2fsck should be run to
5632 * force a large enough s_min_extra_isize.
5634 if ((jbd2_journal_extend(handle,
5635 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5636 ret = ext4_expand_extra_isize(inode,
5637 sbi->s_want_extra_isize,
5641 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5642 ext4_warning(inode->i_sb,
5643 "Unable to expand inode %lu. Delete"
5644 " some EAs or run e2fsck.",
5647 le16_to_cpu(sbi->s_es->s_mnt_count);
5652 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5656 * ext4_dirty_inode() is called from __mark_inode_dirty()
5658 * We're really interested in the case where a file is being extended.
5659 * i_size has been changed by generic_commit_write() and we thus need
5660 * to include the updated inode in the current transaction.
5662 * Also, dquot_alloc_block() will always dirty the inode when blocks
5663 * are allocated to the file.
5665 * If the inode is marked synchronous, we don't honour that here - doing
5666 * so would cause a commit on atime updates, which we don't bother doing.
5667 * We handle synchronous inodes at the highest possible level.
5669 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5670 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5671 * to copy into the on-disk inode structure are the timestamp files.
5673 void ext4_dirty_inode(struct inode *inode, int flags)
5677 if (flags == I_DIRTY_TIME)
5679 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5683 ext4_mark_inode_dirty(handle, inode);
5685 ext4_journal_stop(handle);
5692 * Bind an inode's backing buffer_head into this transaction, to prevent
5693 * it from being flushed to disk early. Unlike
5694 * ext4_reserve_inode_write, this leaves behind no bh reference and
5695 * returns no iloc structure, so the caller needs to repeat the iloc
5696 * lookup to mark the inode dirty later.
5698 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5700 struct ext4_iloc iloc;
5704 err = ext4_get_inode_loc(inode, &iloc);
5706 BUFFER_TRACE(iloc.bh, "get_write_access");
5707 err = jbd2_journal_get_write_access(handle, iloc.bh);
5709 err = ext4_handle_dirty_metadata(handle,
5715 ext4_std_error(inode->i_sb, err);
5720 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5725 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5728 * We have to be very careful here: changing a data block's
5729 * journaling status dynamically is dangerous. If we write a
5730 * data block to the journal, change the status and then delete
5731 * that block, we risk forgetting to revoke the old log record
5732 * from the journal and so a subsequent replay can corrupt data.
5733 * So, first we make sure that the journal is empty and that
5734 * nobody is changing anything.
5737 journal = EXT4_JOURNAL(inode);
5740 if (is_journal_aborted(journal))
5743 /* Wait for all existing dio workers */
5744 ext4_inode_block_unlocked_dio(inode);
5745 inode_dio_wait(inode);
5748 * Before flushing the journal and switching inode's aops, we have
5749 * to flush all dirty data the inode has. There can be outstanding
5750 * delayed allocations, there can be unwritten extents created by
5751 * fallocate or buffered writes in dioread_nolock mode covered by
5752 * dirty data which can be converted only after flushing the dirty
5753 * data (and journalled aops don't know how to handle these cases).
5756 down_write(&EXT4_I(inode)->i_mmap_sem);
5757 err = filemap_write_and_wait(inode->i_mapping);
5759 up_write(&EXT4_I(inode)->i_mmap_sem);
5760 ext4_inode_resume_unlocked_dio(inode);
5765 percpu_down_write(&sbi->s_writepages_rwsem);
5766 jbd2_journal_lock_updates(journal);
5769 * OK, there are no updates running now, and all cached data is
5770 * synced to disk. We are now in a completely consistent state
5771 * which doesn't have anything in the journal, and we know that
5772 * no filesystem updates are running, so it is safe to modify
5773 * the inode's in-core data-journaling state flag now.
5777 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5779 err = jbd2_journal_flush(journal);
5781 jbd2_journal_unlock_updates(journal);
5782 percpu_up_write(&sbi->s_writepages_rwsem);
5783 ext4_inode_resume_unlocked_dio(inode);
5786 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5788 ext4_set_aops(inode);
5790 jbd2_journal_unlock_updates(journal);
5791 percpu_up_write(&sbi->s_writepages_rwsem);
5794 up_write(&EXT4_I(inode)->i_mmap_sem);
5795 ext4_inode_resume_unlocked_dio(inode);
5797 /* Finally we can mark the inode as dirty. */
5799 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5801 return PTR_ERR(handle);
5803 err = ext4_mark_inode_dirty(handle, inode);
5804 ext4_handle_sync(handle);
5805 ext4_journal_stop(handle);
5806 ext4_std_error(inode->i_sb, err);
5811 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5813 return !buffer_mapped(bh);
5816 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5818 struct page *page = vmf->page;
5822 struct file *file = vma->vm_file;
5823 struct inode *inode = file_inode(file);
5824 struct address_space *mapping = inode->i_mapping;
5826 get_block_t *get_block;
5829 sb_start_pagefault(inode->i_sb);
5830 file_update_time(vma->vm_file);
5832 down_read(&EXT4_I(inode)->i_mmap_sem);
5834 ret = ext4_convert_inline_data(inode);
5838 /* Delalloc case is easy... */
5839 if (test_opt(inode->i_sb, DELALLOC) &&
5840 !ext4_should_journal_data(inode) &&
5841 !ext4_nonda_switch(inode->i_sb)) {
5843 ret = block_page_mkwrite(vma, vmf,
5844 ext4_da_get_block_prep);
5845 } while (ret == -ENOSPC &&
5846 ext4_should_retry_alloc(inode->i_sb, &retries));
5851 size = i_size_read(inode);
5852 /* Page got truncated from under us? */
5853 if (page->mapping != mapping || page_offset(page) > size) {
5855 ret = VM_FAULT_NOPAGE;
5859 if (page->index == size >> PAGE_SHIFT)
5860 len = size & ~PAGE_MASK;
5864 * Return if we have all the buffers mapped. This avoids the need to do
5865 * journal_start/journal_stop which can block and take a long time
5867 if (page_has_buffers(page)) {
5868 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5870 ext4_bh_unmapped)) {
5871 /* Wait so that we don't change page under IO */
5872 wait_for_stable_page(page);
5873 ret = VM_FAULT_LOCKED;
5878 /* OK, we need to fill the hole... */
5879 if (ext4_should_dioread_nolock(inode))
5880 get_block = ext4_get_block_unwritten;
5882 get_block = ext4_get_block;
5884 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5885 ext4_writepage_trans_blocks(inode));
5886 if (IS_ERR(handle)) {
5887 ret = VM_FAULT_SIGBUS;
5890 ret = block_page_mkwrite(vma, vmf, get_block);
5891 if (!ret && ext4_should_journal_data(inode)) {
5892 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5893 PAGE_SIZE, NULL, do_journal_get_write_access)) {
5895 ret = VM_FAULT_SIGBUS;
5896 ext4_journal_stop(handle);
5899 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5901 ext4_journal_stop(handle);
5902 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5905 ret = block_page_mkwrite_return(ret);
5907 up_read(&EXT4_I(inode)->i_mmap_sem);
5908 sb_end_pagefault(inode->i_sb);
5912 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5914 struct inode *inode = file_inode(vma->vm_file);
5917 down_read(&EXT4_I(inode)->i_mmap_sem);
5918 err = filemap_fault(vma, vmf);
5919 up_read(&EXT4_I(inode)->i_mmap_sem);
5925 * Find the first extent at or after @lblk in an inode that is not a hole.
5926 * Search for @map_len blocks at most. The extent is returned in @result.
5928 * The function returns 1 if we found an extent. The function returns 0 in
5929 * case there is no extent at or after @lblk and in that case also sets
5930 * @result->es_len to 0. In case of error, the error code is returned.
5932 int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
5933 unsigned int map_len, struct extent_status *result)
5935 struct ext4_map_blocks map;
5936 struct extent_status es = {};
5940 map.m_len = map_len;
5943 * For non-extent based files this loop may iterate several times since
5944 * we do not determine full hole size.
5946 while (map.m_len > 0) {
5947 ret = ext4_map_blocks(NULL, inode, &map, 0);
5950 /* There's extent covering m_lblk? Just return it. */
5954 ext4_es_store_pblock(result, map.m_pblk);
5955 result->es_lblk = map.m_lblk;
5956 result->es_len = map.m_len;
5957 if (map.m_flags & EXT4_MAP_UNWRITTEN)
5958 status = EXTENT_STATUS_UNWRITTEN;
5960 status = EXTENT_STATUS_WRITTEN;
5961 ext4_es_store_status(result, status);
5964 ext4_es_find_delayed_extent_range(inode, map.m_lblk,
5965 map.m_lblk + map.m_len - 1,
5967 /* Is delalloc data before next block in extent tree? */
5968 if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
5969 ext4_lblk_t offset = 0;
5971 if (es.es_lblk < lblk)
5972 offset = lblk - es.es_lblk;
5973 result->es_lblk = es.es_lblk + offset;
5974 ext4_es_store_pblock(result,
5975 ext4_es_pblock(&es) + offset);
5976 result->es_len = es.es_len - offset;
5977 ext4_es_store_status(result, ext4_es_status(&es));
5981 /* There's a hole at m_lblk, advance us after it */
5982 map.m_lblk += map.m_len;
5983 map_len -= map.m_len;
5984 map.m_len = map_len;