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);
1221 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1222 if (IS_ERR(handle)) {
1224 return PTR_ERR(handle);
1228 if (page->mapping != mapping) {
1229 /* The page got truncated from under us */
1232 ext4_journal_stop(handle);
1235 /* In case writeback began while the page was unlocked */
1236 wait_for_stable_page(page);
1238 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1239 if (ext4_should_dioread_nolock(inode))
1240 ret = ext4_block_write_begin(page, pos, len,
1241 ext4_get_block_unwritten);
1243 ret = ext4_block_write_begin(page, pos, len,
1246 if (ext4_should_dioread_nolock(inode))
1247 ret = __block_write_begin(page, pos, len,
1248 ext4_get_block_unwritten);
1250 ret = __block_write_begin(page, pos, len, ext4_get_block);
1252 if (!ret && ext4_should_journal_data(inode)) {
1253 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1255 do_journal_get_write_access);
1261 * __block_write_begin may have instantiated a few blocks
1262 * outside i_size. Trim these off again. Don't need
1263 * i_size_read because we hold i_mutex.
1265 * Add inode to orphan list in case we crash before
1268 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1269 ext4_orphan_add(handle, inode);
1271 ext4_journal_stop(handle);
1272 if (pos + len > inode->i_size) {
1273 ext4_truncate_failed_write(inode);
1275 * If truncate failed early the inode might
1276 * still be on the orphan list; we need to
1277 * make sure the inode is removed from the
1278 * orphan list in that case.
1281 ext4_orphan_del(NULL, inode);
1284 if (ret == -ENOSPC &&
1285 ext4_should_retry_alloc(inode->i_sb, &retries))
1294 /* For write_end() in data=journal mode */
1295 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1298 if (!buffer_mapped(bh) || buffer_freed(bh))
1300 set_buffer_uptodate(bh);
1301 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1302 clear_buffer_meta(bh);
1303 clear_buffer_prio(bh);
1308 * We need to pick up the new inode size which generic_commit_write gave us
1309 * `file' can be NULL - eg, when called from page_symlink().
1311 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1312 * buffers are managed internally.
1314 static int ext4_write_end(struct file *file,
1315 struct address_space *mapping,
1316 loff_t pos, unsigned len, unsigned copied,
1317 struct page *page, void *fsdata)
1319 handle_t *handle = ext4_journal_current_handle();
1320 struct inode *inode = mapping->host;
1321 loff_t old_size = inode->i_size;
1323 int i_size_changed = 0;
1324 int inline_data = ext4_has_inline_data(inode);
1326 trace_ext4_write_end(inode, pos, len, copied);
1328 ret = ext4_write_inline_data_end(inode, pos, len,
1337 copied = block_write_end(file, mapping, pos,
1338 len, copied, page, fsdata);
1340 * it's important to update i_size while still holding page lock:
1341 * page writeout could otherwise come in and zero beyond i_size.
1343 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1348 pagecache_isize_extended(inode, old_size, pos);
1350 * Don't mark the inode dirty under page lock. First, it unnecessarily
1351 * makes the holding time of page lock longer. Second, it forces lock
1352 * ordering of page lock and transaction start for journaling
1355 if (i_size_changed || inline_data)
1356 ext4_mark_inode_dirty(handle, inode);
1358 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1359 /* if we have allocated more blocks and copied
1360 * less. We will have blocks allocated outside
1361 * inode->i_size. So truncate them
1363 ext4_orphan_add(handle, inode);
1365 ret2 = ext4_journal_stop(handle);
1369 if (pos + len > inode->i_size) {
1370 ext4_truncate_failed_write(inode);
1372 * If truncate failed early the inode might still be
1373 * on the orphan list; we need to make sure the inode
1374 * is removed from the orphan list in that case.
1377 ext4_orphan_del(NULL, inode);
1380 return ret ? ret : copied;
1384 * This is a private version of page_zero_new_buffers() which doesn't
1385 * set the buffer to be dirty, since in data=journalled mode we need
1386 * to call ext4_handle_dirty_metadata() instead.
1388 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1390 unsigned from, unsigned to)
1392 unsigned int block_start = 0, block_end;
1393 struct buffer_head *head, *bh;
1395 bh = head = page_buffers(page);
1397 block_end = block_start + bh->b_size;
1398 if (buffer_new(bh)) {
1399 if (block_end > from && block_start < to) {
1400 if (!PageUptodate(page)) {
1401 unsigned start, size;
1403 start = max(from, block_start);
1404 size = min(to, block_end) - start;
1406 zero_user(page, start, size);
1407 write_end_fn(handle, bh);
1409 clear_buffer_new(bh);
1412 block_start = block_end;
1413 bh = bh->b_this_page;
1414 } while (bh != head);
1417 static int ext4_journalled_write_end(struct file *file,
1418 struct address_space *mapping,
1419 loff_t pos, unsigned len, unsigned copied,
1420 struct page *page, void *fsdata)
1422 handle_t *handle = ext4_journal_current_handle();
1423 struct inode *inode = mapping->host;
1424 loff_t old_size = inode->i_size;
1428 int size_changed = 0;
1429 int inline_data = ext4_has_inline_data(inode);
1431 trace_ext4_journalled_write_end(inode, pos, len, copied);
1432 from = pos & (PAGE_SIZE - 1);
1435 BUG_ON(!ext4_handle_valid(handle));
1438 ret = ext4_write_inline_data_end(inode, pos, len,
1446 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1448 ext4_journalled_zero_new_buffers(handle, page, from, to);
1450 if (unlikely(copied < len))
1451 ext4_journalled_zero_new_buffers(handle, page,
1453 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1454 from + copied, &partial,
1457 SetPageUptodate(page);
1459 size_changed = ext4_update_inode_size(inode, pos + copied);
1460 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1461 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1466 pagecache_isize_extended(inode, old_size, pos);
1468 if (size_changed || inline_data) {
1469 ret2 = ext4_mark_inode_dirty(handle, inode);
1474 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1475 /* if we have allocated more blocks and copied
1476 * less. We will have blocks allocated outside
1477 * inode->i_size. So truncate them
1479 ext4_orphan_add(handle, inode);
1482 ret2 = ext4_journal_stop(handle);
1485 if (pos + len > inode->i_size) {
1486 ext4_truncate_failed_write(inode);
1488 * If truncate failed early the inode might still be
1489 * on the orphan list; we need to make sure the inode
1490 * is removed from the orphan list in that case.
1493 ext4_orphan_del(NULL, inode);
1496 return ret ? ret : copied;
1500 * Reserve space for a single cluster
1502 static int ext4_da_reserve_space(struct inode *inode)
1504 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1505 struct ext4_inode_info *ei = EXT4_I(inode);
1509 * We will charge metadata quota at writeout time; this saves
1510 * us from metadata over-estimation, though we may go over by
1511 * a small amount in the end. Here we just reserve for data.
1513 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1517 spin_lock(&ei->i_block_reservation_lock);
1518 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1519 spin_unlock(&ei->i_block_reservation_lock);
1520 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1523 ei->i_reserved_data_blocks++;
1524 trace_ext4_da_reserve_space(inode);
1525 spin_unlock(&ei->i_block_reservation_lock);
1527 return 0; /* success */
1530 static void ext4_da_release_space(struct inode *inode, int to_free)
1532 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1533 struct ext4_inode_info *ei = EXT4_I(inode);
1536 return; /* Nothing to release, exit */
1538 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1540 trace_ext4_da_release_space(inode, to_free);
1541 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1543 * if there aren't enough reserved blocks, then the
1544 * counter is messed up somewhere. Since this
1545 * function is called from invalidate page, it's
1546 * harmless to return without any action.
1548 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1549 "ino %lu, to_free %d with only %d reserved "
1550 "data blocks", inode->i_ino, to_free,
1551 ei->i_reserved_data_blocks);
1553 to_free = ei->i_reserved_data_blocks;
1555 ei->i_reserved_data_blocks -= to_free;
1557 /* update fs dirty data blocks counter */
1558 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1560 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1562 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1565 static void ext4_da_page_release_reservation(struct page *page,
1566 unsigned int offset,
1567 unsigned int length)
1569 int to_release = 0, contiguous_blks = 0;
1570 struct buffer_head *head, *bh;
1571 unsigned int curr_off = 0;
1572 struct inode *inode = page->mapping->host;
1573 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1574 unsigned int stop = offset + length;
1578 BUG_ON(stop > PAGE_SIZE || stop < length);
1580 head = page_buffers(page);
1583 unsigned int next_off = curr_off + bh->b_size;
1585 if (next_off > stop)
1588 if ((offset <= curr_off) && (buffer_delay(bh))) {
1591 clear_buffer_delay(bh);
1592 } else if (contiguous_blks) {
1593 lblk = page->index <<
1594 (PAGE_SHIFT - inode->i_blkbits);
1595 lblk += (curr_off >> inode->i_blkbits) -
1597 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1598 contiguous_blks = 0;
1600 curr_off = next_off;
1601 } while ((bh = bh->b_this_page) != head);
1603 if (contiguous_blks) {
1604 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1605 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1606 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1609 /* If we have released all the blocks belonging to a cluster, then we
1610 * need to release the reserved space for that cluster. */
1611 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1612 while (num_clusters > 0) {
1613 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1614 ((num_clusters - 1) << sbi->s_cluster_bits);
1615 if (sbi->s_cluster_ratio == 1 ||
1616 !ext4_find_delalloc_cluster(inode, lblk))
1617 ext4_da_release_space(inode, 1);
1624 * Delayed allocation stuff
1627 struct mpage_da_data {
1628 struct inode *inode;
1629 struct writeback_control *wbc;
1631 pgoff_t first_page; /* The first page to write */
1632 pgoff_t next_page; /* Current page to examine */
1633 pgoff_t last_page; /* Last page to examine */
1635 * Extent to map - this can be after first_page because that can be
1636 * fully mapped. We somewhat abuse m_flags to store whether the extent
1637 * is delalloc or unwritten.
1639 struct ext4_map_blocks map;
1640 struct ext4_io_submit io_submit; /* IO submission data */
1643 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1648 struct pagevec pvec;
1649 struct inode *inode = mpd->inode;
1650 struct address_space *mapping = inode->i_mapping;
1652 /* This is necessary when next_page == 0. */
1653 if (mpd->first_page >= mpd->next_page)
1656 index = mpd->first_page;
1657 end = mpd->next_page - 1;
1659 ext4_lblk_t start, last;
1660 start = index << (PAGE_SHIFT - inode->i_blkbits);
1661 last = end << (PAGE_SHIFT - inode->i_blkbits);
1662 ext4_es_remove_extent(inode, start, last - start + 1);
1665 pagevec_init(&pvec, 0);
1666 while (index <= end) {
1667 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1670 for (i = 0; i < nr_pages; i++) {
1671 struct page *page = pvec.pages[i];
1672 if (page->index > end)
1674 BUG_ON(!PageLocked(page));
1675 BUG_ON(PageWriteback(page));
1677 if (page_mapped(page))
1678 clear_page_dirty_for_io(page);
1679 block_invalidatepage(page, 0, PAGE_SIZE);
1680 ClearPageUptodate(page);
1684 index = pvec.pages[nr_pages - 1]->index + 1;
1685 pagevec_release(&pvec);
1689 static void ext4_print_free_blocks(struct inode *inode)
1691 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1692 struct super_block *sb = inode->i_sb;
1693 struct ext4_inode_info *ei = EXT4_I(inode);
1695 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1696 EXT4_C2B(EXT4_SB(inode->i_sb),
1697 ext4_count_free_clusters(sb)));
1698 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1699 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1700 (long long) EXT4_C2B(EXT4_SB(sb),
1701 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1702 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1703 (long long) EXT4_C2B(EXT4_SB(sb),
1704 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1705 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1706 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1707 ei->i_reserved_data_blocks);
1711 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1713 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1717 * This function is grabs code from the very beginning of
1718 * ext4_map_blocks, but assumes that the caller is from delayed write
1719 * time. This function looks up the requested blocks and sets the
1720 * buffer delay bit under the protection of i_data_sem.
1722 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1723 struct ext4_map_blocks *map,
1724 struct buffer_head *bh)
1726 struct extent_status es;
1728 sector_t invalid_block = ~((sector_t) 0xffff);
1729 #ifdef ES_AGGRESSIVE_TEST
1730 struct ext4_map_blocks orig_map;
1732 memcpy(&orig_map, map, sizeof(*map));
1735 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1739 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1740 "logical block %lu\n", inode->i_ino, map->m_len,
1741 (unsigned long) map->m_lblk);
1743 /* Lookup extent status tree firstly */
1744 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1745 if (ext4_es_is_hole(&es)) {
1747 down_read(&EXT4_I(inode)->i_data_sem);
1752 * Delayed extent could be allocated by fallocate.
1753 * So we need to check it.
1755 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1756 map_bh(bh, inode->i_sb, invalid_block);
1758 set_buffer_delay(bh);
1762 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1763 retval = es.es_len - (iblock - es.es_lblk);
1764 if (retval > map->m_len)
1765 retval = map->m_len;
1766 map->m_len = retval;
1767 if (ext4_es_is_written(&es))
1768 map->m_flags |= EXT4_MAP_MAPPED;
1769 else if (ext4_es_is_unwritten(&es))
1770 map->m_flags |= EXT4_MAP_UNWRITTEN;
1774 #ifdef ES_AGGRESSIVE_TEST
1775 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1781 * Try to see if we can get the block without requesting a new
1782 * file system block.
1784 down_read(&EXT4_I(inode)->i_data_sem);
1785 if (ext4_has_inline_data(inode))
1787 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1788 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1790 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1796 * XXX: __block_prepare_write() unmaps passed block,
1800 * If the block was allocated from previously allocated cluster,
1801 * then we don't need to reserve it again. However we still need
1802 * to reserve metadata for every block we're going to write.
1804 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1805 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1806 ret = ext4_da_reserve_space(inode);
1808 /* not enough space to reserve */
1814 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1815 ~0, EXTENT_STATUS_DELAYED);
1821 map_bh(bh, inode->i_sb, invalid_block);
1823 set_buffer_delay(bh);
1824 } else if (retval > 0) {
1826 unsigned int status;
1828 if (unlikely(retval != map->m_len)) {
1829 ext4_warning(inode->i_sb,
1830 "ES len assertion failed for inode "
1831 "%lu: retval %d != map->m_len %d",
1832 inode->i_ino, retval, map->m_len);
1836 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1837 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1838 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1839 map->m_pblk, status);
1845 up_read((&EXT4_I(inode)->i_data_sem));
1851 * This is a special get_block_t callback which is used by
1852 * ext4_da_write_begin(). It will either return mapped block or
1853 * reserve space for a single block.
1855 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1856 * We also have b_blocknr = -1 and b_bdev initialized properly
1858 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1859 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1860 * initialized properly.
1862 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1863 struct buffer_head *bh, int create)
1865 struct ext4_map_blocks map;
1868 BUG_ON(create == 0);
1869 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1871 map.m_lblk = iblock;
1875 * first, we need to know whether the block is allocated already
1876 * preallocated blocks are unmapped but should treated
1877 * the same as allocated blocks.
1879 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1883 map_bh(bh, inode->i_sb, map.m_pblk);
1884 ext4_update_bh_state(bh, map.m_flags);
1886 if (buffer_unwritten(bh)) {
1887 /* A delayed write to unwritten bh should be marked
1888 * new and mapped. Mapped ensures that we don't do
1889 * get_block multiple times when we write to the same
1890 * offset and new ensures that we do proper zero out
1891 * for partial write.
1894 set_buffer_mapped(bh);
1899 static int bget_one(handle_t *handle, struct buffer_head *bh)
1905 static int bput_one(handle_t *handle, struct buffer_head *bh)
1911 static int __ext4_journalled_writepage(struct page *page,
1914 struct address_space *mapping = page->mapping;
1915 struct inode *inode = mapping->host;
1916 struct buffer_head *page_bufs = NULL;
1917 handle_t *handle = NULL;
1918 int ret = 0, err = 0;
1919 int inline_data = ext4_has_inline_data(inode);
1920 struct buffer_head *inode_bh = NULL;
1922 ClearPageChecked(page);
1925 BUG_ON(page->index != 0);
1926 BUG_ON(len > ext4_get_max_inline_size(inode));
1927 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1928 if (inode_bh == NULL)
1931 page_bufs = page_buffers(page);
1936 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1940 * We need to release the page lock before we start the
1941 * journal, so grab a reference so the page won't disappear
1942 * out from under us.
1947 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1948 ext4_writepage_trans_blocks(inode));
1949 if (IS_ERR(handle)) {
1950 ret = PTR_ERR(handle);
1952 goto out_no_pagelock;
1954 BUG_ON(!ext4_handle_valid(handle));
1958 if (page->mapping != mapping) {
1959 /* The page got truncated from under us */
1960 ext4_journal_stop(handle);
1966 ret = ext4_mark_inode_dirty(handle, inode);
1968 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1969 do_journal_get_write_access);
1971 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1976 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1977 err = ext4_journal_stop(handle);
1981 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1985 if (!inline_data && page_bufs)
1986 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1993 * Note that we don't need to start a transaction unless we're journaling data
1994 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1995 * need to file the inode to the transaction's list in ordered mode because if
1996 * we are writing back data added by write(), the inode is already there and if
1997 * we are writing back data modified via mmap(), no one guarantees in which
1998 * transaction the data will hit the disk. In case we are journaling data, we
1999 * cannot start transaction directly because transaction start ranks above page
2000 * lock so we have to do some magic.
2002 * This function can get called via...
2003 * - ext4_writepages after taking page lock (have journal handle)
2004 * - journal_submit_inode_data_buffers (no journal handle)
2005 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2006 * - grab_page_cache when doing write_begin (have journal handle)
2008 * We don't do any block allocation in this function. If we have page with
2009 * multiple blocks we need to write those buffer_heads that are mapped. This
2010 * is important for mmaped based write. So if we do with blocksize 1K
2011 * truncate(f, 1024);
2012 * a = mmap(f, 0, 4096);
2014 * truncate(f, 4096);
2015 * we have in the page first buffer_head mapped via page_mkwrite call back
2016 * but other buffer_heads would be unmapped but dirty (dirty done via the
2017 * do_wp_page). So writepage should write the first block. If we modify
2018 * the mmap area beyond 1024 we will again get a page_fault and the
2019 * page_mkwrite callback will do the block allocation and mark the
2020 * buffer_heads mapped.
2022 * We redirty the page if we have any buffer_heads that is either delay or
2023 * unwritten in the page.
2025 * We can get recursively called as show below.
2027 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2030 * But since we don't do any block allocation we should not deadlock.
2031 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2033 static int ext4_writepage(struct page *page,
2034 struct writeback_control *wbc)
2039 struct buffer_head *page_bufs = NULL;
2040 struct inode *inode = page->mapping->host;
2041 struct ext4_io_submit io_submit;
2042 bool keep_towrite = false;
2044 trace_ext4_writepage(page);
2045 size = i_size_read(inode);
2046 if (page->index == size >> PAGE_SHIFT)
2047 len = size & ~PAGE_MASK;
2051 page_bufs = page_buffers(page);
2053 * We cannot do block allocation or other extent handling in this
2054 * function. If there are buffers needing that, we have to redirty
2055 * the page. But we may reach here when we do a journal commit via
2056 * journal_submit_inode_data_buffers() and in that case we must write
2057 * allocated buffers to achieve data=ordered mode guarantees.
2059 * Also, if there is only one buffer per page (the fs block
2060 * size == the page size), if one buffer needs block
2061 * allocation or needs to modify the extent tree to clear the
2062 * unwritten flag, we know that the page can't be written at
2063 * all, so we might as well refuse the write immediately.
2064 * Unfortunately if the block size != page size, we can't as
2065 * easily detect this case using ext4_walk_page_buffers(), but
2066 * for the extremely common case, this is an optimization that
2067 * skips a useless round trip through ext4_bio_write_page().
2069 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2070 ext4_bh_delay_or_unwritten)) {
2071 redirty_page_for_writepage(wbc, page);
2072 if ((current->flags & PF_MEMALLOC) ||
2073 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2075 * For memory cleaning there's no point in writing only
2076 * some buffers. So just bail out. Warn if we came here
2077 * from direct reclaim.
2079 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2084 keep_towrite = true;
2087 if (PageChecked(page) && ext4_should_journal_data(inode))
2089 * It's mmapped pagecache. Add buffers and journal it. There
2090 * doesn't seem much point in redirtying the page here.
2092 return __ext4_journalled_writepage(page, len);
2094 ext4_io_submit_init(&io_submit, wbc);
2095 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2096 if (!io_submit.io_end) {
2097 redirty_page_for_writepage(wbc, page);
2101 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2102 ext4_io_submit(&io_submit);
2103 /* Drop io_end reference we got from init */
2104 ext4_put_io_end_defer(io_submit.io_end);
2108 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2114 BUG_ON(page->index != mpd->first_page);
2115 clear_page_dirty_for_io(page);
2117 * We have to be very careful here! Nothing protects writeback path
2118 * against i_size changes and the page can be writeably mapped into
2119 * page tables. So an application can be growing i_size and writing
2120 * data through mmap while writeback runs. clear_page_dirty_for_io()
2121 * write-protects our page in page tables and the page cannot get
2122 * written to again until we release page lock. So only after
2123 * clear_page_dirty_for_io() we are safe to sample i_size for
2124 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2125 * on the barrier provided by TestClearPageDirty in
2126 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2127 * after page tables are updated.
2129 size = i_size_read(mpd->inode);
2130 if (page->index == size >> PAGE_SHIFT)
2131 len = size & ~PAGE_MASK;
2134 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2136 mpd->wbc->nr_to_write--;
2142 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2145 * mballoc gives us at most this number of blocks...
2146 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2147 * The rest of mballoc seems to handle chunks up to full group size.
2149 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2152 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2154 * @mpd - extent of blocks
2155 * @lblk - logical number of the block in the file
2156 * @bh - buffer head we want to add to the extent
2158 * The function is used to collect contig. blocks in the same state. If the
2159 * buffer doesn't require mapping for writeback and we haven't started the
2160 * extent of buffers to map yet, the function returns 'true' immediately - the
2161 * caller can write the buffer right away. Otherwise the function returns true
2162 * if the block has been added to the extent, false if the block couldn't be
2165 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2166 struct buffer_head *bh)
2168 struct ext4_map_blocks *map = &mpd->map;
2170 /* Buffer that doesn't need mapping for writeback? */
2171 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2172 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2173 /* So far no extent to map => we write the buffer right away */
2174 if (map->m_len == 0)
2179 /* First block in the extent? */
2180 if (map->m_len == 0) {
2183 map->m_flags = bh->b_state & BH_FLAGS;
2187 /* Don't go larger than mballoc is willing to allocate */
2188 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2191 /* Can we merge the block to our big extent? */
2192 if (lblk == map->m_lblk + map->m_len &&
2193 (bh->b_state & BH_FLAGS) == map->m_flags) {
2201 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2203 * @mpd - extent of blocks for mapping
2204 * @head - the first buffer in the page
2205 * @bh - buffer we should start processing from
2206 * @lblk - logical number of the block in the file corresponding to @bh
2208 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2209 * the page for IO if all buffers in this page were mapped and there's no
2210 * accumulated extent of buffers to map or add buffers in the page to the
2211 * extent of buffers to map. The function returns 1 if the caller can continue
2212 * by processing the next page, 0 if it should stop adding buffers to the
2213 * extent to map because we cannot extend it anymore. It can also return value
2214 * < 0 in case of error during IO submission.
2216 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2217 struct buffer_head *head,
2218 struct buffer_head *bh,
2221 struct inode *inode = mpd->inode;
2223 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2224 >> inode->i_blkbits;
2227 BUG_ON(buffer_locked(bh));
2229 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2230 /* Found extent to map? */
2233 /* Everything mapped so far and we hit EOF */
2236 } while (lblk++, (bh = bh->b_this_page) != head);
2237 /* So far everything mapped? Submit the page for IO. */
2238 if (mpd->map.m_len == 0) {
2239 err = mpage_submit_page(mpd, head->b_page);
2243 return lblk < blocks;
2247 * mpage_map_buffers - update buffers corresponding to changed extent and
2248 * submit fully mapped pages for IO
2250 * @mpd - description of extent to map, on return next extent to map
2252 * Scan buffers corresponding to changed extent (we expect corresponding pages
2253 * to be already locked) and update buffer state according to new extent state.
2254 * We map delalloc buffers to their physical location, clear unwritten bits,
2255 * and mark buffers as uninit when we perform writes to unwritten extents
2256 * and do extent conversion after IO is finished. If the last page is not fully
2257 * mapped, we update @map to the next extent in the last page that needs
2258 * mapping. Otherwise we submit the page for IO.
2260 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2262 struct pagevec pvec;
2264 struct inode *inode = mpd->inode;
2265 struct buffer_head *head, *bh;
2266 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2272 start = mpd->map.m_lblk >> bpp_bits;
2273 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2274 lblk = start << bpp_bits;
2275 pblock = mpd->map.m_pblk;
2277 pagevec_init(&pvec, 0);
2278 while (start <= end) {
2279 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2283 for (i = 0; i < nr_pages; i++) {
2284 struct page *page = pvec.pages[i];
2286 if (page->index > end)
2288 /* Up to 'end' pages must be contiguous */
2289 BUG_ON(page->index != start);
2290 bh = head = page_buffers(page);
2292 if (lblk < mpd->map.m_lblk)
2294 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2296 * Buffer after end of mapped extent.
2297 * Find next buffer in the page to map.
2300 mpd->map.m_flags = 0;
2302 * FIXME: If dioread_nolock supports
2303 * blocksize < pagesize, we need to make
2304 * sure we add size mapped so far to
2305 * io_end->size as the following call
2306 * can submit the page for IO.
2308 err = mpage_process_page_bufs(mpd, head,
2310 pagevec_release(&pvec);
2315 if (buffer_delay(bh)) {
2316 clear_buffer_delay(bh);
2317 bh->b_blocknr = pblock++;
2319 clear_buffer_unwritten(bh);
2320 } while (lblk++, (bh = bh->b_this_page) != head);
2323 * FIXME: This is going to break if dioread_nolock
2324 * supports blocksize < pagesize as we will try to
2325 * convert potentially unmapped parts of inode.
2327 mpd->io_submit.io_end->size += PAGE_SIZE;
2328 /* Page fully mapped - let IO run! */
2329 err = mpage_submit_page(mpd, page);
2331 pagevec_release(&pvec);
2336 pagevec_release(&pvec);
2338 /* Extent fully mapped and matches with page boundary. We are done. */
2340 mpd->map.m_flags = 0;
2344 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2346 struct inode *inode = mpd->inode;
2347 struct ext4_map_blocks *map = &mpd->map;
2348 int get_blocks_flags;
2349 int err, dioread_nolock;
2351 trace_ext4_da_write_pages_extent(inode, map);
2353 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2354 * to convert an unwritten extent to be initialized (in the case
2355 * where we have written into one or more preallocated blocks). It is
2356 * possible that we're going to need more metadata blocks than
2357 * previously reserved. However we must not fail because we're in
2358 * writeback and there is nothing we can do about it so it might result
2359 * in data loss. So use reserved blocks to allocate metadata if
2362 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2363 * the blocks in question are delalloc blocks. This indicates
2364 * that the blocks and quotas has already been checked when
2365 * the data was copied into the page cache.
2367 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2368 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2369 EXT4_GET_BLOCKS_IO_SUBMIT;
2370 dioread_nolock = ext4_should_dioread_nolock(inode);
2372 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2373 if (map->m_flags & (1 << BH_Delay))
2374 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2376 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2379 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2380 if (!mpd->io_submit.io_end->handle &&
2381 ext4_handle_valid(handle)) {
2382 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2383 handle->h_rsv_handle = NULL;
2385 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2388 BUG_ON(map->m_len == 0);
2389 if (map->m_flags & EXT4_MAP_NEW) {
2390 struct block_device *bdev = inode->i_sb->s_bdev;
2393 for (i = 0; i < map->m_len; i++)
2394 unmap_underlying_metadata(bdev, map->m_pblk + i);
2400 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2401 * mpd->len and submit pages underlying it for IO
2403 * @handle - handle for journal operations
2404 * @mpd - extent to map
2405 * @give_up_on_write - we set this to true iff there is a fatal error and there
2406 * is no hope of writing the data. The caller should discard
2407 * dirty pages to avoid infinite loops.
2409 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2410 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2411 * them to initialized or split the described range from larger unwritten
2412 * extent. Note that we need not map all the described range since allocation
2413 * can return less blocks or the range is covered by more unwritten extents. We
2414 * cannot map more because we are limited by reserved transaction credits. On
2415 * the other hand we always make sure that the last touched page is fully
2416 * mapped so that it can be written out (and thus forward progress is
2417 * guaranteed). After mapping we submit all mapped pages for IO.
2419 static int mpage_map_and_submit_extent(handle_t *handle,
2420 struct mpage_da_data *mpd,
2421 bool *give_up_on_write)
2423 struct inode *inode = mpd->inode;
2424 struct ext4_map_blocks *map = &mpd->map;
2429 mpd->io_submit.io_end->offset =
2430 ((loff_t)map->m_lblk) << inode->i_blkbits;
2432 err = mpage_map_one_extent(handle, mpd);
2434 struct super_block *sb = inode->i_sb;
2436 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2437 goto invalidate_dirty_pages;
2439 * Let the uper layers retry transient errors.
2440 * In the case of ENOSPC, if ext4_count_free_blocks()
2441 * is non-zero, a commit should free up blocks.
2443 if ((err == -ENOMEM) ||
2444 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2446 goto update_disksize;
2449 ext4_msg(sb, KERN_CRIT,
2450 "Delayed block allocation failed for "
2451 "inode %lu at logical offset %llu with"
2452 " max blocks %u with error %d",
2454 (unsigned long long)map->m_lblk,
2455 (unsigned)map->m_len, -err);
2456 ext4_msg(sb, KERN_CRIT,
2457 "This should not happen!! Data will "
2460 ext4_print_free_blocks(inode);
2461 invalidate_dirty_pages:
2462 *give_up_on_write = true;
2467 * Update buffer state, submit mapped pages, and get us new
2470 err = mpage_map_and_submit_buffers(mpd);
2472 goto update_disksize;
2473 } while (map->m_len);
2477 * Update on-disk size after IO is submitted. Races with
2478 * truncate are avoided by checking i_size under i_data_sem.
2480 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2481 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2485 down_write(&EXT4_I(inode)->i_data_sem);
2486 i_size = i_size_read(inode);
2487 if (disksize > i_size)
2489 if (disksize > EXT4_I(inode)->i_disksize)
2490 EXT4_I(inode)->i_disksize = disksize;
2491 err2 = ext4_mark_inode_dirty(handle, inode);
2492 up_write(&EXT4_I(inode)->i_data_sem);
2494 ext4_error(inode->i_sb,
2495 "Failed to mark inode %lu dirty",
2504 * Calculate the total number of credits to reserve for one writepages
2505 * iteration. This is called from ext4_writepages(). We map an extent of
2506 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2507 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2508 * bpp - 1 blocks in bpp different extents.
2510 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2512 int bpp = ext4_journal_blocks_per_page(inode);
2514 return ext4_meta_trans_blocks(inode,
2515 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2519 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2520 * and underlying extent to map
2522 * @mpd - where to look for pages
2524 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2525 * IO immediately. When we find a page which isn't mapped we start accumulating
2526 * extent of buffers underlying these pages that needs mapping (formed by
2527 * either delayed or unwritten buffers). We also lock the pages containing
2528 * these buffers. The extent found is returned in @mpd structure (starting at
2529 * mpd->lblk with length mpd->len blocks).
2531 * Note that this function can attach bios to one io_end structure which are
2532 * neither logically nor physically contiguous. Although it may seem as an
2533 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2534 * case as we need to track IO to all buffers underlying a page in one io_end.
2536 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2538 struct address_space *mapping = mpd->inode->i_mapping;
2539 struct pagevec pvec;
2540 unsigned int nr_pages;
2541 long left = mpd->wbc->nr_to_write;
2542 pgoff_t index = mpd->first_page;
2543 pgoff_t end = mpd->last_page;
2546 int blkbits = mpd->inode->i_blkbits;
2548 struct buffer_head *head;
2550 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2551 tag = PAGECACHE_TAG_TOWRITE;
2553 tag = PAGECACHE_TAG_DIRTY;
2555 pagevec_init(&pvec, 0);
2557 mpd->next_page = index;
2558 while (index <= end) {
2559 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2560 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2564 for (i = 0; i < nr_pages; i++) {
2565 struct page *page = pvec.pages[i];
2568 * At this point, the page may be truncated or
2569 * invalidated (changing page->mapping to NULL), or
2570 * even swizzled back from swapper_space to tmpfs file
2571 * mapping. However, page->index will not change
2572 * because we have a reference on the page.
2574 if (page->index > end)
2578 * Accumulated enough dirty pages? This doesn't apply
2579 * to WB_SYNC_ALL mode. For integrity sync we have to
2580 * keep going because someone may be concurrently
2581 * dirtying pages, and we might have synced a lot of
2582 * newly appeared dirty pages, but have not synced all
2583 * of the old dirty pages.
2585 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2588 /* If we can't merge this page, we are done. */
2589 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2594 * If the page is no longer dirty, or its mapping no
2595 * longer corresponds to inode we are writing (which
2596 * means it has been truncated or invalidated), or the
2597 * page is already under writeback and we are not doing
2598 * a data integrity writeback, skip the page
2600 if (!PageDirty(page) ||
2601 (PageWriteback(page) &&
2602 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2603 unlikely(page->mapping != mapping)) {
2608 wait_on_page_writeback(page);
2609 BUG_ON(PageWriteback(page));
2611 if (mpd->map.m_len == 0)
2612 mpd->first_page = page->index;
2613 mpd->next_page = page->index + 1;
2614 /* Add all dirty buffers to mpd */
2615 lblk = ((ext4_lblk_t)page->index) <<
2616 (PAGE_SHIFT - blkbits);
2617 head = page_buffers(page);
2618 err = mpage_process_page_bufs(mpd, head, head, lblk);
2624 pagevec_release(&pvec);
2629 pagevec_release(&pvec);
2633 static int __writepage(struct page *page, struct writeback_control *wbc,
2636 struct address_space *mapping = data;
2637 int ret = ext4_writepage(page, wbc);
2638 mapping_set_error(mapping, ret);
2642 static int ext4_writepages(struct address_space *mapping,
2643 struct writeback_control *wbc)
2645 pgoff_t writeback_index = 0;
2646 long nr_to_write = wbc->nr_to_write;
2647 int range_whole = 0;
2649 handle_t *handle = NULL;
2650 struct mpage_da_data mpd;
2651 struct inode *inode = mapping->host;
2652 int needed_blocks, rsv_blocks = 0, ret = 0;
2653 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2655 struct blk_plug plug;
2656 bool give_up_on_write = false;
2658 percpu_down_read(&sbi->s_writepages_rwsem);
2659 trace_ext4_writepages(inode, wbc);
2661 if (dax_mapping(mapping)) {
2662 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2664 goto out_writepages;
2668 * No pages to write? This is mainly a kludge to avoid starting
2669 * a transaction for special inodes like journal inode on last iput()
2670 * because that could violate lock ordering on umount
2672 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2673 goto out_writepages;
2675 if (ext4_should_journal_data(inode)) {
2676 struct blk_plug plug;
2678 blk_start_plug(&plug);
2679 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2680 blk_finish_plug(&plug);
2681 goto out_writepages;
2685 * If the filesystem has aborted, it is read-only, so return
2686 * right away instead of dumping stack traces later on that
2687 * will obscure the real source of the problem. We test
2688 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2689 * the latter could be true if the filesystem is mounted
2690 * read-only, and in that case, ext4_writepages should
2691 * *never* be called, so if that ever happens, we would want
2694 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2696 goto out_writepages;
2699 if (ext4_should_dioread_nolock(inode)) {
2701 * We may need to convert up to one extent per block in
2702 * the page and we may dirty the inode.
2704 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2705 PAGE_SIZE >> inode->i_blkbits);
2709 * If we have inline data and arrive here, it means that
2710 * we will soon create the block for the 1st page, so
2711 * we'd better clear the inline data here.
2713 if (ext4_has_inline_data(inode)) {
2714 /* Just inode will be modified... */
2715 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2716 if (IS_ERR(handle)) {
2717 ret = PTR_ERR(handle);
2718 goto out_writepages;
2720 BUG_ON(ext4_test_inode_state(inode,
2721 EXT4_STATE_MAY_INLINE_DATA));
2722 ext4_destroy_inline_data(handle, inode);
2723 ext4_journal_stop(handle);
2726 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2729 if (wbc->range_cyclic) {
2730 writeback_index = mapping->writeback_index;
2731 if (writeback_index)
2733 mpd.first_page = writeback_index;
2736 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2737 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2742 ext4_io_submit_init(&mpd.io_submit, wbc);
2744 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2745 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2747 blk_start_plug(&plug);
2748 while (!done && mpd.first_page <= mpd.last_page) {
2749 /* For each extent of pages we use new io_end */
2750 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2751 if (!mpd.io_submit.io_end) {
2757 * We have two constraints: We find one extent to map and we
2758 * must always write out whole page (makes a difference when
2759 * blocksize < pagesize) so that we don't block on IO when we
2760 * try to write out the rest of the page. Journalled mode is
2761 * not supported by delalloc.
2763 BUG_ON(ext4_should_journal_data(inode));
2764 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2766 /* start a new transaction */
2767 handle = ext4_journal_start_with_reserve(inode,
2768 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2769 if (IS_ERR(handle)) {
2770 ret = PTR_ERR(handle);
2771 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2772 "%ld pages, ino %lu; err %d", __func__,
2773 wbc->nr_to_write, inode->i_ino, ret);
2774 /* Release allocated io_end */
2775 ext4_put_io_end(mpd.io_submit.io_end);
2779 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2780 ret = mpage_prepare_extent_to_map(&mpd);
2783 ret = mpage_map_and_submit_extent(handle, &mpd,
2787 * We scanned the whole range (or exhausted
2788 * nr_to_write), submitted what was mapped and
2789 * didn't find anything needing mapping. We are
2796 * Caution: If the handle is synchronous,
2797 * ext4_journal_stop() can wait for transaction commit
2798 * to finish which may depend on writeback of pages to
2799 * complete or on page lock to be released. In that
2800 * case, we have to wait until after after we have
2801 * submitted all the IO, released page locks we hold,
2802 * and dropped io_end reference (for extent conversion
2803 * to be able to complete) before stopping the handle.
2805 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2806 ext4_journal_stop(handle);
2809 /* Submit prepared bio */
2810 ext4_io_submit(&mpd.io_submit);
2811 /* Unlock pages we didn't use */
2812 mpage_release_unused_pages(&mpd, give_up_on_write);
2814 * Drop our io_end reference we got from init. We have
2815 * to be careful and use deferred io_end finishing if
2816 * we are still holding the transaction as we can
2817 * release the last reference to io_end which may end
2818 * up doing unwritten extent conversion.
2821 ext4_put_io_end_defer(mpd.io_submit.io_end);
2822 ext4_journal_stop(handle);
2824 ext4_put_io_end(mpd.io_submit.io_end);
2826 if (ret == -ENOSPC && sbi->s_journal) {
2828 * Commit the transaction which would
2829 * free blocks released in the transaction
2832 jbd2_journal_force_commit_nested(sbi->s_journal);
2836 /* Fatal error - ENOMEM, EIO... */
2840 blk_finish_plug(&plug);
2841 if (!ret && !cycled && wbc->nr_to_write > 0) {
2843 mpd.last_page = writeback_index - 1;
2849 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2851 * Set the writeback_index so that range_cyclic
2852 * mode will write it back later
2854 mapping->writeback_index = mpd.first_page;
2857 trace_ext4_writepages_result(inode, wbc, ret,
2858 nr_to_write - wbc->nr_to_write);
2859 percpu_up_read(&sbi->s_writepages_rwsem);
2863 static int ext4_nonda_switch(struct super_block *sb)
2865 s64 free_clusters, dirty_clusters;
2866 struct ext4_sb_info *sbi = EXT4_SB(sb);
2869 * switch to non delalloc mode if we are running low
2870 * on free block. The free block accounting via percpu
2871 * counters can get slightly wrong with percpu_counter_batch getting
2872 * accumulated on each CPU without updating global counters
2873 * Delalloc need an accurate free block accounting. So switch
2874 * to non delalloc when we are near to error range.
2877 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2879 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2881 * Start pushing delalloc when 1/2 of free blocks are dirty.
2883 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2884 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2886 if (2 * free_clusters < 3 * dirty_clusters ||
2887 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2889 * free block count is less than 150% of dirty blocks
2890 * or free blocks is less than watermark
2897 /* We always reserve for an inode update; the superblock could be there too */
2898 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2900 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2903 if (pos + len <= 0x7fffffffULL)
2906 /* We might need to update the superblock to set LARGE_FILE */
2910 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2911 loff_t pos, unsigned len, unsigned flags,
2912 struct page **pagep, void **fsdata)
2914 int ret, retries = 0;
2917 struct inode *inode = mapping->host;
2920 index = pos >> PAGE_SHIFT;
2922 if (ext4_nonda_switch(inode->i_sb)) {
2923 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2924 return ext4_write_begin(file, mapping, pos,
2925 len, flags, pagep, fsdata);
2927 *fsdata = (void *)0;
2928 trace_ext4_da_write_begin(inode, pos, len, flags);
2930 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2931 ret = ext4_da_write_inline_data_begin(mapping, inode,
2941 * grab_cache_page_write_begin() can take a long time if the
2942 * system is thrashing due to memory pressure, or if the page
2943 * is being written back. So grab it first before we start
2944 * the transaction handle. This also allows us to allocate
2945 * the page (if needed) without using GFP_NOFS.
2948 page = grab_cache_page_write_begin(mapping, index, flags);
2954 * With delayed allocation, we don't log the i_disksize update
2955 * if there is delayed block allocation. But we still need
2956 * to journalling the i_disksize update if writes to the end
2957 * of file which has an already mapped buffer.
2960 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2961 ext4_da_write_credits(inode, pos, len));
2962 if (IS_ERR(handle)) {
2964 return PTR_ERR(handle);
2968 if (page->mapping != mapping) {
2969 /* The page got truncated from under us */
2972 ext4_journal_stop(handle);
2975 /* In case writeback began while the page was unlocked */
2976 wait_for_stable_page(page);
2978 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2979 ret = ext4_block_write_begin(page, pos, len,
2980 ext4_da_get_block_prep);
2982 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2986 ext4_journal_stop(handle);
2988 * block_write_begin may have instantiated a few blocks
2989 * outside i_size. Trim these off again. Don't need
2990 * i_size_read because we hold i_mutex.
2992 if (pos + len > inode->i_size)
2993 ext4_truncate_failed_write(inode);
2995 if (ret == -ENOSPC &&
2996 ext4_should_retry_alloc(inode->i_sb, &retries))
3008 * Check if we should update i_disksize
3009 * when write to the end of file but not require block allocation
3011 static int ext4_da_should_update_i_disksize(struct page *page,
3012 unsigned long offset)
3014 struct buffer_head *bh;
3015 struct inode *inode = page->mapping->host;
3019 bh = page_buffers(page);
3020 idx = offset >> inode->i_blkbits;
3022 for (i = 0; i < idx; i++)
3023 bh = bh->b_this_page;
3025 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3030 static int ext4_da_write_end(struct file *file,
3031 struct address_space *mapping,
3032 loff_t pos, unsigned len, unsigned copied,
3033 struct page *page, void *fsdata)
3035 struct inode *inode = mapping->host;
3037 handle_t *handle = ext4_journal_current_handle();
3039 unsigned long start, end;
3040 int write_mode = (int)(unsigned long)fsdata;
3042 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3043 return ext4_write_end(file, mapping, pos,
3044 len, copied, page, fsdata);
3046 trace_ext4_da_write_end(inode, pos, len, copied);
3047 start = pos & (PAGE_SIZE - 1);
3048 end = start + copied - 1;
3051 * generic_write_end() will run mark_inode_dirty() if i_size
3052 * changes. So let's piggyback the i_disksize mark_inode_dirty
3055 new_i_size = pos + copied;
3056 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3057 if (ext4_has_inline_data(inode) ||
3058 ext4_da_should_update_i_disksize(page, end)) {
3059 ext4_update_i_disksize(inode, new_i_size);
3060 /* We need to mark inode dirty even if
3061 * new_i_size is less that inode->i_size
3062 * bu greater than i_disksize.(hint delalloc)
3064 ext4_mark_inode_dirty(handle, inode);
3068 if (write_mode != CONVERT_INLINE_DATA &&
3069 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3070 ext4_has_inline_data(inode))
3071 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3074 ret2 = generic_write_end(file, mapping, pos, len, copied,
3080 ret2 = ext4_journal_stop(handle);
3084 return ret ? ret : copied;
3087 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3088 unsigned int length)
3091 * Drop reserved blocks
3093 BUG_ON(!PageLocked(page));
3094 if (!page_has_buffers(page))
3097 ext4_da_page_release_reservation(page, offset, length);
3100 ext4_invalidatepage(page, offset, length);
3106 * Force all delayed allocation blocks to be allocated for a given inode.
3108 int ext4_alloc_da_blocks(struct inode *inode)
3110 trace_ext4_alloc_da_blocks(inode);
3112 if (!EXT4_I(inode)->i_reserved_data_blocks)
3116 * We do something simple for now. The filemap_flush() will
3117 * also start triggering a write of the data blocks, which is
3118 * not strictly speaking necessary (and for users of
3119 * laptop_mode, not even desirable). However, to do otherwise
3120 * would require replicating code paths in:
3122 * ext4_writepages() ->
3123 * write_cache_pages() ---> (via passed in callback function)
3124 * __mpage_da_writepage() -->
3125 * mpage_add_bh_to_extent()
3126 * mpage_da_map_blocks()
3128 * The problem is that write_cache_pages(), located in
3129 * mm/page-writeback.c, marks pages clean in preparation for
3130 * doing I/O, which is not desirable if we're not planning on
3133 * We could call write_cache_pages(), and then redirty all of
3134 * the pages by calling redirty_page_for_writepage() but that
3135 * would be ugly in the extreme. So instead we would need to
3136 * replicate parts of the code in the above functions,
3137 * simplifying them because we wouldn't actually intend to
3138 * write out the pages, but rather only collect contiguous
3139 * logical block extents, call the multi-block allocator, and
3140 * then update the buffer heads with the block allocations.
3142 * For now, though, we'll cheat by calling filemap_flush(),
3143 * which will map the blocks, and start the I/O, but not
3144 * actually wait for the I/O to complete.
3146 return filemap_flush(inode->i_mapping);
3150 * bmap() is special. It gets used by applications such as lilo and by
3151 * the swapper to find the on-disk block of a specific piece of data.
3153 * Naturally, this is dangerous if the block concerned is still in the
3154 * journal. If somebody makes a swapfile on an ext4 data-journaling
3155 * filesystem and enables swap, then they may get a nasty shock when the
3156 * data getting swapped to that swapfile suddenly gets overwritten by
3157 * the original zero's written out previously to the journal and
3158 * awaiting writeback in the kernel's buffer cache.
3160 * So, if we see any bmap calls here on a modified, data-journaled file,
3161 * take extra steps to flush any blocks which might be in the cache.
3163 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3165 struct inode *inode = mapping->host;
3170 * We can get here for an inline file via the FIBMAP ioctl
3172 if (ext4_has_inline_data(inode))
3175 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3176 test_opt(inode->i_sb, DELALLOC)) {
3178 * With delalloc we want to sync the file
3179 * so that we can make sure we allocate
3182 filemap_write_and_wait(mapping);
3185 if (EXT4_JOURNAL(inode) &&
3186 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3188 * This is a REALLY heavyweight approach, but the use of
3189 * bmap on dirty files is expected to be extremely rare:
3190 * only if we run lilo or swapon on a freshly made file
3191 * do we expect this to happen.
3193 * (bmap requires CAP_SYS_RAWIO so this does not
3194 * represent an unprivileged user DOS attack --- we'd be
3195 * in trouble if mortal users could trigger this path at
3198 * NB. EXT4_STATE_JDATA is not set on files other than
3199 * regular files. If somebody wants to bmap a directory
3200 * or symlink and gets confused because the buffer
3201 * hasn't yet been flushed to disk, they deserve
3202 * everything they get.
3205 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3206 journal = EXT4_JOURNAL(inode);
3207 jbd2_journal_lock_updates(journal);
3208 err = jbd2_journal_flush(journal);
3209 jbd2_journal_unlock_updates(journal);
3215 return generic_block_bmap(mapping, block, ext4_get_block);
3218 static int ext4_readpage(struct file *file, struct page *page)
3221 struct inode *inode = page->mapping->host;
3223 trace_ext4_readpage(page);
3225 if (ext4_has_inline_data(inode))
3226 ret = ext4_readpage_inline(inode, page);
3229 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3235 ext4_readpages(struct file *file, struct address_space *mapping,
3236 struct list_head *pages, unsigned nr_pages)
3238 struct inode *inode = mapping->host;
3240 /* If the file has inline data, no need to do readpages. */
3241 if (ext4_has_inline_data(inode))
3244 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3247 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3248 unsigned int length)
3250 trace_ext4_invalidatepage(page, offset, length);
3252 /* No journalling happens on data buffers when this function is used */
3253 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3255 block_invalidatepage(page, offset, length);
3258 static int __ext4_journalled_invalidatepage(struct page *page,
3259 unsigned int offset,
3260 unsigned int length)
3262 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3264 trace_ext4_journalled_invalidatepage(page, offset, length);
3267 * If it's a full truncate we just forget about the pending dirtying
3269 if (offset == 0 && length == PAGE_SIZE)
3270 ClearPageChecked(page);
3272 return jbd2_journal_invalidatepage(journal, page, offset, length);
3275 /* Wrapper for aops... */
3276 static void ext4_journalled_invalidatepage(struct page *page,
3277 unsigned int offset,
3278 unsigned int length)
3280 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3283 static int ext4_releasepage(struct page *page, gfp_t wait)
3285 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3287 trace_ext4_releasepage(page);
3289 /* Page has dirty journalled data -> cannot release */
3290 if (PageChecked(page))
3293 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3295 return try_to_free_buffers(page);
3298 #ifdef CONFIG_FS_DAX
3300 * Get block function for DAX IO and mmap faults. It takes care of converting
3301 * unwritten extents to written ones and initializes new / converted blocks
3304 int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3305 struct buffer_head *bh_result, int create)
3309 ext4_debug("inode %lu, create flag %d\n", inode->i_ino, create);
3311 return _ext4_get_block(inode, iblock, bh_result, 0);
3313 ret = ext4_get_block_trans(inode, iblock, bh_result,
3314 EXT4_GET_BLOCKS_PRE_IO |
3315 EXT4_GET_BLOCKS_CREATE_ZERO);
3319 if (buffer_unwritten(bh_result)) {
3321 * We are protected by i_mmap_sem or i_mutex so we know block
3322 * cannot go away from under us even though we dropped
3323 * i_data_sem. Convert extent to written and write zeros there.
3325 ret = ext4_get_block_trans(inode, iblock, bh_result,
3326 EXT4_GET_BLOCKS_CONVERT |
3327 EXT4_GET_BLOCKS_CREATE_ZERO);
3332 * At least for now we have to clear BH_New so that DAX code
3333 * doesn't attempt to zero blocks again in a racy way.
3335 clear_buffer_new(bh_result);
3339 /* Just define empty function, it will never get called. */
3340 int ext4_dax_get_block(struct inode *inode, sector_t iblock,
3341 struct buffer_head *bh_result, int create)
3348 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3349 ssize_t size, void *private)
3351 ext4_io_end_t *io_end = private;
3353 /* if not async direct IO just return */
3357 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3358 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3359 io_end, io_end->inode->i_ino, iocb, offset, size);
3362 * Error during AIO DIO. We cannot convert unwritten extents as the
3363 * data was not written. Just clear the unwritten flag and drop io_end.
3366 ext4_clear_io_unwritten_flag(io_end);
3369 io_end->offset = offset;
3370 io_end->size = size;
3371 ext4_put_io_end(io_end);
3377 * Handling of direct IO writes.
3379 * For ext4 extent files, ext4 will do direct-io write even to holes,
3380 * preallocated extents, and those write extend the file, no need to
3381 * fall back to buffered IO.
3383 * For holes, we fallocate those blocks, mark them as unwritten
3384 * If those blocks were preallocated, we mark sure they are split, but
3385 * still keep the range to write as unwritten.
3387 * The unwritten extents will be converted to written when DIO is completed.
3388 * For async direct IO, since the IO may still pending when return, we
3389 * set up an end_io call back function, which will do the conversion
3390 * when async direct IO completed.
3392 * If the O_DIRECT write will extend the file then add this inode to the
3393 * orphan list. So recovery will truncate it back to the original size
3394 * if the machine crashes during the write.
3397 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3399 struct file *file = iocb->ki_filp;
3400 struct inode *inode = file->f_mapping->host;
3402 loff_t offset = iocb->ki_pos;
3403 size_t count = iov_iter_count(iter);
3405 get_block_t *get_block_func = NULL;
3407 loff_t final_size = offset + count;
3411 if (final_size > inode->i_size) {
3412 /* Credits for sb + inode write */
3413 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3414 if (IS_ERR(handle)) {
3415 ret = PTR_ERR(handle);
3418 ret = ext4_orphan_add(handle, inode);
3420 ext4_journal_stop(handle);
3424 ext4_update_i_disksize(inode, inode->i_size);
3425 ext4_journal_stop(handle);
3428 BUG_ON(iocb->private == NULL);
3431 * Make all waiters for direct IO properly wait also for extent
3432 * conversion. This also disallows race between truncate() and
3433 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3435 inode_dio_begin(inode);
3437 /* If we do a overwrite dio, i_mutex locking can be released */
3438 overwrite = *((int *)iocb->private);
3441 inode_unlock(inode);
3444 * For extent mapped files we could direct write to holes and fallocate.
3446 * Allocated blocks to fill the hole are marked as unwritten to prevent
3447 * parallel buffered read to expose the stale data before DIO complete
3450 * As to previously fallocated extents, ext4 get_block will just simply
3451 * mark the buffer mapped but still keep the extents unwritten.
3453 * For non AIO case, we will convert those unwritten extents to written
3454 * after return back from blockdev_direct_IO. That way we save us from
3455 * allocating io_end structure and also the overhead of offloading
3456 * the extent convertion to a workqueue.
3458 * For async DIO, the conversion needs to be deferred when the
3459 * IO is completed. The ext4 end_io callback function will be
3460 * called to take care of the conversion work. Here for async
3461 * case, we allocate an io_end structure to hook to the iocb.
3463 iocb->private = NULL;
3465 get_block_func = ext4_dio_get_block_overwrite;
3466 else if (IS_DAX(inode)) {
3468 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3469 * writes need zeroing either because they can race with page
3470 * faults or because they use partial blocks.
3472 if (round_down(offset, i_blocksize(inode)) >= inode->i_size &&
3473 ext4_aligned_io(inode, offset, count))
3474 get_block_func = ext4_dio_get_block;
3476 get_block_func = ext4_dax_get_block;
3477 dio_flags = DIO_LOCKING;
3478 } else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3479 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3480 get_block_func = ext4_dio_get_block;
3481 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3482 } else if (is_sync_kiocb(iocb)) {
3483 get_block_func = ext4_dio_get_block_unwritten_sync;
3484 dio_flags = DIO_LOCKING;
3486 get_block_func = ext4_dio_get_block_unwritten_async;
3487 dio_flags = DIO_LOCKING;
3489 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3490 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3492 if (IS_DAX(inode)) {
3493 ret = dax_do_io(iocb, inode, iter, get_block_func,
3494 ext4_end_io_dio, dio_flags);
3496 ret = __blockdev_direct_IO(iocb, inode,
3497 inode->i_sb->s_bdev, iter,
3499 ext4_end_io_dio, NULL, dio_flags);
3501 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3502 EXT4_STATE_DIO_UNWRITTEN)) {
3505 * for non AIO case, since the IO is already
3506 * completed, we could do the conversion right here
3508 err = ext4_convert_unwritten_extents(NULL, inode,
3512 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3515 inode_dio_end(inode);
3516 /* take i_mutex locking again if we do a ovewrite dio */
3520 if (ret < 0 && final_size > inode->i_size)
3521 ext4_truncate_failed_write(inode);
3523 /* Handle extending of i_size after direct IO write */
3527 /* Credits for sb + inode write */
3528 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3529 if (IS_ERR(handle)) {
3531 * We wrote the data but cannot extend
3532 * i_size. Bail out. In async io case, we do
3533 * not return error here because we have
3534 * already submmitted the corresponding
3535 * bio. Returning error here makes the caller
3536 * think that this IO is done and failed
3537 * resulting in race with bio's completion
3541 ret = PTR_ERR(handle);
3543 ext4_orphan_del(NULL, inode);
3548 ext4_orphan_del(handle, inode);
3550 loff_t end = offset + ret;
3551 if (end > inode->i_size) {
3552 ext4_update_i_disksize(inode, end);
3553 i_size_write(inode, end);
3555 * We're going to return a positive `ret'
3556 * here due to non-zero-length I/O, so there's
3557 * no way of reporting error returns from
3558 * ext4_mark_inode_dirty() to userspace. So
3561 ext4_mark_inode_dirty(handle, inode);
3564 err = ext4_journal_stop(handle);
3572 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3574 struct address_space *mapping = iocb->ki_filp->f_mapping;
3575 struct inode *inode = mapping->host;
3577 loff_t offset = iocb->ki_pos;
3578 loff_t size = i_size_read(inode);
3584 * Shared inode_lock is enough for us - it protects against concurrent
3585 * writes & truncates and since we take care of writing back page cache,
3586 * we are protected against page writeback as well.
3588 inode_lock_shared(inode);
3589 if (IS_DAX(inode)) {
3590 ret = dax_do_io(iocb, inode, iter, ext4_dio_get_block, NULL, 0);
3592 size_t count = iov_iter_count(iter);
3594 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3595 iocb->ki_pos + count);
3598 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3599 iter, ext4_dio_get_block,
3603 inode_unlock_shared(inode);
3607 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3609 struct file *file = iocb->ki_filp;
3610 struct inode *inode = file->f_mapping->host;
3611 size_t count = iov_iter_count(iter);
3612 loff_t offset = iocb->ki_pos;
3615 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3616 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3621 * If we are doing data journalling we don't support O_DIRECT
3623 if (ext4_should_journal_data(inode))
3626 /* Let buffer I/O handle the inline data case. */
3627 if (ext4_has_inline_data(inode))
3630 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3631 if (iov_iter_rw(iter) == READ)
3632 ret = ext4_direct_IO_read(iocb, iter);
3634 ret = ext4_direct_IO_write(iocb, iter);
3635 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3640 * Pages can be marked dirty completely asynchronously from ext4's journalling
3641 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3642 * much here because ->set_page_dirty is called under VFS locks. The page is
3643 * not necessarily locked.
3645 * We cannot just dirty the page and leave attached buffers clean, because the
3646 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3647 * or jbddirty because all the journalling code will explode.
3649 * So what we do is to mark the page "pending dirty" and next time writepage
3650 * is called, propagate that into the buffers appropriately.
3652 static int ext4_journalled_set_page_dirty(struct page *page)
3654 SetPageChecked(page);
3655 return __set_page_dirty_nobuffers(page);
3658 static const struct address_space_operations ext4_aops = {
3659 .readpage = ext4_readpage,
3660 .readpages = ext4_readpages,
3661 .writepage = ext4_writepage,
3662 .writepages = ext4_writepages,
3663 .write_begin = ext4_write_begin,
3664 .write_end = ext4_write_end,
3666 .invalidatepage = ext4_invalidatepage,
3667 .releasepage = ext4_releasepage,
3668 .direct_IO = ext4_direct_IO,
3669 .migratepage = buffer_migrate_page,
3670 .is_partially_uptodate = block_is_partially_uptodate,
3671 .error_remove_page = generic_error_remove_page,
3674 static const struct address_space_operations ext4_journalled_aops = {
3675 .readpage = ext4_readpage,
3676 .readpages = ext4_readpages,
3677 .writepage = ext4_writepage,
3678 .writepages = ext4_writepages,
3679 .write_begin = ext4_write_begin,
3680 .write_end = ext4_journalled_write_end,
3681 .set_page_dirty = ext4_journalled_set_page_dirty,
3683 .invalidatepage = ext4_journalled_invalidatepage,
3684 .releasepage = ext4_releasepage,
3685 .direct_IO = ext4_direct_IO,
3686 .is_partially_uptodate = block_is_partially_uptodate,
3687 .error_remove_page = generic_error_remove_page,
3690 static const struct address_space_operations ext4_da_aops = {
3691 .readpage = ext4_readpage,
3692 .readpages = ext4_readpages,
3693 .writepage = ext4_writepage,
3694 .writepages = ext4_writepages,
3695 .write_begin = ext4_da_write_begin,
3696 .write_end = ext4_da_write_end,
3698 .invalidatepage = ext4_da_invalidatepage,
3699 .releasepage = ext4_releasepage,
3700 .direct_IO = ext4_direct_IO,
3701 .migratepage = buffer_migrate_page,
3702 .is_partially_uptodate = block_is_partially_uptodate,
3703 .error_remove_page = generic_error_remove_page,
3706 void ext4_set_aops(struct inode *inode)
3708 switch (ext4_inode_journal_mode(inode)) {
3709 case EXT4_INODE_ORDERED_DATA_MODE:
3710 case EXT4_INODE_WRITEBACK_DATA_MODE:
3712 case EXT4_INODE_JOURNAL_DATA_MODE:
3713 inode->i_mapping->a_ops = &ext4_journalled_aops;
3718 if (test_opt(inode->i_sb, DELALLOC))
3719 inode->i_mapping->a_ops = &ext4_da_aops;
3721 inode->i_mapping->a_ops = &ext4_aops;
3724 static int __ext4_block_zero_page_range(handle_t *handle,
3725 struct address_space *mapping, loff_t from, loff_t length)
3727 ext4_fsblk_t index = from >> PAGE_SHIFT;
3728 unsigned offset = from & (PAGE_SIZE-1);
3729 unsigned blocksize, pos;
3731 struct inode *inode = mapping->host;
3732 struct buffer_head *bh;
3736 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3737 mapping_gfp_constraint(mapping, ~__GFP_FS));
3741 blocksize = inode->i_sb->s_blocksize;
3743 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3745 if (!page_has_buffers(page))
3746 create_empty_buffers(page, blocksize, 0);
3748 /* Find the buffer that contains "offset" */
3749 bh = page_buffers(page);
3751 while (offset >= pos) {
3752 bh = bh->b_this_page;
3756 if (buffer_freed(bh)) {
3757 BUFFER_TRACE(bh, "freed: skip");
3760 if (!buffer_mapped(bh)) {
3761 BUFFER_TRACE(bh, "unmapped");
3762 ext4_get_block(inode, iblock, bh, 0);
3763 /* unmapped? It's a hole - nothing to do */
3764 if (!buffer_mapped(bh)) {
3765 BUFFER_TRACE(bh, "still unmapped");
3770 /* Ok, it's mapped. Make sure it's up-to-date */
3771 if (PageUptodate(page))
3772 set_buffer_uptodate(bh);
3774 if (!buffer_uptodate(bh)) {
3776 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3778 /* Uhhuh. Read error. Complain and punt. */
3779 if (!buffer_uptodate(bh))
3781 if (S_ISREG(inode->i_mode) &&
3782 ext4_encrypted_inode(inode)) {
3783 /* We expect the key to be set. */
3784 BUG_ON(!fscrypt_has_encryption_key(inode));
3785 BUG_ON(blocksize != PAGE_SIZE);
3786 WARN_ON_ONCE(fscrypt_decrypt_page(page));
3789 if (ext4_should_journal_data(inode)) {
3790 BUFFER_TRACE(bh, "get write access");
3791 err = ext4_journal_get_write_access(handle, bh);
3795 zero_user(page, offset, length);
3796 BUFFER_TRACE(bh, "zeroed end of block");
3798 if (ext4_should_journal_data(inode)) {
3799 err = ext4_handle_dirty_metadata(handle, inode, bh);
3802 mark_buffer_dirty(bh);
3803 if (ext4_should_order_data(inode))
3804 err = ext4_jbd2_inode_add_write(handle, inode);
3814 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3815 * starting from file offset 'from'. The range to be zero'd must
3816 * be contained with in one block. If the specified range exceeds
3817 * the end of the block it will be shortened to end of the block
3818 * that cooresponds to 'from'
3820 static int ext4_block_zero_page_range(handle_t *handle,
3821 struct address_space *mapping, loff_t from, loff_t length)
3823 struct inode *inode = mapping->host;
3824 unsigned offset = from & (PAGE_SIZE-1);
3825 unsigned blocksize = inode->i_sb->s_blocksize;
3826 unsigned max = blocksize - (offset & (blocksize - 1));
3829 * correct length if it does not fall between
3830 * 'from' and the end of the block
3832 if (length > max || length < 0)
3836 return dax_zero_page_range(inode, from, length, ext4_get_block);
3837 return __ext4_block_zero_page_range(handle, mapping, from, length);
3841 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3842 * up to the end of the block which corresponds to `from'.
3843 * This required during truncate. We need to physically zero the tail end
3844 * of that block so it doesn't yield old data if the file is later grown.
3846 static int ext4_block_truncate_page(handle_t *handle,
3847 struct address_space *mapping, loff_t from)
3849 unsigned offset = from & (PAGE_SIZE-1);
3852 struct inode *inode = mapping->host;
3854 /* If we are processing an encrypted inode during orphan list handling */
3855 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
3858 blocksize = inode->i_sb->s_blocksize;
3859 length = blocksize - (offset & (blocksize - 1));
3861 return ext4_block_zero_page_range(handle, mapping, from, length);
3864 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3865 loff_t lstart, loff_t length)
3867 struct super_block *sb = inode->i_sb;
3868 struct address_space *mapping = inode->i_mapping;
3869 unsigned partial_start, partial_end;
3870 ext4_fsblk_t start, end;
3871 loff_t byte_end = (lstart + length - 1);
3874 partial_start = lstart & (sb->s_blocksize - 1);
3875 partial_end = byte_end & (sb->s_blocksize - 1);
3877 start = lstart >> sb->s_blocksize_bits;
3878 end = byte_end >> sb->s_blocksize_bits;
3880 /* Handle partial zero within the single block */
3882 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3883 err = ext4_block_zero_page_range(handle, mapping,
3887 /* Handle partial zero out on the start of the range */
3888 if (partial_start) {
3889 err = ext4_block_zero_page_range(handle, mapping,
3890 lstart, sb->s_blocksize);
3894 /* Handle partial zero out on the end of the range */
3895 if (partial_end != sb->s_blocksize - 1)
3896 err = ext4_block_zero_page_range(handle, mapping,
3897 byte_end - partial_end,
3902 int ext4_can_truncate(struct inode *inode)
3904 if (S_ISREG(inode->i_mode))
3906 if (S_ISDIR(inode->i_mode))
3908 if (S_ISLNK(inode->i_mode))
3909 return !ext4_inode_is_fast_symlink(inode);
3914 * We have to make sure i_disksize gets properly updated before we truncate
3915 * page cache due to hole punching or zero range. Otherwise i_disksize update
3916 * can get lost as it may have been postponed to submission of writeback but
3917 * that will never happen after we truncate page cache.
3919 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3923 loff_t size = i_size_read(inode);
3925 WARN_ON(!inode_is_locked(inode));
3926 if (offset > size || offset + len < size)
3929 if (EXT4_I(inode)->i_disksize >= size)
3932 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3934 return PTR_ERR(handle);
3935 ext4_update_i_disksize(inode, size);
3936 ext4_mark_inode_dirty(handle, inode);
3937 ext4_journal_stop(handle);
3943 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3944 * associated with the given offset and length
3946 * @inode: File inode
3947 * @offset: The offset where the hole will begin
3948 * @len: The length of the hole
3950 * Returns: 0 on success or negative on failure
3953 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3955 struct super_block *sb = inode->i_sb;
3956 ext4_lblk_t first_block, stop_block;
3957 struct address_space *mapping = inode->i_mapping;
3958 loff_t first_block_offset, last_block_offset;
3960 unsigned int credits;
3963 if (!S_ISREG(inode->i_mode))
3966 trace_ext4_punch_hole(inode, offset, length, 0);
3968 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
3969 if (ext4_has_inline_data(inode)) {
3970 down_write(&EXT4_I(inode)->i_mmap_sem);
3971 ret = ext4_convert_inline_data(inode);
3972 up_write(&EXT4_I(inode)->i_mmap_sem);
3978 * Write out all dirty pages to avoid race conditions
3979 * Then release them.
3981 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3982 ret = filemap_write_and_wait_range(mapping, offset,
3983 offset + length - 1);
3990 /* No need to punch hole beyond i_size */
3991 if (offset >= inode->i_size)
3995 * If the hole extends beyond i_size, set the hole
3996 * to end after the page that contains i_size
3998 if (offset + length > inode->i_size) {
3999 length = inode->i_size +
4000 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4004 if (offset & (sb->s_blocksize - 1) ||
4005 (offset + length) & (sb->s_blocksize - 1)) {
4007 * Attach jinode to inode for jbd2 if we do any zeroing of
4010 ret = ext4_inode_attach_jinode(inode);
4016 /* Wait all existing dio workers, newcomers will block on i_mutex */
4017 ext4_inode_block_unlocked_dio(inode);
4018 inode_dio_wait(inode);
4021 * Prevent page faults from reinstantiating pages we have released from
4024 down_write(&EXT4_I(inode)->i_mmap_sem);
4025 first_block_offset = round_up(offset, sb->s_blocksize);
4026 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4028 /* Now release the pages and zero block aligned part of pages*/
4029 if (last_block_offset > first_block_offset) {
4030 ret = ext4_update_disksize_before_punch(inode, offset, length);
4033 truncate_pagecache_range(inode, first_block_offset,
4037 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4038 credits = ext4_writepage_trans_blocks(inode);
4040 credits = ext4_blocks_for_truncate(inode);
4041 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4042 if (IS_ERR(handle)) {
4043 ret = PTR_ERR(handle);
4044 ext4_std_error(sb, ret);
4048 ret = ext4_zero_partial_blocks(handle, inode, offset,
4053 first_block = (offset + sb->s_blocksize - 1) >>
4054 EXT4_BLOCK_SIZE_BITS(sb);
4055 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4057 /* If there are blocks to remove, do it */
4058 if (stop_block > first_block) {
4060 down_write(&EXT4_I(inode)->i_data_sem);
4061 ext4_discard_preallocations(inode);
4063 ret = ext4_es_remove_extent(inode, first_block,
4064 stop_block - first_block);
4066 up_write(&EXT4_I(inode)->i_data_sem);
4070 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4071 ret = ext4_ext_remove_space(inode, first_block,
4074 ret = ext4_ind_remove_space(handle, inode, first_block,
4077 up_write(&EXT4_I(inode)->i_data_sem);
4080 ext4_handle_sync(handle);
4082 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4083 ext4_mark_inode_dirty(handle, inode);
4085 ext4_update_inode_fsync_trans(handle, inode, 1);
4087 ext4_journal_stop(handle);
4089 up_write(&EXT4_I(inode)->i_mmap_sem);
4090 ext4_inode_resume_unlocked_dio(inode);
4092 inode_unlock(inode);
4096 int ext4_inode_attach_jinode(struct inode *inode)
4098 struct ext4_inode_info *ei = EXT4_I(inode);
4099 struct jbd2_inode *jinode;
4101 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4104 jinode = jbd2_alloc_inode(GFP_KERNEL);
4105 spin_lock(&inode->i_lock);
4108 spin_unlock(&inode->i_lock);
4111 ei->jinode = jinode;
4112 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4115 spin_unlock(&inode->i_lock);
4116 if (unlikely(jinode != NULL))
4117 jbd2_free_inode(jinode);
4124 * We block out ext4_get_block() block instantiations across the entire
4125 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4126 * simultaneously on behalf of the same inode.
4128 * As we work through the truncate and commit bits of it to the journal there
4129 * is one core, guiding principle: the file's tree must always be consistent on
4130 * disk. We must be able to restart the truncate after a crash.
4132 * The file's tree may be transiently inconsistent in memory (although it
4133 * probably isn't), but whenever we close off and commit a journal transaction,
4134 * the contents of (the filesystem + the journal) must be consistent and
4135 * restartable. It's pretty simple, really: bottom up, right to left (although
4136 * left-to-right works OK too).
4138 * Note that at recovery time, journal replay occurs *before* the restart of
4139 * truncate against the orphan inode list.
4141 * The committed inode has the new, desired i_size (which is the same as
4142 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4143 * that this inode's truncate did not complete and it will again call
4144 * ext4_truncate() to have another go. So there will be instantiated blocks
4145 * to the right of the truncation point in a crashed ext4 filesystem. But
4146 * that's fine - as long as they are linked from the inode, the post-crash
4147 * ext4_truncate() run will find them and release them.
4149 void ext4_truncate(struct inode *inode)
4151 struct ext4_inode_info *ei = EXT4_I(inode);
4152 unsigned int credits;
4154 struct address_space *mapping = inode->i_mapping;
4157 * There is a possibility that we're either freeing the inode
4158 * or it's a completely new inode. In those cases we might not
4159 * have i_mutex locked because it's not necessary.
4161 if (!(inode->i_state & (I_NEW|I_FREEING)))
4162 WARN_ON(!inode_is_locked(inode));
4163 trace_ext4_truncate_enter(inode);
4165 if (!ext4_can_truncate(inode))
4168 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4170 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4171 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4173 if (ext4_has_inline_data(inode)) {
4176 ext4_inline_data_truncate(inode, &has_inline);
4181 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4182 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4183 if (ext4_inode_attach_jinode(inode) < 0)
4187 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4188 credits = ext4_writepage_trans_blocks(inode);
4190 credits = ext4_blocks_for_truncate(inode);
4192 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4193 if (IS_ERR(handle)) {
4194 ext4_std_error(inode->i_sb, PTR_ERR(handle));
4198 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4199 ext4_block_truncate_page(handle, mapping, inode->i_size);
4202 * We add the inode to the orphan list, so that if this
4203 * truncate spans multiple transactions, and we crash, we will
4204 * resume the truncate when the filesystem recovers. It also
4205 * marks the inode dirty, to catch the new size.
4207 * Implication: the file must always be in a sane, consistent
4208 * truncatable state while each transaction commits.
4210 if (ext4_orphan_add(handle, inode))
4213 down_write(&EXT4_I(inode)->i_data_sem);
4215 ext4_discard_preallocations(inode);
4217 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4218 ext4_ext_truncate(handle, inode);
4220 ext4_ind_truncate(handle, inode);
4222 up_write(&ei->i_data_sem);
4225 ext4_handle_sync(handle);
4229 * If this was a simple ftruncate() and the file will remain alive,
4230 * then we need to clear up the orphan record which we created above.
4231 * However, if this was a real unlink then we were called by
4232 * ext4_evict_inode(), and we allow that function to clean up the
4233 * orphan info for us.
4236 ext4_orphan_del(handle, inode);
4238 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4239 ext4_mark_inode_dirty(handle, inode);
4240 ext4_journal_stop(handle);
4242 trace_ext4_truncate_exit(inode);
4246 * ext4_get_inode_loc returns with an extra refcount against the inode's
4247 * underlying buffer_head on success. If 'in_mem' is true, we have all
4248 * data in memory that is needed to recreate the on-disk version of this
4251 static int __ext4_get_inode_loc(struct inode *inode,
4252 struct ext4_iloc *iloc, int in_mem)
4254 struct ext4_group_desc *gdp;
4255 struct buffer_head *bh;
4256 struct super_block *sb = inode->i_sb;
4258 int inodes_per_block, inode_offset;
4261 if (inode->i_ino < EXT4_ROOT_INO ||
4262 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4263 return -EFSCORRUPTED;
4265 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4266 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4271 * Figure out the offset within the block group inode table
4273 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4274 inode_offset = ((inode->i_ino - 1) %
4275 EXT4_INODES_PER_GROUP(sb));
4276 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4277 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4279 bh = sb_getblk(sb, block);
4282 if (!buffer_uptodate(bh)) {
4286 * If the buffer has the write error flag, we have failed
4287 * to write out another inode in the same block. In this
4288 * case, we don't have to read the block because we may
4289 * read the old inode data successfully.
4291 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4292 set_buffer_uptodate(bh);
4294 if (buffer_uptodate(bh)) {
4295 /* someone brought it uptodate while we waited */
4301 * If we have all information of the inode in memory and this
4302 * is the only valid inode in the block, we need not read the
4306 struct buffer_head *bitmap_bh;
4309 start = inode_offset & ~(inodes_per_block - 1);
4311 /* Is the inode bitmap in cache? */
4312 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4313 if (unlikely(!bitmap_bh))
4317 * If the inode bitmap isn't in cache then the
4318 * optimisation may end up performing two reads instead
4319 * of one, so skip it.
4321 if (!buffer_uptodate(bitmap_bh)) {
4325 for (i = start; i < start + inodes_per_block; i++) {
4326 if (i == inode_offset)
4328 if (ext4_test_bit(i, bitmap_bh->b_data))
4332 if (i == start + inodes_per_block) {
4333 /* all other inodes are free, so skip I/O */
4334 memset(bh->b_data, 0, bh->b_size);
4335 set_buffer_uptodate(bh);
4343 * If we need to do any I/O, try to pre-readahead extra
4344 * blocks from the inode table.
4346 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4347 ext4_fsblk_t b, end, table;
4349 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4351 table = ext4_inode_table(sb, gdp);
4352 /* s_inode_readahead_blks is always a power of 2 */
4353 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4357 num = EXT4_INODES_PER_GROUP(sb);
4358 if (ext4_has_group_desc_csum(sb))
4359 num -= ext4_itable_unused_count(sb, gdp);
4360 table += num / inodes_per_block;
4364 sb_breadahead(sb, b++);
4368 * There are other valid inodes in the buffer, this inode
4369 * has in-inode xattrs, or we don't have this inode in memory.
4370 * Read the block from disk.
4372 trace_ext4_load_inode(inode);
4374 bh->b_end_io = end_buffer_read_sync;
4375 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4377 if (!buffer_uptodate(bh)) {
4378 EXT4_ERROR_INODE_BLOCK(inode, block,
4379 "unable to read itable block");
4389 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4391 /* We have all inode data except xattrs in memory here. */
4392 return __ext4_get_inode_loc(inode, iloc,
4393 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4396 void ext4_set_inode_flags(struct inode *inode)
4398 unsigned int flags = EXT4_I(inode)->i_flags;
4399 unsigned int new_fl = 0;
4401 if (flags & EXT4_SYNC_FL)
4403 if (flags & EXT4_APPEND_FL)
4405 if (flags & EXT4_IMMUTABLE_FL)
4406 new_fl |= S_IMMUTABLE;
4407 if (flags & EXT4_NOATIME_FL)
4408 new_fl |= S_NOATIME;
4409 if (flags & EXT4_DIRSYNC_FL)
4410 new_fl |= S_DIRSYNC;
4411 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
4413 inode_set_flags(inode, new_fl,
4414 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4417 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4418 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4420 unsigned int vfs_fl;
4421 unsigned long old_fl, new_fl;
4424 vfs_fl = ei->vfs_inode.i_flags;
4425 old_fl = ei->i_flags;
4426 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4427 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4429 if (vfs_fl & S_SYNC)
4430 new_fl |= EXT4_SYNC_FL;
4431 if (vfs_fl & S_APPEND)
4432 new_fl |= EXT4_APPEND_FL;
4433 if (vfs_fl & S_IMMUTABLE)
4434 new_fl |= EXT4_IMMUTABLE_FL;
4435 if (vfs_fl & S_NOATIME)
4436 new_fl |= EXT4_NOATIME_FL;
4437 if (vfs_fl & S_DIRSYNC)
4438 new_fl |= EXT4_DIRSYNC_FL;
4439 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4442 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4443 struct ext4_inode_info *ei)
4446 struct inode *inode = &(ei->vfs_inode);
4447 struct super_block *sb = inode->i_sb;
4449 if (ext4_has_feature_huge_file(sb)) {
4450 /* we are using combined 48 bit field */
4451 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4452 le32_to_cpu(raw_inode->i_blocks_lo);
4453 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4454 /* i_blocks represent file system block size */
4455 return i_blocks << (inode->i_blkbits - 9);
4460 return le32_to_cpu(raw_inode->i_blocks_lo);
4464 static inline void ext4_iget_extra_inode(struct inode *inode,
4465 struct ext4_inode *raw_inode,
4466 struct ext4_inode_info *ei)
4468 __le32 *magic = (void *)raw_inode +
4469 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4470 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4471 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4472 ext4_find_inline_data_nolock(inode);
4474 EXT4_I(inode)->i_inline_off = 0;
4477 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4479 if (!ext4_has_feature_project(inode->i_sb))
4481 *projid = EXT4_I(inode)->i_projid;
4485 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4486 ext4_iget_flags flags, const char *function,
4489 struct ext4_iloc iloc;
4490 struct ext4_inode *raw_inode;
4491 struct ext4_inode_info *ei;
4492 struct inode *inode;
4493 journal_t *journal = EXT4_SB(sb)->s_journal;
4501 if ((!(flags & EXT4_IGET_SPECIAL) &&
4502 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4503 (ino < EXT4_ROOT_INO) ||
4504 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4505 if (flags & EXT4_IGET_HANDLE)
4506 return ERR_PTR(-ESTALE);
4507 __ext4_error(sb, function, line,
4508 "inode #%lu: comm %s: iget: illegal inode #",
4509 ino, current->comm);
4510 return ERR_PTR(-EFSCORRUPTED);
4513 inode = iget_locked(sb, ino);
4515 return ERR_PTR(-ENOMEM);
4516 if (!(inode->i_state & I_NEW))
4522 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4525 raw_inode = ext4_raw_inode(&iloc);
4527 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4528 ext4_error_inode(inode, function, line, 0,
4529 "iget: root inode unallocated");
4530 ret = -EFSCORRUPTED;
4534 if ((flags & EXT4_IGET_HANDLE) &&
4535 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4540 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4541 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4542 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4543 EXT4_INODE_SIZE(inode->i_sb)) {
4544 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4545 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4546 EXT4_INODE_SIZE(inode->i_sb));
4547 ret = -EFSCORRUPTED;
4551 ei->i_extra_isize = 0;
4553 /* Precompute checksum seed for inode metadata */
4554 if (ext4_has_metadata_csum(sb)) {
4555 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4557 __le32 inum = cpu_to_le32(inode->i_ino);
4558 __le32 gen = raw_inode->i_generation;
4559 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4561 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4565 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4566 ext4_error_inode(inode, function, line, 0,
4567 "iget: checksum invalid");
4572 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4573 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4574 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4575 if (ext4_has_feature_project(sb) &&
4576 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4577 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4578 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4580 i_projid = EXT4_DEF_PROJID;
4582 if (!(test_opt(inode->i_sb, NO_UID32))) {
4583 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4584 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4586 i_uid_write(inode, i_uid);
4587 i_gid_write(inode, i_gid);
4588 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4589 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4591 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4592 ei->i_inline_off = 0;
4593 ei->i_dir_start_lookup = 0;
4594 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4595 /* We now have enough fields to check if the inode was active or not.
4596 * This is needed because nfsd might try to access dead inodes
4597 * the test is that same one that e2fsck uses
4598 * NeilBrown 1999oct15
4600 if (inode->i_nlink == 0) {
4601 if ((inode->i_mode == 0 ||
4602 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4603 ino != EXT4_BOOT_LOADER_INO) {
4604 /* this inode is deleted */
4608 /* The only unlinked inodes we let through here have
4609 * valid i_mode and are being read by the orphan
4610 * recovery code: that's fine, we're about to complete
4611 * the process of deleting those.
4612 * OR it is the EXT4_BOOT_LOADER_INO which is
4613 * not initialized on a new filesystem. */
4615 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4616 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4617 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4618 if (ext4_has_feature_64bit(sb))
4620 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4621 inode->i_size = ext4_isize(raw_inode);
4622 if ((size = i_size_read(inode)) < 0) {
4623 ext4_error_inode(inode, function, line, 0,
4624 "iget: bad i_size value: %lld", size);
4625 ret = -EFSCORRUPTED;
4629 * If dir_index is not enabled but there's dir with INDEX flag set,
4630 * we'd normally treat htree data as empty space. But with metadata
4631 * checksumming that corrupts checksums so forbid that.
4633 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4634 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4635 EXT4_ERROR_INODE(inode,
4636 "iget: Dir with htree data on filesystem without dir_index feature.");
4637 ret = -EFSCORRUPTED;
4640 ei->i_disksize = inode->i_size;
4642 ei->i_reserved_quota = 0;
4644 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4645 ei->i_block_group = iloc.block_group;
4646 ei->i_last_alloc_group = ~0;
4648 * NOTE! The in-memory inode i_data array is in little-endian order
4649 * even on big-endian machines: we do NOT byteswap the block numbers!
4651 for (block = 0; block < EXT4_N_BLOCKS; block++)
4652 ei->i_data[block] = raw_inode->i_block[block];
4653 INIT_LIST_HEAD(&ei->i_orphan);
4656 * Set transaction id's of transactions that have to be committed
4657 * to finish f[data]sync. We set them to currently running transaction
4658 * as we cannot be sure that the inode or some of its metadata isn't
4659 * part of the transaction - the inode could have been reclaimed and
4660 * now it is reread from disk.
4663 transaction_t *transaction;
4666 read_lock(&journal->j_state_lock);
4667 if (journal->j_running_transaction)
4668 transaction = journal->j_running_transaction;
4670 transaction = journal->j_committing_transaction;
4672 tid = transaction->t_tid;
4674 tid = journal->j_commit_sequence;
4675 read_unlock(&journal->j_state_lock);
4676 ei->i_sync_tid = tid;
4677 ei->i_datasync_tid = tid;
4680 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4681 if (ei->i_extra_isize == 0) {
4682 /* The extra space is currently unused. Use it. */
4683 ei->i_extra_isize = sizeof(struct ext4_inode) -
4684 EXT4_GOOD_OLD_INODE_SIZE;
4686 ext4_iget_extra_inode(inode, raw_inode, ei);
4690 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4691 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4692 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4693 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4695 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4696 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4697 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4698 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4700 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4705 if (ei->i_file_acl &&
4706 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4707 ext4_error_inode(inode, function, line, 0,
4708 "iget: bad extended attribute block %llu",
4710 ret = -EFSCORRUPTED;
4712 } else if (!ext4_has_inline_data(inode)) {
4713 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4714 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4715 (S_ISLNK(inode->i_mode) &&
4716 !ext4_inode_is_fast_symlink(inode))))
4717 /* Validate extent which is part of inode */
4718 ret = ext4_ext_check_inode(inode);
4719 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4720 (S_ISLNK(inode->i_mode) &&
4721 !ext4_inode_is_fast_symlink(inode))) {
4722 /* Validate block references which are part of inode */
4723 ret = ext4_ind_check_inode(inode);
4729 if (S_ISREG(inode->i_mode)) {
4730 inode->i_op = &ext4_file_inode_operations;
4731 inode->i_fop = &ext4_file_operations;
4732 ext4_set_aops(inode);
4733 } else if (S_ISDIR(inode->i_mode)) {
4734 inode->i_op = &ext4_dir_inode_operations;
4735 inode->i_fop = &ext4_dir_operations;
4736 } else if (S_ISLNK(inode->i_mode)) {
4737 if (ext4_encrypted_inode(inode)) {
4738 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4739 ext4_set_aops(inode);
4740 } else if (ext4_inode_is_fast_symlink(inode)) {
4741 inode->i_link = (char *)ei->i_data;
4742 inode->i_op = &ext4_fast_symlink_inode_operations;
4743 nd_terminate_link(ei->i_data, inode->i_size,
4744 sizeof(ei->i_data) - 1);
4746 inode->i_op = &ext4_symlink_inode_operations;
4747 ext4_set_aops(inode);
4749 inode_nohighmem(inode);
4750 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4751 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4752 inode->i_op = &ext4_special_inode_operations;
4753 if (raw_inode->i_block[0])
4754 init_special_inode(inode, inode->i_mode,
4755 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4757 init_special_inode(inode, inode->i_mode,
4758 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4759 } else if (ino == EXT4_BOOT_LOADER_INO) {
4760 make_bad_inode(inode);
4762 ret = -EFSCORRUPTED;
4763 ext4_error_inode(inode, function, line, 0,
4764 "iget: bogus i_mode (%o)", inode->i_mode);
4768 ext4_set_inode_flags(inode);
4769 unlock_new_inode(inode);
4775 return ERR_PTR(ret);
4778 static int ext4_inode_blocks_set(handle_t *handle,
4779 struct ext4_inode *raw_inode,
4780 struct ext4_inode_info *ei)
4782 struct inode *inode = &(ei->vfs_inode);
4783 u64 i_blocks = READ_ONCE(inode->i_blocks);
4784 struct super_block *sb = inode->i_sb;
4786 if (i_blocks <= ~0U) {
4788 * i_blocks can be represented in a 32 bit variable
4789 * as multiple of 512 bytes
4791 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4792 raw_inode->i_blocks_high = 0;
4793 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4796 if (!ext4_has_feature_huge_file(sb))
4799 if (i_blocks <= 0xffffffffffffULL) {
4801 * i_blocks can be represented in a 48 bit variable
4802 * as multiple of 512 bytes
4804 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4805 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4806 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4808 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4809 /* i_block is stored in file system block size */
4810 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4811 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4812 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4817 struct other_inode {
4818 unsigned long orig_ino;
4819 struct ext4_inode *raw_inode;
4822 static int other_inode_match(struct inode * inode, unsigned long ino,
4825 struct other_inode *oi = (struct other_inode *) data;
4827 if ((inode->i_ino != ino) ||
4828 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4829 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4830 ((inode->i_state & I_DIRTY_TIME) == 0))
4832 spin_lock(&inode->i_lock);
4833 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4834 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4835 (inode->i_state & I_DIRTY_TIME)) {
4836 struct ext4_inode_info *ei = EXT4_I(inode);
4838 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4839 spin_unlock(&inode->i_lock);
4841 spin_lock(&ei->i_raw_lock);
4842 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4843 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4844 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4845 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4846 spin_unlock(&ei->i_raw_lock);
4847 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4850 spin_unlock(&inode->i_lock);
4855 * Opportunistically update the other time fields for other inodes in
4856 * the same inode table block.
4858 static void ext4_update_other_inodes_time(struct super_block *sb,
4859 unsigned long orig_ino, char *buf)
4861 struct other_inode oi;
4863 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4864 int inode_size = EXT4_INODE_SIZE(sb);
4866 oi.orig_ino = orig_ino;
4868 * Calculate the first inode in the inode table block. Inode
4869 * numbers are one-based. That is, the first inode in a block
4870 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4872 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4873 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4874 if (ino == orig_ino)
4876 oi.raw_inode = (struct ext4_inode *) buf;
4877 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4882 * Post the struct inode info into an on-disk inode location in the
4883 * buffer-cache. This gobbles the caller's reference to the
4884 * buffer_head in the inode location struct.
4886 * The caller must have write access to iloc->bh.
4888 static int ext4_do_update_inode(handle_t *handle,
4889 struct inode *inode,
4890 struct ext4_iloc *iloc)
4892 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4893 struct ext4_inode_info *ei = EXT4_I(inode);
4894 struct buffer_head *bh = iloc->bh;
4895 struct super_block *sb = inode->i_sb;
4897 int need_datasync = 0, set_large_file = 0;
4902 spin_lock(&ei->i_raw_lock);
4904 /* For fields not tracked in the in-memory inode,
4905 * initialise them to zero for new inodes. */
4906 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4907 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4909 ext4_get_inode_flags(ei);
4910 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4911 i_uid = i_uid_read(inode);
4912 i_gid = i_gid_read(inode);
4913 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4914 if (!(test_opt(inode->i_sb, NO_UID32))) {
4915 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4916 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4918 * Fix up interoperability with old kernels. Otherwise, old inodes get
4919 * re-used with the upper 16 bits of the uid/gid intact
4921 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4922 raw_inode->i_uid_high = 0;
4923 raw_inode->i_gid_high = 0;
4925 raw_inode->i_uid_high =
4926 cpu_to_le16(high_16_bits(i_uid));
4927 raw_inode->i_gid_high =
4928 cpu_to_le16(high_16_bits(i_gid));
4931 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4932 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4933 raw_inode->i_uid_high = 0;
4934 raw_inode->i_gid_high = 0;
4936 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4938 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4939 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4940 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4941 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4943 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4945 spin_unlock(&ei->i_raw_lock);
4948 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4949 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4950 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4951 raw_inode->i_file_acl_high =
4952 cpu_to_le16(ei->i_file_acl >> 32);
4953 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4954 if (ei->i_disksize != ext4_isize(raw_inode)) {
4955 ext4_isize_set(raw_inode, ei->i_disksize);
4958 if (ei->i_disksize > 0x7fffffffULL) {
4959 if (!ext4_has_feature_large_file(sb) ||
4960 EXT4_SB(sb)->s_es->s_rev_level ==
4961 cpu_to_le32(EXT4_GOOD_OLD_REV))
4964 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4965 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4966 if (old_valid_dev(inode->i_rdev)) {
4967 raw_inode->i_block[0] =
4968 cpu_to_le32(old_encode_dev(inode->i_rdev));
4969 raw_inode->i_block[1] = 0;
4971 raw_inode->i_block[0] = 0;
4972 raw_inode->i_block[1] =
4973 cpu_to_le32(new_encode_dev(inode->i_rdev));
4974 raw_inode->i_block[2] = 0;
4976 } else if (!ext4_has_inline_data(inode)) {
4977 for (block = 0; block < EXT4_N_BLOCKS; block++)
4978 raw_inode->i_block[block] = ei->i_data[block];
4981 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4982 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4983 if (ei->i_extra_isize) {
4984 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4985 raw_inode->i_version_hi =
4986 cpu_to_le32(inode->i_version >> 32);
4987 raw_inode->i_extra_isize =
4988 cpu_to_le16(ei->i_extra_isize);
4992 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
4993 i_projid != EXT4_DEF_PROJID);
4995 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4996 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4997 raw_inode->i_projid = cpu_to_le32(i_projid);
4999 ext4_inode_csum_set(inode, raw_inode, ei);
5000 spin_unlock(&ei->i_raw_lock);
5001 if (inode->i_sb->s_flags & MS_LAZYTIME)
5002 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5005 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5006 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5009 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5010 if (set_large_file) {
5011 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5012 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5015 ext4_update_dynamic_rev(sb);
5016 ext4_set_feature_large_file(sb);
5017 ext4_handle_sync(handle);
5018 err = ext4_handle_dirty_super(handle, sb);
5020 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5023 ext4_std_error(inode->i_sb, err);
5028 * ext4_write_inode()
5030 * We are called from a few places:
5032 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5033 * Here, there will be no transaction running. We wait for any running
5034 * transaction to commit.
5036 * - Within flush work (sys_sync(), kupdate and such).
5037 * We wait on commit, if told to.
5039 * - Within iput_final() -> write_inode_now()
5040 * We wait on commit, if told to.
5042 * In all cases it is actually safe for us to return without doing anything,
5043 * because the inode has been copied into a raw inode buffer in
5044 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5047 * Note that we are absolutely dependent upon all inode dirtiers doing the
5048 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5049 * which we are interested.
5051 * It would be a bug for them to not do this. The code:
5053 * mark_inode_dirty(inode)
5055 * inode->i_size = expr;
5057 * is in error because write_inode() could occur while `stuff()' is running,
5058 * and the new i_size will be lost. Plus the inode will no longer be on the
5059 * superblock's dirty inode list.
5061 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5065 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5068 if (EXT4_SB(inode->i_sb)->s_journal) {
5069 if (ext4_journal_current_handle()) {
5070 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5076 * No need to force transaction in WB_SYNC_NONE mode. Also
5077 * ext4_sync_fs() will force the commit after everything is
5080 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5083 err = ext4_force_commit(inode->i_sb);
5085 struct ext4_iloc iloc;
5087 err = __ext4_get_inode_loc(inode, &iloc, 0);
5091 * sync(2) will flush the whole buffer cache. No need to do
5092 * it here separately for each inode.
5094 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5095 sync_dirty_buffer(iloc.bh);
5096 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5097 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5098 "IO error syncing inode");
5107 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5108 * buffers that are attached to a page stradding i_size and are undergoing
5109 * commit. In that case we have to wait for commit to finish and try again.
5111 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5115 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5116 tid_t commit_tid = 0;
5119 offset = inode->i_size & (PAGE_SIZE - 1);
5121 * If the page is fully truncated, we don't need to wait for any commit
5122 * (and we even should not as __ext4_journalled_invalidatepage() may
5123 * strip all buffers from the page but keep the page dirty which can then
5124 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5125 * buffers). Also we don't need to wait for any commit if all buffers in
5126 * the page remain valid. This is most beneficial for the common case of
5127 * blocksize == PAGESIZE.
5129 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5132 page = find_lock_page(inode->i_mapping,
5133 inode->i_size >> PAGE_SHIFT);
5136 ret = __ext4_journalled_invalidatepage(page, offset,
5137 PAGE_SIZE - offset);
5143 read_lock(&journal->j_state_lock);
5144 if (journal->j_committing_transaction)
5145 commit_tid = journal->j_committing_transaction->t_tid;
5146 read_unlock(&journal->j_state_lock);
5148 jbd2_log_wait_commit(journal, commit_tid);
5155 * Called from notify_change.
5157 * We want to trap VFS attempts to truncate the file as soon as
5158 * possible. In particular, we want to make sure that when the VFS
5159 * shrinks i_size, we put the inode on the orphan list and modify
5160 * i_disksize immediately, so that during the subsequent flushing of
5161 * dirty pages and freeing of disk blocks, we can guarantee that any
5162 * commit will leave the blocks being flushed in an unused state on
5163 * disk. (On recovery, the inode will get truncated and the blocks will
5164 * be freed, so we have a strong guarantee that no future commit will
5165 * leave these blocks visible to the user.)
5167 * Another thing we have to assure is that if we are in ordered mode
5168 * and inode is still attached to the committing transaction, we must
5169 * we start writeout of all the dirty pages which are being truncated.
5170 * This way we are sure that all the data written in the previous
5171 * transaction are already on disk (truncate waits for pages under
5174 * Called with inode->i_mutex down.
5176 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5178 struct inode *inode = d_inode(dentry);
5181 const unsigned int ia_valid = attr->ia_valid;
5183 error = setattr_prepare(dentry, attr);
5187 if (is_quota_modification(inode, attr)) {
5188 error = dquot_initialize(inode);
5192 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5193 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5196 /* (user+group)*(old+new) structure, inode write (sb,
5197 * inode block, ? - but truncate inode update has it) */
5198 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5199 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5200 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5201 if (IS_ERR(handle)) {
5202 error = PTR_ERR(handle);
5205 error = dquot_transfer(inode, attr);
5207 ext4_journal_stop(handle);
5210 /* Update corresponding info in inode so that everything is in
5211 * one transaction */
5212 if (attr->ia_valid & ATTR_UID)
5213 inode->i_uid = attr->ia_uid;
5214 if (attr->ia_valid & ATTR_GID)
5215 inode->i_gid = attr->ia_gid;
5216 error = ext4_mark_inode_dirty(handle, inode);
5217 ext4_journal_stop(handle);
5220 if (attr->ia_valid & ATTR_SIZE) {
5222 loff_t oldsize = inode->i_size;
5223 int shrink = (attr->ia_size <= inode->i_size);
5225 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5226 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5228 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5231 if (!S_ISREG(inode->i_mode))
5234 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5235 inode_inc_iversion(inode);
5237 if (ext4_should_order_data(inode) &&
5238 (attr->ia_size < inode->i_size)) {
5239 error = ext4_begin_ordered_truncate(inode,
5244 if (attr->ia_size != inode->i_size) {
5245 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5246 if (IS_ERR(handle)) {
5247 error = PTR_ERR(handle);
5250 if (ext4_handle_valid(handle) && shrink) {
5251 error = ext4_orphan_add(handle, inode);
5255 * Update c/mtime on truncate up, ext4_truncate() will
5256 * update c/mtime in shrink case below
5259 inode->i_mtime = ext4_current_time(inode);
5260 inode->i_ctime = inode->i_mtime;
5262 down_write(&EXT4_I(inode)->i_data_sem);
5263 EXT4_I(inode)->i_disksize = attr->ia_size;
5264 rc = ext4_mark_inode_dirty(handle, inode);
5268 * We have to update i_size under i_data_sem together
5269 * with i_disksize to avoid races with writeback code
5270 * running ext4_wb_update_i_disksize().
5273 i_size_write(inode, attr->ia_size);
5274 up_write(&EXT4_I(inode)->i_data_sem);
5275 ext4_journal_stop(handle);
5277 if (orphan && inode->i_nlink)
5278 ext4_orphan_del(NULL, inode);
5283 pagecache_isize_extended(inode, oldsize, inode->i_size);
5286 * Blocks are going to be removed from the inode. Wait
5287 * for dio in flight. Temporarily disable
5288 * dioread_nolock to prevent livelock.
5291 if (!ext4_should_journal_data(inode)) {
5292 ext4_inode_block_unlocked_dio(inode);
5293 inode_dio_wait(inode);
5294 ext4_inode_resume_unlocked_dio(inode);
5296 ext4_wait_for_tail_page_commit(inode);
5298 down_write(&EXT4_I(inode)->i_mmap_sem);
5300 * Truncate pagecache after we've waited for commit
5301 * in data=journal mode to make pages freeable.
5303 truncate_pagecache(inode, inode->i_size);
5305 ext4_truncate(inode);
5306 up_write(&EXT4_I(inode)->i_mmap_sem);
5310 setattr_copy(inode, attr);
5311 mark_inode_dirty(inode);
5315 * If the call to ext4_truncate failed to get a transaction handle at
5316 * all, we need to clean up the in-core orphan list manually.
5318 if (orphan && inode->i_nlink)
5319 ext4_orphan_del(NULL, inode);
5321 if (!rc && (ia_valid & ATTR_MODE))
5322 rc = posix_acl_chmod(inode, inode->i_mode);
5325 ext4_std_error(inode->i_sb, error);
5331 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5334 struct inode *inode;
5335 unsigned long long delalloc_blocks;
5337 inode = d_inode(dentry);
5338 generic_fillattr(inode, stat);
5341 * If there is inline data in the inode, the inode will normally not
5342 * have data blocks allocated (it may have an external xattr block).
5343 * Report at least one sector for such files, so tools like tar, rsync,
5344 * others doen't incorrectly think the file is completely sparse.
5346 if (unlikely(ext4_has_inline_data(inode)))
5347 stat->blocks += (stat->size + 511) >> 9;
5350 * We can't update i_blocks if the block allocation is delayed
5351 * otherwise in the case of system crash before the real block
5352 * allocation is done, we will have i_blocks inconsistent with
5353 * on-disk file blocks.
5354 * We always keep i_blocks updated together with real
5355 * allocation. But to not confuse with user, stat
5356 * will return the blocks that include the delayed allocation
5357 * blocks for this file.
5359 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5360 EXT4_I(inode)->i_reserved_data_blocks);
5361 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5365 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5368 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5369 return ext4_ind_trans_blocks(inode, lblocks);
5370 return ext4_ext_index_trans_blocks(inode, pextents);
5374 * Account for index blocks, block groups bitmaps and block group
5375 * descriptor blocks if modify datablocks and index blocks
5376 * worse case, the indexs blocks spread over different block groups
5378 * If datablocks are discontiguous, they are possible to spread over
5379 * different block groups too. If they are contiguous, with flexbg,
5380 * they could still across block group boundary.
5382 * Also account for superblock, inode, quota and xattr blocks
5384 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5387 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5393 * How many index blocks need to touch to map @lblocks logical blocks
5394 * to @pextents physical extents?
5396 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5401 * Now let's see how many group bitmaps and group descriptors need
5404 groups = idxblocks + pextents;
5406 if (groups > ngroups)
5408 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5409 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5411 /* bitmaps and block group descriptor blocks */
5412 ret += groups + gdpblocks;
5414 /* Blocks for super block, inode, quota and xattr blocks */
5415 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5421 * Calculate the total number of credits to reserve to fit
5422 * the modification of a single pages into a single transaction,
5423 * which may include multiple chunks of block allocations.
5425 * This could be called via ext4_write_begin()
5427 * We need to consider the worse case, when
5428 * one new block per extent.
5430 int ext4_writepage_trans_blocks(struct inode *inode)
5432 int bpp = ext4_journal_blocks_per_page(inode);
5435 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5437 /* Account for data blocks for journalled mode */
5438 if (ext4_should_journal_data(inode))
5444 * Calculate the journal credits for a chunk of data modification.
5446 * This is called from DIO, fallocate or whoever calling
5447 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5449 * journal buffers for data blocks are not included here, as DIO
5450 * and fallocate do no need to journal data buffers.
5452 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5454 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5458 * The caller must have previously called ext4_reserve_inode_write().
5459 * Give this, we know that the caller already has write access to iloc->bh.
5461 int ext4_mark_iloc_dirty(handle_t *handle,
5462 struct inode *inode, struct ext4_iloc *iloc)
5466 if (IS_I_VERSION(inode))
5467 inode_inc_iversion(inode);
5469 /* the do_update_inode consumes one bh->b_count */
5472 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5473 err = ext4_do_update_inode(handle, inode, iloc);
5479 * On success, We end up with an outstanding reference count against
5480 * iloc->bh. This _must_ be cleaned up later.
5484 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5485 struct ext4_iloc *iloc)
5489 err = ext4_get_inode_loc(inode, iloc);
5491 BUFFER_TRACE(iloc->bh, "get_write_access");
5492 err = ext4_journal_get_write_access(handle, iloc->bh);
5498 ext4_std_error(inode->i_sb, err);
5503 * Expand an inode by new_extra_isize bytes.
5504 * Returns 0 on success or negative error number on failure.
5506 static int ext4_expand_extra_isize(struct inode *inode,
5507 unsigned int new_extra_isize,
5508 struct ext4_iloc iloc,
5511 struct ext4_inode *raw_inode;
5512 struct ext4_xattr_ibody_header *header;
5513 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5514 struct ext4_inode_info *ei = EXT4_I(inode);
5516 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5519 /* this was checked at iget time, but double check for good measure */
5520 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5521 (ei->i_extra_isize & 3)) {
5522 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5524 EXT4_INODE_SIZE(inode->i_sb));
5525 return -EFSCORRUPTED;
5527 if ((new_extra_isize < ei->i_extra_isize) ||
5528 (new_extra_isize < 4) ||
5529 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5530 return -EINVAL; /* Should never happen */
5532 raw_inode = ext4_raw_inode(&iloc);
5534 header = IHDR(inode, raw_inode);
5536 /* No extended attributes present */
5537 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5538 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5539 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5540 EXT4_I(inode)->i_extra_isize, 0,
5541 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5542 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5546 /* try to expand with EAs present */
5547 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5552 * What we do here is to mark the in-core inode as clean with respect to inode
5553 * dirtiness (it may still be data-dirty).
5554 * This means that the in-core inode may be reaped by prune_icache
5555 * without having to perform any I/O. This is a very good thing,
5556 * because *any* task may call prune_icache - even ones which
5557 * have a transaction open against a different journal.
5559 * Is this cheating? Not really. Sure, we haven't written the
5560 * inode out, but prune_icache isn't a user-visible syncing function.
5561 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5562 * we start and wait on commits.
5564 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5566 struct ext4_iloc iloc;
5567 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5568 static unsigned int mnt_count;
5572 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5573 err = ext4_reserve_inode_write(handle, inode, &iloc);
5576 if (ext4_handle_valid(handle) &&
5577 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5578 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5580 * We need extra buffer credits since we may write into EA block
5581 * with this same handle. If journal_extend fails, then it will
5582 * only result in a minor loss of functionality for that inode.
5583 * If this is felt to be critical, then e2fsck should be run to
5584 * force a large enough s_min_extra_isize.
5586 if ((jbd2_journal_extend(handle,
5587 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5588 ret = ext4_expand_extra_isize(inode,
5589 sbi->s_want_extra_isize,
5593 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5594 ext4_warning(inode->i_sb,
5595 "Unable to expand inode %lu. Delete"
5596 " some EAs or run e2fsck.",
5599 le16_to_cpu(sbi->s_es->s_mnt_count);
5604 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5608 * ext4_dirty_inode() is called from __mark_inode_dirty()
5610 * We're really interested in the case where a file is being extended.
5611 * i_size has been changed by generic_commit_write() and we thus need
5612 * to include the updated inode in the current transaction.
5614 * Also, dquot_alloc_block() will always dirty the inode when blocks
5615 * are allocated to the file.
5617 * If the inode is marked synchronous, we don't honour that here - doing
5618 * so would cause a commit on atime updates, which we don't bother doing.
5619 * We handle synchronous inodes at the highest possible level.
5621 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5622 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5623 * to copy into the on-disk inode structure are the timestamp files.
5625 void ext4_dirty_inode(struct inode *inode, int flags)
5629 if (flags == I_DIRTY_TIME)
5631 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5635 ext4_mark_inode_dirty(handle, inode);
5637 ext4_journal_stop(handle);
5644 * Bind an inode's backing buffer_head into this transaction, to prevent
5645 * it from being flushed to disk early. Unlike
5646 * ext4_reserve_inode_write, this leaves behind no bh reference and
5647 * returns no iloc structure, so the caller needs to repeat the iloc
5648 * lookup to mark the inode dirty later.
5650 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5652 struct ext4_iloc iloc;
5656 err = ext4_get_inode_loc(inode, &iloc);
5658 BUFFER_TRACE(iloc.bh, "get_write_access");
5659 err = jbd2_journal_get_write_access(handle, iloc.bh);
5661 err = ext4_handle_dirty_metadata(handle,
5667 ext4_std_error(inode->i_sb, err);
5672 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5677 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5680 * We have to be very careful here: changing a data block's
5681 * journaling status dynamically is dangerous. If we write a
5682 * data block to the journal, change the status and then delete
5683 * that block, we risk forgetting to revoke the old log record
5684 * from the journal and so a subsequent replay can corrupt data.
5685 * So, first we make sure that the journal is empty and that
5686 * nobody is changing anything.
5689 journal = EXT4_JOURNAL(inode);
5692 if (is_journal_aborted(journal))
5695 /* Wait for all existing dio workers */
5696 ext4_inode_block_unlocked_dio(inode);
5697 inode_dio_wait(inode);
5700 * Before flushing the journal and switching inode's aops, we have
5701 * to flush all dirty data the inode has. There can be outstanding
5702 * delayed allocations, there can be unwritten extents created by
5703 * fallocate or buffered writes in dioread_nolock mode covered by
5704 * dirty data which can be converted only after flushing the dirty
5705 * data (and journalled aops don't know how to handle these cases).
5708 down_write(&EXT4_I(inode)->i_mmap_sem);
5709 err = filemap_write_and_wait(inode->i_mapping);
5711 up_write(&EXT4_I(inode)->i_mmap_sem);
5712 ext4_inode_resume_unlocked_dio(inode);
5717 percpu_down_write(&sbi->s_writepages_rwsem);
5718 jbd2_journal_lock_updates(journal);
5721 * OK, there are no updates running now, and all cached data is
5722 * synced to disk. We are now in a completely consistent state
5723 * which doesn't have anything in the journal, and we know that
5724 * no filesystem updates are running, so it is safe to modify
5725 * the inode's in-core data-journaling state flag now.
5729 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5731 err = jbd2_journal_flush(journal);
5733 jbd2_journal_unlock_updates(journal);
5734 percpu_up_write(&sbi->s_writepages_rwsem);
5735 ext4_inode_resume_unlocked_dio(inode);
5738 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5740 ext4_set_aops(inode);
5742 jbd2_journal_unlock_updates(journal);
5743 percpu_up_write(&sbi->s_writepages_rwsem);
5746 up_write(&EXT4_I(inode)->i_mmap_sem);
5747 ext4_inode_resume_unlocked_dio(inode);
5749 /* Finally we can mark the inode as dirty. */
5751 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5753 return PTR_ERR(handle);
5755 err = ext4_mark_inode_dirty(handle, inode);
5756 ext4_handle_sync(handle);
5757 ext4_journal_stop(handle);
5758 ext4_std_error(inode->i_sb, err);
5763 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5765 return !buffer_mapped(bh);
5768 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5770 struct page *page = vmf->page;
5774 struct file *file = vma->vm_file;
5775 struct inode *inode = file_inode(file);
5776 struct address_space *mapping = inode->i_mapping;
5778 get_block_t *get_block;
5781 sb_start_pagefault(inode->i_sb);
5782 file_update_time(vma->vm_file);
5784 down_read(&EXT4_I(inode)->i_mmap_sem);
5786 ret = ext4_convert_inline_data(inode);
5790 /* Delalloc case is easy... */
5791 if (test_opt(inode->i_sb, DELALLOC) &&
5792 !ext4_should_journal_data(inode) &&
5793 !ext4_nonda_switch(inode->i_sb)) {
5795 ret = block_page_mkwrite(vma, vmf,
5796 ext4_da_get_block_prep);
5797 } while (ret == -ENOSPC &&
5798 ext4_should_retry_alloc(inode->i_sb, &retries));
5803 size = i_size_read(inode);
5804 /* Page got truncated from under us? */
5805 if (page->mapping != mapping || page_offset(page) > size) {
5807 ret = VM_FAULT_NOPAGE;
5811 if (page->index == size >> PAGE_SHIFT)
5812 len = size & ~PAGE_MASK;
5816 * Return if we have all the buffers mapped. This avoids the need to do
5817 * journal_start/journal_stop which can block and take a long time
5819 if (page_has_buffers(page)) {
5820 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5822 ext4_bh_unmapped)) {
5823 /* Wait so that we don't change page under IO */
5824 wait_for_stable_page(page);
5825 ret = VM_FAULT_LOCKED;
5830 /* OK, we need to fill the hole... */
5831 if (ext4_should_dioread_nolock(inode))
5832 get_block = ext4_get_block_unwritten;
5834 get_block = ext4_get_block;
5836 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5837 ext4_writepage_trans_blocks(inode));
5838 if (IS_ERR(handle)) {
5839 ret = VM_FAULT_SIGBUS;
5842 ret = block_page_mkwrite(vma, vmf, get_block);
5843 if (!ret && ext4_should_journal_data(inode)) {
5844 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5845 PAGE_SIZE, NULL, do_journal_get_write_access)) {
5847 ret = VM_FAULT_SIGBUS;
5848 ext4_journal_stop(handle);
5851 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5853 ext4_journal_stop(handle);
5854 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5857 ret = block_page_mkwrite_return(ret);
5859 up_read(&EXT4_I(inode)->i_mmap_sem);
5860 sb_end_pagefault(inode->i_sb);
5864 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5866 struct inode *inode = file_inode(vma->vm_file);
5869 down_read(&EXT4_I(inode)->i_mmap_sem);
5870 err = filemap_fault(vma, vmf);
5871 up_read(&EXT4_I(inode)->i_mmap_sem);
5877 * Find the first extent at or after @lblk in an inode that is not a hole.
5878 * Search for @map_len blocks at most. The extent is returned in @result.
5880 * The function returns 1 if we found an extent. The function returns 0 in
5881 * case there is no extent at or after @lblk and in that case also sets
5882 * @result->es_len to 0. In case of error, the error code is returned.
5884 int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
5885 unsigned int map_len, struct extent_status *result)
5887 struct ext4_map_blocks map;
5888 struct extent_status es = {};
5892 map.m_len = map_len;
5895 * For non-extent based files this loop may iterate several times since
5896 * we do not determine full hole size.
5898 while (map.m_len > 0) {
5899 ret = ext4_map_blocks(NULL, inode, &map, 0);
5902 /* There's extent covering m_lblk? Just return it. */
5906 ext4_es_store_pblock(result, map.m_pblk);
5907 result->es_lblk = map.m_lblk;
5908 result->es_len = map.m_len;
5909 if (map.m_flags & EXT4_MAP_UNWRITTEN)
5910 status = EXTENT_STATUS_UNWRITTEN;
5912 status = EXTENT_STATUS_WRITTEN;
5913 ext4_es_store_status(result, status);
5916 ext4_es_find_delayed_extent_range(inode, map.m_lblk,
5917 map.m_lblk + map.m_len - 1,
5919 /* Is delalloc data before next block in extent tree? */
5920 if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
5921 ext4_lblk_t offset = 0;
5923 if (es.es_lblk < lblk)
5924 offset = lblk - es.es_lblk;
5925 result->es_lblk = es.es_lblk + offset;
5926 ext4_es_store_pblock(result,
5927 ext4_es_pblock(&es) + offset);
5928 result->es_len = es.es_len - offset;
5929 ext4_es_store_status(result, ext4_es_status(&es));
5933 /* There's a hole at m_lblk, advance us after it */
5934 map.m_lblk += map.m_len;
5935 map_len -= map.m_len;
5936 map.m_len = map_len;