1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
54 struct ext4_inode_info *ei)
56 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 int offset = offsetof(struct ext4_inode, i_checksum_lo);
60 unsigned int csum_size = sizeof(dummy_csum);
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
63 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
65 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
66 EXT4_GOOD_OLD_INODE_SIZE - offset);
68 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
69 offset = offsetof(struct ext4_inode, i_checksum_hi);
70 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
71 EXT4_GOOD_OLD_INODE_SIZE,
72 offset - EXT4_GOOD_OLD_INODE_SIZE);
73 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
74 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
78 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
79 EXT4_INODE_SIZE(inode->i_sb) - offset);
85 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
86 struct ext4_inode_info *ei)
88 __u32 provided, calculated;
90 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
91 cpu_to_le32(EXT4_OS_LINUX) ||
92 !ext4_has_metadata_csum(inode->i_sb))
95 provided = le16_to_cpu(raw->i_checksum_lo);
96 calculated = ext4_inode_csum(inode, raw, ei);
97 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
98 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
99 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 calculated &= 0xFFFF;
103 return provided == calculated;
106 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107 struct ext4_inode_info *ei)
111 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112 cpu_to_le32(EXT4_OS_LINUX) ||
113 !ext4_has_metadata_csum(inode->i_sb))
116 csum = ext4_inode_csum(inode, raw, ei);
117 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
123 static inline int ext4_begin_ordered_truncate(struct inode *inode,
126 trace_ext4_begin_ordered_truncate(inode, new_size);
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
133 if (!EXT4_I(inode)->jinode)
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136 EXT4_I(inode)->jinode,
140 static void ext4_invalidatepage(struct page *page, unsigned int offset,
141 unsigned int length);
142 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
144 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
151 int ext4_inode_is_fast_symlink(struct inode *inode)
153 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
157 if (ext4_has_inline_data(inode))
160 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
162 return S_ISLNK(inode->i_mode) && inode->i_size &&
163 (inode->i_size < EXT4_N_BLOCKS * 4);
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
177 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
178 * moment, get_block can be called only for blocks inside i_size since
179 * page cache has been already dropped and writes are blocked by
180 * i_mutex. So we can safely drop the i_data_sem here.
182 BUG_ON(EXT4_JOURNAL(inode) == NULL);
183 jbd_debug(2, "restarting handle %p\n", handle);
184 up_write(&EXT4_I(inode)->i_data_sem);
185 ret = ext4_journal_restart(handle, nblocks);
186 down_write(&EXT4_I(inode)->i_data_sem);
187 ext4_discard_preallocations(inode);
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_evict_inode(struct inode *inode)
199 int extra_credits = 3;
200 struct ext4_xattr_inode_array *ea_inode_array = NULL;
202 trace_ext4_evict_inode(inode);
204 if (inode->i_nlink) {
206 * When journalling data dirty buffers are tracked only in the
207 * journal. So although mm thinks everything is clean and
208 * ready for reaping the inode might still have some pages to
209 * write in the running transaction or waiting to be
210 * checkpointed. Thus calling jbd2_journal_invalidatepage()
211 * (via truncate_inode_pages()) to discard these buffers can
212 * cause data loss. Also even if we did not discard these
213 * buffers, we would have no way to find them after the inode
214 * is reaped and thus user could see stale data if he tries to
215 * read them before the transaction is checkpointed. So be
216 * careful and force everything to disk here... We use
217 * ei->i_datasync_tid to store the newest transaction
218 * containing inode's data.
220 * Note that directories do not have this problem because they
221 * don't use page cache.
223 if (inode->i_ino != EXT4_JOURNAL_INO &&
224 ext4_should_journal_data(inode) &&
225 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
226 inode->i_data.nrpages) {
227 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
228 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
230 jbd2_complete_transaction(journal, commit_tid);
231 filemap_write_and_wait(&inode->i_data);
233 truncate_inode_pages_final(&inode->i_data);
238 if (is_bad_inode(inode))
240 dquot_initialize(inode);
242 if (ext4_should_order_data(inode))
243 ext4_begin_ordered_truncate(inode, 0);
244 truncate_inode_pages_final(&inode->i_data);
247 * Protect us against freezing - iput() caller didn't have to have any
248 * protection against it
250 sb_start_intwrite(inode->i_sb);
252 if (!IS_NOQUOTA(inode))
253 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
255 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
256 ext4_blocks_for_truncate(inode)+extra_credits);
257 if (IS_ERR(handle)) {
258 ext4_std_error(inode->i_sb, PTR_ERR(handle));
260 * If we're going to skip the normal cleanup, we still need to
261 * make sure that the in-core orphan linked list is properly
264 ext4_orphan_del(NULL, inode);
265 sb_end_intwrite(inode->i_sb);
270 ext4_handle_sync(handle);
273 * Set inode->i_size to 0 before calling ext4_truncate(). We need
274 * special handling of symlinks here because i_size is used to
275 * determine whether ext4_inode_info->i_data contains symlink data or
276 * block mappings. Setting i_size to 0 will remove its fast symlink
277 * status. Erase i_data so that it becomes a valid empty block map.
279 if (ext4_inode_is_fast_symlink(inode))
280 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
282 err = ext4_mark_inode_dirty(handle, inode);
284 ext4_warning(inode->i_sb,
285 "couldn't mark inode dirty (err %d)", err);
288 if (inode->i_blocks) {
289 err = ext4_truncate(inode);
291 ext4_error(inode->i_sb,
292 "couldn't truncate inode %lu (err %d)",
298 /* Remove xattr references. */
299 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
302 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
304 ext4_journal_stop(handle);
305 ext4_orphan_del(NULL, inode);
306 sb_end_intwrite(inode->i_sb);
307 ext4_xattr_inode_array_free(ea_inode_array);
312 * Kill off the orphan record which ext4_truncate created.
313 * AKPM: I think this can be inside the above `if'.
314 * Note that ext4_orphan_del() has to be able to cope with the
315 * deletion of a non-existent orphan - this is because we don't
316 * know if ext4_truncate() actually created an orphan record.
317 * (Well, we could do this if we need to, but heck - it works)
319 ext4_orphan_del(handle, inode);
320 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
323 * One subtle ordering requirement: if anything has gone wrong
324 * (transaction abort, IO errors, whatever), then we can still
325 * do these next steps (the fs will already have been marked as
326 * having errors), but we can't free the inode if the mark_dirty
329 if (ext4_mark_inode_dirty(handle, inode))
330 /* If that failed, just do the required in-core inode clear. */
331 ext4_clear_inode(inode);
333 ext4_free_inode(handle, inode);
334 ext4_journal_stop(handle);
335 sb_end_intwrite(inode->i_sb);
336 ext4_xattr_inode_array_free(ea_inode_array);
339 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
343 qsize_t *ext4_get_reserved_space(struct inode *inode)
345 return &EXT4_I(inode)->i_reserved_quota;
350 * Called with i_data_sem down, which is important since we can call
351 * ext4_discard_preallocations() from here.
353 void ext4_da_update_reserve_space(struct inode *inode,
354 int used, int quota_claim)
356 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
357 struct ext4_inode_info *ei = EXT4_I(inode);
359 spin_lock(&ei->i_block_reservation_lock);
360 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
361 if (unlikely(used > ei->i_reserved_data_blocks)) {
362 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
363 "with only %d reserved data blocks",
364 __func__, inode->i_ino, used,
365 ei->i_reserved_data_blocks);
367 used = ei->i_reserved_data_blocks;
370 /* Update per-inode reservations */
371 ei->i_reserved_data_blocks -= used;
372 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
374 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
376 /* Update quota subsystem for data blocks */
378 dquot_claim_block(inode, EXT4_C2B(sbi, used));
381 * We did fallocate with an offset that is already delayed
382 * allocated. So on delayed allocated writeback we should
383 * not re-claim the quota for fallocated blocks.
385 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
389 * If we have done all the pending block allocations and if
390 * there aren't any writers on the inode, we can discard the
391 * inode's preallocations.
393 if ((ei->i_reserved_data_blocks == 0) &&
394 !inode_is_open_for_write(inode))
395 ext4_discard_preallocations(inode);
398 static int __check_block_validity(struct inode *inode, const char *func,
400 struct ext4_map_blocks *map)
402 if (ext4_has_feature_journal(inode->i_sb) &&
404 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
406 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
408 ext4_error_inode(inode, func, line, map->m_pblk,
409 "lblock %lu mapped to illegal pblock %llu "
410 "(length %d)", (unsigned long) map->m_lblk,
411 map->m_pblk, map->m_len);
412 return -EFSCORRUPTED;
417 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
422 if (IS_ENCRYPTED(inode))
423 return fscrypt_zeroout_range(inode, lblk, pblk, len);
425 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
432 #define check_block_validity(inode, map) \
433 __check_block_validity((inode), __func__, __LINE__, (map))
435 #ifdef ES_AGGRESSIVE_TEST
436 static void ext4_map_blocks_es_recheck(handle_t *handle,
438 struct ext4_map_blocks *es_map,
439 struct ext4_map_blocks *map,
446 * There is a race window that the result is not the same.
447 * e.g. xfstests #223 when dioread_nolock enables. The reason
448 * is that we lookup a block mapping in extent status tree with
449 * out taking i_data_sem. So at the time the unwritten extent
450 * could be converted.
452 down_read(&EXT4_I(inode)->i_data_sem);
453 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
454 retval = ext4_ext_map_blocks(handle, inode, map, flags &
455 EXT4_GET_BLOCKS_KEEP_SIZE);
457 retval = ext4_ind_map_blocks(handle, inode, map, flags &
458 EXT4_GET_BLOCKS_KEEP_SIZE);
460 up_read((&EXT4_I(inode)->i_data_sem));
463 * We don't check m_len because extent will be collpased in status
464 * tree. So the m_len might not equal.
466 if (es_map->m_lblk != map->m_lblk ||
467 es_map->m_flags != map->m_flags ||
468 es_map->m_pblk != map->m_pblk) {
469 printk("ES cache assertion failed for inode: %lu "
470 "es_cached ex [%d/%d/%llu/%x] != "
471 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472 inode->i_ino, es_map->m_lblk, es_map->m_len,
473 es_map->m_pblk, es_map->m_flags, map->m_lblk,
474 map->m_len, map->m_pblk, map->m_flags,
478 #endif /* ES_AGGRESSIVE_TEST */
481 * The ext4_map_blocks() function tries to look up the requested blocks,
482 * and returns if the blocks are already mapped.
484 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485 * and store the allocated blocks in the result buffer head and mark it
488 * If file type is extents based, it will call ext4_ext_map_blocks(),
489 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
492 * On success, it returns the number of blocks being mapped or allocated. if
493 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
494 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
496 * It returns 0 if plain look up failed (blocks have not been allocated), in
497 * that case, @map is returned as unmapped but we still do fill map->m_len to
498 * indicate the length of a hole starting at map->m_lblk.
500 * It returns the error in case of allocation failure.
502 int ext4_map_blocks(handle_t *handle, struct inode *inode,
503 struct ext4_map_blocks *map, int flags)
505 struct extent_status es;
508 #ifdef ES_AGGRESSIVE_TEST
509 struct ext4_map_blocks orig_map;
511 memcpy(&orig_map, map, sizeof(*map));
515 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
516 "logical block %lu\n", inode->i_ino, flags, map->m_len,
517 (unsigned long) map->m_lblk);
520 * ext4_map_blocks returns an int, and m_len is an unsigned int
522 if (unlikely(map->m_len > INT_MAX))
523 map->m_len = INT_MAX;
525 /* We can handle the block number less than EXT_MAX_BLOCKS */
526 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
527 return -EFSCORRUPTED;
529 /* Lookup extent status tree firstly */
530 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
531 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
532 map->m_pblk = ext4_es_pblock(&es) +
533 map->m_lblk - es.es_lblk;
534 map->m_flags |= ext4_es_is_written(&es) ?
535 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
536 retval = es.es_len - (map->m_lblk - es.es_lblk);
537 if (retval > map->m_len)
540 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
542 retval = es.es_len - (map->m_lblk - es.es_lblk);
543 if (retval > map->m_len)
550 #ifdef ES_AGGRESSIVE_TEST
551 ext4_map_blocks_es_recheck(handle, inode, map,
558 * Try to see if we can get the block without requesting a new
561 down_read(&EXT4_I(inode)->i_data_sem);
562 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
563 retval = ext4_ext_map_blocks(handle, inode, map, flags &
564 EXT4_GET_BLOCKS_KEEP_SIZE);
566 retval = ext4_ind_map_blocks(handle, inode, map, flags &
567 EXT4_GET_BLOCKS_KEEP_SIZE);
572 if (unlikely(retval != map->m_len)) {
573 ext4_warning(inode->i_sb,
574 "ES len assertion failed for inode "
575 "%lu: retval %d != map->m_len %d",
576 inode->i_ino, retval, map->m_len);
580 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
581 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
582 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
583 !(status & EXTENT_STATUS_WRITTEN) &&
584 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
585 map->m_lblk + map->m_len - 1))
586 status |= EXTENT_STATUS_DELAYED;
587 ret = ext4_es_insert_extent(inode, map->m_lblk,
588 map->m_len, map->m_pblk, status);
592 up_read((&EXT4_I(inode)->i_data_sem));
595 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
596 ret = check_block_validity(inode, map);
601 /* If it is only a block(s) look up */
602 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
606 * Returns if the blocks have already allocated
608 * Note that if blocks have been preallocated
609 * ext4_ext_get_block() returns the create = 0
610 * with buffer head unmapped.
612 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
614 * If we need to convert extent to unwritten
615 * we continue and do the actual work in
616 * ext4_ext_map_blocks()
618 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
622 * Here we clear m_flags because after allocating an new extent,
623 * it will be set again.
625 map->m_flags &= ~EXT4_MAP_FLAGS;
628 * New blocks allocate and/or writing to unwritten extent
629 * will possibly result in updating i_data, so we take
630 * the write lock of i_data_sem, and call get_block()
631 * with create == 1 flag.
633 down_write(&EXT4_I(inode)->i_data_sem);
636 * We need to check for EXT4 here because migrate
637 * could have changed the inode type in between
639 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
640 retval = ext4_ext_map_blocks(handle, inode, map, flags);
642 retval = ext4_ind_map_blocks(handle, inode, map, flags);
644 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
646 * We allocated new blocks which will result in
647 * i_data's format changing. Force the migrate
648 * to fail by clearing migrate flags
650 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
654 * Update reserved blocks/metadata blocks after successful
655 * block allocation which had been deferred till now. We don't
656 * support fallocate for non extent files. So we can update
657 * reserve space here.
660 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
661 ext4_da_update_reserve_space(inode, retval, 1);
667 if (unlikely(retval != map->m_len)) {
668 ext4_warning(inode->i_sb,
669 "ES len assertion failed for inode "
670 "%lu: retval %d != map->m_len %d",
671 inode->i_ino, retval, map->m_len);
676 * We have to zeroout blocks before inserting them into extent
677 * status tree. Otherwise someone could look them up there and
678 * use them before they are really zeroed. We also have to
679 * unmap metadata before zeroing as otherwise writeback can
680 * overwrite zeros with stale data from block device.
682 if (flags & EXT4_GET_BLOCKS_ZERO &&
683 map->m_flags & EXT4_MAP_MAPPED &&
684 map->m_flags & EXT4_MAP_NEW) {
685 ret = ext4_issue_zeroout(inode, map->m_lblk,
686 map->m_pblk, map->m_len);
694 * If the extent has been zeroed out, we don't need to update
695 * extent status tree.
697 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
698 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
699 if (ext4_es_is_written(&es))
702 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
703 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
704 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
705 !(status & EXTENT_STATUS_WRITTEN) &&
706 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
707 map->m_lblk + map->m_len - 1))
708 status |= EXTENT_STATUS_DELAYED;
709 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
710 map->m_pblk, status);
718 up_write((&EXT4_I(inode)->i_data_sem));
719 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
720 ret = check_block_validity(inode, map);
725 * Inodes with freshly allocated blocks where contents will be
726 * visible after transaction commit must be on transaction's
729 if (map->m_flags & EXT4_MAP_NEW &&
730 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
731 !(flags & EXT4_GET_BLOCKS_ZERO) &&
732 !ext4_is_quota_file(inode) &&
733 ext4_should_order_data(inode)) {
735 (loff_t)map->m_lblk << inode->i_blkbits;
736 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
738 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
739 ret = ext4_jbd2_inode_add_wait(handle, inode,
742 ret = ext4_jbd2_inode_add_write(handle, inode,
752 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
753 * we have to be careful as someone else may be manipulating b_state as well.
755 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
757 unsigned long old_state;
758 unsigned long new_state;
760 flags &= EXT4_MAP_FLAGS;
762 /* Dummy buffer_head? Set non-atomically. */
764 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
768 * Someone else may be modifying b_state. Be careful! This is ugly but
769 * once we get rid of using bh as a container for mapping information
770 * to pass to / from get_block functions, this can go away.
773 old_state = READ_ONCE(bh->b_state);
774 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
776 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
779 static int _ext4_get_block(struct inode *inode, sector_t iblock,
780 struct buffer_head *bh, int flags)
782 struct ext4_map_blocks map;
785 if (ext4_has_inline_data(inode))
789 map.m_len = bh->b_size >> inode->i_blkbits;
791 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
794 map_bh(bh, inode->i_sb, map.m_pblk);
795 ext4_update_bh_state(bh, map.m_flags);
796 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
798 } else if (ret == 0) {
799 /* hole case, need to fill in bh->b_size */
800 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
805 int ext4_get_block(struct inode *inode, sector_t iblock,
806 struct buffer_head *bh, int create)
808 return _ext4_get_block(inode, iblock, bh,
809 create ? EXT4_GET_BLOCKS_CREATE : 0);
813 * Get block function used when preparing for buffered write if we require
814 * creating an unwritten extent if blocks haven't been allocated. The extent
815 * will be converted to written after the IO is complete.
817 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
818 struct buffer_head *bh_result, int create)
820 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
821 inode->i_ino, create);
822 return _ext4_get_block(inode, iblock, bh_result,
823 EXT4_GET_BLOCKS_IO_CREATE_EXT);
826 /* Maximum number of blocks we map for direct IO at once. */
827 #define DIO_MAX_BLOCKS 4096
830 * Get blocks function for the cases that need to start a transaction -
831 * generally difference cases of direct IO and DAX IO. It also handles retries
834 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
835 struct buffer_head *bh_result, int flags)
842 /* Trim mapping request to maximum we can map at once for DIO */
843 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
844 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
845 dio_credits = ext4_chunk_trans_blocks(inode,
846 bh_result->b_size >> inode->i_blkbits);
848 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
850 return PTR_ERR(handle);
852 ret = _ext4_get_block(inode, iblock, bh_result, flags);
853 ext4_journal_stop(handle);
855 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
860 /* Get block function for DIO reads and writes to inodes without extents */
861 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
862 struct buffer_head *bh, int create)
864 /* We don't expect handle for direct IO */
865 WARN_ON_ONCE(ext4_journal_current_handle());
868 return _ext4_get_block(inode, iblock, bh, 0);
869 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
873 * Get block function for AIO DIO writes when we create unwritten extent if
874 * blocks are not allocated yet. The extent will be converted to written
875 * after IO is complete.
877 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
878 sector_t iblock, struct buffer_head *bh_result, int create)
882 /* We don't expect handle for direct IO */
883 WARN_ON_ONCE(ext4_journal_current_handle());
885 ret = ext4_get_block_trans(inode, iblock, bh_result,
886 EXT4_GET_BLOCKS_IO_CREATE_EXT);
889 * When doing DIO using unwritten extents, we need io_end to convert
890 * unwritten extents to written on IO completion. We allocate io_end
891 * once we spot unwritten extent and store it in b_private. Generic
892 * DIO code keeps b_private set and furthermore passes the value to
893 * our completion callback in 'private' argument.
895 if (!ret && buffer_unwritten(bh_result)) {
896 if (!bh_result->b_private) {
897 ext4_io_end_t *io_end;
899 io_end = ext4_init_io_end(inode, GFP_KERNEL);
902 bh_result->b_private = io_end;
903 ext4_set_io_unwritten_flag(inode, io_end);
905 set_buffer_defer_completion(bh_result);
912 * Get block function for non-AIO DIO writes when we create unwritten extent if
913 * blocks are not allocated yet. The extent will be converted to written
914 * after IO is complete by ext4_direct_IO_write().
916 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
917 sector_t iblock, struct buffer_head *bh_result, int create)
921 /* We don't expect handle for direct IO */
922 WARN_ON_ONCE(ext4_journal_current_handle());
924 ret = ext4_get_block_trans(inode, iblock, bh_result,
925 EXT4_GET_BLOCKS_IO_CREATE_EXT);
928 * Mark inode as having pending DIO writes to unwritten extents.
929 * ext4_direct_IO_write() checks this flag and converts extents to
932 if (!ret && buffer_unwritten(bh_result))
933 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
938 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
939 struct buffer_head *bh_result, int create)
943 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
944 inode->i_ino, create);
945 /* We don't expect handle for direct IO */
946 WARN_ON_ONCE(ext4_journal_current_handle());
948 ret = _ext4_get_block(inode, iblock, bh_result, 0);
950 * Blocks should have been preallocated! ext4_file_write_iter() checks
953 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
960 * `handle' can be NULL if create is zero
962 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
963 ext4_lblk_t block, int map_flags)
965 struct ext4_map_blocks map;
966 struct buffer_head *bh;
967 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
970 J_ASSERT(handle != NULL || create == 0);
974 err = ext4_map_blocks(handle, inode, &map, map_flags);
977 return create ? ERR_PTR(-ENOSPC) : NULL;
981 bh = sb_getblk(inode->i_sb, map.m_pblk);
983 return ERR_PTR(-ENOMEM);
984 if (map.m_flags & EXT4_MAP_NEW) {
985 J_ASSERT(create != 0);
986 J_ASSERT(handle != NULL);
989 * Now that we do not always journal data, we should
990 * keep in mind whether this should always journal the
991 * new buffer as metadata. For now, regular file
992 * writes use ext4_get_block instead, so it's not a
996 BUFFER_TRACE(bh, "call get_create_access");
997 err = ext4_journal_get_create_access(handle, bh);
1002 if (!buffer_uptodate(bh)) {
1003 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1004 set_buffer_uptodate(bh);
1007 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1008 err = ext4_handle_dirty_metadata(handle, inode, bh);
1012 BUFFER_TRACE(bh, "not a new buffer");
1016 return ERR_PTR(err);
1019 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1020 ext4_lblk_t block, int map_flags)
1022 struct buffer_head *bh;
1024 bh = ext4_getblk(handle, inode, block, map_flags);
1027 if (!bh || buffer_uptodate(bh))
1029 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1031 if (buffer_uptodate(bh))
1034 return ERR_PTR(-EIO);
1037 /* Read a contiguous batch of blocks. */
1038 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1039 bool wait, struct buffer_head **bhs)
1043 for (i = 0; i < bh_count; i++) {
1044 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1045 if (IS_ERR(bhs[i])) {
1046 err = PTR_ERR(bhs[i]);
1052 for (i = 0; i < bh_count; i++)
1053 /* Note that NULL bhs[i] is valid because of holes. */
1054 if (bhs[i] && !buffer_uptodate(bhs[i]))
1055 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1061 for (i = 0; i < bh_count; i++)
1063 wait_on_buffer(bhs[i]);
1065 for (i = 0; i < bh_count; i++) {
1066 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1074 for (i = 0; i < bh_count; i++) {
1081 int ext4_walk_page_buffers(handle_t *handle,
1082 struct buffer_head *head,
1086 int (*fn)(handle_t *handle,
1087 struct buffer_head *bh))
1089 struct buffer_head *bh;
1090 unsigned block_start, block_end;
1091 unsigned blocksize = head->b_size;
1093 struct buffer_head *next;
1095 for (bh = head, block_start = 0;
1096 ret == 0 && (bh != head || !block_start);
1097 block_start = block_end, bh = next) {
1098 next = bh->b_this_page;
1099 block_end = block_start + blocksize;
1100 if (block_end <= from || block_start >= to) {
1101 if (partial && !buffer_uptodate(bh))
1105 err = (*fn)(handle, bh);
1113 * To preserve ordering, it is essential that the hole instantiation and
1114 * the data write be encapsulated in a single transaction. We cannot
1115 * close off a transaction and start a new one between the ext4_get_block()
1116 * and the commit_write(). So doing the jbd2_journal_start at the start of
1117 * prepare_write() is the right place.
1119 * Also, this function can nest inside ext4_writepage(). In that case, we
1120 * *know* that ext4_writepage() has generated enough buffer credits to do the
1121 * whole page. So we won't block on the journal in that case, which is good,
1122 * because the caller may be PF_MEMALLOC.
1124 * By accident, ext4 can be reentered when a transaction is open via
1125 * quota file writes. If we were to commit the transaction while thus
1126 * reentered, there can be a deadlock - we would be holding a quota
1127 * lock, and the commit would never complete if another thread had a
1128 * transaction open and was blocking on the quota lock - a ranking
1131 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1132 * will _not_ run commit under these circumstances because handle->h_ref
1133 * is elevated. We'll still have enough credits for the tiny quotafile
1136 int do_journal_get_write_access(handle_t *handle,
1137 struct buffer_head *bh)
1139 int dirty = buffer_dirty(bh);
1142 if (!buffer_mapped(bh) || buffer_freed(bh))
1145 * __block_write_begin() could have dirtied some buffers. Clean
1146 * the dirty bit as jbd2_journal_get_write_access() could complain
1147 * otherwise about fs integrity issues. Setting of the dirty bit
1148 * by __block_write_begin() isn't a real problem here as we clear
1149 * the bit before releasing a page lock and thus writeback cannot
1150 * ever write the buffer.
1153 clear_buffer_dirty(bh);
1154 BUFFER_TRACE(bh, "get write access");
1155 ret = ext4_journal_get_write_access(handle, bh);
1157 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1161 #ifdef CONFIG_FS_ENCRYPTION
1162 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1163 get_block_t *get_block)
1165 unsigned from = pos & (PAGE_SIZE - 1);
1166 unsigned to = from + len;
1167 struct inode *inode = page->mapping->host;
1168 unsigned block_start, block_end;
1171 unsigned blocksize = inode->i_sb->s_blocksize;
1173 struct buffer_head *bh, *head, *wait[2];
1177 BUG_ON(!PageLocked(page));
1178 BUG_ON(from > PAGE_SIZE);
1179 BUG_ON(to > PAGE_SIZE);
1182 if (!page_has_buffers(page))
1183 create_empty_buffers(page, blocksize, 0);
1184 head = page_buffers(page);
1185 bbits = ilog2(blocksize);
1186 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1188 for (bh = head, block_start = 0; bh != head || !block_start;
1189 block++, block_start = block_end, bh = bh->b_this_page) {
1190 block_end = block_start + blocksize;
1191 if (block_end <= from || block_start >= to) {
1192 if (PageUptodate(page)) {
1193 if (!buffer_uptodate(bh))
1194 set_buffer_uptodate(bh);
1199 clear_buffer_new(bh);
1200 if (!buffer_mapped(bh)) {
1201 WARN_ON(bh->b_size != blocksize);
1202 err = get_block(inode, block, bh, 1);
1205 if (buffer_new(bh)) {
1206 if (PageUptodate(page)) {
1207 clear_buffer_new(bh);
1208 set_buffer_uptodate(bh);
1209 mark_buffer_dirty(bh);
1212 if (block_end > to || block_start < from)
1213 zero_user_segments(page, to, block_end,
1218 if (PageUptodate(page)) {
1219 if (!buffer_uptodate(bh))
1220 set_buffer_uptodate(bh);
1223 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1224 !buffer_unwritten(bh) &&
1225 (block_start < from || block_end > to)) {
1226 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1227 wait[nr_wait++] = bh;
1231 * If we issued read requests, let them complete.
1233 for (i = 0; i < nr_wait; i++) {
1234 wait_on_buffer(wait[i]);
1235 if (!buffer_uptodate(wait[i]))
1238 if (unlikely(err)) {
1239 page_zero_new_buffers(page, from, to);
1240 } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
1241 for (i = 0; i < nr_wait; i++) {
1244 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1245 bh_offset(wait[i]));
1247 clear_buffer_uptodate(wait[i]);
1257 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1258 loff_t pos, unsigned len, unsigned flags,
1259 struct page **pagep, void **fsdata)
1261 struct inode *inode = mapping->host;
1262 int ret, needed_blocks;
1269 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1272 trace_ext4_write_begin(inode, pos, len, flags);
1274 * Reserve one block more for addition to orphan list in case
1275 * we allocate blocks but write fails for some reason
1277 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1278 index = pos >> PAGE_SHIFT;
1279 from = pos & (PAGE_SIZE - 1);
1282 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1283 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1292 * grab_cache_page_write_begin() can take a long time if the
1293 * system is thrashing due to memory pressure, or if the page
1294 * is being written back. So grab it first before we start
1295 * the transaction handle. This also allows us to allocate
1296 * the page (if needed) without using GFP_NOFS.
1299 page = grab_cache_page_write_begin(mapping, index, flags);
1305 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1306 if (IS_ERR(handle)) {
1308 return PTR_ERR(handle);
1312 if (page->mapping != mapping) {
1313 /* The page got truncated from under us */
1316 ext4_journal_stop(handle);
1319 /* In case writeback began while the page was unlocked */
1320 wait_for_stable_page(page);
1322 #ifdef CONFIG_FS_ENCRYPTION
1323 if (ext4_should_dioread_nolock(inode))
1324 ret = ext4_block_write_begin(page, pos, len,
1325 ext4_get_block_unwritten);
1327 ret = ext4_block_write_begin(page, pos, len,
1330 if (ext4_should_dioread_nolock(inode))
1331 ret = __block_write_begin(page, pos, len,
1332 ext4_get_block_unwritten);
1334 ret = __block_write_begin(page, pos, len, ext4_get_block);
1336 if (!ret && ext4_should_journal_data(inode)) {
1337 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1339 do_journal_get_write_access);
1343 bool extended = (pos + len > inode->i_size) &&
1344 !ext4_verity_in_progress(inode);
1348 * __block_write_begin may have instantiated a few blocks
1349 * outside i_size. Trim these off again. Don't need
1350 * i_size_read because we hold i_mutex.
1352 * Add inode to orphan list in case we crash before
1355 if (extended && ext4_can_truncate(inode))
1356 ext4_orphan_add(handle, inode);
1358 ext4_journal_stop(handle);
1360 ext4_truncate_failed_write(inode);
1362 * If truncate failed early the inode might
1363 * still be on the orphan list; we need to
1364 * make sure the inode is removed from the
1365 * orphan list in that case.
1368 ext4_orphan_del(NULL, inode);
1371 if (ret == -ENOSPC &&
1372 ext4_should_retry_alloc(inode->i_sb, &retries))
1381 /* For write_end() in data=journal mode */
1382 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1385 if (!buffer_mapped(bh) || buffer_freed(bh))
1387 set_buffer_uptodate(bh);
1388 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1389 clear_buffer_meta(bh);
1390 clear_buffer_prio(bh);
1395 * We need to pick up the new inode size which generic_commit_write gave us
1396 * `file' can be NULL - eg, when called from page_symlink().
1398 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1399 * buffers are managed internally.
1401 static int ext4_write_end(struct file *file,
1402 struct address_space *mapping,
1403 loff_t pos, unsigned len, unsigned copied,
1404 struct page *page, void *fsdata)
1406 handle_t *handle = ext4_journal_current_handle();
1407 struct inode *inode = mapping->host;
1408 loff_t old_size = inode->i_size;
1410 int i_size_changed = 0;
1411 int inline_data = ext4_has_inline_data(inode);
1412 bool verity = ext4_verity_in_progress(inode);
1414 trace_ext4_write_end(inode, pos, len, copied);
1416 ret = ext4_write_inline_data_end(inode, pos, len,
1425 copied = block_write_end(file, mapping, pos,
1426 len, copied, page, fsdata);
1428 * it's important to update i_size while still holding page lock:
1429 * page writeout could otherwise come in and zero beyond i_size.
1431 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1432 * blocks are being written past EOF, so skip the i_size update.
1435 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1439 if (old_size < pos && !verity)
1440 pagecache_isize_extended(inode, old_size, pos);
1442 * Don't mark the inode dirty under page lock. First, it unnecessarily
1443 * makes the holding time of page lock longer. Second, it forces lock
1444 * ordering of page lock and transaction start for journaling
1447 if (i_size_changed || inline_data)
1448 ext4_mark_inode_dirty(handle, inode);
1450 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1451 /* if we have allocated more blocks and copied
1452 * less. We will have blocks allocated outside
1453 * inode->i_size. So truncate them
1455 ext4_orphan_add(handle, inode);
1457 ret2 = ext4_journal_stop(handle);
1461 if (pos + len > inode->i_size && !verity) {
1462 ext4_truncate_failed_write(inode);
1464 * If truncate failed early the inode might still be
1465 * on the orphan list; we need to make sure the inode
1466 * is removed from the orphan list in that case.
1469 ext4_orphan_del(NULL, inode);
1472 return ret ? ret : copied;
1476 * This is a private version of page_zero_new_buffers() which doesn't
1477 * set the buffer to be dirty, since in data=journalled mode we need
1478 * to call ext4_handle_dirty_metadata() instead.
1480 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1482 unsigned from, unsigned to)
1484 unsigned int block_start = 0, block_end;
1485 struct buffer_head *head, *bh;
1487 bh = head = page_buffers(page);
1489 block_end = block_start + bh->b_size;
1490 if (buffer_new(bh)) {
1491 if (block_end > from && block_start < to) {
1492 if (!PageUptodate(page)) {
1493 unsigned start, size;
1495 start = max(from, block_start);
1496 size = min(to, block_end) - start;
1498 zero_user(page, start, size);
1499 write_end_fn(handle, bh);
1501 clear_buffer_new(bh);
1504 block_start = block_end;
1505 bh = bh->b_this_page;
1506 } while (bh != head);
1509 static int ext4_journalled_write_end(struct file *file,
1510 struct address_space *mapping,
1511 loff_t pos, unsigned len, unsigned copied,
1512 struct page *page, void *fsdata)
1514 handle_t *handle = ext4_journal_current_handle();
1515 struct inode *inode = mapping->host;
1516 loff_t old_size = inode->i_size;
1520 int size_changed = 0;
1521 int inline_data = ext4_has_inline_data(inode);
1522 bool verity = ext4_verity_in_progress(inode);
1524 trace_ext4_journalled_write_end(inode, pos, len, copied);
1525 from = pos & (PAGE_SIZE - 1);
1528 BUG_ON(!ext4_handle_valid(handle));
1531 ret = ext4_write_inline_data_end(inode, pos, len,
1539 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1541 ext4_journalled_zero_new_buffers(handle, page, from, to);
1543 if (unlikely(copied < len))
1544 ext4_journalled_zero_new_buffers(handle, page,
1546 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1547 from + copied, &partial,
1550 SetPageUptodate(page);
1553 size_changed = ext4_update_inode_size(inode, pos + copied);
1554 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1555 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1559 if (old_size < pos && !verity)
1560 pagecache_isize_extended(inode, old_size, pos);
1562 if (size_changed || inline_data) {
1563 ret2 = ext4_mark_inode_dirty(handle, inode);
1568 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1569 /* if we have allocated more blocks and copied
1570 * less. We will have blocks allocated outside
1571 * inode->i_size. So truncate them
1573 ext4_orphan_add(handle, inode);
1576 ret2 = ext4_journal_stop(handle);
1579 if (pos + len > inode->i_size && !verity) {
1580 ext4_truncate_failed_write(inode);
1582 * If truncate failed early the inode might still be
1583 * on the orphan list; we need to make sure the inode
1584 * is removed from the orphan list in that case.
1587 ext4_orphan_del(NULL, inode);
1590 return ret ? ret : copied;
1594 * Reserve space for a single cluster
1596 static int ext4_da_reserve_space(struct inode *inode)
1598 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1599 struct ext4_inode_info *ei = EXT4_I(inode);
1603 * We will charge metadata quota at writeout time; this saves
1604 * us from metadata over-estimation, though we may go over by
1605 * a small amount in the end. Here we just reserve for data.
1607 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1611 spin_lock(&ei->i_block_reservation_lock);
1612 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1613 spin_unlock(&ei->i_block_reservation_lock);
1614 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1617 ei->i_reserved_data_blocks++;
1618 trace_ext4_da_reserve_space(inode);
1619 spin_unlock(&ei->i_block_reservation_lock);
1621 return 0; /* success */
1624 void ext4_da_release_space(struct inode *inode, int to_free)
1626 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1627 struct ext4_inode_info *ei = EXT4_I(inode);
1630 return; /* Nothing to release, exit */
1632 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1634 trace_ext4_da_release_space(inode, to_free);
1635 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1637 * if there aren't enough reserved blocks, then the
1638 * counter is messed up somewhere. Since this
1639 * function is called from invalidate page, it's
1640 * harmless to return without any action.
1642 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1643 "ino %lu, to_free %d with only %d reserved "
1644 "data blocks", inode->i_ino, to_free,
1645 ei->i_reserved_data_blocks);
1647 to_free = ei->i_reserved_data_blocks;
1649 ei->i_reserved_data_blocks -= to_free;
1651 /* update fs dirty data blocks counter */
1652 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1654 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1656 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1659 static void ext4_da_page_release_reservation(struct page *page,
1660 unsigned int offset,
1661 unsigned int length)
1663 int contiguous_blks = 0;
1664 struct buffer_head *head, *bh;
1665 unsigned int curr_off = 0;
1666 struct inode *inode = page->mapping->host;
1667 unsigned int stop = offset + length;
1670 BUG_ON(stop > PAGE_SIZE || stop < length);
1672 head = page_buffers(page);
1675 unsigned int next_off = curr_off + bh->b_size;
1677 if (next_off > stop)
1680 if ((offset <= curr_off) && (buffer_delay(bh))) {
1682 clear_buffer_delay(bh);
1683 } else if (contiguous_blks) {
1684 lblk = page->index <<
1685 (PAGE_SHIFT - inode->i_blkbits);
1686 lblk += (curr_off >> inode->i_blkbits) -
1688 ext4_es_remove_blks(inode, lblk, contiguous_blks);
1689 contiguous_blks = 0;
1691 curr_off = next_off;
1692 } while ((bh = bh->b_this_page) != head);
1694 if (contiguous_blks) {
1695 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1696 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1697 ext4_es_remove_blks(inode, lblk, contiguous_blks);
1703 * Delayed allocation stuff
1706 struct mpage_da_data {
1707 struct inode *inode;
1708 struct writeback_control *wbc;
1710 pgoff_t first_page; /* The first page to write */
1711 pgoff_t next_page; /* Current page to examine */
1712 pgoff_t last_page; /* Last page to examine */
1714 * Extent to map - this can be after first_page because that can be
1715 * fully mapped. We somewhat abuse m_flags to store whether the extent
1716 * is delalloc or unwritten.
1718 struct ext4_map_blocks map;
1719 struct ext4_io_submit io_submit; /* IO submission data */
1720 unsigned int do_map:1;
1723 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1728 struct pagevec pvec;
1729 struct inode *inode = mpd->inode;
1730 struct address_space *mapping = inode->i_mapping;
1732 /* This is necessary when next_page == 0. */
1733 if (mpd->first_page >= mpd->next_page)
1736 index = mpd->first_page;
1737 end = mpd->next_page - 1;
1739 ext4_lblk_t start, last;
1740 start = index << (PAGE_SHIFT - inode->i_blkbits);
1741 last = end << (PAGE_SHIFT - inode->i_blkbits);
1742 ext4_es_remove_extent(inode, start, last - start + 1);
1745 pagevec_init(&pvec);
1746 while (index <= end) {
1747 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1750 for (i = 0; i < nr_pages; i++) {
1751 struct page *page = pvec.pages[i];
1753 BUG_ON(!PageLocked(page));
1754 BUG_ON(PageWriteback(page));
1756 if (page_mapped(page))
1757 clear_page_dirty_for_io(page);
1758 block_invalidatepage(page, 0, PAGE_SIZE);
1759 ClearPageUptodate(page);
1763 pagevec_release(&pvec);
1767 static void ext4_print_free_blocks(struct inode *inode)
1769 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1770 struct super_block *sb = inode->i_sb;
1771 struct ext4_inode_info *ei = EXT4_I(inode);
1773 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1774 EXT4_C2B(EXT4_SB(inode->i_sb),
1775 ext4_count_free_clusters(sb)));
1776 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1777 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1778 (long long) EXT4_C2B(EXT4_SB(sb),
1779 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1780 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1781 (long long) EXT4_C2B(EXT4_SB(sb),
1782 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1783 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1784 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1785 ei->i_reserved_data_blocks);
1789 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1791 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1795 * ext4_insert_delayed_block - adds a delayed block to the extents status
1796 * tree, incrementing the reserved cluster/block
1797 * count or making a pending reservation
1800 * @inode - file containing the newly added block
1801 * @lblk - logical block to be added
1803 * Returns 0 on success, negative error code on failure.
1805 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1807 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1809 bool allocated = false;
1812 * If the cluster containing lblk is shared with a delayed,
1813 * written, or unwritten extent in a bigalloc file system, it's
1814 * already been accounted for and does not need to be reserved.
1815 * A pending reservation must be made for the cluster if it's
1816 * shared with a written or unwritten extent and doesn't already
1817 * have one. Written and unwritten extents can be purged from the
1818 * extents status tree if the system is under memory pressure, so
1819 * it's necessary to examine the extent tree if a search of the
1820 * extents status tree doesn't get a match.
1822 if (sbi->s_cluster_ratio == 1) {
1823 ret = ext4_da_reserve_space(inode);
1824 if (ret != 0) /* ENOSPC */
1826 } else { /* bigalloc */
1827 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1828 if (!ext4_es_scan_clu(inode,
1829 &ext4_es_is_mapped, lblk)) {
1830 ret = ext4_clu_mapped(inode,
1831 EXT4_B2C(sbi, lblk));
1835 ret = ext4_da_reserve_space(inode);
1836 if (ret != 0) /* ENOSPC */
1847 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1854 * This function is grabs code from the very beginning of
1855 * ext4_map_blocks, but assumes that the caller is from delayed write
1856 * time. This function looks up the requested blocks and sets the
1857 * buffer delay bit under the protection of i_data_sem.
1859 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1860 struct ext4_map_blocks *map,
1861 struct buffer_head *bh)
1863 struct extent_status es;
1865 sector_t invalid_block = ~((sector_t) 0xffff);
1866 #ifdef ES_AGGRESSIVE_TEST
1867 struct ext4_map_blocks orig_map;
1869 memcpy(&orig_map, map, sizeof(*map));
1872 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1876 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1877 "logical block %lu\n", inode->i_ino, map->m_len,
1878 (unsigned long) map->m_lblk);
1880 /* Lookup extent status tree firstly */
1881 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1882 if (ext4_es_is_hole(&es)) {
1884 down_read(&EXT4_I(inode)->i_data_sem);
1889 * Delayed extent could be allocated by fallocate.
1890 * So we need to check it.
1892 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1893 map_bh(bh, inode->i_sb, invalid_block);
1895 set_buffer_delay(bh);
1899 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1900 retval = es.es_len - (iblock - es.es_lblk);
1901 if (retval > map->m_len)
1902 retval = map->m_len;
1903 map->m_len = retval;
1904 if (ext4_es_is_written(&es))
1905 map->m_flags |= EXT4_MAP_MAPPED;
1906 else if (ext4_es_is_unwritten(&es))
1907 map->m_flags |= EXT4_MAP_UNWRITTEN;
1911 #ifdef ES_AGGRESSIVE_TEST
1912 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1918 * Try to see if we can get the block without requesting a new
1919 * file system block.
1921 down_read(&EXT4_I(inode)->i_data_sem);
1922 if (ext4_has_inline_data(inode))
1924 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1925 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1927 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1934 * XXX: __block_prepare_write() unmaps passed block,
1938 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1944 map_bh(bh, inode->i_sb, invalid_block);
1946 set_buffer_delay(bh);
1947 } else if (retval > 0) {
1949 unsigned int status;
1951 if (unlikely(retval != map->m_len)) {
1952 ext4_warning(inode->i_sb,
1953 "ES len assertion failed for inode "
1954 "%lu: retval %d != map->m_len %d",
1955 inode->i_ino, retval, map->m_len);
1959 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1960 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1961 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1962 map->m_pblk, status);
1968 up_read((&EXT4_I(inode)->i_data_sem));
1974 * This is a special get_block_t callback which is used by
1975 * ext4_da_write_begin(). It will either return mapped block or
1976 * reserve space for a single block.
1978 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1979 * We also have b_blocknr = -1 and b_bdev initialized properly
1981 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1982 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1983 * initialized properly.
1985 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1986 struct buffer_head *bh, int create)
1988 struct ext4_map_blocks map;
1991 BUG_ON(create == 0);
1992 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1994 map.m_lblk = iblock;
1998 * first, we need to know whether the block is allocated already
1999 * preallocated blocks are unmapped but should treated
2000 * the same as allocated blocks.
2002 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2006 map_bh(bh, inode->i_sb, map.m_pblk);
2007 ext4_update_bh_state(bh, map.m_flags);
2009 if (buffer_unwritten(bh)) {
2010 /* A delayed write to unwritten bh should be marked
2011 * new and mapped. Mapped ensures that we don't do
2012 * get_block multiple times when we write to the same
2013 * offset and new ensures that we do proper zero out
2014 * for partial write.
2017 set_buffer_mapped(bh);
2022 static int bget_one(handle_t *handle, struct buffer_head *bh)
2028 static int bput_one(handle_t *handle, struct buffer_head *bh)
2034 static int __ext4_journalled_writepage(struct page *page,
2037 struct address_space *mapping = page->mapping;
2038 struct inode *inode = mapping->host;
2039 struct buffer_head *page_bufs = NULL;
2040 handle_t *handle = NULL;
2041 int ret = 0, err = 0;
2042 int inline_data = ext4_has_inline_data(inode);
2043 struct buffer_head *inode_bh = NULL;
2045 ClearPageChecked(page);
2048 BUG_ON(page->index != 0);
2049 BUG_ON(len > ext4_get_max_inline_size(inode));
2050 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2051 if (inode_bh == NULL)
2054 page_bufs = page_buffers(page);
2059 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2063 * We need to release the page lock before we start the
2064 * journal, so grab a reference so the page won't disappear
2065 * out from under us.
2070 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2071 ext4_writepage_trans_blocks(inode));
2072 if (IS_ERR(handle)) {
2073 ret = PTR_ERR(handle);
2075 goto out_no_pagelock;
2077 BUG_ON(!ext4_handle_valid(handle));
2081 if (page->mapping != mapping) {
2082 /* The page got truncated from under us */
2083 ext4_journal_stop(handle);
2089 ret = ext4_mark_inode_dirty(handle, inode);
2091 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2092 do_journal_get_write_access);
2094 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2099 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2100 err = ext4_journal_stop(handle);
2104 if (!ext4_has_inline_data(inode))
2105 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2107 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2116 * Note that we don't need to start a transaction unless we're journaling data
2117 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2118 * need to file the inode to the transaction's list in ordered mode because if
2119 * we are writing back data added by write(), the inode is already there and if
2120 * we are writing back data modified via mmap(), no one guarantees in which
2121 * transaction the data will hit the disk. In case we are journaling data, we
2122 * cannot start transaction directly because transaction start ranks above page
2123 * lock so we have to do some magic.
2125 * This function can get called via...
2126 * - ext4_writepages after taking page lock (have journal handle)
2127 * - journal_submit_inode_data_buffers (no journal handle)
2128 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2129 * - grab_page_cache when doing write_begin (have journal handle)
2131 * We don't do any block allocation in this function. If we have page with
2132 * multiple blocks we need to write those buffer_heads that are mapped. This
2133 * is important for mmaped based write. So if we do with blocksize 1K
2134 * truncate(f, 1024);
2135 * a = mmap(f, 0, 4096);
2137 * truncate(f, 4096);
2138 * we have in the page first buffer_head mapped via page_mkwrite call back
2139 * but other buffer_heads would be unmapped but dirty (dirty done via the
2140 * do_wp_page). So writepage should write the first block. If we modify
2141 * the mmap area beyond 1024 we will again get a page_fault and the
2142 * page_mkwrite callback will do the block allocation and mark the
2143 * buffer_heads mapped.
2145 * We redirty the page if we have any buffer_heads that is either delay or
2146 * unwritten in the page.
2148 * We can get recursively called as show below.
2150 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2153 * But since we don't do any block allocation we should not deadlock.
2154 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2156 static int ext4_writepage(struct page *page,
2157 struct writeback_control *wbc)
2162 struct buffer_head *page_bufs = NULL;
2163 struct inode *inode = page->mapping->host;
2164 struct ext4_io_submit io_submit;
2165 bool keep_towrite = false;
2167 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2168 ext4_invalidatepage(page, 0, PAGE_SIZE);
2173 trace_ext4_writepage(page);
2174 size = i_size_read(inode);
2175 if (page->index == size >> PAGE_SHIFT &&
2176 !ext4_verity_in_progress(inode))
2177 len = size & ~PAGE_MASK;
2181 page_bufs = page_buffers(page);
2183 * We cannot do block allocation or other extent handling in this
2184 * function. If there are buffers needing that, we have to redirty
2185 * the page. But we may reach here when we do a journal commit via
2186 * journal_submit_inode_data_buffers() and in that case we must write
2187 * allocated buffers to achieve data=ordered mode guarantees.
2189 * Also, if there is only one buffer per page (the fs block
2190 * size == the page size), if one buffer needs block
2191 * allocation or needs to modify the extent tree to clear the
2192 * unwritten flag, we know that the page can't be written at
2193 * all, so we might as well refuse the write immediately.
2194 * Unfortunately if the block size != page size, we can't as
2195 * easily detect this case using ext4_walk_page_buffers(), but
2196 * for the extremely common case, this is an optimization that
2197 * skips a useless round trip through ext4_bio_write_page().
2199 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2200 ext4_bh_delay_or_unwritten)) {
2201 redirty_page_for_writepage(wbc, page);
2202 if ((current->flags & PF_MEMALLOC) ||
2203 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2205 * For memory cleaning there's no point in writing only
2206 * some buffers. So just bail out. Warn if we came here
2207 * from direct reclaim.
2209 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2214 keep_towrite = true;
2217 if (PageChecked(page) && ext4_should_journal_data(inode))
2219 * It's mmapped pagecache. Add buffers and journal it. There
2220 * doesn't seem much point in redirtying the page here.
2222 return __ext4_journalled_writepage(page, len);
2224 ext4_io_submit_init(&io_submit, wbc);
2225 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2226 if (!io_submit.io_end) {
2227 redirty_page_for_writepage(wbc, page);
2231 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2232 ext4_io_submit(&io_submit);
2233 /* Drop io_end reference we got from init */
2234 ext4_put_io_end_defer(io_submit.io_end);
2238 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2244 BUG_ON(page->index != mpd->first_page);
2245 clear_page_dirty_for_io(page);
2247 * We have to be very careful here! Nothing protects writeback path
2248 * against i_size changes and the page can be writeably mapped into
2249 * page tables. So an application can be growing i_size and writing
2250 * data through mmap while writeback runs. clear_page_dirty_for_io()
2251 * write-protects our page in page tables and the page cannot get
2252 * written to again until we release page lock. So only after
2253 * clear_page_dirty_for_io() we are safe to sample i_size for
2254 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2255 * on the barrier provided by TestClearPageDirty in
2256 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2257 * after page tables are updated.
2259 size = i_size_read(mpd->inode);
2260 if (page->index == size >> PAGE_SHIFT &&
2261 !ext4_verity_in_progress(mpd->inode))
2262 len = size & ~PAGE_MASK;
2265 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2267 mpd->wbc->nr_to_write--;
2273 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2276 * mballoc gives us at most this number of blocks...
2277 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2278 * The rest of mballoc seems to handle chunks up to full group size.
2280 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2283 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2285 * @mpd - extent of blocks
2286 * @lblk - logical number of the block in the file
2287 * @bh - buffer head we want to add to the extent
2289 * The function is used to collect contig. blocks in the same state. If the
2290 * buffer doesn't require mapping for writeback and we haven't started the
2291 * extent of buffers to map yet, the function returns 'true' immediately - the
2292 * caller can write the buffer right away. Otherwise the function returns true
2293 * if the block has been added to the extent, false if the block couldn't be
2296 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2297 struct buffer_head *bh)
2299 struct ext4_map_blocks *map = &mpd->map;
2301 /* Buffer that doesn't need mapping for writeback? */
2302 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2303 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2304 /* So far no extent to map => we write the buffer right away */
2305 if (map->m_len == 0)
2310 /* First block in the extent? */
2311 if (map->m_len == 0) {
2312 /* We cannot map unless handle is started... */
2317 map->m_flags = bh->b_state & BH_FLAGS;
2321 /* Don't go larger than mballoc is willing to allocate */
2322 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2325 /* Can we merge the block to our big extent? */
2326 if (lblk == map->m_lblk + map->m_len &&
2327 (bh->b_state & BH_FLAGS) == map->m_flags) {
2335 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2337 * @mpd - extent of blocks for mapping
2338 * @head - the first buffer in the page
2339 * @bh - buffer we should start processing from
2340 * @lblk - logical number of the block in the file corresponding to @bh
2342 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2343 * the page for IO if all buffers in this page were mapped and there's no
2344 * accumulated extent of buffers to map or add buffers in the page to the
2345 * extent of buffers to map. The function returns 1 if the caller can continue
2346 * by processing the next page, 0 if it should stop adding buffers to the
2347 * extent to map because we cannot extend it anymore. It can also return value
2348 * < 0 in case of error during IO submission.
2350 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2351 struct buffer_head *head,
2352 struct buffer_head *bh,
2355 struct inode *inode = mpd->inode;
2357 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2358 >> inode->i_blkbits;
2360 if (ext4_verity_in_progress(inode))
2361 blocks = EXT_MAX_BLOCKS;
2364 BUG_ON(buffer_locked(bh));
2366 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2367 /* Found extent to map? */
2370 /* Buffer needs mapping and handle is not started? */
2373 /* Everything mapped so far and we hit EOF */
2376 } while (lblk++, (bh = bh->b_this_page) != head);
2377 /* So far everything mapped? Submit the page for IO. */
2378 if (mpd->map.m_len == 0) {
2379 err = mpage_submit_page(mpd, head->b_page);
2383 return lblk < blocks;
2387 * mpage_map_buffers - update buffers corresponding to changed extent and
2388 * submit fully mapped pages for IO
2390 * @mpd - description of extent to map, on return next extent to map
2392 * Scan buffers corresponding to changed extent (we expect corresponding pages
2393 * to be already locked) and update buffer state according to new extent state.
2394 * We map delalloc buffers to their physical location, clear unwritten bits,
2395 * and mark buffers as uninit when we perform writes to unwritten extents
2396 * and do extent conversion after IO is finished. If the last page is not fully
2397 * mapped, we update @map to the next extent in the last page that needs
2398 * mapping. Otherwise we submit the page for IO.
2400 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2402 struct pagevec pvec;
2404 struct inode *inode = mpd->inode;
2405 struct buffer_head *head, *bh;
2406 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2412 start = mpd->map.m_lblk >> bpp_bits;
2413 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2414 lblk = start << bpp_bits;
2415 pblock = mpd->map.m_pblk;
2417 pagevec_init(&pvec);
2418 while (start <= end) {
2419 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2423 for (i = 0; i < nr_pages; i++) {
2424 struct page *page = pvec.pages[i];
2426 bh = head = page_buffers(page);
2428 if (lblk < mpd->map.m_lblk)
2430 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2432 * Buffer after end of mapped extent.
2433 * Find next buffer in the page to map.
2436 mpd->map.m_flags = 0;
2438 * FIXME: If dioread_nolock supports
2439 * blocksize < pagesize, we need to make
2440 * sure we add size mapped so far to
2441 * io_end->size as the following call
2442 * can submit the page for IO.
2444 err = mpage_process_page_bufs(mpd, head,
2446 pagevec_release(&pvec);
2451 if (buffer_delay(bh)) {
2452 clear_buffer_delay(bh);
2453 bh->b_blocknr = pblock++;
2455 clear_buffer_unwritten(bh);
2456 } while (lblk++, (bh = bh->b_this_page) != head);
2459 * FIXME: This is going to break if dioread_nolock
2460 * supports blocksize < pagesize as we will try to
2461 * convert potentially unmapped parts of inode.
2463 mpd->io_submit.io_end->size += PAGE_SIZE;
2464 /* Page fully mapped - let IO run! */
2465 err = mpage_submit_page(mpd, page);
2467 pagevec_release(&pvec);
2471 pagevec_release(&pvec);
2473 /* Extent fully mapped and matches with page boundary. We are done. */
2475 mpd->map.m_flags = 0;
2479 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2481 struct inode *inode = mpd->inode;
2482 struct ext4_map_blocks *map = &mpd->map;
2483 int get_blocks_flags;
2484 int err, dioread_nolock;
2486 trace_ext4_da_write_pages_extent(inode, map);
2488 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2489 * to convert an unwritten extent to be initialized (in the case
2490 * where we have written into one or more preallocated blocks). It is
2491 * possible that we're going to need more metadata blocks than
2492 * previously reserved. However we must not fail because we're in
2493 * writeback and there is nothing we can do about it so it might result
2494 * in data loss. So use reserved blocks to allocate metadata if
2497 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2498 * the blocks in question are delalloc blocks. This indicates
2499 * that the blocks and quotas has already been checked when
2500 * the data was copied into the page cache.
2502 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2503 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2504 EXT4_GET_BLOCKS_IO_SUBMIT;
2505 dioread_nolock = ext4_should_dioread_nolock(inode);
2507 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2508 if (map->m_flags & (1 << BH_Delay))
2509 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2511 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2514 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2515 if (!mpd->io_submit.io_end->handle &&
2516 ext4_handle_valid(handle)) {
2517 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2518 handle->h_rsv_handle = NULL;
2520 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2523 BUG_ON(map->m_len == 0);
2528 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2529 * mpd->len and submit pages underlying it for IO
2531 * @handle - handle for journal operations
2532 * @mpd - extent to map
2533 * @give_up_on_write - we set this to true iff there is a fatal error and there
2534 * is no hope of writing the data. The caller should discard
2535 * dirty pages to avoid infinite loops.
2537 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2538 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2539 * them to initialized or split the described range from larger unwritten
2540 * extent. Note that we need not map all the described range since allocation
2541 * can return less blocks or the range is covered by more unwritten extents. We
2542 * cannot map more because we are limited by reserved transaction credits. On
2543 * the other hand we always make sure that the last touched page is fully
2544 * mapped so that it can be written out (and thus forward progress is
2545 * guaranteed). After mapping we submit all mapped pages for IO.
2547 static int mpage_map_and_submit_extent(handle_t *handle,
2548 struct mpage_da_data *mpd,
2549 bool *give_up_on_write)
2551 struct inode *inode = mpd->inode;
2552 struct ext4_map_blocks *map = &mpd->map;
2557 mpd->io_submit.io_end->offset =
2558 ((loff_t)map->m_lblk) << inode->i_blkbits;
2560 err = mpage_map_one_extent(handle, mpd);
2562 struct super_block *sb = inode->i_sb;
2564 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2565 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2566 goto invalidate_dirty_pages;
2568 * Let the uper layers retry transient errors.
2569 * In the case of ENOSPC, if ext4_count_free_blocks()
2570 * is non-zero, a commit should free up blocks.
2572 if ((err == -ENOMEM) ||
2573 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2575 goto update_disksize;
2578 ext4_msg(sb, KERN_CRIT,
2579 "Delayed block allocation failed for "
2580 "inode %lu at logical offset %llu with"
2581 " max blocks %u with error %d",
2583 (unsigned long long)map->m_lblk,
2584 (unsigned)map->m_len, -err);
2585 ext4_msg(sb, KERN_CRIT,
2586 "This should not happen!! Data will "
2589 ext4_print_free_blocks(inode);
2590 invalidate_dirty_pages:
2591 *give_up_on_write = true;
2596 * Update buffer state, submit mapped pages, and get us new
2599 err = mpage_map_and_submit_buffers(mpd);
2601 goto update_disksize;
2602 } while (map->m_len);
2606 * Update on-disk size after IO is submitted. Races with
2607 * truncate are avoided by checking i_size under i_data_sem.
2609 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2610 if (disksize > EXT4_I(inode)->i_disksize) {
2614 down_write(&EXT4_I(inode)->i_data_sem);
2615 i_size = i_size_read(inode);
2616 if (disksize > i_size)
2618 if (disksize > EXT4_I(inode)->i_disksize)
2619 EXT4_I(inode)->i_disksize = disksize;
2620 up_write(&EXT4_I(inode)->i_data_sem);
2621 err2 = ext4_mark_inode_dirty(handle, inode);
2623 ext4_error(inode->i_sb,
2624 "Failed to mark inode %lu dirty",
2633 * Calculate the total number of credits to reserve for one writepages
2634 * iteration. This is called from ext4_writepages(). We map an extent of
2635 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2636 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2637 * bpp - 1 blocks in bpp different extents.
2639 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2641 int bpp = ext4_journal_blocks_per_page(inode);
2643 return ext4_meta_trans_blocks(inode,
2644 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2648 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2649 * and underlying extent to map
2651 * @mpd - where to look for pages
2653 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2654 * IO immediately. When we find a page which isn't mapped we start accumulating
2655 * extent of buffers underlying these pages that needs mapping (formed by
2656 * either delayed or unwritten buffers). We also lock the pages containing
2657 * these buffers. The extent found is returned in @mpd structure (starting at
2658 * mpd->lblk with length mpd->len blocks).
2660 * Note that this function can attach bios to one io_end structure which are
2661 * neither logically nor physically contiguous. Although it may seem as an
2662 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2663 * case as we need to track IO to all buffers underlying a page in one io_end.
2665 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2667 struct address_space *mapping = mpd->inode->i_mapping;
2668 struct pagevec pvec;
2669 unsigned int nr_pages;
2670 long left = mpd->wbc->nr_to_write;
2671 pgoff_t index = mpd->first_page;
2672 pgoff_t end = mpd->last_page;
2675 int blkbits = mpd->inode->i_blkbits;
2677 struct buffer_head *head;
2679 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2680 tag = PAGECACHE_TAG_TOWRITE;
2682 tag = PAGECACHE_TAG_DIRTY;
2684 pagevec_init(&pvec);
2686 mpd->next_page = index;
2687 while (index <= end) {
2688 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2693 for (i = 0; i < nr_pages; i++) {
2694 struct page *page = pvec.pages[i];
2697 * Accumulated enough dirty pages? This doesn't apply
2698 * to WB_SYNC_ALL mode. For integrity sync we have to
2699 * keep going because someone may be concurrently
2700 * dirtying pages, and we might have synced a lot of
2701 * newly appeared dirty pages, but have not synced all
2702 * of the old dirty pages.
2704 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2707 /* If we can't merge this page, we are done. */
2708 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2713 * If the page is no longer dirty, or its mapping no
2714 * longer corresponds to inode we are writing (which
2715 * means it has been truncated or invalidated), or the
2716 * page is already under writeback and we are not doing
2717 * a data integrity writeback, skip the page
2719 if (!PageDirty(page) ||
2720 (PageWriteback(page) &&
2721 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2722 unlikely(page->mapping != mapping)) {
2727 wait_on_page_writeback(page);
2728 BUG_ON(PageWriteback(page));
2730 if (mpd->map.m_len == 0)
2731 mpd->first_page = page->index;
2732 mpd->next_page = page->index + 1;
2733 /* Add all dirty buffers to mpd */
2734 lblk = ((ext4_lblk_t)page->index) <<
2735 (PAGE_SHIFT - blkbits);
2736 head = page_buffers(page);
2737 err = mpage_process_page_bufs(mpd, head, head, lblk);
2743 pagevec_release(&pvec);
2748 pagevec_release(&pvec);
2752 static int ext4_writepages(struct address_space *mapping,
2753 struct writeback_control *wbc)
2755 pgoff_t writeback_index = 0;
2756 long nr_to_write = wbc->nr_to_write;
2757 int range_whole = 0;
2759 handle_t *handle = NULL;
2760 struct mpage_da_data mpd;
2761 struct inode *inode = mapping->host;
2762 int needed_blocks, rsv_blocks = 0, ret = 0;
2763 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2765 struct blk_plug plug;
2766 bool give_up_on_write = false;
2768 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2771 percpu_down_read(&sbi->s_journal_flag_rwsem);
2772 trace_ext4_writepages(inode, wbc);
2775 * No pages to write? This is mainly a kludge to avoid starting
2776 * a transaction for special inodes like journal inode on last iput()
2777 * because that could violate lock ordering on umount
2779 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2780 goto out_writepages;
2782 if (ext4_should_journal_data(inode)) {
2783 ret = generic_writepages(mapping, wbc);
2784 goto out_writepages;
2788 * If the filesystem has aborted, it is read-only, so return
2789 * right away instead of dumping stack traces later on that
2790 * will obscure the real source of the problem. We test
2791 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2792 * the latter could be true if the filesystem is mounted
2793 * read-only, and in that case, ext4_writepages should
2794 * *never* be called, so if that ever happens, we would want
2797 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2798 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2800 goto out_writepages;
2803 if (ext4_should_dioread_nolock(inode)) {
2805 * We may need to convert up to one extent per block in
2806 * the page and we may dirty the inode.
2808 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2809 PAGE_SIZE >> inode->i_blkbits);
2813 * If we have inline data and arrive here, it means that
2814 * we will soon create the block for the 1st page, so
2815 * we'd better clear the inline data here.
2817 if (ext4_has_inline_data(inode)) {
2818 /* Just inode will be modified... */
2819 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2820 if (IS_ERR(handle)) {
2821 ret = PTR_ERR(handle);
2822 goto out_writepages;
2824 BUG_ON(ext4_test_inode_state(inode,
2825 EXT4_STATE_MAY_INLINE_DATA));
2826 ext4_destroy_inline_data(handle, inode);
2827 ext4_journal_stop(handle);
2830 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2833 if (wbc->range_cyclic) {
2834 writeback_index = mapping->writeback_index;
2835 if (writeback_index)
2837 mpd.first_page = writeback_index;
2840 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2841 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2846 ext4_io_submit_init(&mpd.io_submit, wbc);
2848 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2849 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2851 blk_start_plug(&plug);
2854 * First writeback pages that don't need mapping - we can avoid
2855 * starting a transaction unnecessarily and also avoid being blocked
2856 * in the block layer on device congestion while having transaction
2860 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2861 if (!mpd.io_submit.io_end) {
2865 ret = mpage_prepare_extent_to_map(&mpd);
2866 /* Unlock pages we didn't use */
2867 mpage_release_unused_pages(&mpd, false);
2868 /* Submit prepared bio */
2869 ext4_io_submit(&mpd.io_submit);
2870 ext4_put_io_end_defer(mpd.io_submit.io_end);
2871 mpd.io_submit.io_end = NULL;
2875 while (!done && mpd.first_page <= mpd.last_page) {
2876 /* For each extent of pages we use new io_end */
2877 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2878 if (!mpd.io_submit.io_end) {
2884 * We have two constraints: We find one extent to map and we
2885 * must always write out whole page (makes a difference when
2886 * blocksize < pagesize) so that we don't block on IO when we
2887 * try to write out the rest of the page. Journalled mode is
2888 * not supported by delalloc.
2890 BUG_ON(ext4_should_journal_data(inode));
2891 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2893 /* start a new transaction */
2894 handle = ext4_journal_start_with_reserve(inode,
2895 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2896 if (IS_ERR(handle)) {
2897 ret = PTR_ERR(handle);
2898 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2899 "%ld pages, ino %lu; err %d", __func__,
2900 wbc->nr_to_write, inode->i_ino, ret);
2901 /* Release allocated io_end */
2902 ext4_put_io_end(mpd.io_submit.io_end);
2903 mpd.io_submit.io_end = NULL;
2908 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2909 ret = mpage_prepare_extent_to_map(&mpd);
2912 ret = mpage_map_and_submit_extent(handle, &mpd,
2916 * We scanned the whole range (or exhausted
2917 * nr_to_write), submitted what was mapped and
2918 * didn't find anything needing mapping. We are
2925 * Caution: If the handle is synchronous,
2926 * ext4_journal_stop() can wait for transaction commit
2927 * to finish which may depend on writeback of pages to
2928 * complete or on page lock to be released. In that
2929 * case, we have to wait until after after we have
2930 * submitted all the IO, released page locks we hold,
2931 * and dropped io_end reference (for extent conversion
2932 * to be able to complete) before stopping the handle.
2934 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2935 ext4_journal_stop(handle);
2939 /* Unlock pages we didn't use */
2940 mpage_release_unused_pages(&mpd, give_up_on_write);
2941 /* Submit prepared bio */
2942 ext4_io_submit(&mpd.io_submit);
2945 * Drop our io_end reference we got from init. We have
2946 * to be careful and use deferred io_end finishing if
2947 * we are still holding the transaction as we can
2948 * release the last reference to io_end which may end
2949 * up doing unwritten extent conversion.
2952 ext4_put_io_end_defer(mpd.io_submit.io_end);
2953 ext4_journal_stop(handle);
2955 ext4_put_io_end(mpd.io_submit.io_end);
2956 mpd.io_submit.io_end = NULL;
2958 if (ret == -ENOSPC && sbi->s_journal) {
2960 * Commit the transaction which would
2961 * free blocks released in the transaction
2964 jbd2_journal_force_commit_nested(sbi->s_journal);
2968 /* Fatal error - ENOMEM, EIO... */
2973 blk_finish_plug(&plug);
2974 if (!ret && !cycled && wbc->nr_to_write > 0) {
2976 mpd.last_page = writeback_index - 1;
2982 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2984 * Set the writeback_index so that range_cyclic
2985 * mode will write it back later
2987 mapping->writeback_index = mpd.first_page;
2990 trace_ext4_writepages_result(inode, wbc, ret,
2991 nr_to_write - wbc->nr_to_write);
2992 percpu_up_read(&sbi->s_journal_flag_rwsem);
2996 static int ext4_dax_writepages(struct address_space *mapping,
2997 struct writeback_control *wbc)
3000 long nr_to_write = wbc->nr_to_write;
3001 struct inode *inode = mapping->host;
3002 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
3004 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3007 percpu_down_read(&sbi->s_journal_flag_rwsem);
3008 trace_ext4_writepages(inode, wbc);
3010 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
3011 trace_ext4_writepages_result(inode, wbc, ret,
3012 nr_to_write - wbc->nr_to_write);
3013 percpu_up_read(&sbi->s_journal_flag_rwsem);
3017 static int ext4_nonda_switch(struct super_block *sb)
3019 s64 free_clusters, dirty_clusters;
3020 struct ext4_sb_info *sbi = EXT4_SB(sb);
3023 * switch to non delalloc mode if we are running low
3024 * on free block. The free block accounting via percpu
3025 * counters can get slightly wrong with percpu_counter_batch getting
3026 * accumulated on each CPU without updating global counters
3027 * Delalloc need an accurate free block accounting. So switch
3028 * to non delalloc when we are near to error range.
3031 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3033 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3035 * Start pushing delalloc when 1/2 of free blocks are dirty.
3037 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3038 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3040 if (2 * free_clusters < 3 * dirty_clusters ||
3041 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3043 * free block count is less than 150% of dirty blocks
3044 * or free blocks is less than watermark
3051 /* We always reserve for an inode update; the superblock could be there too */
3052 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3054 if (likely(ext4_has_feature_large_file(inode->i_sb)))
3057 if (pos + len <= 0x7fffffffULL)
3060 /* We might need to update the superblock to set LARGE_FILE */
3064 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3065 loff_t pos, unsigned len, unsigned flags,
3066 struct page **pagep, void **fsdata)
3068 int ret, retries = 0;
3071 struct inode *inode = mapping->host;
3074 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3077 index = pos >> PAGE_SHIFT;
3079 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
3080 ext4_verity_in_progress(inode)) {
3081 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3082 return ext4_write_begin(file, mapping, pos,
3083 len, flags, pagep, fsdata);
3085 *fsdata = (void *)0;
3086 trace_ext4_da_write_begin(inode, pos, len, flags);
3088 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3089 ret = ext4_da_write_inline_data_begin(mapping, inode,
3099 * grab_cache_page_write_begin() can take a long time if the
3100 * system is thrashing due to memory pressure, or if the page
3101 * is being written back. So grab it first before we start
3102 * the transaction handle. This also allows us to allocate
3103 * the page (if needed) without using GFP_NOFS.
3106 page = grab_cache_page_write_begin(mapping, index, flags);
3112 * With delayed allocation, we don't log the i_disksize update
3113 * if there is delayed block allocation. But we still need
3114 * to journalling the i_disksize update if writes to the end
3115 * of file which has an already mapped buffer.
3118 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3119 ext4_da_write_credits(inode, pos, len));
3120 if (IS_ERR(handle)) {
3122 return PTR_ERR(handle);
3126 if (page->mapping != mapping) {
3127 /* The page got truncated from under us */
3130 ext4_journal_stop(handle);
3133 /* In case writeback began while the page was unlocked */
3134 wait_for_stable_page(page);
3136 #ifdef CONFIG_FS_ENCRYPTION
3137 ret = ext4_block_write_begin(page, pos, len,
3138 ext4_da_get_block_prep);
3140 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3144 ext4_journal_stop(handle);
3146 * block_write_begin may have instantiated a few blocks
3147 * outside i_size. Trim these off again. Don't need
3148 * i_size_read because we hold i_mutex.
3150 if (pos + len > inode->i_size)
3151 ext4_truncate_failed_write(inode);
3153 if (ret == -ENOSPC &&
3154 ext4_should_retry_alloc(inode->i_sb, &retries))
3166 * Check if we should update i_disksize
3167 * when write to the end of file but not require block allocation
3169 static int ext4_da_should_update_i_disksize(struct page *page,
3170 unsigned long offset)
3172 struct buffer_head *bh;
3173 struct inode *inode = page->mapping->host;
3177 bh = page_buffers(page);
3178 idx = offset >> inode->i_blkbits;
3180 for (i = 0; i < idx; i++)
3181 bh = bh->b_this_page;
3183 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3188 static int ext4_da_write_end(struct file *file,
3189 struct address_space *mapping,
3190 loff_t pos, unsigned len, unsigned copied,
3191 struct page *page, void *fsdata)
3193 struct inode *inode = mapping->host;
3195 handle_t *handle = ext4_journal_current_handle();
3197 unsigned long start, end;
3198 int write_mode = (int)(unsigned long)fsdata;
3200 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3201 return ext4_write_end(file, mapping, pos,
3202 len, copied, page, fsdata);
3204 trace_ext4_da_write_end(inode, pos, len, copied);
3205 start = pos & (PAGE_SIZE - 1);
3206 end = start + copied - 1;
3209 * generic_write_end() will run mark_inode_dirty() if i_size
3210 * changes. So let's piggyback the i_disksize mark_inode_dirty
3213 new_i_size = pos + copied;
3214 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3215 if (ext4_has_inline_data(inode) ||
3216 ext4_da_should_update_i_disksize(page, end)) {
3217 ext4_update_i_disksize(inode, new_i_size);
3218 /* We need to mark inode dirty even if
3219 * new_i_size is less that inode->i_size
3220 * bu greater than i_disksize.(hint delalloc)
3222 ext4_mark_inode_dirty(handle, inode);
3226 if (write_mode != CONVERT_INLINE_DATA &&
3227 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3228 ext4_has_inline_data(inode))
3229 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3232 ret2 = generic_write_end(file, mapping, pos, len, copied,
3238 ret2 = ext4_journal_stop(handle);
3242 return ret ? ret : copied;
3245 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3246 unsigned int length)
3249 * Drop reserved blocks
3251 BUG_ON(!PageLocked(page));
3252 if (!page_has_buffers(page))
3255 ext4_da_page_release_reservation(page, offset, length);
3258 ext4_invalidatepage(page, offset, length);
3264 * Force all delayed allocation blocks to be allocated for a given inode.
3266 int ext4_alloc_da_blocks(struct inode *inode)
3268 trace_ext4_alloc_da_blocks(inode);
3270 if (!EXT4_I(inode)->i_reserved_data_blocks)
3274 * We do something simple for now. The filemap_flush() will
3275 * also start triggering a write of the data blocks, which is
3276 * not strictly speaking necessary (and for users of
3277 * laptop_mode, not even desirable). However, to do otherwise
3278 * would require replicating code paths in:
3280 * ext4_writepages() ->
3281 * write_cache_pages() ---> (via passed in callback function)
3282 * __mpage_da_writepage() -->
3283 * mpage_add_bh_to_extent()
3284 * mpage_da_map_blocks()
3286 * The problem is that write_cache_pages(), located in
3287 * mm/page-writeback.c, marks pages clean in preparation for
3288 * doing I/O, which is not desirable if we're not planning on
3291 * We could call write_cache_pages(), and then redirty all of
3292 * the pages by calling redirty_page_for_writepage() but that
3293 * would be ugly in the extreme. So instead we would need to
3294 * replicate parts of the code in the above functions,
3295 * simplifying them because we wouldn't actually intend to
3296 * write out the pages, but rather only collect contiguous
3297 * logical block extents, call the multi-block allocator, and
3298 * then update the buffer heads with the block allocations.
3300 * For now, though, we'll cheat by calling filemap_flush(),
3301 * which will map the blocks, and start the I/O, but not
3302 * actually wait for the I/O to complete.
3304 return filemap_flush(inode->i_mapping);
3308 * bmap() is special. It gets used by applications such as lilo and by
3309 * the swapper to find the on-disk block of a specific piece of data.
3311 * Naturally, this is dangerous if the block concerned is still in the
3312 * journal. If somebody makes a swapfile on an ext4 data-journaling
3313 * filesystem and enables swap, then they may get a nasty shock when the
3314 * data getting swapped to that swapfile suddenly gets overwritten by
3315 * the original zero's written out previously to the journal and
3316 * awaiting writeback in the kernel's buffer cache.
3318 * So, if we see any bmap calls here on a modified, data-journaled file,
3319 * take extra steps to flush any blocks which might be in the cache.
3321 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3323 struct inode *inode = mapping->host;
3328 * We can get here for an inline file via the FIBMAP ioctl
3330 if (ext4_has_inline_data(inode))
3333 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3334 test_opt(inode->i_sb, DELALLOC)) {
3336 * With delalloc we want to sync the file
3337 * so that we can make sure we allocate
3340 filemap_write_and_wait(mapping);
3343 if (EXT4_JOURNAL(inode) &&
3344 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3346 * This is a REALLY heavyweight approach, but the use of
3347 * bmap on dirty files is expected to be extremely rare:
3348 * only if we run lilo or swapon on a freshly made file
3349 * do we expect this to happen.
3351 * (bmap requires CAP_SYS_RAWIO so this does not
3352 * represent an unprivileged user DOS attack --- we'd be
3353 * in trouble if mortal users could trigger this path at
3356 * NB. EXT4_STATE_JDATA is not set on files other than
3357 * regular files. If somebody wants to bmap a directory
3358 * or symlink and gets confused because the buffer
3359 * hasn't yet been flushed to disk, they deserve
3360 * everything they get.
3363 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3364 journal = EXT4_JOURNAL(inode);
3365 jbd2_journal_lock_updates(journal);
3366 err = jbd2_journal_flush(journal);
3367 jbd2_journal_unlock_updates(journal);
3373 return generic_block_bmap(mapping, block, ext4_get_block);
3376 static int ext4_readpage(struct file *file, struct page *page)
3379 struct inode *inode = page->mapping->host;
3381 trace_ext4_readpage(page);
3383 if (ext4_has_inline_data(inode))
3384 ret = ext4_readpage_inline(inode, page);
3387 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3394 ext4_readpages(struct file *file, struct address_space *mapping,
3395 struct list_head *pages, unsigned nr_pages)
3397 struct inode *inode = mapping->host;
3399 /* If the file has inline data, no need to do readpages. */
3400 if (ext4_has_inline_data(inode))
3403 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3406 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3407 unsigned int length)
3409 trace_ext4_invalidatepage(page, offset, length);
3411 /* No journalling happens on data buffers when this function is used */
3412 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3414 block_invalidatepage(page, offset, length);
3417 static int __ext4_journalled_invalidatepage(struct page *page,
3418 unsigned int offset,
3419 unsigned int length)
3421 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3423 trace_ext4_journalled_invalidatepage(page, offset, length);
3426 * If it's a full truncate we just forget about the pending dirtying
3428 if (offset == 0 && length == PAGE_SIZE)
3429 ClearPageChecked(page);
3431 return jbd2_journal_invalidatepage(journal, page, offset, length);
3434 /* Wrapper for aops... */
3435 static void ext4_journalled_invalidatepage(struct page *page,
3436 unsigned int offset,
3437 unsigned int length)
3439 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3442 static int ext4_releasepage(struct page *page, gfp_t wait)
3444 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3446 trace_ext4_releasepage(page);
3448 /* Page has dirty journalled data -> cannot release */
3449 if (PageChecked(page))
3452 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3454 return try_to_free_buffers(page);
3457 static bool ext4_inode_datasync_dirty(struct inode *inode)
3459 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3462 return !jbd2_transaction_committed(journal,
3463 EXT4_I(inode)->i_datasync_tid);
3464 /* Any metadata buffers to write? */
3465 if (!list_empty(&inode->i_mapping->private_list))
3467 return inode->i_state & I_DIRTY_DATASYNC;
3470 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3471 unsigned flags, struct iomap *iomap)
3473 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3474 unsigned int blkbits = inode->i_blkbits;
3475 unsigned long first_block, last_block;
3476 struct ext4_map_blocks map;
3477 bool delalloc = false;
3480 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3482 first_block = offset >> blkbits;
3483 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3484 EXT4_MAX_LOGICAL_BLOCK);
3486 if (flags & IOMAP_REPORT) {
3487 if (ext4_has_inline_data(inode)) {
3488 ret = ext4_inline_data_iomap(inode, iomap);
3489 if (ret != -EAGAIN) {
3490 if (ret == 0 && offset >= iomap->length)
3496 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3500 map.m_lblk = first_block;
3501 map.m_len = last_block - first_block + 1;
3503 if (flags & IOMAP_REPORT) {
3504 ret = ext4_map_blocks(NULL, inode, &map, 0);
3509 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3510 struct extent_status es;
3512 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3513 map.m_lblk, end, &es);
3515 if (!es.es_len || es.es_lblk > end) {
3516 /* entire range is a hole */
3517 } else if (es.es_lblk > map.m_lblk) {
3518 /* range starts with a hole */
3519 map.m_len = es.es_lblk - map.m_lblk;
3521 ext4_lblk_t offs = 0;
3523 if (es.es_lblk < map.m_lblk)
3524 offs = map.m_lblk - es.es_lblk;
3525 map.m_lblk = es.es_lblk + offs;
3526 map.m_len = es.es_len - offs;
3530 } else if (flags & IOMAP_WRITE) {
3535 /* Trim mapping request to maximum we can map at once for DIO */
3536 if (map.m_len > DIO_MAX_BLOCKS)
3537 map.m_len = DIO_MAX_BLOCKS;
3538 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3541 * Either we allocate blocks and then we don't get unwritten
3542 * extent so we have reserved enough credits, or the blocks
3543 * are already allocated and unwritten and in that case
3544 * extent conversion fits in the credits as well.
3546 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3549 return PTR_ERR(handle);
3551 ret = ext4_map_blocks(handle, inode, &map,
3552 EXT4_GET_BLOCKS_CREATE_ZERO);
3554 ext4_journal_stop(handle);
3555 if (ret == -ENOSPC &&
3556 ext4_should_retry_alloc(inode->i_sb, &retries))
3562 * If we added blocks beyond i_size, we need to make sure they
3563 * will get truncated if we crash before updating i_size in
3564 * ext4_iomap_end(). For faults we don't need to do that (and
3565 * even cannot because for orphan list operations inode_lock is
3566 * required) - if we happen to instantiate block beyond i_size,
3567 * it is because we race with truncate which has already added
3568 * the inode to the orphan list.
3570 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3571 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3574 err = ext4_orphan_add(handle, inode);
3576 ext4_journal_stop(handle);
3580 ext4_journal_stop(handle);
3582 ret = ext4_map_blocks(NULL, inode, &map, 0);
3588 if (ext4_inode_datasync_dirty(inode))
3589 iomap->flags |= IOMAP_F_DIRTY;
3590 iomap->bdev = inode->i_sb->s_bdev;
3591 iomap->dax_dev = sbi->s_daxdev;
3592 iomap->offset = (u64)first_block << blkbits;
3593 iomap->length = (u64)map.m_len << blkbits;
3596 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3597 iomap->addr = IOMAP_NULL_ADDR;
3599 if (map.m_flags & EXT4_MAP_MAPPED) {
3600 iomap->type = IOMAP_MAPPED;
3601 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3602 iomap->type = IOMAP_UNWRITTEN;
3607 iomap->addr = (u64)map.m_pblk << blkbits;
3610 if (map.m_flags & EXT4_MAP_NEW)
3611 iomap->flags |= IOMAP_F_NEW;
3616 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3617 ssize_t written, unsigned flags, struct iomap *iomap)
3621 int blkbits = inode->i_blkbits;
3622 bool truncate = false;
3624 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3627 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3628 if (IS_ERR(handle)) {
3629 ret = PTR_ERR(handle);
3632 if (ext4_update_inode_size(inode, offset + written))
3633 ext4_mark_inode_dirty(handle, inode);
3635 * We may need to truncate allocated but not written blocks beyond EOF.
3637 if (iomap->offset + iomap->length >
3638 ALIGN(inode->i_size, 1 << blkbits)) {
3639 ext4_lblk_t written_blk, end_blk;
3641 written_blk = (offset + written) >> blkbits;
3642 end_blk = (offset + length) >> blkbits;
3643 if (written_blk < end_blk && ext4_can_truncate(inode))
3647 * Remove inode from orphan list if we were extending a inode and
3648 * everything went fine.
3650 if (!truncate && inode->i_nlink &&
3651 !list_empty(&EXT4_I(inode)->i_orphan))
3652 ext4_orphan_del(handle, inode);
3653 ext4_journal_stop(handle);
3655 ext4_truncate_failed_write(inode);
3658 * If truncate failed early the inode might still be on the
3659 * orphan list; we need to make sure the inode is removed from
3660 * the orphan list in that case.
3663 ext4_orphan_del(NULL, inode);
3668 const struct iomap_ops ext4_iomap_ops = {
3669 .iomap_begin = ext4_iomap_begin,
3670 .iomap_end = ext4_iomap_end,
3673 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3674 ssize_t size, void *private)
3676 ext4_io_end_t *io_end = private;
3678 /* if not async direct IO just return */
3682 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3683 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3684 io_end, io_end->inode->i_ino, iocb, offset, size);
3687 * Error during AIO DIO. We cannot convert unwritten extents as the
3688 * data was not written. Just clear the unwritten flag and drop io_end.
3691 ext4_clear_io_unwritten_flag(io_end);
3694 io_end->offset = offset;
3695 io_end->size = size;
3696 ext4_put_io_end(io_end);
3702 * Handling of direct IO writes.
3704 * For ext4 extent files, ext4 will do direct-io write even to holes,
3705 * preallocated extents, and those write extend the file, no need to
3706 * fall back to buffered IO.
3708 * For holes, we fallocate those blocks, mark them as unwritten
3709 * If those blocks were preallocated, we mark sure they are split, but
3710 * still keep the range to write as unwritten.
3712 * The unwritten extents will be converted to written when DIO is completed.
3713 * For async direct IO, since the IO may still pending when return, we
3714 * set up an end_io call back function, which will do the conversion
3715 * when async direct IO completed.
3717 * If the O_DIRECT write will extend the file then add this inode to the
3718 * orphan list. So recovery will truncate it back to the original size
3719 * if the machine crashes during the write.
3722 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3724 struct file *file = iocb->ki_filp;
3725 struct inode *inode = file->f_mapping->host;
3726 struct ext4_inode_info *ei = EXT4_I(inode);
3728 loff_t offset = iocb->ki_pos;
3729 size_t count = iov_iter_count(iter);
3731 get_block_t *get_block_func = NULL;
3733 loff_t final_size = offset + count;
3737 if (final_size > inode->i_size || final_size > ei->i_disksize) {
3738 /* Credits for sb + inode write */
3739 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3740 if (IS_ERR(handle)) {
3741 ret = PTR_ERR(handle);
3744 ret = ext4_orphan_add(handle, inode);
3746 ext4_journal_stop(handle);
3750 ext4_update_i_disksize(inode, inode->i_size);
3751 ext4_journal_stop(handle);
3754 BUG_ON(iocb->private == NULL);
3757 * Make all waiters for direct IO properly wait also for extent
3758 * conversion. This also disallows race between truncate() and
3759 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3761 inode_dio_begin(inode);
3763 /* If we do a overwrite dio, i_mutex locking can be released */
3764 overwrite = *((int *)iocb->private);
3767 inode_unlock(inode);
3770 * For extent mapped files we could direct write to holes and fallocate.
3772 * Allocated blocks to fill the hole are marked as unwritten to prevent
3773 * parallel buffered read to expose the stale data before DIO complete
3776 * As to previously fallocated extents, ext4 get_block will just simply
3777 * mark the buffer mapped but still keep the extents unwritten.
3779 * For non AIO case, we will convert those unwritten extents to written
3780 * after return back from blockdev_direct_IO. That way we save us from
3781 * allocating io_end structure and also the overhead of offloading
3782 * the extent convertion to a workqueue.
3784 * For async DIO, the conversion needs to be deferred when the
3785 * IO is completed. The ext4 end_io callback function will be
3786 * called to take care of the conversion work. Here for async
3787 * case, we allocate an io_end structure to hook to the iocb.
3789 iocb->private = NULL;
3791 get_block_func = ext4_dio_get_block_overwrite;
3792 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3793 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3794 get_block_func = ext4_dio_get_block;
3795 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3796 } else if (is_sync_kiocb(iocb)) {
3797 get_block_func = ext4_dio_get_block_unwritten_sync;
3798 dio_flags = DIO_LOCKING;
3800 get_block_func = ext4_dio_get_block_unwritten_async;
3801 dio_flags = DIO_LOCKING;
3803 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3804 get_block_func, ext4_end_io_dio, NULL,
3807 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3808 EXT4_STATE_DIO_UNWRITTEN)) {
3811 * for non AIO case, since the IO is already
3812 * completed, we could do the conversion right here
3814 err = ext4_convert_unwritten_extents(NULL, inode,
3818 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3821 inode_dio_end(inode);
3822 /* take i_mutex locking again if we do a ovewrite dio */
3826 if (ret < 0 && final_size > inode->i_size)
3827 ext4_truncate_failed_write(inode);
3829 /* Handle extending of i_size after direct IO write */
3833 /* Credits for sb + inode write */
3834 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3835 if (IS_ERR(handle)) {
3837 * We wrote the data but cannot extend
3838 * i_size. Bail out. In async io case, we do
3839 * not return error here because we have
3840 * already submmitted the corresponding
3841 * bio. Returning error here makes the caller
3842 * think that this IO is done and failed
3843 * resulting in race with bio's completion
3847 ret = PTR_ERR(handle);
3849 ext4_orphan_del(NULL, inode);
3854 ext4_orphan_del(handle, inode);
3856 loff_t end = offset + ret;
3857 if (end > inode->i_size || end > ei->i_disksize) {
3858 ext4_update_i_disksize(inode, end);
3859 if (end > inode->i_size)
3860 i_size_write(inode, end);
3862 * We're going to return a positive `ret'
3863 * here due to non-zero-length I/O, so there's
3864 * no way of reporting error returns from
3865 * ext4_mark_inode_dirty() to userspace. So
3868 ext4_mark_inode_dirty(handle, inode);
3871 err = ext4_journal_stop(handle);
3879 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3881 struct address_space *mapping = iocb->ki_filp->f_mapping;
3882 struct inode *inode = mapping->host;
3883 size_t count = iov_iter_count(iter);
3887 * Shared inode_lock is enough for us - it protects against concurrent
3888 * writes & truncates and since we take care of writing back page cache,
3889 * we are protected against page writeback as well.
3891 inode_lock_shared(inode);
3892 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3893 iocb->ki_pos + count - 1);
3896 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3897 iter, ext4_dio_get_block, NULL, NULL, 0);
3899 inode_unlock_shared(inode);
3903 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3905 struct file *file = iocb->ki_filp;
3906 struct inode *inode = file->f_mapping->host;
3907 size_t count = iov_iter_count(iter);
3908 loff_t offset = iocb->ki_pos;
3911 #ifdef CONFIG_FS_ENCRYPTION
3912 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
3915 if (fsverity_active(inode))
3919 * If we are doing data journalling we don't support O_DIRECT
3921 if (ext4_should_journal_data(inode))
3924 /* Let buffer I/O handle the inline data case. */
3925 if (ext4_has_inline_data(inode))
3928 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3929 if (iov_iter_rw(iter) == READ)
3930 ret = ext4_direct_IO_read(iocb, iter);
3932 ret = ext4_direct_IO_write(iocb, iter);
3933 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3938 * Pages can be marked dirty completely asynchronously from ext4's journalling
3939 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3940 * much here because ->set_page_dirty is called under VFS locks. The page is
3941 * not necessarily locked.
3943 * We cannot just dirty the page and leave attached buffers clean, because the
3944 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3945 * or jbddirty because all the journalling code will explode.
3947 * So what we do is to mark the page "pending dirty" and next time writepage
3948 * is called, propagate that into the buffers appropriately.
3950 static int ext4_journalled_set_page_dirty(struct page *page)
3952 SetPageChecked(page);
3953 return __set_page_dirty_nobuffers(page);
3956 static int ext4_set_page_dirty(struct page *page)
3958 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3959 WARN_ON_ONCE(!page_has_buffers(page));
3960 return __set_page_dirty_buffers(page);
3963 static const struct address_space_operations ext4_aops = {
3964 .readpage = ext4_readpage,
3965 .readpages = ext4_readpages,
3966 .writepage = ext4_writepage,
3967 .writepages = ext4_writepages,
3968 .write_begin = ext4_write_begin,
3969 .write_end = ext4_write_end,
3970 .set_page_dirty = ext4_set_page_dirty,
3972 .invalidatepage = ext4_invalidatepage,
3973 .releasepage = ext4_releasepage,
3974 .direct_IO = ext4_direct_IO,
3975 .migratepage = buffer_migrate_page,
3976 .is_partially_uptodate = block_is_partially_uptodate,
3977 .error_remove_page = generic_error_remove_page,
3980 static const struct address_space_operations ext4_journalled_aops = {
3981 .readpage = ext4_readpage,
3982 .readpages = ext4_readpages,
3983 .writepage = ext4_writepage,
3984 .writepages = ext4_writepages,
3985 .write_begin = ext4_write_begin,
3986 .write_end = ext4_journalled_write_end,
3987 .set_page_dirty = ext4_journalled_set_page_dirty,
3989 .invalidatepage = ext4_journalled_invalidatepage,
3990 .releasepage = ext4_releasepage,
3991 .direct_IO = ext4_direct_IO,
3992 .is_partially_uptodate = block_is_partially_uptodate,
3993 .error_remove_page = generic_error_remove_page,
3996 static const struct address_space_operations ext4_da_aops = {
3997 .readpage = ext4_readpage,
3998 .readpages = ext4_readpages,
3999 .writepage = ext4_writepage,
4000 .writepages = ext4_writepages,
4001 .write_begin = ext4_da_write_begin,
4002 .write_end = ext4_da_write_end,
4003 .set_page_dirty = ext4_set_page_dirty,
4005 .invalidatepage = ext4_da_invalidatepage,
4006 .releasepage = ext4_releasepage,
4007 .direct_IO = ext4_direct_IO,
4008 .migratepage = buffer_migrate_page,
4009 .is_partially_uptodate = block_is_partially_uptodate,
4010 .error_remove_page = generic_error_remove_page,
4013 static const struct address_space_operations ext4_dax_aops = {
4014 .writepages = ext4_dax_writepages,
4015 .direct_IO = noop_direct_IO,
4016 .set_page_dirty = noop_set_page_dirty,
4018 .invalidatepage = noop_invalidatepage,
4021 void ext4_set_aops(struct inode *inode)
4023 switch (ext4_inode_journal_mode(inode)) {
4024 case EXT4_INODE_ORDERED_DATA_MODE:
4025 case EXT4_INODE_WRITEBACK_DATA_MODE:
4027 case EXT4_INODE_JOURNAL_DATA_MODE:
4028 inode->i_mapping->a_ops = &ext4_journalled_aops;
4034 inode->i_mapping->a_ops = &ext4_dax_aops;
4035 else if (test_opt(inode->i_sb, DELALLOC))
4036 inode->i_mapping->a_ops = &ext4_da_aops;
4038 inode->i_mapping->a_ops = &ext4_aops;
4041 static int __ext4_block_zero_page_range(handle_t *handle,
4042 struct address_space *mapping, loff_t from, loff_t length)
4044 ext4_fsblk_t index = from >> PAGE_SHIFT;
4045 unsigned offset = from & (PAGE_SIZE-1);
4046 unsigned blocksize, pos;
4048 struct inode *inode = mapping->host;
4049 struct buffer_head *bh;
4053 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4054 mapping_gfp_constraint(mapping, ~__GFP_FS));
4058 blocksize = inode->i_sb->s_blocksize;
4060 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4062 if (!page_has_buffers(page))
4063 create_empty_buffers(page, blocksize, 0);
4065 /* Find the buffer that contains "offset" */
4066 bh = page_buffers(page);
4068 while (offset >= pos) {
4069 bh = bh->b_this_page;
4073 if (buffer_freed(bh)) {
4074 BUFFER_TRACE(bh, "freed: skip");
4077 if (!buffer_mapped(bh)) {
4078 BUFFER_TRACE(bh, "unmapped");
4079 ext4_get_block(inode, iblock, bh, 0);
4080 /* unmapped? It's a hole - nothing to do */
4081 if (!buffer_mapped(bh)) {
4082 BUFFER_TRACE(bh, "still unmapped");
4087 /* Ok, it's mapped. Make sure it's up-to-date */
4088 if (PageUptodate(page))
4089 set_buffer_uptodate(bh);
4091 if (!buffer_uptodate(bh)) {
4093 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4095 /* Uhhuh. Read error. Complain and punt. */
4096 if (!buffer_uptodate(bh))
4098 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
4099 /* We expect the key to be set. */
4100 BUG_ON(!fscrypt_has_encryption_key(inode));
4101 WARN_ON_ONCE(fscrypt_decrypt_pagecache_blocks(
4102 page, blocksize, bh_offset(bh)));
4105 if (ext4_should_journal_data(inode)) {
4106 BUFFER_TRACE(bh, "get write access");
4107 err = ext4_journal_get_write_access(handle, bh);
4111 zero_user(page, offset, length);
4112 BUFFER_TRACE(bh, "zeroed end of block");
4114 if (ext4_should_journal_data(inode)) {
4115 err = ext4_handle_dirty_metadata(handle, inode, bh);
4118 mark_buffer_dirty(bh);
4119 if (ext4_should_order_data(inode))
4120 err = ext4_jbd2_inode_add_write(handle, inode, from,
4131 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4132 * starting from file offset 'from'. The range to be zero'd must
4133 * be contained with in one block. If the specified range exceeds
4134 * the end of the block it will be shortened to end of the block
4135 * that cooresponds to 'from'
4137 static int ext4_block_zero_page_range(handle_t *handle,
4138 struct address_space *mapping, loff_t from, loff_t length)
4140 struct inode *inode = mapping->host;
4141 unsigned offset = from & (PAGE_SIZE-1);
4142 unsigned blocksize = inode->i_sb->s_blocksize;
4143 unsigned max = blocksize - (offset & (blocksize - 1));
4146 * correct length if it does not fall between
4147 * 'from' and the end of the block
4149 if (length > max || length < 0)
4152 if (IS_DAX(inode)) {
4153 return iomap_zero_range(inode, from, length, NULL,
4156 return __ext4_block_zero_page_range(handle, mapping, from, length);
4160 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4161 * up to the end of the block which corresponds to `from'.
4162 * This required during truncate. We need to physically zero the tail end
4163 * of that block so it doesn't yield old data if the file is later grown.
4165 static int ext4_block_truncate_page(handle_t *handle,
4166 struct address_space *mapping, loff_t from)
4168 unsigned offset = from & (PAGE_SIZE-1);
4171 struct inode *inode = mapping->host;
4173 /* If we are processing an encrypted inode during orphan list handling */
4174 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
4177 blocksize = inode->i_sb->s_blocksize;
4178 length = blocksize - (offset & (blocksize - 1));
4180 return ext4_block_zero_page_range(handle, mapping, from, length);
4183 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4184 loff_t lstart, loff_t length)
4186 struct super_block *sb = inode->i_sb;
4187 struct address_space *mapping = inode->i_mapping;
4188 unsigned partial_start, partial_end;
4189 ext4_fsblk_t start, end;
4190 loff_t byte_end = (lstart + length - 1);
4193 partial_start = lstart & (sb->s_blocksize - 1);
4194 partial_end = byte_end & (sb->s_blocksize - 1);
4196 start = lstart >> sb->s_blocksize_bits;
4197 end = byte_end >> sb->s_blocksize_bits;
4199 /* Handle partial zero within the single block */
4201 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4202 err = ext4_block_zero_page_range(handle, mapping,
4206 /* Handle partial zero out on the start of the range */
4207 if (partial_start) {
4208 err = ext4_block_zero_page_range(handle, mapping,
4209 lstart, sb->s_blocksize);
4213 /* Handle partial zero out on the end of the range */
4214 if (partial_end != sb->s_blocksize - 1)
4215 err = ext4_block_zero_page_range(handle, mapping,
4216 byte_end - partial_end,
4221 int ext4_can_truncate(struct inode *inode)
4223 if (S_ISREG(inode->i_mode))
4225 if (S_ISDIR(inode->i_mode))
4227 if (S_ISLNK(inode->i_mode))
4228 return !ext4_inode_is_fast_symlink(inode);
4233 * We have to make sure i_disksize gets properly updated before we truncate
4234 * page cache due to hole punching or zero range. Otherwise i_disksize update
4235 * can get lost as it may have been postponed to submission of writeback but
4236 * that will never happen after we truncate page cache.
4238 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4242 loff_t size = i_size_read(inode);
4244 WARN_ON(!inode_is_locked(inode));
4245 if (offset > size || offset + len < size)
4248 if (EXT4_I(inode)->i_disksize >= size)
4251 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4253 return PTR_ERR(handle);
4254 ext4_update_i_disksize(inode, size);
4255 ext4_mark_inode_dirty(handle, inode);
4256 ext4_journal_stop(handle);
4261 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4263 up_write(&ei->i_mmap_sem);
4265 down_write(&ei->i_mmap_sem);
4268 int ext4_break_layouts(struct inode *inode)
4270 struct ext4_inode_info *ei = EXT4_I(inode);
4274 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4278 page = dax_layout_busy_page(inode->i_mapping);
4282 error = ___wait_var_event(&page->_refcount,
4283 atomic_read(&page->_refcount) == 1,
4284 TASK_INTERRUPTIBLE, 0, 0,
4285 ext4_wait_dax_page(ei));
4286 } while (error == 0);
4292 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4293 * associated with the given offset and length
4295 * @inode: File inode
4296 * @offset: The offset where the hole will begin
4297 * @len: The length of the hole
4299 * Returns: 0 on success or negative on failure
4302 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4304 struct super_block *sb = inode->i_sb;
4305 ext4_lblk_t first_block, stop_block;
4306 struct address_space *mapping = inode->i_mapping;
4307 loff_t first_block_offset, last_block_offset;
4309 unsigned int credits;
4312 if (!S_ISREG(inode->i_mode))
4315 trace_ext4_punch_hole(inode, offset, length, 0);
4318 * Write out all dirty pages to avoid race conditions
4319 * Then release them.
4321 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4322 ret = filemap_write_and_wait_range(mapping, offset,
4323 offset + length - 1);
4330 /* No need to punch hole beyond i_size */
4331 if (offset >= inode->i_size)
4335 * If the hole extends beyond i_size, set the hole
4336 * to end after the page that contains i_size
4338 if (offset + length > inode->i_size) {
4339 length = inode->i_size +
4340 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4344 if (offset & (sb->s_blocksize - 1) ||
4345 (offset + length) & (sb->s_blocksize - 1)) {
4347 * Attach jinode to inode for jbd2 if we do any zeroing of
4350 ret = ext4_inode_attach_jinode(inode);
4356 /* Wait all existing dio workers, newcomers will block on i_mutex */
4357 inode_dio_wait(inode);
4360 * Prevent page faults from reinstantiating pages we have released from
4363 down_write(&EXT4_I(inode)->i_mmap_sem);
4365 ret = ext4_break_layouts(inode);
4369 first_block_offset = round_up(offset, sb->s_blocksize);
4370 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4372 /* Now release the pages and zero block aligned part of pages*/
4373 if (last_block_offset > first_block_offset) {
4374 ret = ext4_update_disksize_before_punch(inode, offset, length);
4377 truncate_pagecache_range(inode, first_block_offset,
4381 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4382 credits = ext4_writepage_trans_blocks(inode);
4384 credits = ext4_blocks_for_truncate(inode);
4385 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4386 if (IS_ERR(handle)) {
4387 ret = PTR_ERR(handle);
4388 ext4_std_error(sb, ret);
4392 ret = ext4_zero_partial_blocks(handle, inode, offset,
4397 first_block = (offset + sb->s_blocksize - 1) >>
4398 EXT4_BLOCK_SIZE_BITS(sb);
4399 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4401 /* If there are blocks to remove, do it */
4402 if (stop_block > first_block) {
4404 down_write(&EXT4_I(inode)->i_data_sem);
4405 ext4_discard_preallocations(inode);
4407 ret = ext4_es_remove_extent(inode, first_block,
4408 stop_block - first_block);
4410 up_write(&EXT4_I(inode)->i_data_sem);
4414 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4415 ret = ext4_ext_remove_space(inode, first_block,
4418 ret = ext4_ind_remove_space(handle, inode, first_block,
4421 up_write(&EXT4_I(inode)->i_data_sem);
4424 ext4_handle_sync(handle);
4426 inode->i_mtime = inode->i_ctime = current_time(inode);
4427 ext4_mark_inode_dirty(handle, inode);
4429 ext4_update_inode_fsync_trans(handle, inode, 1);
4431 ext4_journal_stop(handle);
4433 up_write(&EXT4_I(inode)->i_mmap_sem);
4435 inode_unlock(inode);
4439 int ext4_inode_attach_jinode(struct inode *inode)
4441 struct ext4_inode_info *ei = EXT4_I(inode);
4442 struct jbd2_inode *jinode;
4444 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4447 jinode = jbd2_alloc_inode(GFP_KERNEL);
4448 spin_lock(&inode->i_lock);
4451 spin_unlock(&inode->i_lock);
4454 ei->jinode = jinode;
4455 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4458 spin_unlock(&inode->i_lock);
4459 if (unlikely(jinode != NULL))
4460 jbd2_free_inode(jinode);
4467 * We block out ext4_get_block() block instantiations across the entire
4468 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4469 * simultaneously on behalf of the same inode.
4471 * As we work through the truncate and commit bits of it to the journal there
4472 * is one core, guiding principle: the file's tree must always be consistent on
4473 * disk. We must be able to restart the truncate after a crash.
4475 * The file's tree may be transiently inconsistent in memory (although it
4476 * probably isn't), but whenever we close off and commit a journal transaction,
4477 * the contents of (the filesystem + the journal) must be consistent and
4478 * restartable. It's pretty simple, really: bottom up, right to left (although
4479 * left-to-right works OK too).
4481 * Note that at recovery time, journal replay occurs *before* the restart of
4482 * truncate against the orphan inode list.
4484 * The committed inode has the new, desired i_size (which is the same as
4485 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4486 * that this inode's truncate did not complete and it will again call
4487 * ext4_truncate() to have another go. So there will be instantiated blocks
4488 * to the right of the truncation point in a crashed ext4 filesystem. But
4489 * that's fine - as long as they are linked from the inode, the post-crash
4490 * ext4_truncate() run will find them and release them.
4492 int ext4_truncate(struct inode *inode)
4494 struct ext4_inode_info *ei = EXT4_I(inode);
4495 unsigned int credits;
4498 struct address_space *mapping = inode->i_mapping;
4501 * There is a possibility that we're either freeing the inode
4502 * or it's a completely new inode. In those cases we might not
4503 * have i_mutex locked because it's not necessary.
4505 if (!(inode->i_state & (I_NEW|I_FREEING)))
4506 WARN_ON(!inode_is_locked(inode));
4507 trace_ext4_truncate_enter(inode);
4509 if (!ext4_can_truncate(inode))
4512 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4514 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4515 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4517 if (ext4_has_inline_data(inode)) {
4520 err = ext4_inline_data_truncate(inode, &has_inline);
4527 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4528 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4529 if (ext4_inode_attach_jinode(inode) < 0)
4533 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4534 credits = ext4_writepage_trans_blocks(inode);
4536 credits = ext4_blocks_for_truncate(inode);
4538 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4540 return PTR_ERR(handle);
4542 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4543 ext4_block_truncate_page(handle, mapping, inode->i_size);
4546 * We add the inode to the orphan list, so that if this
4547 * truncate spans multiple transactions, and we crash, we will
4548 * resume the truncate when the filesystem recovers. It also
4549 * marks the inode dirty, to catch the new size.
4551 * Implication: the file must always be in a sane, consistent
4552 * truncatable state while each transaction commits.
4554 err = ext4_orphan_add(handle, inode);
4558 down_write(&EXT4_I(inode)->i_data_sem);
4560 ext4_discard_preallocations(inode);
4562 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4563 err = ext4_ext_truncate(handle, inode);
4565 ext4_ind_truncate(handle, inode);
4567 up_write(&ei->i_data_sem);
4572 ext4_handle_sync(handle);
4576 * If this was a simple ftruncate() and the file will remain alive,
4577 * then we need to clear up the orphan record which we created above.
4578 * However, if this was a real unlink then we were called by
4579 * ext4_evict_inode(), and we allow that function to clean up the
4580 * orphan info for us.
4583 ext4_orphan_del(handle, inode);
4585 inode->i_mtime = inode->i_ctime = current_time(inode);
4586 ext4_mark_inode_dirty(handle, inode);
4587 ext4_journal_stop(handle);
4589 trace_ext4_truncate_exit(inode);
4594 * ext4_get_inode_loc returns with an extra refcount against the inode's
4595 * underlying buffer_head on success. If 'in_mem' is true, we have all
4596 * data in memory that is needed to recreate the on-disk version of this
4599 static int __ext4_get_inode_loc(struct inode *inode,
4600 struct ext4_iloc *iloc, int in_mem)
4602 struct ext4_group_desc *gdp;
4603 struct buffer_head *bh;
4604 struct super_block *sb = inode->i_sb;
4606 int inodes_per_block, inode_offset;
4609 if (inode->i_ino < EXT4_ROOT_INO ||
4610 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4611 return -EFSCORRUPTED;
4613 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4614 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4619 * Figure out the offset within the block group inode table
4621 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4622 inode_offset = ((inode->i_ino - 1) %
4623 EXT4_INODES_PER_GROUP(sb));
4624 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4625 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4627 bh = sb_getblk(sb, block);
4630 if (!buffer_uptodate(bh)) {
4634 * If the buffer has the write error flag, we have failed
4635 * to write out another inode in the same block. In this
4636 * case, we don't have to read the block because we may
4637 * read the old inode data successfully.
4639 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4640 set_buffer_uptodate(bh);
4642 if (buffer_uptodate(bh)) {
4643 /* someone brought it uptodate while we waited */
4649 * If we have all information of the inode in memory and this
4650 * is the only valid inode in the block, we need not read the
4654 struct buffer_head *bitmap_bh;
4657 start = inode_offset & ~(inodes_per_block - 1);
4659 /* Is the inode bitmap in cache? */
4660 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4661 if (unlikely(!bitmap_bh))
4665 * If the inode bitmap isn't in cache then the
4666 * optimisation may end up performing two reads instead
4667 * of one, so skip it.
4669 if (!buffer_uptodate(bitmap_bh)) {
4673 for (i = start; i < start + inodes_per_block; i++) {
4674 if (i == inode_offset)
4676 if (ext4_test_bit(i, bitmap_bh->b_data))
4680 if (i == start + inodes_per_block) {
4681 /* all other inodes are free, so skip I/O */
4682 memset(bh->b_data, 0, bh->b_size);
4683 set_buffer_uptodate(bh);
4691 * If we need to do any I/O, try to pre-readahead extra
4692 * blocks from the inode table.
4694 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4695 ext4_fsblk_t b, end, table;
4697 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4699 table = ext4_inode_table(sb, gdp);
4700 /* s_inode_readahead_blks is always a power of 2 */
4701 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4705 num = EXT4_INODES_PER_GROUP(sb);
4706 if (ext4_has_group_desc_csum(sb))
4707 num -= ext4_itable_unused_count(sb, gdp);
4708 table += num / inodes_per_block;
4712 sb_breadahead(sb, b++);
4716 * There are other valid inodes in the buffer, this inode
4717 * has in-inode xattrs, or we don't have this inode in memory.
4718 * Read the block from disk.
4720 trace_ext4_load_inode(inode);
4722 bh->b_end_io = end_buffer_read_sync;
4723 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4725 if (!buffer_uptodate(bh)) {
4726 EXT4_ERROR_INODE_BLOCK(inode, block,
4727 "unable to read itable block");
4737 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4739 /* We have all inode data except xattrs in memory here. */
4740 return __ext4_get_inode_loc(inode, iloc,
4741 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4744 static bool ext4_should_use_dax(struct inode *inode)
4746 if (!test_opt(inode->i_sb, DAX))
4748 if (!S_ISREG(inode->i_mode))
4750 if (ext4_should_journal_data(inode))
4752 if (ext4_has_inline_data(inode))
4754 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4756 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4761 void ext4_set_inode_flags(struct inode *inode)
4763 unsigned int flags = EXT4_I(inode)->i_flags;
4764 unsigned int new_fl = 0;
4766 if (flags & EXT4_SYNC_FL)
4768 if (flags & EXT4_APPEND_FL)
4770 if (flags & EXT4_IMMUTABLE_FL)
4771 new_fl |= S_IMMUTABLE;
4772 if (flags & EXT4_NOATIME_FL)
4773 new_fl |= S_NOATIME;
4774 if (flags & EXT4_DIRSYNC_FL)
4775 new_fl |= S_DIRSYNC;
4776 if (ext4_should_use_dax(inode))
4778 if (flags & EXT4_ENCRYPT_FL)
4779 new_fl |= S_ENCRYPTED;
4780 if (flags & EXT4_CASEFOLD_FL)
4781 new_fl |= S_CASEFOLD;
4782 if (flags & EXT4_VERITY_FL)
4784 inode_set_flags(inode, new_fl,
4785 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4786 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4789 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4790 struct ext4_inode_info *ei)
4793 struct inode *inode = &(ei->vfs_inode);
4794 struct super_block *sb = inode->i_sb;
4796 if (ext4_has_feature_huge_file(sb)) {
4797 /* we are using combined 48 bit field */
4798 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4799 le32_to_cpu(raw_inode->i_blocks_lo);
4800 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4801 /* i_blocks represent file system block size */
4802 return i_blocks << (inode->i_blkbits - 9);
4807 return le32_to_cpu(raw_inode->i_blocks_lo);
4811 static inline int ext4_iget_extra_inode(struct inode *inode,
4812 struct ext4_inode *raw_inode,
4813 struct ext4_inode_info *ei)
4815 __le32 *magic = (void *)raw_inode +
4816 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4818 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4819 EXT4_INODE_SIZE(inode->i_sb) &&
4820 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4821 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4822 return ext4_find_inline_data_nolock(inode);
4824 EXT4_I(inode)->i_inline_off = 0;
4828 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4830 if (!ext4_has_feature_project(inode->i_sb))
4832 *projid = EXT4_I(inode)->i_projid;
4837 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4838 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4841 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4843 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4844 inode_set_iversion_raw(inode, val);
4846 inode_set_iversion_queried(inode, val);
4848 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4850 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4851 return inode_peek_iversion_raw(inode);
4853 return inode_peek_iversion(inode);
4856 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4857 ext4_iget_flags flags, const char *function,
4860 struct ext4_iloc iloc;
4861 struct ext4_inode *raw_inode;
4862 struct ext4_inode_info *ei;
4863 struct inode *inode;
4864 journal_t *journal = EXT4_SB(sb)->s_journal;
4872 if ((!(flags & EXT4_IGET_SPECIAL) &&
4873 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4874 (ino < EXT4_ROOT_INO) ||
4875 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4876 if (flags & EXT4_IGET_HANDLE)
4877 return ERR_PTR(-ESTALE);
4878 __ext4_error(sb, function, line,
4879 "inode #%lu: comm %s: iget: illegal inode #",
4880 ino, current->comm);
4881 return ERR_PTR(-EFSCORRUPTED);
4884 inode = iget_locked(sb, ino);
4886 return ERR_PTR(-ENOMEM);
4887 if (!(inode->i_state & I_NEW))
4893 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4896 raw_inode = ext4_raw_inode(&iloc);
4898 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4899 ext4_error_inode(inode, function, line, 0,
4900 "iget: root inode unallocated");
4901 ret = -EFSCORRUPTED;
4905 if ((flags & EXT4_IGET_HANDLE) &&
4906 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4911 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4912 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4913 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4914 EXT4_INODE_SIZE(inode->i_sb) ||
4915 (ei->i_extra_isize & 3)) {
4916 ext4_error_inode(inode, function, line, 0,
4917 "iget: bad extra_isize %u "
4920 EXT4_INODE_SIZE(inode->i_sb));
4921 ret = -EFSCORRUPTED;
4925 ei->i_extra_isize = 0;
4927 /* Precompute checksum seed for inode metadata */
4928 if (ext4_has_metadata_csum(sb)) {
4929 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4931 __le32 inum = cpu_to_le32(inode->i_ino);
4932 __le32 gen = raw_inode->i_generation;
4933 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4935 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4939 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4940 ext4_error_inode(inode, function, line, 0,
4941 "iget: checksum invalid");
4946 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4947 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4948 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4949 if (ext4_has_feature_project(sb) &&
4950 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4951 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4952 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4954 i_projid = EXT4_DEF_PROJID;
4956 if (!(test_opt(inode->i_sb, NO_UID32))) {
4957 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4958 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4960 i_uid_write(inode, i_uid);
4961 i_gid_write(inode, i_gid);
4962 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4963 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4965 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4966 ei->i_inline_off = 0;
4967 ei->i_dir_start_lookup = 0;
4968 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4969 /* We now have enough fields to check if the inode was active or not.
4970 * This is needed because nfsd might try to access dead inodes
4971 * the test is that same one that e2fsck uses
4972 * NeilBrown 1999oct15
4974 if (inode->i_nlink == 0) {
4975 if ((inode->i_mode == 0 ||
4976 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4977 ino != EXT4_BOOT_LOADER_INO) {
4978 /* this inode is deleted */
4982 /* The only unlinked inodes we let through here have
4983 * valid i_mode and are being read by the orphan
4984 * recovery code: that's fine, we're about to complete
4985 * the process of deleting those.
4986 * OR it is the EXT4_BOOT_LOADER_INO which is
4987 * not initialized on a new filesystem. */
4989 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4990 ext4_set_inode_flags(inode);
4991 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4992 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4993 if (ext4_has_feature_64bit(sb))
4995 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4996 inode->i_size = ext4_isize(sb, raw_inode);
4997 if ((size = i_size_read(inode)) < 0) {
4998 ext4_error_inode(inode, function, line, 0,
4999 "iget: bad i_size value: %lld", size);
5000 ret = -EFSCORRUPTED;
5003 ei->i_disksize = inode->i_size;
5005 ei->i_reserved_quota = 0;
5007 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5008 ei->i_block_group = iloc.block_group;
5009 ei->i_last_alloc_group = ~0;
5011 * NOTE! The in-memory inode i_data array is in little-endian order
5012 * even on big-endian machines: we do NOT byteswap the block numbers!
5014 for (block = 0; block < EXT4_N_BLOCKS; block++)
5015 ei->i_data[block] = raw_inode->i_block[block];
5016 INIT_LIST_HEAD(&ei->i_orphan);
5019 * Set transaction id's of transactions that have to be committed
5020 * to finish f[data]sync. We set them to currently running transaction
5021 * as we cannot be sure that the inode or some of its metadata isn't
5022 * part of the transaction - the inode could have been reclaimed and
5023 * now it is reread from disk.
5026 transaction_t *transaction;
5029 read_lock(&journal->j_state_lock);
5030 if (journal->j_running_transaction)
5031 transaction = journal->j_running_transaction;
5033 transaction = journal->j_committing_transaction;
5035 tid = transaction->t_tid;
5037 tid = journal->j_commit_sequence;
5038 read_unlock(&journal->j_state_lock);
5039 ei->i_sync_tid = tid;
5040 ei->i_datasync_tid = tid;
5043 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5044 if (ei->i_extra_isize == 0) {
5045 /* The extra space is currently unused. Use it. */
5046 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5047 ei->i_extra_isize = sizeof(struct ext4_inode) -
5048 EXT4_GOOD_OLD_INODE_SIZE;
5050 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5056 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5057 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5058 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5059 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5061 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5062 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5064 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5065 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5067 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5069 ext4_inode_set_iversion_queried(inode, ivers);
5073 if (ei->i_file_acl &&
5074 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5075 ext4_error_inode(inode, function, line, 0,
5076 "iget: bad extended attribute block %llu",
5078 ret = -EFSCORRUPTED;
5080 } else if (!ext4_has_inline_data(inode)) {
5081 /* validate the block references in the inode */
5082 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5083 (S_ISLNK(inode->i_mode) &&
5084 !ext4_inode_is_fast_symlink(inode))) {
5085 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5086 ret = ext4_ext_check_inode(inode);
5088 ret = ext4_ind_check_inode(inode);
5094 if (S_ISREG(inode->i_mode)) {
5095 inode->i_op = &ext4_file_inode_operations;
5096 inode->i_fop = &ext4_file_operations;
5097 ext4_set_aops(inode);
5098 } else if (S_ISDIR(inode->i_mode)) {
5099 inode->i_op = &ext4_dir_inode_operations;
5100 inode->i_fop = &ext4_dir_operations;
5101 } else if (S_ISLNK(inode->i_mode)) {
5102 /* VFS does not allow setting these so must be corruption */
5103 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5104 ext4_error_inode(inode, function, line, 0,
5105 "iget: immutable or append flags "
5106 "not allowed on symlinks");
5107 ret = -EFSCORRUPTED;
5110 if (IS_ENCRYPTED(inode)) {
5111 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5112 ext4_set_aops(inode);
5113 } else if (ext4_inode_is_fast_symlink(inode)) {
5114 inode->i_link = (char *)ei->i_data;
5115 inode->i_op = &ext4_fast_symlink_inode_operations;
5116 nd_terminate_link(ei->i_data, inode->i_size,
5117 sizeof(ei->i_data) - 1);
5119 inode->i_op = &ext4_symlink_inode_operations;
5120 ext4_set_aops(inode);
5122 inode_nohighmem(inode);
5123 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5124 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5125 inode->i_op = &ext4_special_inode_operations;
5126 if (raw_inode->i_block[0])
5127 init_special_inode(inode, inode->i_mode,
5128 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5130 init_special_inode(inode, inode->i_mode,
5131 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5132 } else if (ino == EXT4_BOOT_LOADER_INO) {
5133 make_bad_inode(inode);
5135 ret = -EFSCORRUPTED;
5136 ext4_error_inode(inode, function, line, 0,
5137 "iget: bogus i_mode (%o)", inode->i_mode);
5142 unlock_new_inode(inode);
5148 return ERR_PTR(ret);
5151 static int ext4_inode_blocks_set(handle_t *handle,
5152 struct ext4_inode *raw_inode,
5153 struct ext4_inode_info *ei)
5155 struct inode *inode = &(ei->vfs_inode);
5156 u64 i_blocks = inode->i_blocks;
5157 struct super_block *sb = inode->i_sb;
5159 if (i_blocks <= ~0U) {
5161 * i_blocks can be represented in a 32 bit variable
5162 * as multiple of 512 bytes
5164 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5165 raw_inode->i_blocks_high = 0;
5166 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5169 if (!ext4_has_feature_huge_file(sb))
5172 if (i_blocks <= 0xffffffffffffULL) {
5174 * i_blocks can be represented in a 48 bit variable
5175 * as multiple of 512 bytes
5177 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5178 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5179 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5181 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5182 /* i_block is stored in file system block size */
5183 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5184 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5185 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5190 struct other_inode {
5191 unsigned long orig_ino;
5192 struct ext4_inode *raw_inode;
5195 static int other_inode_match(struct inode * inode, unsigned long ino,
5198 struct other_inode *oi = (struct other_inode *) data;
5200 if ((inode->i_ino != ino) ||
5201 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5203 ((inode->i_state & I_DIRTY_TIME) == 0))
5205 spin_lock(&inode->i_lock);
5206 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5207 I_DIRTY_INODE)) == 0) &&
5208 (inode->i_state & I_DIRTY_TIME)) {
5209 struct ext4_inode_info *ei = EXT4_I(inode);
5211 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5212 spin_unlock(&inode->i_lock);
5214 spin_lock(&ei->i_raw_lock);
5215 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5216 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5217 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5218 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5219 spin_unlock(&ei->i_raw_lock);
5220 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5223 spin_unlock(&inode->i_lock);
5228 * Opportunistically update the other time fields for other inodes in
5229 * the same inode table block.
5231 static void ext4_update_other_inodes_time(struct super_block *sb,
5232 unsigned long orig_ino, char *buf)
5234 struct other_inode oi;
5236 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5237 int inode_size = EXT4_INODE_SIZE(sb);
5239 oi.orig_ino = orig_ino;
5241 * Calculate the first inode in the inode table block. Inode
5242 * numbers are one-based. That is, the first inode in a block
5243 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5245 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5246 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5247 if (ino == orig_ino)
5249 oi.raw_inode = (struct ext4_inode *) buf;
5250 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5255 * Post the struct inode info into an on-disk inode location in the
5256 * buffer-cache. This gobbles the caller's reference to the
5257 * buffer_head in the inode location struct.
5259 * The caller must have write access to iloc->bh.
5261 static int ext4_do_update_inode(handle_t *handle,
5262 struct inode *inode,
5263 struct ext4_iloc *iloc)
5265 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5266 struct ext4_inode_info *ei = EXT4_I(inode);
5267 struct buffer_head *bh = iloc->bh;
5268 struct super_block *sb = inode->i_sb;
5269 int err = 0, rc, block;
5270 int need_datasync = 0, set_large_file = 0;
5275 spin_lock(&ei->i_raw_lock);
5277 /* For fields not tracked in the in-memory inode,
5278 * initialise them to zero for new inodes. */
5279 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5280 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5282 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5283 i_uid = i_uid_read(inode);
5284 i_gid = i_gid_read(inode);
5285 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5286 if (!(test_opt(inode->i_sb, NO_UID32))) {
5287 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5288 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5290 * Fix up interoperability with old kernels. Otherwise, old inodes get
5291 * re-used with the upper 16 bits of the uid/gid intact
5293 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5294 raw_inode->i_uid_high = 0;
5295 raw_inode->i_gid_high = 0;
5297 raw_inode->i_uid_high =
5298 cpu_to_le16(high_16_bits(i_uid));
5299 raw_inode->i_gid_high =
5300 cpu_to_le16(high_16_bits(i_gid));
5303 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5304 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5305 raw_inode->i_uid_high = 0;
5306 raw_inode->i_gid_high = 0;
5308 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5310 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5311 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5312 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5313 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5315 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5317 spin_unlock(&ei->i_raw_lock);
5320 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5321 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5322 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5323 raw_inode->i_file_acl_high =
5324 cpu_to_le16(ei->i_file_acl >> 32);
5325 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5326 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5327 ext4_isize_set(raw_inode, ei->i_disksize);
5330 if (ei->i_disksize > 0x7fffffffULL) {
5331 if (!ext4_has_feature_large_file(sb) ||
5332 EXT4_SB(sb)->s_es->s_rev_level ==
5333 cpu_to_le32(EXT4_GOOD_OLD_REV))
5336 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5337 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5338 if (old_valid_dev(inode->i_rdev)) {
5339 raw_inode->i_block[0] =
5340 cpu_to_le32(old_encode_dev(inode->i_rdev));
5341 raw_inode->i_block[1] = 0;
5343 raw_inode->i_block[0] = 0;
5344 raw_inode->i_block[1] =
5345 cpu_to_le32(new_encode_dev(inode->i_rdev));
5346 raw_inode->i_block[2] = 0;
5348 } else if (!ext4_has_inline_data(inode)) {
5349 for (block = 0; block < EXT4_N_BLOCKS; block++)
5350 raw_inode->i_block[block] = ei->i_data[block];
5353 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5354 u64 ivers = ext4_inode_peek_iversion(inode);
5356 raw_inode->i_disk_version = cpu_to_le32(ivers);
5357 if (ei->i_extra_isize) {
5358 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5359 raw_inode->i_version_hi =
5360 cpu_to_le32(ivers >> 32);
5361 raw_inode->i_extra_isize =
5362 cpu_to_le16(ei->i_extra_isize);
5366 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5367 i_projid != EXT4_DEF_PROJID);
5369 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5370 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5371 raw_inode->i_projid = cpu_to_le32(i_projid);
5373 ext4_inode_csum_set(inode, raw_inode, ei);
5374 spin_unlock(&ei->i_raw_lock);
5375 if (inode->i_sb->s_flags & SB_LAZYTIME)
5376 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5379 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5380 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5383 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5384 if (set_large_file) {
5385 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5386 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5389 ext4_set_feature_large_file(sb);
5390 ext4_handle_sync(handle);
5391 err = ext4_handle_dirty_super(handle, sb);
5393 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5396 ext4_std_error(inode->i_sb, err);
5401 * ext4_write_inode()
5403 * We are called from a few places:
5405 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5406 * Here, there will be no transaction running. We wait for any running
5407 * transaction to commit.
5409 * - Within flush work (sys_sync(), kupdate and such).
5410 * We wait on commit, if told to.
5412 * - Within iput_final() -> write_inode_now()
5413 * We wait on commit, if told to.
5415 * In all cases it is actually safe for us to return without doing anything,
5416 * because the inode has been copied into a raw inode buffer in
5417 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5420 * Note that we are absolutely dependent upon all inode dirtiers doing the
5421 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5422 * which we are interested.
5424 * It would be a bug for them to not do this. The code:
5426 * mark_inode_dirty(inode)
5428 * inode->i_size = expr;
5430 * is in error because write_inode() could occur while `stuff()' is running,
5431 * and the new i_size will be lost. Plus the inode will no longer be on the
5432 * superblock's dirty inode list.
5434 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5438 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5439 sb_rdonly(inode->i_sb))
5442 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5445 if (EXT4_SB(inode->i_sb)->s_journal) {
5446 if (ext4_journal_current_handle()) {
5447 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5453 * No need to force transaction in WB_SYNC_NONE mode. Also
5454 * ext4_sync_fs() will force the commit after everything is
5457 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5460 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5461 EXT4_I(inode)->i_sync_tid);
5463 struct ext4_iloc iloc;
5465 err = __ext4_get_inode_loc(inode, &iloc, 0);
5469 * sync(2) will flush the whole buffer cache. No need to do
5470 * it here separately for each inode.
5472 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5473 sync_dirty_buffer(iloc.bh);
5474 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5475 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5476 "IO error syncing inode");
5485 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5486 * buffers that are attached to a page stradding i_size and are undergoing
5487 * commit. In that case we have to wait for commit to finish and try again.
5489 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5493 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5494 tid_t commit_tid = 0;
5497 offset = inode->i_size & (PAGE_SIZE - 1);
5499 * All buffers in the last page remain valid? Then there's nothing to
5500 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5503 if (offset > PAGE_SIZE - i_blocksize(inode))
5506 page = find_lock_page(inode->i_mapping,
5507 inode->i_size >> PAGE_SHIFT);
5510 ret = __ext4_journalled_invalidatepage(page, offset,
5511 PAGE_SIZE - offset);
5517 read_lock(&journal->j_state_lock);
5518 if (journal->j_committing_transaction)
5519 commit_tid = journal->j_committing_transaction->t_tid;
5520 read_unlock(&journal->j_state_lock);
5522 jbd2_log_wait_commit(journal, commit_tid);
5529 * Called from notify_change.
5531 * We want to trap VFS attempts to truncate the file as soon as
5532 * possible. In particular, we want to make sure that when the VFS
5533 * shrinks i_size, we put the inode on the orphan list and modify
5534 * i_disksize immediately, so that during the subsequent flushing of
5535 * dirty pages and freeing of disk blocks, we can guarantee that any
5536 * commit will leave the blocks being flushed in an unused state on
5537 * disk. (On recovery, the inode will get truncated and the blocks will
5538 * be freed, so we have a strong guarantee that no future commit will
5539 * leave these blocks visible to the user.)
5541 * Another thing we have to assure is that if we are in ordered mode
5542 * and inode is still attached to the committing transaction, we must
5543 * we start writeout of all the dirty pages which are being truncated.
5544 * This way we are sure that all the data written in the previous
5545 * transaction are already on disk (truncate waits for pages under
5548 * Called with inode->i_mutex down.
5550 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5552 struct inode *inode = d_inode(dentry);
5555 const unsigned int ia_valid = attr->ia_valid;
5557 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5560 if (unlikely(IS_IMMUTABLE(inode)))
5563 if (unlikely(IS_APPEND(inode) &&
5564 (ia_valid & (ATTR_MODE | ATTR_UID |
5565 ATTR_GID | ATTR_TIMES_SET))))
5568 error = setattr_prepare(dentry, attr);
5572 error = fscrypt_prepare_setattr(dentry, attr);
5576 error = fsverity_prepare_setattr(dentry, attr);
5580 if (is_quota_modification(inode, attr)) {
5581 error = dquot_initialize(inode);
5585 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5586 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5589 /* (user+group)*(old+new) structure, inode write (sb,
5590 * inode block, ? - but truncate inode update has it) */
5591 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5592 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5593 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5594 if (IS_ERR(handle)) {
5595 error = PTR_ERR(handle);
5599 /* dquot_transfer() calls back ext4_get_inode_usage() which
5600 * counts xattr inode references.
5602 down_read(&EXT4_I(inode)->xattr_sem);
5603 error = dquot_transfer(inode, attr);
5604 up_read(&EXT4_I(inode)->xattr_sem);
5607 ext4_journal_stop(handle);
5610 /* Update corresponding info in inode so that everything is in
5611 * one transaction */
5612 if (attr->ia_valid & ATTR_UID)
5613 inode->i_uid = attr->ia_uid;
5614 if (attr->ia_valid & ATTR_GID)
5615 inode->i_gid = attr->ia_gid;
5616 error = ext4_mark_inode_dirty(handle, inode);
5617 ext4_journal_stop(handle);
5620 if (attr->ia_valid & ATTR_SIZE) {
5622 loff_t oldsize = inode->i_size;
5623 int shrink = (attr->ia_size < inode->i_size);
5625 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5626 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5628 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5631 if (!S_ISREG(inode->i_mode))
5634 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5635 inode_inc_iversion(inode);
5638 if (ext4_should_order_data(inode)) {
5639 error = ext4_begin_ordered_truncate(inode,
5645 * Blocks are going to be removed from the inode. Wait
5646 * for dio in flight.
5648 inode_dio_wait(inode);
5651 down_write(&EXT4_I(inode)->i_mmap_sem);
5653 rc = ext4_break_layouts(inode);
5655 up_write(&EXT4_I(inode)->i_mmap_sem);
5659 if (attr->ia_size != inode->i_size) {
5660 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5661 if (IS_ERR(handle)) {
5662 error = PTR_ERR(handle);
5665 if (ext4_handle_valid(handle) && shrink) {
5666 error = ext4_orphan_add(handle, inode);
5670 * Update c/mtime on truncate up, ext4_truncate() will
5671 * update c/mtime in shrink case below
5674 inode->i_mtime = current_time(inode);
5675 inode->i_ctime = inode->i_mtime;
5677 down_write(&EXT4_I(inode)->i_data_sem);
5678 EXT4_I(inode)->i_disksize = attr->ia_size;
5679 rc = ext4_mark_inode_dirty(handle, inode);
5683 * We have to update i_size under i_data_sem together
5684 * with i_disksize to avoid races with writeback code
5685 * running ext4_wb_update_i_disksize().
5688 i_size_write(inode, attr->ia_size);
5689 up_write(&EXT4_I(inode)->i_data_sem);
5690 ext4_journal_stop(handle);
5694 pagecache_isize_extended(inode, oldsize,
5696 } else if (ext4_should_journal_data(inode)) {
5697 ext4_wait_for_tail_page_commit(inode);
5702 * Truncate pagecache after we've waited for commit
5703 * in data=journal mode to make pages freeable.
5705 truncate_pagecache(inode, inode->i_size);
5707 * Call ext4_truncate() even if i_size didn't change to
5708 * truncate possible preallocated blocks.
5710 if (attr->ia_size <= oldsize) {
5711 rc = ext4_truncate(inode);
5716 up_write(&EXT4_I(inode)->i_mmap_sem);
5720 setattr_copy(inode, attr);
5721 mark_inode_dirty(inode);
5725 * If the call to ext4_truncate failed to get a transaction handle at
5726 * all, we need to clean up the in-core orphan list manually.
5728 if (orphan && inode->i_nlink)
5729 ext4_orphan_del(NULL, inode);
5731 if (!error && (ia_valid & ATTR_MODE))
5732 rc = posix_acl_chmod(inode, inode->i_mode);
5735 ext4_std_error(inode->i_sb, error);
5741 int ext4_getattr(const struct path *path, struct kstat *stat,
5742 u32 request_mask, unsigned int query_flags)
5744 struct inode *inode = d_inode(path->dentry);
5745 struct ext4_inode *raw_inode;
5746 struct ext4_inode_info *ei = EXT4_I(inode);
5749 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5750 stat->result_mask |= STATX_BTIME;
5751 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5752 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5755 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5756 if (flags & EXT4_APPEND_FL)
5757 stat->attributes |= STATX_ATTR_APPEND;
5758 if (flags & EXT4_COMPR_FL)
5759 stat->attributes |= STATX_ATTR_COMPRESSED;
5760 if (flags & EXT4_ENCRYPT_FL)
5761 stat->attributes |= STATX_ATTR_ENCRYPTED;
5762 if (flags & EXT4_IMMUTABLE_FL)
5763 stat->attributes |= STATX_ATTR_IMMUTABLE;
5764 if (flags & EXT4_NODUMP_FL)
5765 stat->attributes |= STATX_ATTR_NODUMP;
5767 stat->attributes_mask |= (STATX_ATTR_APPEND |
5768 STATX_ATTR_COMPRESSED |
5769 STATX_ATTR_ENCRYPTED |
5770 STATX_ATTR_IMMUTABLE |
5773 generic_fillattr(inode, stat);
5777 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5778 u32 request_mask, unsigned int query_flags)
5780 struct inode *inode = d_inode(path->dentry);
5781 u64 delalloc_blocks;
5783 ext4_getattr(path, stat, request_mask, query_flags);
5786 * If there is inline data in the inode, the inode will normally not
5787 * have data blocks allocated (it may have an external xattr block).
5788 * Report at least one sector for such files, so tools like tar, rsync,
5789 * others don't incorrectly think the file is completely sparse.
5791 if (unlikely(ext4_has_inline_data(inode)))
5792 stat->blocks += (stat->size + 511) >> 9;
5795 * We can't update i_blocks if the block allocation is delayed
5796 * otherwise in the case of system crash before the real block
5797 * allocation is done, we will have i_blocks inconsistent with
5798 * on-disk file blocks.
5799 * We always keep i_blocks updated together with real
5800 * allocation. But to not confuse with user, stat
5801 * will return the blocks that include the delayed allocation
5802 * blocks for this file.
5804 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5805 EXT4_I(inode)->i_reserved_data_blocks);
5806 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5810 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5813 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5814 return ext4_ind_trans_blocks(inode, lblocks);
5815 return ext4_ext_index_trans_blocks(inode, pextents);
5819 * Account for index blocks, block groups bitmaps and block group
5820 * descriptor blocks if modify datablocks and index blocks
5821 * worse case, the indexs blocks spread over different block groups
5823 * If datablocks are discontiguous, they are possible to spread over
5824 * different block groups too. If they are contiguous, with flexbg,
5825 * they could still across block group boundary.
5827 * Also account for superblock, inode, quota and xattr blocks
5829 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5832 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5838 * How many index blocks need to touch to map @lblocks logical blocks
5839 * to @pextents physical extents?
5841 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5846 * Now let's see how many group bitmaps and group descriptors need
5849 groups = idxblocks + pextents;
5851 if (groups > ngroups)
5853 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5854 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5856 /* bitmaps and block group descriptor blocks */
5857 ret += groups + gdpblocks;
5859 /* Blocks for super block, inode, quota and xattr blocks */
5860 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5866 * Calculate the total number of credits to reserve to fit
5867 * the modification of a single pages into a single transaction,
5868 * which may include multiple chunks of block allocations.
5870 * This could be called via ext4_write_begin()
5872 * We need to consider the worse case, when
5873 * one new block per extent.
5875 int ext4_writepage_trans_blocks(struct inode *inode)
5877 int bpp = ext4_journal_blocks_per_page(inode);
5880 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5882 /* Account for data blocks for journalled mode */
5883 if (ext4_should_journal_data(inode))
5889 * Calculate the journal credits for a chunk of data modification.
5891 * This is called from DIO, fallocate or whoever calling
5892 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5894 * journal buffers for data blocks are not included here, as DIO
5895 * and fallocate do no need to journal data buffers.
5897 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5899 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5903 * The caller must have previously called ext4_reserve_inode_write().
5904 * Give this, we know that the caller already has write access to iloc->bh.
5906 int ext4_mark_iloc_dirty(handle_t *handle,
5907 struct inode *inode, struct ext4_iloc *iloc)
5911 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5915 if (IS_I_VERSION(inode))
5916 inode_inc_iversion(inode);
5918 /* the do_update_inode consumes one bh->b_count */
5921 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5922 err = ext4_do_update_inode(handle, inode, iloc);
5928 * On success, We end up with an outstanding reference count against
5929 * iloc->bh. This _must_ be cleaned up later.
5933 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5934 struct ext4_iloc *iloc)
5938 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5941 err = ext4_get_inode_loc(inode, iloc);
5943 BUFFER_TRACE(iloc->bh, "get_write_access");
5944 err = ext4_journal_get_write_access(handle, iloc->bh);
5950 ext4_std_error(inode->i_sb, err);
5954 static int __ext4_expand_extra_isize(struct inode *inode,
5955 unsigned int new_extra_isize,
5956 struct ext4_iloc *iloc,
5957 handle_t *handle, int *no_expand)
5959 struct ext4_inode *raw_inode;
5960 struct ext4_xattr_ibody_header *header;
5963 raw_inode = ext4_raw_inode(iloc);
5965 header = IHDR(inode, raw_inode);
5967 /* No extended attributes present */
5968 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5969 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5970 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5971 EXT4_I(inode)->i_extra_isize, 0,
5972 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5973 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5977 /* try to expand with EAs present */
5978 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5982 * Inode size expansion failed; don't try again
5991 * Expand an inode by new_extra_isize bytes.
5992 * Returns 0 on success or negative error number on failure.
5994 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5995 unsigned int new_extra_isize,
5996 struct ext4_iloc iloc,
6002 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
6006 * In nojournal mode, we can immediately attempt to expand
6007 * the inode. When journaled, we first need to obtain extra
6008 * buffer credits since we may write into the EA block
6009 * with this same handle. If journal_extend fails, then it will
6010 * only result in a minor loss of functionality for that inode.
6011 * If this is felt to be critical, then e2fsck should be run to
6012 * force a large enough s_min_extra_isize.
6014 if (ext4_handle_valid(handle) &&
6015 jbd2_journal_extend(handle,
6016 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
6019 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
6022 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
6023 handle, &no_expand);
6024 ext4_write_unlock_xattr(inode, &no_expand);
6029 int ext4_expand_extra_isize(struct inode *inode,
6030 unsigned int new_extra_isize,
6031 struct ext4_iloc *iloc)
6037 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
6042 handle = ext4_journal_start(inode, EXT4_HT_INODE,
6043 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6044 if (IS_ERR(handle)) {
6045 error = PTR_ERR(handle);
6050 ext4_write_lock_xattr(inode, &no_expand);
6052 BUFFER_TRACE(iloc->bh, "get_write_access");
6053 error = ext4_journal_get_write_access(handle, iloc->bh);
6059 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6060 handle, &no_expand);
6062 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6066 ext4_write_unlock_xattr(inode, &no_expand);
6068 ext4_journal_stop(handle);
6073 * What we do here is to mark the in-core inode as clean with respect to inode
6074 * dirtiness (it may still be data-dirty).
6075 * This means that the in-core inode may be reaped by prune_icache
6076 * without having to perform any I/O. This is a very good thing,
6077 * because *any* task may call prune_icache - even ones which
6078 * have a transaction open against a different journal.
6080 * Is this cheating? Not really. Sure, we haven't written the
6081 * inode out, but prune_icache isn't a user-visible syncing function.
6082 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6083 * we start and wait on commits.
6085 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6087 struct ext4_iloc iloc;
6088 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6092 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6093 err = ext4_reserve_inode_write(handle, inode, &iloc);
6097 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6098 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6101 return ext4_mark_iloc_dirty(handle, inode, &iloc);
6105 * ext4_dirty_inode() is called from __mark_inode_dirty()
6107 * We're really interested in the case where a file is being extended.
6108 * i_size has been changed by generic_commit_write() and we thus need
6109 * to include the updated inode in the current transaction.
6111 * Also, dquot_alloc_block() will always dirty the inode when blocks
6112 * are allocated to the file.
6114 * If the inode is marked synchronous, we don't honour that here - doing
6115 * so would cause a commit on atime updates, which we don't bother doing.
6116 * We handle synchronous inodes at the highest possible level.
6118 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6119 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6120 * to copy into the on-disk inode structure are the timestamp files.
6122 void ext4_dirty_inode(struct inode *inode, int flags)
6126 if (flags == I_DIRTY_TIME)
6128 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6132 ext4_mark_inode_dirty(handle, inode);
6134 ext4_journal_stop(handle);
6139 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6144 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6147 * We have to be very careful here: changing a data block's
6148 * journaling status dynamically is dangerous. If we write a
6149 * data block to the journal, change the status and then delete
6150 * that block, we risk forgetting to revoke the old log record
6151 * from the journal and so a subsequent replay can corrupt data.
6152 * So, first we make sure that the journal is empty and that
6153 * nobody is changing anything.
6156 journal = EXT4_JOURNAL(inode);
6159 if (is_journal_aborted(journal))
6162 /* Wait for all existing dio workers */
6163 inode_dio_wait(inode);
6166 * Before flushing the journal and switching inode's aops, we have
6167 * to flush all dirty data the inode has. There can be outstanding
6168 * delayed allocations, there can be unwritten extents created by
6169 * fallocate or buffered writes in dioread_nolock mode covered by
6170 * dirty data which can be converted only after flushing the dirty
6171 * data (and journalled aops don't know how to handle these cases).
6174 down_write(&EXT4_I(inode)->i_mmap_sem);
6175 err = filemap_write_and_wait(inode->i_mapping);
6177 up_write(&EXT4_I(inode)->i_mmap_sem);
6182 percpu_down_write(&sbi->s_journal_flag_rwsem);
6183 jbd2_journal_lock_updates(journal);
6186 * OK, there are no updates running now, and all cached data is
6187 * synced to disk. We are now in a completely consistent state
6188 * which doesn't have anything in the journal, and we know that
6189 * no filesystem updates are running, so it is safe to modify
6190 * the inode's in-core data-journaling state flag now.
6194 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6196 err = jbd2_journal_flush(journal);
6198 jbd2_journal_unlock_updates(journal);
6199 percpu_up_write(&sbi->s_journal_flag_rwsem);
6202 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6204 ext4_set_aops(inode);
6206 jbd2_journal_unlock_updates(journal);
6207 percpu_up_write(&sbi->s_journal_flag_rwsem);
6210 up_write(&EXT4_I(inode)->i_mmap_sem);
6212 /* Finally we can mark the inode as dirty. */
6214 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6216 return PTR_ERR(handle);
6218 err = ext4_mark_inode_dirty(handle, inode);
6219 ext4_handle_sync(handle);
6220 ext4_journal_stop(handle);
6221 ext4_std_error(inode->i_sb, err);
6226 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6228 return !buffer_mapped(bh);
6231 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6233 struct vm_area_struct *vma = vmf->vma;
6234 struct page *page = vmf->page;
6239 struct file *file = vma->vm_file;
6240 struct inode *inode = file_inode(file);
6241 struct address_space *mapping = inode->i_mapping;
6243 get_block_t *get_block;
6246 if (unlikely(IS_IMMUTABLE(inode)))
6247 return VM_FAULT_SIGBUS;
6249 sb_start_pagefault(inode->i_sb);
6250 file_update_time(vma->vm_file);
6252 down_read(&EXT4_I(inode)->i_mmap_sem);
6254 err = ext4_convert_inline_data(inode);
6258 /* Delalloc case is easy... */
6259 if (test_opt(inode->i_sb, DELALLOC) &&
6260 !ext4_should_journal_data(inode) &&
6261 !ext4_nonda_switch(inode->i_sb)) {
6263 err = block_page_mkwrite(vma, vmf,
6264 ext4_da_get_block_prep);
6265 } while (err == -ENOSPC &&
6266 ext4_should_retry_alloc(inode->i_sb, &retries));
6271 size = i_size_read(inode);
6272 /* Page got truncated from under us? */
6273 if (page->mapping != mapping || page_offset(page) > size) {
6275 ret = VM_FAULT_NOPAGE;
6279 if (page->index == size >> PAGE_SHIFT)
6280 len = size & ~PAGE_MASK;
6284 * Return if we have all the buffers mapped. This avoids the need to do
6285 * journal_start/journal_stop which can block and take a long time
6287 if (page_has_buffers(page)) {
6288 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6290 ext4_bh_unmapped)) {
6291 /* Wait so that we don't change page under IO */
6292 wait_for_stable_page(page);
6293 ret = VM_FAULT_LOCKED;
6298 /* OK, we need to fill the hole... */
6299 if (ext4_should_dioread_nolock(inode))
6300 get_block = ext4_get_block_unwritten;
6302 get_block = ext4_get_block;
6304 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6305 ext4_writepage_trans_blocks(inode));
6306 if (IS_ERR(handle)) {
6307 ret = VM_FAULT_SIGBUS;
6310 err = block_page_mkwrite(vma, vmf, get_block);
6311 if (!err && ext4_should_journal_data(inode)) {
6312 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6313 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6315 ret = VM_FAULT_SIGBUS;
6316 ext4_journal_stop(handle);
6319 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6321 ext4_journal_stop(handle);
6322 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6325 ret = block_page_mkwrite_return(err);
6327 up_read(&EXT4_I(inode)->i_mmap_sem);
6328 sb_end_pagefault(inode->i_sb);
6332 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6334 struct inode *inode = file_inode(vmf->vma->vm_file);
6337 down_read(&EXT4_I(inode)->i_mmap_sem);
6338 ret = filemap_fault(vmf);
6339 up_read(&EXT4_I(inode)->i_mmap_sem);