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/mount.h>
24 #include <linux/time.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/dax.h>
28 #include <linux/quotaops.h>
29 #include <linux/string.h>
30 #include <linux/buffer_head.h>
31 #include <linux/writeback.h>
32 #include <linux/pagevec.h>
33 #include <linux/mpage.h>
34 #include <linux/namei.h>
35 #include <linux/uio.h>
36 #include <linux/bio.h>
37 #include <linux/workqueue.h>
38 #include <linux/kernel.h>
39 #include <linux/printk.h>
40 #include <linux/slab.h>
41 #include <linux/bitops.h>
42 #include <linux/iomap.h>
43 #include <linux/iversion.h>
45 #include "ext4_jbd2.h"
50 #include <trace/events/ext4.h>
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 int offset = offsetof(struct ext4_inode, i_checksum_lo);
59 unsigned int csum_size = sizeof(dummy_csum);
61 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
62 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
64 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
65 EXT4_GOOD_OLD_INODE_SIZE - offset);
67 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
68 offset = offsetof(struct ext4_inode, i_checksum_hi);
69 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
70 EXT4_GOOD_OLD_INODE_SIZE,
71 offset - EXT4_GOOD_OLD_INODE_SIZE);
72 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
73 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
77 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
78 EXT4_INODE_SIZE(inode->i_sb) - offset);
84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 struct ext4_inode_info *ei)
87 __u32 provided, calculated;
89 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 cpu_to_le32(EXT4_OS_LINUX) ||
91 !ext4_has_metadata_csum(inode->i_sb))
94 provided = le16_to_cpu(raw->i_checksum_lo);
95 calculated = ext4_inode_csum(inode, raw, ei);
96 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100 calculated &= 0xFFFF;
102 return provided == calculated;
105 void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 struct ext4_inode_info *ei)
110 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 cpu_to_le32(EXT4_OS_LINUX) ||
112 !ext4_has_metadata_csum(inode->i_sb))
115 csum = ext4_inode_csum(inode, raw, ei);
116 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
125 trace_ext4_begin_ordered_truncate(inode, new_size);
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
132 if (!EXT4_I(inode)->jinode)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 EXT4_I(inode)->jinode,
139 static void ext4_invalidatepage(struct page *page, unsigned int offset,
140 unsigned int length);
141 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
142 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
143 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
147 * Test whether an inode is a fast symlink.
148 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 int ext4_inode_is_fast_symlink(struct inode *inode)
152 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
153 int ea_blocks = EXT4_I(inode)->i_file_acl ?
154 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
156 if (ext4_has_inline_data(inode))
159 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 return S_ISLNK(inode->i_mode) && inode->i_size &&
162 (inode->i_size < EXT4_N_BLOCKS * 4);
166 * Called at the last iput() if i_nlink is zero.
168 void ext4_evict_inode(struct inode *inode)
173 * Credits for final inode cleanup and freeing:
174 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
175 * (xattr block freeing), bitmap, group descriptor (inode freeing)
177 int extra_credits = 6;
178 struct ext4_xattr_inode_array *ea_inode_array = NULL;
179 bool freeze_protected = false;
181 trace_ext4_evict_inode(inode);
183 if (inode->i_nlink) {
185 * When journalling data dirty buffers are tracked only in the
186 * journal. So although mm thinks everything is clean and
187 * ready for reaping the inode might still have some pages to
188 * write in the running transaction or waiting to be
189 * checkpointed. Thus calling jbd2_journal_invalidatepage()
190 * (via truncate_inode_pages()) to discard these buffers can
191 * cause data loss. Also even if we did not discard these
192 * buffers, we would have no way to find them after the inode
193 * is reaped and thus user could see stale data if he tries to
194 * read them before the transaction is checkpointed. So be
195 * careful and force everything to disk here... We use
196 * ei->i_datasync_tid to store the newest transaction
197 * containing inode's data.
199 * Note that directories do not have this problem because they
200 * don't use page cache.
202 if (inode->i_ino != EXT4_JOURNAL_INO &&
203 ext4_should_journal_data(inode) &&
204 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
205 inode->i_data.nrpages) {
206 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
207 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
209 jbd2_complete_transaction(journal, commit_tid);
210 filemap_write_and_wait(&inode->i_data);
212 truncate_inode_pages_final(&inode->i_data);
217 if (is_bad_inode(inode))
219 dquot_initialize(inode);
221 if (ext4_should_order_data(inode))
222 ext4_begin_ordered_truncate(inode, 0);
223 truncate_inode_pages_final(&inode->i_data);
226 * For inodes with journalled data, transaction commit could have
227 * dirtied the inode. Flush worker is ignoring it because of I_FREEING
228 * flag but we still need to remove the inode from the writeback lists.
230 if (!list_empty_careful(&inode->i_io_list)) {
231 WARN_ON_ONCE(!ext4_should_journal_data(inode));
232 inode_io_list_del(inode);
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it. When we are in a running transaction though,
238 * we are already protected against freezing and we cannot grab further
239 * protection due to lock ordering constraints.
241 if (!ext4_journal_current_handle()) {
242 sb_start_intwrite(inode->i_sb);
243 freeze_protected = true;
246 if (!IS_NOQUOTA(inode))
247 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
250 * Block bitmap, group descriptor, and inode are accounted in both
251 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
253 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
254 ext4_blocks_for_truncate(inode) + extra_credits - 3);
255 if (IS_ERR(handle)) {
256 ext4_std_error(inode->i_sb, PTR_ERR(handle));
258 * If we're going to skip the normal cleanup, we still need to
259 * make sure that the in-core orphan linked list is properly
262 ext4_orphan_del(NULL, inode);
263 if (freeze_protected)
264 sb_end_intwrite(inode->i_sb);
269 ext4_handle_sync(handle);
272 * Set inode->i_size to 0 before calling ext4_truncate(). We need
273 * special handling of symlinks here because i_size is used to
274 * determine whether ext4_inode_info->i_data contains symlink data or
275 * block mappings. Setting i_size to 0 will remove its fast symlink
276 * status. Erase i_data so that it becomes a valid empty block map.
278 if (ext4_inode_is_fast_symlink(inode))
279 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
281 err = ext4_mark_inode_dirty(handle, inode);
283 ext4_warning(inode->i_sb,
284 "couldn't mark inode dirty (err %d)", err);
287 if (inode->i_blocks) {
288 err = ext4_truncate(inode);
290 ext4_error_err(inode->i_sb, -err,
291 "couldn't truncate inode %lu (err %d)",
297 /* Remove xattr references. */
298 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
301 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
303 ext4_journal_stop(handle);
304 ext4_orphan_del(NULL, inode);
305 if (freeze_protected)
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 if (freeze_protected)
336 sb_end_intwrite(inode->i_sb);
337 ext4_xattr_inode_array_free(ea_inode_array);
340 if (!list_empty(&EXT4_I(inode)->i_fc_list))
341 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM);
342 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
346 qsize_t *ext4_get_reserved_space(struct inode *inode)
348 return &EXT4_I(inode)->i_reserved_quota;
353 * Called with i_data_sem down, which is important since we can call
354 * ext4_discard_preallocations() from here.
356 void ext4_da_update_reserve_space(struct inode *inode,
357 int used, int quota_claim)
359 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
360 struct ext4_inode_info *ei = EXT4_I(inode);
362 spin_lock(&ei->i_block_reservation_lock);
363 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
364 if (unlikely(used > ei->i_reserved_data_blocks)) {
365 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
366 "with only %d reserved data blocks",
367 __func__, inode->i_ino, used,
368 ei->i_reserved_data_blocks);
370 used = ei->i_reserved_data_blocks;
373 /* Update per-inode reservations */
374 ei->i_reserved_data_blocks -= used;
375 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
377 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
379 /* Update quota subsystem for data blocks */
381 dquot_claim_block(inode, EXT4_C2B(sbi, used));
384 * We did fallocate with an offset that is already delayed
385 * allocated. So on delayed allocated writeback we should
386 * not re-claim the quota for fallocated blocks.
388 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
392 * If we have done all the pending block allocations and if
393 * there aren't any writers on the inode, we can discard the
394 * inode's preallocations.
396 if ((ei->i_reserved_data_blocks == 0) &&
397 !inode_is_open_for_write(inode))
398 ext4_discard_preallocations(inode, 0);
401 static int __check_block_validity(struct inode *inode, const char *func,
403 struct ext4_map_blocks *map)
405 if (ext4_has_feature_journal(inode->i_sb) &&
407 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
409 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
410 ext4_error_inode(inode, func, line, map->m_pblk,
411 "lblock %lu mapped to illegal pblock %llu "
412 "(length %d)", (unsigned long) map->m_lblk,
413 map->m_pblk, map->m_len);
414 return -EFSCORRUPTED;
419 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
424 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
425 return fscrypt_zeroout_range(inode, lblk, pblk, len);
427 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
434 #define check_block_validity(inode, map) \
435 __check_block_validity((inode), __func__, __LINE__, (map))
437 #ifdef ES_AGGRESSIVE_TEST
438 static void ext4_map_blocks_es_recheck(handle_t *handle,
440 struct ext4_map_blocks *es_map,
441 struct ext4_map_blocks *map,
448 * There is a race window that the result is not the same.
449 * e.g. xfstests #223 when dioread_nolock enables. The reason
450 * is that we lookup a block mapping in extent status tree with
451 * out taking i_data_sem. So at the time the unwritten extent
452 * could be converted.
454 down_read(&EXT4_I(inode)->i_data_sem);
455 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
456 retval = ext4_ext_map_blocks(handle, inode, map, 0);
458 retval = ext4_ind_map_blocks(handle, inode, map, 0);
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(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
516 flags, map->m_len, (unsigned long) map->m_lblk);
519 * ext4_map_blocks returns an int, and m_len is an unsigned int
521 if (unlikely(map->m_len > INT_MAX))
522 map->m_len = INT_MAX;
524 /* We can handle the block number less than EXT_MAX_BLOCKS */
525 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
526 return -EFSCORRUPTED;
528 /* Lookup extent status tree firstly */
529 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
530 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &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, 0);
565 retval = ext4_ind_map_blocks(handle, inode, map, 0);
570 if (unlikely(retval != map->m_len)) {
571 ext4_warning(inode->i_sb,
572 "ES len assertion failed for inode "
573 "%lu: retval %d != map->m_len %d",
574 inode->i_ino, retval, map->m_len);
578 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
579 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
580 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
581 !(status & EXTENT_STATUS_WRITTEN) &&
582 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
583 map->m_lblk + map->m_len - 1))
584 status |= EXTENT_STATUS_DELAYED;
585 ret = ext4_es_insert_extent(inode, map->m_lblk,
586 map->m_len, map->m_pblk, status);
590 up_read((&EXT4_I(inode)->i_data_sem));
593 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
594 ret = check_block_validity(inode, map);
599 /* If it is only a block(s) look up */
600 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
604 * Returns if the blocks have already allocated
606 * Note that if blocks have been preallocated
607 * ext4_ext_get_block() returns the create = 0
608 * with buffer head unmapped.
610 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
612 * If we need to convert extent to unwritten
613 * we continue and do the actual work in
614 * ext4_ext_map_blocks()
616 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
620 * Here we clear m_flags because after allocating an new extent,
621 * it will be set again.
623 map->m_flags &= ~EXT4_MAP_FLAGS;
626 * New blocks allocate and/or writing to unwritten extent
627 * will possibly result in updating i_data, so we take
628 * the write lock of i_data_sem, and call get_block()
629 * with create == 1 flag.
631 down_write(&EXT4_I(inode)->i_data_sem);
634 * We need to check for EXT4 here because migrate
635 * could have changed the inode type in between
637 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
638 retval = ext4_ext_map_blocks(handle, inode, map, flags);
640 retval = ext4_ind_map_blocks(handle, inode, map, flags);
642 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
644 * We allocated new blocks which will result in
645 * i_data's format changing. Force the migrate
646 * to fail by clearing migrate flags
648 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
652 * Update reserved blocks/metadata blocks after successful
653 * block allocation which had been deferred till now. We don't
654 * support fallocate for non extent files. So we can update
655 * reserve space here.
658 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
659 ext4_da_update_reserve_space(inode, retval, 1);
665 if (unlikely(retval != map->m_len)) {
666 ext4_warning(inode->i_sb,
667 "ES len assertion failed for inode "
668 "%lu: retval %d != map->m_len %d",
669 inode->i_ino, retval, map->m_len);
674 * We have to zeroout blocks before inserting them into extent
675 * status tree. Otherwise someone could look them up there and
676 * use them before they are really zeroed. We also have to
677 * unmap metadata before zeroing as otherwise writeback can
678 * overwrite zeros with stale data from block device.
680 if (flags & EXT4_GET_BLOCKS_ZERO &&
681 map->m_flags & EXT4_MAP_MAPPED &&
682 map->m_flags & EXT4_MAP_NEW) {
683 ret = ext4_issue_zeroout(inode, map->m_lblk,
684 map->m_pblk, map->m_len);
692 * If the extent has been zeroed out, we don't need to update
693 * extent status tree.
695 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
696 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
697 if (ext4_es_is_written(&es))
700 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
701 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
702 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
703 !(status & EXTENT_STATUS_WRITTEN) &&
704 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
705 map->m_lblk + map->m_len - 1))
706 status |= EXTENT_STATUS_DELAYED;
707 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
708 map->m_pblk, status);
716 up_write((&EXT4_I(inode)->i_data_sem));
717 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
718 ret = check_block_validity(inode, map);
723 * Inodes with freshly allocated blocks where contents will be
724 * visible after transaction commit must be on transaction's
727 if (map->m_flags & EXT4_MAP_NEW &&
728 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
729 !(flags & EXT4_GET_BLOCKS_ZERO) &&
730 !ext4_is_quota_file(inode) &&
731 ext4_should_order_data(inode)) {
733 (loff_t)map->m_lblk << inode->i_blkbits;
734 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
736 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
737 ret = ext4_jbd2_inode_add_wait(handle, inode,
740 ret = ext4_jbd2_inode_add_write(handle, inode,
745 ext4_fc_track_range(handle, inode, map->m_lblk,
746 map->m_lblk + map->m_len - 1);
750 ext_debug(inode, "failed with err %d\n", retval);
755 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
756 * we have to be careful as someone else may be manipulating b_state as well.
758 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
760 unsigned long old_state;
761 unsigned long new_state;
763 flags &= EXT4_MAP_FLAGS;
765 /* Dummy buffer_head? Set non-atomically. */
767 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
771 * Someone else may be modifying b_state. Be careful! This is ugly but
772 * once we get rid of using bh as a container for mapping information
773 * to pass to / from get_block functions, this can go away.
776 old_state = READ_ONCE(bh->b_state);
777 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
779 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
782 static int _ext4_get_block(struct inode *inode, sector_t iblock,
783 struct buffer_head *bh, int flags)
785 struct ext4_map_blocks map;
788 if (ext4_has_inline_data(inode))
792 map.m_len = bh->b_size >> inode->i_blkbits;
794 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
797 map_bh(bh, inode->i_sb, map.m_pblk);
798 ext4_update_bh_state(bh, map.m_flags);
799 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
801 } else if (ret == 0) {
802 /* hole case, need to fill in bh->b_size */
803 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
808 int ext4_get_block(struct inode *inode, sector_t iblock,
809 struct buffer_head *bh, int create)
811 return _ext4_get_block(inode, iblock, bh,
812 create ? EXT4_GET_BLOCKS_CREATE : 0);
816 * Get block function used when preparing for buffered write if we require
817 * creating an unwritten extent if blocks haven't been allocated. The extent
818 * will be converted to written after the IO is complete.
820 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
821 struct buffer_head *bh_result, int create)
823 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
824 inode->i_ino, create);
825 return _ext4_get_block(inode, iblock, bh_result,
826 EXT4_GET_BLOCKS_IO_CREATE_EXT);
829 /* Maximum number of blocks we map for direct IO at once. */
830 #define DIO_MAX_BLOCKS 4096
833 * `handle' can be NULL if create is zero
835 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
836 ext4_lblk_t block, int map_flags)
838 struct ext4_map_blocks map;
839 struct buffer_head *bh;
840 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
843 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
844 || handle != NULL || create == 0);
848 err = ext4_map_blocks(handle, inode, &map, map_flags);
851 return create ? ERR_PTR(-ENOSPC) : NULL;
855 bh = sb_getblk(inode->i_sb, map.m_pblk);
857 return ERR_PTR(-ENOMEM);
858 if (map.m_flags & EXT4_MAP_NEW) {
860 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
861 || (handle != NULL));
864 * Now that we do not always journal data, we should
865 * keep in mind whether this should always journal the
866 * new buffer as metadata. For now, regular file
867 * writes use ext4_get_block instead, so it's not a
871 BUFFER_TRACE(bh, "call get_create_access");
872 err = ext4_journal_get_create_access(handle, bh);
877 if (!buffer_uptodate(bh)) {
878 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
879 set_buffer_uptodate(bh);
882 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
883 err = ext4_handle_dirty_metadata(handle, inode, bh);
887 BUFFER_TRACE(bh, "not a new buffer");
894 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
895 ext4_lblk_t block, int map_flags)
897 struct buffer_head *bh;
900 bh = ext4_getblk(handle, inode, block, map_flags);
903 if (!bh || ext4_buffer_uptodate(bh))
906 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
914 /* Read a contiguous batch of blocks. */
915 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
916 bool wait, struct buffer_head **bhs)
920 for (i = 0; i < bh_count; i++) {
921 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
922 if (IS_ERR(bhs[i])) {
923 err = PTR_ERR(bhs[i]);
929 for (i = 0; i < bh_count; i++)
930 /* Note that NULL bhs[i] is valid because of holes. */
931 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
932 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
937 for (i = 0; i < bh_count; i++)
939 wait_on_buffer(bhs[i]);
941 for (i = 0; i < bh_count; i++) {
942 if (bhs[i] && !buffer_uptodate(bhs[i])) {
950 for (i = 0; i < bh_count; i++) {
957 int ext4_walk_page_buffers(handle_t *handle,
958 struct buffer_head *head,
962 int (*fn)(handle_t *handle,
963 struct buffer_head *bh))
965 struct buffer_head *bh;
966 unsigned block_start, block_end;
967 unsigned blocksize = head->b_size;
969 struct buffer_head *next;
971 for (bh = head, block_start = 0;
972 ret == 0 && (bh != head || !block_start);
973 block_start = block_end, bh = next) {
974 next = bh->b_this_page;
975 block_end = block_start + blocksize;
976 if (block_end <= from || block_start >= to) {
977 if (partial && !buffer_uptodate(bh))
981 err = (*fn)(handle, bh);
989 * To preserve ordering, it is essential that the hole instantiation and
990 * the data write be encapsulated in a single transaction. We cannot
991 * close off a transaction and start a new one between the ext4_get_block()
992 * and the commit_write(). So doing the jbd2_journal_start at the start of
993 * prepare_write() is the right place.
995 * Also, this function can nest inside ext4_writepage(). In that case, we
996 * *know* that ext4_writepage() has generated enough buffer credits to do the
997 * whole page. So we won't block on the journal in that case, which is good,
998 * because the caller may be PF_MEMALLOC.
1000 * By accident, ext4 can be reentered when a transaction is open via
1001 * quota file writes. If we were to commit the transaction while thus
1002 * reentered, there can be a deadlock - we would be holding a quota
1003 * lock, and the commit would never complete if another thread had a
1004 * transaction open and was blocking on the quota lock - a ranking
1007 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1008 * will _not_ run commit under these circumstances because handle->h_ref
1009 * is elevated. We'll still have enough credits for the tiny quotafile
1012 int do_journal_get_write_access(handle_t *handle,
1013 struct buffer_head *bh)
1015 int dirty = buffer_dirty(bh);
1018 if (!buffer_mapped(bh) || buffer_freed(bh))
1021 * __block_write_begin() could have dirtied some buffers. Clean
1022 * the dirty bit as jbd2_journal_get_write_access() could complain
1023 * otherwise about fs integrity issues. Setting of the dirty bit
1024 * by __block_write_begin() isn't a real problem here as we clear
1025 * the bit before releasing a page lock and thus writeback cannot
1026 * ever write the buffer.
1029 clear_buffer_dirty(bh);
1030 BUFFER_TRACE(bh, "get write access");
1031 ret = ext4_journal_get_write_access(handle, bh);
1033 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1037 #ifdef CONFIG_FS_ENCRYPTION
1038 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1039 get_block_t *get_block)
1041 unsigned from = pos & (PAGE_SIZE - 1);
1042 unsigned to = from + len;
1043 struct inode *inode = page->mapping->host;
1044 unsigned block_start, block_end;
1047 unsigned blocksize = inode->i_sb->s_blocksize;
1049 struct buffer_head *bh, *head, *wait[2];
1053 BUG_ON(!PageLocked(page));
1054 BUG_ON(from > PAGE_SIZE);
1055 BUG_ON(to > PAGE_SIZE);
1058 if (!page_has_buffers(page))
1059 create_empty_buffers(page, blocksize, 0);
1060 head = page_buffers(page);
1061 bbits = ilog2(blocksize);
1062 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1064 for (bh = head, block_start = 0; bh != head || !block_start;
1065 block++, block_start = block_end, bh = bh->b_this_page) {
1066 block_end = block_start + blocksize;
1067 if (block_end <= from || block_start >= to) {
1068 if (PageUptodate(page)) {
1069 if (!buffer_uptodate(bh))
1070 set_buffer_uptodate(bh);
1075 clear_buffer_new(bh);
1076 if (!buffer_mapped(bh)) {
1077 WARN_ON(bh->b_size != blocksize);
1078 err = get_block(inode, block, bh, 1);
1081 if (buffer_new(bh)) {
1082 if (PageUptodate(page)) {
1083 clear_buffer_new(bh);
1084 set_buffer_uptodate(bh);
1085 mark_buffer_dirty(bh);
1088 if (block_end > to || block_start < from)
1089 zero_user_segments(page, to, block_end,
1094 if (PageUptodate(page)) {
1095 if (!buffer_uptodate(bh))
1096 set_buffer_uptodate(bh);
1099 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1100 !buffer_unwritten(bh) &&
1101 (block_start < from || block_end > to)) {
1102 ext4_read_bh_lock(bh, 0, false);
1103 wait[nr_wait++] = bh;
1107 * If we issued read requests, let them complete.
1109 for (i = 0; i < nr_wait; i++) {
1110 wait_on_buffer(wait[i]);
1111 if (!buffer_uptodate(wait[i]))
1114 if (unlikely(err)) {
1115 page_zero_new_buffers(page, from, to);
1116 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1117 for (i = 0; i < nr_wait; i++) {
1120 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1121 bh_offset(wait[i]));
1123 clear_buffer_uptodate(wait[i]);
1133 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1134 loff_t pos, unsigned len, unsigned flags,
1135 struct page **pagep, void **fsdata)
1137 struct inode *inode = mapping->host;
1138 int ret, needed_blocks;
1145 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1148 trace_ext4_write_begin(inode, pos, len, flags);
1150 * Reserve one block more for addition to orphan list in case
1151 * we allocate blocks but write fails for some reason
1153 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1154 index = pos >> PAGE_SHIFT;
1155 from = pos & (PAGE_SIZE - 1);
1158 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1159 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1168 * grab_cache_page_write_begin() can take a long time if the
1169 * system is thrashing due to memory pressure, or if the page
1170 * is being written back. So grab it first before we start
1171 * the transaction handle. This also allows us to allocate
1172 * the page (if needed) without using GFP_NOFS.
1175 page = grab_cache_page_write_begin(mapping, index, flags);
1181 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1182 if (IS_ERR(handle)) {
1184 return PTR_ERR(handle);
1188 if (page->mapping != mapping) {
1189 /* The page got truncated from under us */
1192 ext4_journal_stop(handle);
1195 /* In case writeback began while the page was unlocked */
1196 wait_for_stable_page(page);
1198 #ifdef CONFIG_FS_ENCRYPTION
1199 if (ext4_should_dioread_nolock(inode))
1200 ret = ext4_block_write_begin(page, pos, len,
1201 ext4_get_block_unwritten);
1203 ret = ext4_block_write_begin(page, pos, len,
1206 if (ext4_should_dioread_nolock(inode))
1207 ret = __block_write_begin(page, pos, len,
1208 ext4_get_block_unwritten);
1210 ret = __block_write_begin(page, pos, len, ext4_get_block);
1212 if (!ret && ext4_should_journal_data(inode)) {
1213 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1215 do_journal_get_write_access);
1219 bool extended = (pos + len > inode->i_size) &&
1220 !ext4_verity_in_progress(inode);
1224 * __block_write_begin may have instantiated a few blocks
1225 * outside i_size. Trim these off again. Don't need
1226 * i_size_read because we hold i_mutex.
1228 * Add inode to orphan list in case we crash before
1231 if (extended && ext4_can_truncate(inode))
1232 ext4_orphan_add(handle, inode);
1234 ext4_journal_stop(handle);
1236 ext4_truncate_failed_write(inode);
1238 * If truncate failed early the inode might
1239 * still be on the orphan list; we need to
1240 * make sure the inode is removed from the
1241 * orphan list in that case.
1244 ext4_orphan_del(NULL, inode);
1247 if (ret == -ENOSPC &&
1248 ext4_should_retry_alloc(inode->i_sb, &retries))
1257 /* For write_end() in data=journal mode */
1258 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1261 if (!buffer_mapped(bh) || buffer_freed(bh))
1263 set_buffer_uptodate(bh);
1264 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1265 clear_buffer_meta(bh);
1266 clear_buffer_prio(bh);
1271 * We need to pick up the new inode size which generic_commit_write gave us
1272 * `file' can be NULL - eg, when called from page_symlink().
1274 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1275 * buffers are managed internally.
1277 static int ext4_write_end(struct file *file,
1278 struct address_space *mapping,
1279 loff_t pos, unsigned len, unsigned copied,
1280 struct page *page, void *fsdata)
1282 handle_t *handle = ext4_journal_current_handle();
1283 struct inode *inode = mapping->host;
1284 loff_t old_size = inode->i_size;
1286 int i_size_changed = 0;
1287 int inline_data = ext4_has_inline_data(inode);
1288 bool verity = ext4_verity_in_progress(inode);
1290 trace_ext4_write_end(inode, pos, len, copied);
1292 ret = ext4_write_inline_data_end(inode, pos, len,
1301 copied = block_write_end(file, mapping, pos,
1302 len, copied, page, fsdata);
1304 * it's important to update i_size while still holding page lock:
1305 * page writeout could otherwise come in and zero beyond i_size.
1307 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1308 * blocks are being written past EOF, so skip the i_size update.
1311 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1315 if (old_size < pos && !verity)
1316 pagecache_isize_extended(inode, old_size, pos);
1318 * Don't mark the inode dirty under page lock. First, it unnecessarily
1319 * makes the holding time of page lock longer. Second, it forces lock
1320 * ordering of page lock and transaction start for journaling
1323 if (i_size_changed || inline_data)
1324 ret = ext4_mark_inode_dirty(handle, inode);
1326 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1327 /* if we have allocated more blocks and copied
1328 * less. We will have blocks allocated outside
1329 * inode->i_size. So truncate them
1331 ext4_orphan_add(handle, inode);
1333 ret2 = ext4_journal_stop(handle);
1337 if (pos + len > inode->i_size && !verity) {
1338 ext4_truncate_failed_write(inode);
1340 * If truncate failed early the inode might still be
1341 * on the orphan list; we need to make sure the inode
1342 * is removed from the orphan list in that case.
1345 ext4_orphan_del(NULL, inode);
1348 return ret ? ret : copied;
1352 * This is a private version of page_zero_new_buffers() which doesn't
1353 * set the buffer to be dirty, since in data=journalled mode we need
1354 * to call ext4_handle_dirty_metadata() instead.
1356 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1358 unsigned from, unsigned to)
1360 unsigned int block_start = 0, block_end;
1361 struct buffer_head *head, *bh;
1363 bh = head = page_buffers(page);
1365 block_end = block_start + bh->b_size;
1366 if (buffer_new(bh)) {
1367 if (block_end > from && block_start < to) {
1368 if (!PageUptodate(page)) {
1369 unsigned start, size;
1371 start = max(from, block_start);
1372 size = min(to, block_end) - start;
1374 zero_user(page, start, size);
1375 write_end_fn(handle, bh);
1377 clear_buffer_new(bh);
1380 block_start = block_end;
1381 bh = bh->b_this_page;
1382 } while (bh != head);
1385 static int ext4_journalled_write_end(struct file *file,
1386 struct address_space *mapping,
1387 loff_t pos, unsigned len, unsigned copied,
1388 struct page *page, void *fsdata)
1390 handle_t *handle = ext4_journal_current_handle();
1391 struct inode *inode = mapping->host;
1392 loff_t old_size = inode->i_size;
1396 int size_changed = 0;
1397 int inline_data = ext4_has_inline_data(inode);
1398 bool verity = ext4_verity_in_progress(inode);
1400 trace_ext4_journalled_write_end(inode, pos, len, copied);
1401 from = pos & (PAGE_SIZE - 1);
1404 BUG_ON(!ext4_handle_valid(handle));
1407 ret = ext4_write_inline_data_end(inode, pos, len,
1415 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1417 ext4_journalled_zero_new_buffers(handle, page, from, to);
1419 if (unlikely(copied < len))
1420 ext4_journalled_zero_new_buffers(handle, page,
1422 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1423 from + copied, &partial,
1426 SetPageUptodate(page);
1429 size_changed = ext4_update_inode_size(inode, pos + copied);
1430 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1431 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1435 if (old_size < pos && !verity)
1436 pagecache_isize_extended(inode, old_size, pos);
1438 if (size_changed || inline_data) {
1439 ret2 = ext4_mark_inode_dirty(handle, inode);
1444 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1445 /* if we have allocated more blocks and copied
1446 * less. We will have blocks allocated outside
1447 * inode->i_size. So truncate them
1449 ext4_orphan_add(handle, inode);
1452 ret2 = ext4_journal_stop(handle);
1455 if (pos + len > inode->i_size && !verity) {
1456 ext4_truncate_failed_write(inode);
1458 * If truncate failed early the inode might still be
1459 * on the orphan list; we need to make sure the inode
1460 * is removed from the orphan list in that case.
1463 ext4_orphan_del(NULL, inode);
1466 return ret ? ret : copied;
1470 * Reserve space for a single cluster
1472 static int ext4_da_reserve_space(struct inode *inode)
1474 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1475 struct ext4_inode_info *ei = EXT4_I(inode);
1479 * We will charge metadata quota at writeout time; this saves
1480 * us from metadata over-estimation, though we may go over by
1481 * a small amount in the end. Here we just reserve for data.
1483 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1487 spin_lock(&ei->i_block_reservation_lock);
1488 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1489 spin_unlock(&ei->i_block_reservation_lock);
1490 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1493 ei->i_reserved_data_blocks++;
1494 trace_ext4_da_reserve_space(inode);
1495 spin_unlock(&ei->i_block_reservation_lock);
1497 return 0; /* success */
1500 void ext4_da_release_space(struct inode *inode, int to_free)
1502 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1503 struct ext4_inode_info *ei = EXT4_I(inode);
1506 return; /* Nothing to release, exit */
1508 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1510 trace_ext4_da_release_space(inode, to_free);
1511 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1513 * if there aren't enough reserved blocks, then the
1514 * counter is messed up somewhere. Since this
1515 * function is called from invalidate page, it's
1516 * harmless to return without any action.
1518 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1519 "ino %lu, to_free %d with only %d reserved "
1520 "data blocks", inode->i_ino, to_free,
1521 ei->i_reserved_data_blocks);
1523 to_free = ei->i_reserved_data_blocks;
1525 ei->i_reserved_data_blocks -= to_free;
1527 /* update fs dirty data blocks counter */
1528 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1530 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1532 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1536 * Delayed allocation stuff
1539 struct mpage_da_data {
1540 struct inode *inode;
1541 struct writeback_control *wbc;
1543 pgoff_t first_page; /* The first page to write */
1544 pgoff_t next_page; /* Current page to examine */
1545 pgoff_t last_page; /* Last page to examine */
1547 * Extent to map - this can be after first_page because that can be
1548 * fully mapped. We somewhat abuse m_flags to store whether the extent
1549 * is delalloc or unwritten.
1551 struct ext4_map_blocks map;
1552 struct ext4_io_submit io_submit; /* IO submission data */
1553 unsigned int do_map:1;
1554 unsigned int scanned_until_end:1;
1557 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1562 struct pagevec pvec;
1563 struct inode *inode = mpd->inode;
1564 struct address_space *mapping = inode->i_mapping;
1566 /* This is necessary when next_page == 0. */
1567 if (mpd->first_page >= mpd->next_page)
1570 mpd->scanned_until_end = 0;
1571 index = mpd->first_page;
1572 end = mpd->next_page - 1;
1574 ext4_lblk_t start, last;
1575 start = index << (PAGE_SHIFT - inode->i_blkbits);
1576 last = end << (PAGE_SHIFT - inode->i_blkbits);
1577 ext4_es_remove_extent(inode, start, last - start + 1);
1580 pagevec_init(&pvec);
1581 while (index <= end) {
1582 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1585 for (i = 0; i < nr_pages; i++) {
1586 struct page *page = pvec.pages[i];
1588 BUG_ON(!PageLocked(page));
1589 BUG_ON(PageWriteback(page));
1591 if (page_mapped(page))
1592 clear_page_dirty_for_io(page);
1593 block_invalidatepage(page, 0, PAGE_SIZE);
1594 ClearPageUptodate(page);
1598 pagevec_release(&pvec);
1602 static void ext4_print_free_blocks(struct inode *inode)
1604 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1605 struct super_block *sb = inode->i_sb;
1606 struct ext4_inode_info *ei = EXT4_I(inode);
1608 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1609 EXT4_C2B(EXT4_SB(inode->i_sb),
1610 ext4_count_free_clusters(sb)));
1611 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1612 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1613 (long long) EXT4_C2B(EXT4_SB(sb),
1614 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1615 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1616 (long long) EXT4_C2B(EXT4_SB(sb),
1617 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1618 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1619 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1620 ei->i_reserved_data_blocks);
1624 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1626 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1630 * ext4_insert_delayed_block - adds a delayed block to the extents status
1631 * tree, incrementing the reserved cluster/block
1632 * count or making a pending reservation
1635 * @inode - file containing the newly added block
1636 * @lblk - logical block to be added
1638 * Returns 0 on success, negative error code on failure.
1640 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1642 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1644 bool allocated = false;
1647 * If the cluster containing lblk is shared with a delayed,
1648 * written, or unwritten extent in a bigalloc file system, it's
1649 * already been accounted for and does not need to be reserved.
1650 * A pending reservation must be made for the cluster if it's
1651 * shared with a written or unwritten extent and doesn't already
1652 * have one. Written and unwritten extents can be purged from the
1653 * extents status tree if the system is under memory pressure, so
1654 * it's necessary to examine the extent tree if a search of the
1655 * extents status tree doesn't get a match.
1657 if (sbi->s_cluster_ratio == 1) {
1658 ret = ext4_da_reserve_space(inode);
1659 if (ret != 0) /* ENOSPC */
1661 } else { /* bigalloc */
1662 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1663 if (!ext4_es_scan_clu(inode,
1664 &ext4_es_is_mapped, lblk)) {
1665 ret = ext4_clu_mapped(inode,
1666 EXT4_B2C(sbi, lblk));
1670 ret = ext4_da_reserve_space(inode);
1671 if (ret != 0) /* ENOSPC */
1682 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1689 * This function is grabs code from the very beginning of
1690 * ext4_map_blocks, but assumes that the caller is from delayed write
1691 * time. This function looks up the requested blocks and sets the
1692 * buffer delay bit under the protection of i_data_sem.
1694 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1695 struct ext4_map_blocks *map,
1696 struct buffer_head *bh)
1698 struct extent_status es;
1700 sector_t invalid_block = ~((sector_t) 0xffff);
1701 #ifdef ES_AGGRESSIVE_TEST
1702 struct ext4_map_blocks orig_map;
1704 memcpy(&orig_map, map, sizeof(*map));
1707 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1711 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1712 (unsigned long) map->m_lblk);
1714 /* Lookup extent status tree firstly */
1715 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1716 if (ext4_es_is_hole(&es)) {
1718 down_read(&EXT4_I(inode)->i_data_sem);
1723 * Delayed extent could be allocated by fallocate.
1724 * So we need to check it.
1726 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1727 map_bh(bh, inode->i_sb, invalid_block);
1729 set_buffer_delay(bh);
1733 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1734 retval = es.es_len - (iblock - es.es_lblk);
1735 if (retval > map->m_len)
1736 retval = map->m_len;
1737 map->m_len = retval;
1738 if (ext4_es_is_written(&es))
1739 map->m_flags |= EXT4_MAP_MAPPED;
1740 else if (ext4_es_is_unwritten(&es))
1741 map->m_flags |= EXT4_MAP_UNWRITTEN;
1745 #ifdef ES_AGGRESSIVE_TEST
1746 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1752 * Try to see if we can get the block without requesting a new
1753 * file system block.
1755 down_read(&EXT4_I(inode)->i_data_sem);
1756 if (ext4_has_inline_data(inode))
1758 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1759 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1761 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1768 * XXX: __block_prepare_write() unmaps passed block,
1772 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1778 map_bh(bh, inode->i_sb, invalid_block);
1780 set_buffer_delay(bh);
1781 } else if (retval > 0) {
1783 unsigned int status;
1785 if (unlikely(retval != map->m_len)) {
1786 ext4_warning(inode->i_sb,
1787 "ES len assertion failed for inode "
1788 "%lu: retval %d != map->m_len %d",
1789 inode->i_ino, retval, map->m_len);
1793 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1794 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1795 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1796 map->m_pblk, status);
1802 up_read((&EXT4_I(inode)->i_data_sem));
1808 * This is a special get_block_t callback which is used by
1809 * ext4_da_write_begin(). It will either return mapped block or
1810 * reserve space for a single block.
1812 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1813 * We also have b_blocknr = -1 and b_bdev initialized properly
1815 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1816 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1817 * initialized properly.
1819 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1820 struct buffer_head *bh, int create)
1822 struct ext4_map_blocks map;
1825 BUG_ON(create == 0);
1826 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1828 map.m_lblk = iblock;
1832 * first, we need to know whether the block is allocated already
1833 * preallocated blocks are unmapped but should treated
1834 * the same as allocated blocks.
1836 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1840 map_bh(bh, inode->i_sb, map.m_pblk);
1841 ext4_update_bh_state(bh, map.m_flags);
1843 if (buffer_unwritten(bh)) {
1844 /* A delayed write to unwritten bh should be marked
1845 * new and mapped. Mapped ensures that we don't do
1846 * get_block multiple times when we write to the same
1847 * offset and new ensures that we do proper zero out
1848 * for partial write.
1851 set_buffer_mapped(bh);
1856 static int bget_one(handle_t *handle, struct buffer_head *bh)
1862 static int bput_one(handle_t *handle, struct buffer_head *bh)
1868 static int __ext4_journalled_writepage(struct page *page,
1871 struct address_space *mapping = page->mapping;
1872 struct inode *inode = mapping->host;
1873 struct buffer_head *page_bufs = NULL;
1874 handle_t *handle = NULL;
1875 int ret = 0, err = 0;
1876 int inline_data = ext4_has_inline_data(inode);
1877 struct buffer_head *inode_bh = NULL;
1879 ClearPageChecked(page);
1882 BUG_ON(page->index != 0);
1883 BUG_ON(len > ext4_get_max_inline_size(inode));
1884 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1885 if (inode_bh == NULL)
1888 page_bufs = page_buffers(page);
1893 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1897 * We need to release the page lock before we start the
1898 * journal, so grab a reference so the page won't disappear
1899 * out from under us.
1904 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1905 ext4_writepage_trans_blocks(inode));
1906 if (IS_ERR(handle)) {
1907 ret = PTR_ERR(handle);
1909 goto out_no_pagelock;
1911 BUG_ON(!ext4_handle_valid(handle));
1915 if (page->mapping != mapping) {
1916 /* The page got truncated from under us */
1917 ext4_journal_stop(handle);
1923 ret = ext4_mark_inode_dirty(handle, inode);
1925 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1926 do_journal_get_write_access);
1928 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1933 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1936 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1937 err = ext4_journal_stop(handle);
1941 if (!ext4_has_inline_data(inode))
1942 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1944 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1953 * Note that we don't need to start a transaction unless we're journaling data
1954 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1955 * need to file the inode to the transaction's list in ordered mode because if
1956 * we are writing back data added by write(), the inode is already there and if
1957 * we are writing back data modified via mmap(), no one guarantees in which
1958 * transaction the data will hit the disk. In case we are journaling data, we
1959 * cannot start transaction directly because transaction start ranks above page
1960 * lock so we have to do some magic.
1962 * This function can get called via...
1963 * - ext4_writepages after taking page lock (have journal handle)
1964 * - journal_submit_inode_data_buffers (no journal handle)
1965 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1966 * - grab_page_cache when doing write_begin (have journal handle)
1968 * We don't do any block allocation in this function. If we have page with
1969 * multiple blocks we need to write those buffer_heads that are mapped. This
1970 * is important for mmaped based write. So if we do with blocksize 1K
1971 * truncate(f, 1024);
1972 * a = mmap(f, 0, 4096);
1974 * truncate(f, 4096);
1975 * we have in the page first buffer_head mapped via page_mkwrite call back
1976 * but other buffer_heads would be unmapped but dirty (dirty done via the
1977 * do_wp_page). So writepage should write the first block. If we modify
1978 * the mmap area beyond 1024 we will again get a page_fault and the
1979 * page_mkwrite callback will do the block allocation and mark the
1980 * buffer_heads mapped.
1982 * We redirty the page if we have any buffer_heads that is either delay or
1983 * unwritten in the page.
1985 * We can get recursively called as show below.
1987 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1990 * But since we don't do any block allocation we should not deadlock.
1991 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1993 static int ext4_writepage(struct page *page,
1994 struct writeback_control *wbc)
1999 struct buffer_head *page_bufs = NULL;
2000 struct inode *inode = page->mapping->host;
2001 struct ext4_io_submit io_submit;
2002 bool keep_towrite = false;
2004 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2005 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
2010 trace_ext4_writepage(page);
2011 size = i_size_read(inode);
2012 if (page->index == size >> PAGE_SHIFT &&
2013 !ext4_verity_in_progress(inode))
2014 len = size & ~PAGE_MASK;
2018 page_bufs = page_buffers(page);
2020 * We cannot do block allocation or other extent handling in this
2021 * function. If there are buffers needing that, we have to redirty
2022 * the page. But we may reach here when we do a journal commit via
2023 * journal_submit_inode_data_buffers() and in that case we must write
2024 * allocated buffers to achieve data=ordered mode guarantees.
2026 * Also, if there is only one buffer per page (the fs block
2027 * size == the page size), if one buffer needs block
2028 * allocation or needs to modify the extent tree to clear the
2029 * unwritten flag, we know that the page can't be written at
2030 * all, so we might as well refuse the write immediately.
2031 * Unfortunately if the block size != page size, we can't as
2032 * easily detect this case using ext4_walk_page_buffers(), but
2033 * for the extremely common case, this is an optimization that
2034 * skips a useless round trip through ext4_bio_write_page().
2036 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2037 ext4_bh_delay_or_unwritten)) {
2038 redirty_page_for_writepage(wbc, page);
2039 if ((current->flags & PF_MEMALLOC) ||
2040 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2042 * For memory cleaning there's no point in writing only
2043 * some buffers. So just bail out. Warn if we came here
2044 * from direct reclaim.
2046 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2051 keep_towrite = true;
2054 if (PageChecked(page) && ext4_should_journal_data(inode))
2056 * It's mmapped pagecache. Add buffers and journal it. There
2057 * doesn't seem much point in redirtying the page here.
2059 return __ext4_journalled_writepage(page, len);
2061 ext4_io_submit_init(&io_submit, wbc);
2062 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2063 if (!io_submit.io_end) {
2064 redirty_page_for_writepage(wbc, page);
2068 ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite);
2069 ext4_io_submit(&io_submit);
2070 /* Drop io_end reference we got from init */
2071 ext4_put_io_end_defer(io_submit.io_end);
2075 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2081 BUG_ON(page->index != mpd->first_page);
2082 clear_page_dirty_for_io(page);
2084 * We have to be very careful here! Nothing protects writeback path
2085 * against i_size changes and the page can be writeably mapped into
2086 * page tables. So an application can be growing i_size and writing
2087 * data through mmap while writeback runs. clear_page_dirty_for_io()
2088 * write-protects our page in page tables and the page cannot get
2089 * written to again until we release page lock. So only after
2090 * clear_page_dirty_for_io() we are safe to sample i_size for
2091 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2092 * on the barrier provided by TestClearPageDirty in
2093 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2094 * after page tables are updated.
2096 size = i_size_read(mpd->inode);
2097 if (page->index == size >> PAGE_SHIFT &&
2098 !ext4_verity_in_progress(mpd->inode))
2099 len = size & ~PAGE_MASK;
2102 err = ext4_bio_write_page(&mpd->io_submit, page, len, false);
2104 mpd->wbc->nr_to_write--;
2110 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2113 * mballoc gives us at most this number of blocks...
2114 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2115 * The rest of mballoc seems to handle chunks up to full group size.
2117 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2120 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2122 * @mpd - extent of blocks
2123 * @lblk - logical number of the block in the file
2124 * @bh - buffer head we want to add to the extent
2126 * The function is used to collect contig. blocks in the same state. If the
2127 * buffer doesn't require mapping for writeback and we haven't started the
2128 * extent of buffers to map yet, the function returns 'true' immediately - the
2129 * caller can write the buffer right away. Otherwise the function returns true
2130 * if the block has been added to the extent, false if the block couldn't be
2133 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2134 struct buffer_head *bh)
2136 struct ext4_map_blocks *map = &mpd->map;
2138 /* Buffer that doesn't need mapping for writeback? */
2139 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2140 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2141 /* So far no extent to map => we write the buffer right away */
2142 if (map->m_len == 0)
2147 /* First block in the extent? */
2148 if (map->m_len == 0) {
2149 /* We cannot map unless handle is started... */
2154 map->m_flags = bh->b_state & BH_FLAGS;
2158 /* Don't go larger than mballoc is willing to allocate */
2159 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2162 /* Can we merge the block to our big extent? */
2163 if (lblk == map->m_lblk + map->m_len &&
2164 (bh->b_state & BH_FLAGS) == map->m_flags) {
2172 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2174 * @mpd - extent of blocks for mapping
2175 * @head - the first buffer in the page
2176 * @bh - buffer we should start processing from
2177 * @lblk - logical number of the block in the file corresponding to @bh
2179 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2180 * the page for IO if all buffers in this page were mapped and there's no
2181 * accumulated extent of buffers to map or add buffers in the page to the
2182 * extent of buffers to map. The function returns 1 if the caller can continue
2183 * by processing the next page, 0 if it should stop adding buffers to the
2184 * extent to map because we cannot extend it anymore. It can also return value
2185 * < 0 in case of error during IO submission.
2187 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2188 struct buffer_head *head,
2189 struct buffer_head *bh,
2192 struct inode *inode = mpd->inode;
2194 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2195 >> inode->i_blkbits;
2197 if (ext4_verity_in_progress(inode))
2198 blocks = EXT_MAX_BLOCKS;
2201 BUG_ON(buffer_locked(bh));
2203 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2204 /* Found extent to map? */
2207 /* Buffer needs mapping and handle is not started? */
2210 /* Everything mapped so far and we hit EOF */
2213 } while (lblk++, (bh = bh->b_this_page) != head);
2214 /* So far everything mapped? Submit the page for IO. */
2215 if (mpd->map.m_len == 0) {
2216 err = mpage_submit_page(mpd, head->b_page);
2220 if (lblk >= blocks) {
2221 mpd->scanned_until_end = 1;
2228 * mpage_process_page - update page buffers corresponding to changed extent and
2229 * may submit fully mapped page for IO
2231 * @mpd - description of extent to map, on return next extent to map
2232 * @m_lblk - logical block mapping.
2233 * @m_pblk - corresponding physical mapping.
2234 * @map_bh - determines on return whether this page requires any further
2236 * Scan given page buffers corresponding to changed extent and update buffer
2237 * state according to new extent state.
2238 * We map delalloc buffers to their physical location, clear unwritten bits.
2239 * If the given page is not fully mapped, we update @map to the next extent in
2240 * the given page that needs mapping & return @map_bh as true.
2242 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2243 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2246 struct buffer_head *head, *bh;
2247 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2248 ext4_lblk_t lblk = *m_lblk;
2249 ext4_fsblk_t pblock = *m_pblk;
2251 int blkbits = mpd->inode->i_blkbits;
2252 ssize_t io_end_size = 0;
2253 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2255 bh = head = page_buffers(page);
2257 if (lblk < mpd->map.m_lblk)
2259 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2261 * Buffer after end of mapped extent.
2262 * Find next buffer in the page to map.
2265 mpd->map.m_flags = 0;
2266 io_end_vec->size += io_end_size;
2269 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2272 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2273 io_end_vec = ext4_alloc_io_end_vec(io_end);
2274 if (IS_ERR(io_end_vec)) {
2275 err = PTR_ERR(io_end_vec);
2278 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2283 if (buffer_delay(bh)) {
2284 clear_buffer_delay(bh);
2285 bh->b_blocknr = pblock++;
2287 clear_buffer_unwritten(bh);
2288 io_end_size += (1 << blkbits);
2289 } while (lblk++, (bh = bh->b_this_page) != head);
2291 io_end_vec->size += io_end_size;
2301 * mpage_map_buffers - update buffers corresponding to changed extent and
2302 * submit fully mapped pages for IO
2304 * @mpd - description of extent to map, on return next extent to map
2306 * Scan buffers corresponding to changed extent (we expect corresponding pages
2307 * to be already locked) and update buffer state according to new extent state.
2308 * We map delalloc buffers to their physical location, clear unwritten bits,
2309 * and mark buffers as uninit when we perform writes to unwritten extents
2310 * and do extent conversion after IO is finished. If the last page is not fully
2311 * mapped, we update @map to the next extent in the last page that needs
2312 * mapping. Otherwise we submit the page for IO.
2314 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2316 struct pagevec pvec;
2318 struct inode *inode = mpd->inode;
2319 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2322 ext4_fsblk_t pblock;
2324 bool map_bh = false;
2326 start = mpd->map.m_lblk >> bpp_bits;
2327 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2328 lblk = start << bpp_bits;
2329 pblock = mpd->map.m_pblk;
2331 pagevec_init(&pvec);
2332 while (start <= end) {
2333 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2337 for (i = 0; i < nr_pages; i++) {
2338 struct page *page = pvec.pages[i];
2340 err = mpage_process_page(mpd, page, &lblk, &pblock,
2343 * If map_bh is true, means page may require further bh
2344 * mapping, or maybe the page was submitted for IO.
2345 * So we return to call further extent mapping.
2347 if (err < 0 || map_bh)
2349 /* Page fully mapped - let IO run! */
2350 err = mpage_submit_page(mpd, page);
2354 pagevec_release(&pvec);
2356 /* Extent fully mapped and matches with page boundary. We are done. */
2358 mpd->map.m_flags = 0;
2361 pagevec_release(&pvec);
2365 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2367 struct inode *inode = mpd->inode;
2368 struct ext4_map_blocks *map = &mpd->map;
2369 int get_blocks_flags;
2370 int err, dioread_nolock;
2372 trace_ext4_da_write_pages_extent(inode, map);
2374 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2375 * to convert an unwritten extent to be initialized (in the case
2376 * where we have written into one or more preallocated blocks). It is
2377 * possible that we're going to need more metadata blocks than
2378 * previously reserved. However we must not fail because we're in
2379 * writeback and there is nothing we can do about it so it might result
2380 * in data loss. So use reserved blocks to allocate metadata if
2383 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2384 * the blocks in question are delalloc blocks. This indicates
2385 * that the blocks and quotas has already been checked when
2386 * the data was copied into the page cache.
2388 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2389 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2390 EXT4_GET_BLOCKS_IO_SUBMIT;
2391 dioread_nolock = ext4_should_dioread_nolock(inode);
2393 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2394 if (map->m_flags & BIT(BH_Delay))
2395 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2397 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2400 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2401 if (!mpd->io_submit.io_end->handle &&
2402 ext4_handle_valid(handle)) {
2403 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2404 handle->h_rsv_handle = NULL;
2406 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2409 BUG_ON(map->m_len == 0);
2414 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2415 * mpd->len and submit pages underlying it for IO
2417 * @handle - handle for journal operations
2418 * @mpd - extent to map
2419 * @give_up_on_write - we set this to true iff there is a fatal error and there
2420 * is no hope of writing the data. The caller should discard
2421 * dirty pages to avoid infinite loops.
2423 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2424 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2425 * them to initialized or split the described range from larger unwritten
2426 * extent. Note that we need not map all the described range since allocation
2427 * can return less blocks or the range is covered by more unwritten extents. We
2428 * cannot map more because we are limited by reserved transaction credits. On
2429 * the other hand we always make sure that the last touched page is fully
2430 * mapped so that it can be written out (and thus forward progress is
2431 * guaranteed). After mapping we submit all mapped pages for IO.
2433 static int mpage_map_and_submit_extent(handle_t *handle,
2434 struct mpage_da_data *mpd,
2435 bool *give_up_on_write)
2437 struct inode *inode = mpd->inode;
2438 struct ext4_map_blocks *map = &mpd->map;
2442 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2443 struct ext4_io_end_vec *io_end_vec;
2445 io_end_vec = ext4_alloc_io_end_vec(io_end);
2446 if (IS_ERR(io_end_vec))
2447 return PTR_ERR(io_end_vec);
2448 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2450 err = mpage_map_one_extent(handle, mpd);
2452 struct super_block *sb = inode->i_sb;
2454 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2455 ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2456 goto invalidate_dirty_pages;
2458 * Let the uper layers retry transient errors.
2459 * In the case of ENOSPC, if ext4_count_free_blocks()
2460 * is non-zero, a commit should free up blocks.
2462 if ((err == -ENOMEM) ||
2463 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2465 goto update_disksize;
2468 ext4_msg(sb, KERN_CRIT,
2469 "Delayed block allocation failed for "
2470 "inode %lu at logical offset %llu with"
2471 " max blocks %u with error %d",
2473 (unsigned long long)map->m_lblk,
2474 (unsigned)map->m_len, -err);
2475 ext4_msg(sb, KERN_CRIT,
2476 "This should not happen!! Data will "
2479 ext4_print_free_blocks(inode);
2480 invalidate_dirty_pages:
2481 *give_up_on_write = true;
2486 * Update buffer state, submit mapped pages, and get us new
2489 err = mpage_map_and_submit_buffers(mpd);
2491 goto update_disksize;
2492 } while (map->m_len);
2496 * Update on-disk size after IO is submitted. Races with
2497 * truncate are avoided by checking i_size under i_data_sem.
2499 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2500 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2504 down_write(&EXT4_I(inode)->i_data_sem);
2505 i_size = i_size_read(inode);
2506 if (disksize > i_size)
2508 if (disksize > EXT4_I(inode)->i_disksize)
2509 EXT4_I(inode)->i_disksize = disksize;
2510 up_write(&EXT4_I(inode)->i_data_sem);
2511 err2 = ext4_mark_inode_dirty(handle, inode);
2513 ext4_error_err(inode->i_sb, -err2,
2514 "Failed to mark inode %lu dirty",
2524 * Calculate the total number of credits to reserve for one writepages
2525 * iteration. This is called from ext4_writepages(). We map an extent of
2526 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2527 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2528 * bpp - 1 blocks in bpp different extents.
2530 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2532 int bpp = ext4_journal_blocks_per_page(inode);
2534 return ext4_meta_trans_blocks(inode,
2535 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2539 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2540 * and underlying extent to map
2542 * @mpd - where to look for pages
2544 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2545 * IO immediately. When we find a page which isn't mapped we start accumulating
2546 * extent of buffers underlying these pages that needs mapping (formed by
2547 * either delayed or unwritten buffers). We also lock the pages containing
2548 * these buffers. The extent found is returned in @mpd structure (starting at
2549 * mpd->lblk with length mpd->len blocks).
2551 * Note that this function can attach bios to one io_end structure which are
2552 * neither logically nor physically contiguous. Although it may seem as an
2553 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2554 * case as we need to track IO to all buffers underlying a page in one io_end.
2556 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2558 struct address_space *mapping = mpd->inode->i_mapping;
2559 struct pagevec pvec;
2560 unsigned int nr_pages;
2561 long left = mpd->wbc->nr_to_write;
2562 pgoff_t index = mpd->first_page;
2563 pgoff_t end = mpd->last_page;
2566 int blkbits = mpd->inode->i_blkbits;
2568 struct buffer_head *head;
2570 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2571 tag = PAGECACHE_TAG_TOWRITE;
2573 tag = PAGECACHE_TAG_DIRTY;
2575 pagevec_init(&pvec);
2577 mpd->next_page = index;
2578 while (index <= end) {
2579 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2584 for (i = 0; i < nr_pages; i++) {
2585 struct page *page = pvec.pages[i];
2588 * Accumulated enough dirty pages? This doesn't apply
2589 * to WB_SYNC_ALL mode. For integrity sync we have to
2590 * keep going because someone may be concurrently
2591 * dirtying pages, and we might have synced a lot of
2592 * newly appeared dirty pages, but have not synced all
2593 * of the old dirty pages.
2595 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2598 /* If we can't merge this page, we are done. */
2599 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2604 * If the page is no longer dirty, or its mapping no
2605 * longer corresponds to inode we are writing (which
2606 * means it has been truncated or invalidated), or the
2607 * page is already under writeback and we are not doing
2608 * a data integrity writeback, skip the page
2610 if (!PageDirty(page) ||
2611 (PageWriteback(page) &&
2612 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2613 unlikely(page->mapping != mapping)) {
2618 wait_on_page_writeback(page);
2619 BUG_ON(PageWriteback(page));
2621 if (mpd->map.m_len == 0)
2622 mpd->first_page = page->index;
2623 mpd->next_page = page->index + 1;
2624 /* Add all dirty buffers to mpd */
2625 lblk = ((ext4_lblk_t)page->index) <<
2626 (PAGE_SHIFT - blkbits);
2627 head = page_buffers(page);
2628 err = mpage_process_page_bufs(mpd, head, head, lblk);
2634 pagevec_release(&pvec);
2637 mpd->scanned_until_end = 1;
2640 pagevec_release(&pvec);
2644 static int ext4_writepages(struct address_space *mapping,
2645 struct writeback_control *wbc)
2647 pgoff_t writeback_index = 0;
2648 long nr_to_write = wbc->nr_to_write;
2649 int range_whole = 0;
2651 handle_t *handle = NULL;
2652 struct mpage_da_data mpd;
2653 struct inode *inode = mapping->host;
2654 int needed_blocks, rsv_blocks = 0, ret = 0;
2655 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2656 struct blk_plug plug;
2657 bool give_up_on_write = false;
2659 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2662 percpu_down_read(&sbi->s_writepages_rwsem);
2663 trace_ext4_writepages(inode, wbc);
2666 * No pages to write? This is mainly a kludge to avoid starting
2667 * a transaction for special inodes like journal inode on last iput()
2668 * because that could violate lock ordering on umount
2670 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2671 goto out_writepages;
2673 if (ext4_should_journal_data(inode)) {
2674 ret = generic_writepages(mapping, wbc);
2675 goto out_writepages;
2679 * If the filesystem has aborted, it is read-only, so return
2680 * right away instead of dumping stack traces later on that
2681 * will obscure the real source of the problem. We test
2682 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2683 * the latter could be true if the filesystem is mounted
2684 * read-only, and in that case, ext4_writepages should
2685 * *never* be called, so if that ever happens, we would want
2688 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2689 ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2691 goto out_writepages;
2695 * If we have inline data and arrive here, it means that
2696 * we will soon create the block for the 1st page, so
2697 * we'd better clear the inline data here.
2699 if (ext4_has_inline_data(inode)) {
2700 /* Just inode will be modified... */
2701 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2702 if (IS_ERR(handle)) {
2703 ret = PTR_ERR(handle);
2704 goto out_writepages;
2706 BUG_ON(ext4_test_inode_state(inode,
2707 EXT4_STATE_MAY_INLINE_DATA));
2708 ext4_destroy_inline_data(handle, inode);
2709 ext4_journal_stop(handle);
2712 if (ext4_should_dioread_nolock(inode)) {
2714 * We may need to convert up to one extent per block in
2715 * the page and we may dirty the inode.
2717 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2718 PAGE_SIZE >> inode->i_blkbits);
2721 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2724 if (wbc->range_cyclic) {
2725 writeback_index = mapping->writeback_index;
2726 if (writeback_index)
2728 mpd.first_page = writeback_index;
2731 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2732 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2737 ext4_io_submit_init(&mpd.io_submit, wbc);
2739 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2740 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2741 blk_start_plug(&plug);
2744 * First writeback pages that don't need mapping - we can avoid
2745 * starting a transaction unnecessarily and also avoid being blocked
2746 * in the block layer on device congestion while having transaction
2750 mpd.scanned_until_end = 0;
2751 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2752 if (!mpd.io_submit.io_end) {
2756 ret = mpage_prepare_extent_to_map(&mpd);
2757 /* Unlock pages we didn't use */
2758 mpage_release_unused_pages(&mpd, false);
2759 /* Submit prepared bio */
2760 ext4_io_submit(&mpd.io_submit);
2761 ext4_put_io_end_defer(mpd.io_submit.io_end);
2762 mpd.io_submit.io_end = NULL;
2766 while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2767 /* For each extent of pages we use new io_end */
2768 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2769 if (!mpd.io_submit.io_end) {
2775 * We have two constraints: We find one extent to map and we
2776 * must always write out whole page (makes a difference when
2777 * blocksize < pagesize) so that we don't block on IO when we
2778 * try to write out the rest of the page. Journalled mode is
2779 * not supported by delalloc.
2781 BUG_ON(ext4_should_journal_data(inode));
2782 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2784 /* start a new transaction */
2785 handle = ext4_journal_start_with_reserve(inode,
2786 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2787 if (IS_ERR(handle)) {
2788 ret = PTR_ERR(handle);
2789 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2790 "%ld pages, ino %lu; err %d", __func__,
2791 wbc->nr_to_write, inode->i_ino, ret);
2792 /* Release allocated io_end */
2793 ext4_put_io_end(mpd.io_submit.io_end);
2794 mpd.io_submit.io_end = NULL;
2799 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2800 ret = mpage_prepare_extent_to_map(&mpd);
2801 if (!ret && mpd.map.m_len)
2802 ret = mpage_map_and_submit_extent(handle, &mpd,
2805 * Caution: If the handle is synchronous,
2806 * ext4_journal_stop() can wait for transaction commit
2807 * to finish which may depend on writeback of pages to
2808 * complete or on page lock to be released. In that
2809 * case, we have to wait until after we have
2810 * submitted all the IO, released page locks we hold,
2811 * and dropped io_end reference (for extent conversion
2812 * to be able to complete) before stopping the handle.
2814 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2815 ext4_journal_stop(handle);
2819 /* Unlock pages we didn't use */
2820 mpage_release_unused_pages(&mpd, give_up_on_write);
2821 /* Submit prepared bio */
2822 ext4_io_submit(&mpd.io_submit);
2825 * Drop our io_end reference we got from init. We have
2826 * to be careful and use deferred io_end finishing if
2827 * we are still holding the transaction as we can
2828 * release the last reference to io_end which may end
2829 * up doing unwritten extent conversion.
2832 ext4_put_io_end_defer(mpd.io_submit.io_end);
2833 ext4_journal_stop(handle);
2835 ext4_put_io_end(mpd.io_submit.io_end);
2836 mpd.io_submit.io_end = NULL;
2838 if (ret == -ENOSPC && sbi->s_journal) {
2840 * Commit the transaction which would
2841 * free blocks released in the transaction
2844 jbd2_journal_force_commit_nested(sbi->s_journal);
2848 /* Fatal error - ENOMEM, EIO... */
2853 blk_finish_plug(&plug);
2854 if (!ret && !cycled && wbc->nr_to_write > 0) {
2856 mpd.last_page = writeback_index - 1;
2862 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2864 * Set the writeback_index so that range_cyclic
2865 * mode will write it back later
2867 mapping->writeback_index = mpd.first_page;
2870 trace_ext4_writepages_result(inode, wbc, ret,
2871 nr_to_write - wbc->nr_to_write);
2872 percpu_up_read(&sbi->s_writepages_rwsem);
2876 static int ext4_dax_writepages(struct address_space *mapping,
2877 struct writeback_control *wbc)
2880 long nr_to_write = wbc->nr_to_write;
2881 struct inode *inode = mapping->host;
2882 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2884 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2887 percpu_down_read(&sbi->s_writepages_rwsem);
2888 trace_ext4_writepages(inode, wbc);
2890 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2891 trace_ext4_writepages_result(inode, wbc, ret,
2892 nr_to_write - wbc->nr_to_write);
2893 percpu_up_read(&sbi->s_writepages_rwsem);
2897 static int ext4_nonda_switch(struct super_block *sb)
2899 s64 free_clusters, dirty_clusters;
2900 struct ext4_sb_info *sbi = EXT4_SB(sb);
2903 * switch to non delalloc mode if we are running low
2904 * on free block. The free block accounting via percpu
2905 * counters can get slightly wrong with percpu_counter_batch getting
2906 * accumulated on each CPU without updating global counters
2907 * Delalloc need an accurate free block accounting. So switch
2908 * to non delalloc when we are near to error range.
2911 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2913 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2915 * Start pushing delalloc when 1/2 of free blocks are dirty.
2917 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2918 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2920 if (2 * free_clusters < 3 * dirty_clusters ||
2921 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2923 * free block count is less than 150% of dirty blocks
2924 * or free blocks is less than watermark
2931 /* We always reserve for an inode update; the superblock could be there too */
2932 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2934 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2937 if (pos + len <= 0x7fffffffULL)
2940 /* We might need to update the superblock to set LARGE_FILE */
2944 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2945 loff_t pos, unsigned len, unsigned flags,
2946 struct page **pagep, void **fsdata)
2948 int ret, retries = 0;
2951 struct inode *inode = mapping->host;
2954 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2957 index = pos >> PAGE_SHIFT;
2959 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2960 ext4_verity_in_progress(inode)) {
2961 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2962 return ext4_write_begin(file, mapping, pos,
2963 len, flags, pagep, fsdata);
2965 *fsdata = (void *)0;
2966 trace_ext4_da_write_begin(inode, pos, len, flags);
2968 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2969 ret = ext4_da_write_inline_data_begin(mapping, inode,
2979 * grab_cache_page_write_begin() can take a long time if the
2980 * system is thrashing due to memory pressure, or if the page
2981 * is being written back. So grab it first before we start
2982 * the transaction handle. This also allows us to allocate
2983 * the page (if needed) without using GFP_NOFS.
2986 page = grab_cache_page_write_begin(mapping, index, flags);
2992 * With delayed allocation, we don't log the i_disksize update
2993 * if there is delayed block allocation. But we still need
2994 * to journalling the i_disksize update if writes to the end
2995 * of file which has an already mapped buffer.
2998 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2999 ext4_da_write_credits(inode, pos, len));
3000 if (IS_ERR(handle)) {
3002 return PTR_ERR(handle);
3006 if (page->mapping != mapping) {
3007 /* The page got truncated from under us */
3010 ext4_journal_stop(handle);
3013 /* In case writeback began while the page was unlocked */
3014 wait_for_stable_page(page);
3016 #ifdef CONFIG_FS_ENCRYPTION
3017 ret = ext4_block_write_begin(page, pos, len,
3018 ext4_da_get_block_prep);
3020 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3024 ext4_journal_stop(handle);
3026 * block_write_begin may have instantiated a few blocks
3027 * outside i_size. Trim these off again. Don't need
3028 * i_size_read because we hold i_mutex.
3030 if (pos + len > inode->i_size)
3031 ext4_truncate_failed_write(inode);
3033 if (ret == -ENOSPC &&
3034 ext4_should_retry_alloc(inode->i_sb, &retries))
3046 * Check if we should update i_disksize
3047 * when write to the end of file but not require block allocation
3049 static int ext4_da_should_update_i_disksize(struct page *page,
3050 unsigned long offset)
3052 struct buffer_head *bh;
3053 struct inode *inode = page->mapping->host;
3057 bh = page_buffers(page);
3058 idx = offset >> inode->i_blkbits;
3060 for (i = 0; i < idx; i++)
3061 bh = bh->b_this_page;
3063 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3068 static int ext4_da_write_end(struct file *file,
3069 struct address_space *mapping,
3070 loff_t pos, unsigned len, unsigned copied,
3071 struct page *page, void *fsdata)
3073 struct inode *inode = mapping->host;
3075 handle_t *handle = ext4_journal_current_handle();
3077 unsigned long start, end;
3078 int write_mode = (int)(unsigned long)fsdata;
3080 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3081 return ext4_write_end(file, mapping, pos,
3082 len, copied, page, fsdata);
3084 trace_ext4_da_write_end(inode, pos, len, copied);
3085 start = pos & (PAGE_SIZE - 1);
3086 end = start + copied - 1;
3089 * generic_write_end() will run mark_inode_dirty() if i_size
3090 * changes. So let's piggyback the i_disksize mark_inode_dirty
3093 new_i_size = pos + copied;
3094 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3095 if (ext4_has_inline_data(inode) ||
3096 ext4_da_should_update_i_disksize(page, end)) {
3097 ext4_update_i_disksize(inode, new_i_size);
3098 /* We need to mark inode dirty even if
3099 * new_i_size is less that inode->i_size
3100 * bu greater than i_disksize.(hint delalloc)
3102 ret = ext4_mark_inode_dirty(handle, inode);
3106 if (write_mode != CONVERT_INLINE_DATA &&
3107 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3108 ext4_has_inline_data(inode))
3109 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3112 ret2 = generic_write_end(file, mapping, pos, len, copied,
3118 ret2 = ext4_journal_stop(handle);
3119 if (unlikely(ret2 && !ret))
3122 return ret ? ret : copied;
3126 * Force all delayed allocation blocks to be allocated for a given inode.
3128 int ext4_alloc_da_blocks(struct inode *inode)
3130 trace_ext4_alloc_da_blocks(inode);
3132 if (!EXT4_I(inode)->i_reserved_data_blocks)
3136 * We do something simple for now. The filemap_flush() will
3137 * also start triggering a write of the data blocks, which is
3138 * not strictly speaking necessary (and for users of
3139 * laptop_mode, not even desirable). However, to do otherwise
3140 * would require replicating code paths in:
3142 * ext4_writepages() ->
3143 * write_cache_pages() ---> (via passed in callback function)
3144 * __mpage_da_writepage() -->
3145 * mpage_add_bh_to_extent()
3146 * mpage_da_map_blocks()
3148 * The problem is that write_cache_pages(), located in
3149 * mm/page-writeback.c, marks pages clean in preparation for
3150 * doing I/O, which is not desirable if we're not planning on
3153 * We could call write_cache_pages(), and then redirty all of
3154 * the pages by calling redirty_page_for_writepage() but that
3155 * would be ugly in the extreme. So instead we would need to
3156 * replicate parts of the code in the above functions,
3157 * simplifying them because we wouldn't actually intend to
3158 * write out the pages, but rather only collect contiguous
3159 * logical block extents, call the multi-block allocator, and
3160 * then update the buffer heads with the block allocations.
3162 * For now, though, we'll cheat by calling filemap_flush(),
3163 * which will map the blocks, and start the I/O, but not
3164 * actually wait for the I/O to complete.
3166 return filemap_flush(inode->i_mapping);
3170 * bmap() is special. It gets used by applications such as lilo and by
3171 * the swapper to find the on-disk block of a specific piece of data.
3173 * Naturally, this is dangerous if the block concerned is still in the
3174 * journal. If somebody makes a swapfile on an ext4 data-journaling
3175 * filesystem and enables swap, then they may get a nasty shock when the
3176 * data getting swapped to that swapfile suddenly gets overwritten by
3177 * the original zero's written out previously to the journal and
3178 * awaiting writeback in the kernel's buffer cache.
3180 * So, if we see any bmap calls here on a modified, data-journaled file,
3181 * take extra steps to flush any blocks which might be in the cache.
3183 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3185 struct inode *inode = mapping->host;
3190 * We can get here for an inline file via the FIBMAP ioctl
3192 if (ext4_has_inline_data(inode))
3195 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3196 test_opt(inode->i_sb, DELALLOC)) {
3198 * With delalloc we want to sync the file
3199 * so that we can make sure we allocate
3202 filemap_write_and_wait(mapping);
3205 if (EXT4_JOURNAL(inode) &&
3206 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3208 * This is a REALLY heavyweight approach, but the use of
3209 * bmap on dirty files is expected to be extremely rare:
3210 * only if we run lilo or swapon on a freshly made file
3211 * do we expect this to happen.
3213 * (bmap requires CAP_SYS_RAWIO so this does not
3214 * represent an unprivileged user DOS attack --- we'd be
3215 * in trouble if mortal users could trigger this path at
3218 * NB. EXT4_STATE_JDATA is not set on files other than
3219 * regular files. If somebody wants to bmap a directory
3220 * or symlink and gets confused because the buffer
3221 * hasn't yet been flushed to disk, they deserve
3222 * everything they get.
3225 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3226 journal = EXT4_JOURNAL(inode);
3227 jbd2_journal_lock_updates(journal);
3228 err = jbd2_journal_flush(journal);
3229 jbd2_journal_unlock_updates(journal);
3235 return iomap_bmap(mapping, block, &ext4_iomap_ops);
3238 static int ext4_readpage(struct file *file, struct page *page)
3241 struct inode *inode = page->mapping->host;
3243 trace_ext4_readpage(page);
3245 if (ext4_has_inline_data(inode))
3246 ret = ext4_readpage_inline(inode, page);
3249 return ext4_mpage_readpages(inode, NULL, page);
3254 static void ext4_readahead(struct readahead_control *rac)
3256 struct inode *inode = rac->mapping->host;
3258 /* If the file has inline data, no need to do readahead. */
3259 if (ext4_has_inline_data(inode))
3262 ext4_mpage_readpages(inode, rac, NULL);
3265 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3266 unsigned int length)
3268 trace_ext4_invalidatepage(page, offset, length);
3270 /* No journalling happens on data buffers when this function is used */
3271 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3273 block_invalidatepage(page, offset, length);
3276 static int __ext4_journalled_invalidatepage(struct page *page,
3277 unsigned int offset,
3278 unsigned int length)
3280 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3282 trace_ext4_journalled_invalidatepage(page, offset, length);
3285 * If it's a full truncate we just forget about the pending dirtying
3287 if (offset == 0 && length == PAGE_SIZE)
3288 ClearPageChecked(page);
3290 return jbd2_journal_invalidatepage(journal, page, offset, length);
3293 /* Wrapper for aops... */
3294 static void ext4_journalled_invalidatepage(struct page *page,
3295 unsigned int offset,
3296 unsigned int length)
3298 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3301 static int ext4_releasepage(struct page *page, gfp_t wait)
3303 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3305 trace_ext4_releasepage(page);
3307 /* Page has dirty journalled data -> cannot release */
3308 if (PageChecked(page))
3311 return jbd2_journal_try_to_free_buffers(journal, page);
3313 return try_to_free_buffers(page);
3316 static bool ext4_inode_datasync_dirty(struct inode *inode)
3318 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3321 if (jbd2_transaction_committed(journal,
3322 EXT4_I(inode)->i_datasync_tid))
3324 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3325 return !list_empty(&EXT4_I(inode)->i_fc_list);
3329 /* Any metadata buffers to write? */
3330 if (!list_empty(&inode->i_mapping->private_list))
3332 return inode->i_state & I_DIRTY_DATASYNC;
3335 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3336 struct ext4_map_blocks *map, loff_t offset,
3339 u8 blkbits = inode->i_blkbits;
3342 * Writes that span EOF might trigger an I/O size update on completion,
3343 * so consider them to be dirty for the purpose of O_DSYNC, even if
3344 * there is no other metadata changes being made or are pending.
3347 if (ext4_inode_datasync_dirty(inode) ||
3348 offset + length > i_size_read(inode))
3349 iomap->flags |= IOMAP_F_DIRTY;
3351 if (map->m_flags & EXT4_MAP_NEW)
3352 iomap->flags |= IOMAP_F_NEW;
3354 iomap->bdev = inode->i_sb->s_bdev;
3355 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3356 iomap->offset = (u64) map->m_lblk << blkbits;
3357 iomap->length = (u64) map->m_len << blkbits;
3359 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3360 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3361 iomap->flags |= IOMAP_F_MERGED;
3364 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3365 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3366 * set. In order for any allocated unwritten extents to be converted
3367 * into written extents correctly within the ->end_io() handler, we
3368 * need to ensure that the iomap->type is set appropriately. Hence, the
3369 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3372 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3373 iomap->type = IOMAP_UNWRITTEN;
3374 iomap->addr = (u64) map->m_pblk << blkbits;
3375 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3376 iomap->type = IOMAP_MAPPED;
3377 iomap->addr = (u64) map->m_pblk << blkbits;
3379 iomap->type = IOMAP_HOLE;
3380 iomap->addr = IOMAP_NULL_ADDR;
3384 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3388 u8 blkbits = inode->i_blkbits;
3389 int ret, dio_credits, m_flags = 0, retries = 0;
3392 * Trim the mapping request to the maximum value that we can map at
3393 * once for direct I/O.
3395 if (map->m_len > DIO_MAX_BLOCKS)
3396 map->m_len = DIO_MAX_BLOCKS;
3397 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3401 * Either we allocate blocks and then don't get an unwritten extent, so
3402 * in that case we have reserved enough credits. Or, the blocks are
3403 * already allocated and unwritten. In that case, the extent conversion
3404 * fits into the credits as well.
3406 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3408 return PTR_ERR(handle);
3411 * DAX and direct I/O are the only two operations that are currently
3412 * supported with IOMAP_WRITE.
3414 WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3416 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3418 * We use i_size instead of i_disksize here because delalloc writeback
3419 * can complete at any point during the I/O and subsequently push the
3420 * i_disksize out to i_size. This could be beyond where direct I/O is
3421 * happening and thus expose allocated blocks to direct I/O reads.
3423 else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode))
3424 m_flags = EXT4_GET_BLOCKS_CREATE;
3425 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3426 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3428 ret = ext4_map_blocks(handle, inode, map, m_flags);
3431 * We cannot fill holes in indirect tree based inodes as that could
3432 * expose stale data in the case of a crash. Use the magic error code
3433 * to fallback to buffered I/O.
3435 if (!m_flags && !ret)
3438 ext4_journal_stop(handle);
3439 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3446 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3447 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3450 struct ext4_map_blocks map;
3451 u8 blkbits = inode->i_blkbits;
3453 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3456 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3460 * Calculate the first and last logical blocks respectively.
3462 map.m_lblk = offset >> blkbits;
3463 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3464 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3466 if (flags & IOMAP_WRITE) {
3468 * We check here if the blocks are already allocated, then we
3469 * don't need to start a journal txn and we can directly return
3470 * the mapping information. This could boost performance
3471 * especially in multi-threaded overwrite requests.
3473 if (offset + length <= i_size_read(inode)) {
3474 ret = ext4_map_blocks(NULL, inode, &map, 0);
3475 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3478 ret = ext4_iomap_alloc(inode, &map, flags);
3480 ret = ext4_map_blocks(NULL, inode, &map, 0);
3486 ext4_set_iomap(inode, iomap, &map, offset, length);
3491 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3492 loff_t length, unsigned flags, struct iomap *iomap,
3493 struct iomap *srcmap)
3498 * Even for writes we don't need to allocate blocks, so just pretend
3499 * we are reading to save overhead of starting a transaction.
3501 flags &= ~IOMAP_WRITE;
3502 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3503 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3507 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3508 ssize_t written, unsigned flags, struct iomap *iomap)
3511 * Check to see whether an error occurred while writing out the data to
3512 * the allocated blocks. If so, return the magic error code so that we
3513 * fallback to buffered I/O and attempt to complete the remainder of
3514 * the I/O. Any blocks that may have been allocated in preparation for
3515 * the direct I/O will be reused during buffered I/O.
3517 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3523 const struct iomap_ops ext4_iomap_ops = {
3524 .iomap_begin = ext4_iomap_begin,
3525 .iomap_end = ext4_iomap_end,
3528 const struct iomap_ops ext4_iomap_overwrite_ops = {
3529 .iomap_begin = ext4_iomap_overwrite_begin,
3530 .iomap_end = ext4_iomap_end,
3533 static bool ext4_iomap_is_delalloc(struct inode *inode,
3534 struct ext4_map_blocks *map)
3536 struct extent_status es;
3537 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3539 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3540 map->m_lblk, end, &es);
3542 if (!es.es_len || es.es_lblk > end)
3545 if (es.es_lblk > map->m_lblk) {
3546 map->m_len = es.es_lblk - map->m_lblk;
3550 offset = map->m_lblk - es.es_lblk;
3551 map->m_len = es.es_len - offset;
3556 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3557 loff_t length, unsigned int flags,
3558 struct iomap *iomap, struct iomap *srcmap)
3561 bool delalloc = false;
3562 struct ext4_map_blocks map;
3563 u8 blkbits = inode->i_blkbits;
3565 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3568 if (ext4_has_inline_data(inode)) {
3569 ret = ext4_inline_data_iomap(inode, iomap);
3570 if (ret != -EAGAIN) {
3571 if (ret == 0 && offset >= iomap->length)
3578 * Calculate the first and last logical block respectively.
3580 map.m_lblk = offset >> blkbits;
3581 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3582 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3585 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3586 * So handle it here itself instead of querying ext4_map_blocks().
3587 * Since ext4_map_blocks() will warn about it and will return
3590 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3591 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3593 if (offset >= sbi->s_bitmap_maxbytes) {
3599 ret = ext4_map_blocks(NULL, inode, &map, 0);
3603 delalloc = ext4_iomap_is_delalloc(inode, &map);
3606 ext4_set_iomap(inode, iomap, &map, offset, length);
3607 if (delalloc && iomap->type == IOMAP_HOLE)
3608 iomap->type = IOMAP_DELALLOC;
3613 const struct iomap_ops ext4_iomap_report_ops = {
3614 .iomap_begin = ext4_iomap_begin_report,
3618 * Pages can be marked dirty completely asynchronously from ext4's journalling
3619 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3620 * much here because ->set_page_dirty is called under VFS locks. The page is
3621 * not necessarily locked.
3623 * We cannot just dirty the page and leave attached buffers clean, because the
3624 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3625 * or jbddirty because all the journalling code will explode.
3627 * So what we do is to mark the page "pending dirty" and next time writepage
3628 * is called, propagate that into the buffers appropriately.
3630 static int ext4_journalled_set_page_dirty(struct page *page)
3632 SetPageChecked(page);
3633 return __set_page_dirty_nobuffers(page);
3636 static int ext4_set_page_dirty(struct page *page)
3638 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3639 WARN_ON_ONCE(!page_has_buffers(page));
3640 return __set_page_dirty_buffers(page);
3643 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3644 struct file *file, sector_t *span)
3646 return iomap_swapfile_activate(sis, file, span,
3647 &ext4_iomap_report_ops);
3650 static const struct address_space_operations ext4_aops = {
3651 .readpage = ext4_readpage,
3652 .readahead = ext4_readahead,
3653 .writepage = ext4_writepage,
3654 .writepages = ext4_writepages,
3655 .write_begin = ext4_write_begin,
3656 .write_end = ext4_write_end,
3657 .set_page_dirty = ext4_set_page_dirty,
3659 .invalidatepage = ext4_invalidatepage,
3660 .releasepage = ext4_releasepage,
3661 .direct_IO = noop_direct_IO,
3662 .migratepage = buffer_migrate_page,
3663 .is_partially_uptodate = block_is_partially_uptodate,
3664 .error_remove_page = generic_error_remove_page,
3665 .swap_activate = ext4_iomap_swap_activate,
3668 static const struct address_space_operations ext4_journalled_aops = {
3669 .readpage = ext4_readpage,
3670 .readahead = ext4_readahead,
3671 .writepage = ext4_writepage,
3672 .writepages = ext4_writepages,
3673 .write_begin = ext4_write_begin,
3674 .write_end = ext4_journalled_write_end,
3675 .set_page_dirty = ext4_journalled_set_page_dirty,
3677 .invalidatepage = ext4_journalled_invalidatepage,
3678 .releasepage = ext4_releasepage,
3679 .direct_IO = noop_direct_IO,
3680 .is_partially_uptodate = block_is_partially_uptodate,
3681 .error_remove_page = generic_error_remove_page,
3682 .swap_activate = ext4_iomap_swap_activate,
3685 static const struct address_space_operations ext4_da_aops = {
3686 .readpage = ext4_readpage,
3687 .readahead = ext4_readahead,
3688 .writepage = ext4_writepage,
3689 .writepages = ext4_writepages,
3690 .write_begin = ext4_da_write_begin,
3691 .write_end = ext4_da_write_end,
3692 .set_page_dirty = ext4_set_page_dirty,
3694 .invalidatepage = ext4_invalidatepage,
3695 .releasepage = ext4_releasepage,
3696 .direct_IO = noop_direct_IO,
3697 .migratepage = buffer_migrate_page,
3698 .is_partially_uptodate = block_is_partially_uptodate,
3699 .error_remove_page = generic_error_remove_page,
3700 .swap_activate = ext4_iomap_swap_activate,
3703 static const struct address_space_operations ext4_dax_aops = {
3704 .writepages = ext4_dax_writepages,
3705 .direct_IO = noop_direct_IO,
3706 .set_page_dirty = noop_set_page_dirty,
3708 .invalidatepage = noop_invalidatepage,
3709 .swap_activate = ext4_iomap_swap_activate,
3712 void ext4_set_aops(struct inode *inode)
3714 switch (ext4_inode_journal_mode(inode)) {
3715 case EXT4_INODE_ORDERED_DATA_MODE:
3716 case EXT4_INODE_WRITEBACK_DATA_MODE:
3718 case EXT4_INODE_JOURNAL_DATA_MODE:
3719 inode->i_mapping->a_ops = &ext4_journalled_aops;
3725 inode->i_mapping->a_ops = &ext4_dax_aops;
3726 else if (test_opt(inode->i_sb, DELALLOC))
3727 inode->i_mapping->a_ops = &ext4_da_aops;
3729 inode->i_mapping->a_ops = &ext4_aops;
3732 static int __ext4_block_zero_page_range(handle_t *handle,
3733 struct address_space *mapping, loff_t from, loff_t length)
3735 ext4_fsblk_t index = from >> PAGE_SHIFT;
3736 unsigned offset = from & (PAGE_SIZE-1);
3737 unsigned blocksize, pos;
3739 struct inode *inode = mapping->host;
3740 struct buffer_head *bh;
3744 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3745 mapping_gfp_constraint(mapping, ~__GFP_FS));
3749 blocksize = inode->i_sb->s_blocksize;
3751 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3753 if (!page_has_buffers(page))
3754 create_empty_buffers(page, blocksize, 0);
3756 /* Find the buffer that contains "offset" */
3757 bh = page_buffers(page);
3759 while (offset >= pos) {
3760 bh = bh->b_this_page;
3764 if (buffer_freed(bh)) {
3765 BUFFER_TRACE(bh, "freed: skip");
3768 if (!buffer_mapped(bh)) {
3769 BUFFER_TRACE(bh, "unmapped");
3770 ext4_get_block(inode, iblock, bh, 0);
3771 /* unmapped? It's a hole - nothing to do */
3772 if (!buffer_mapped(bh)) {
3773 BUFFER_TRACE(bh, "still unmapped");
3778 /* Ok, it's mapped. Make sure it's up-to-date */
3779 if (PageUptodate(page))
3780 set_buffer_uptodate(bh);
3782 if (!buffer_uptodate(bh)) {
3783 err = ext4_read_bh_lock(bh, 0, true);
3786 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3787 /* We expect the key to be set. */
3788 BUG_ON(!fscrypt_has_encryption_key(inode));
3789 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3792 clear_buffer_uptodate(bh);
3797 if (ext4_should_journal_data(inode)) {
3798 BUFFER_TRACE(bh, "get write access");
3799 err = ext4_journal_get_write_access(handle, bh);
3803 zero_user(page, offset, length);
3804 BUFFER_TRACE(bh, "zeroed end of block");
3806 if (ext4_should_journal_data(inode)) {
3807 err = ext4_handle_dirty_metadata(handle, inode, bh);
3810 mark_buffer_dirty(bh);
3811 if (ext4_should_order_data(inode))
3812 err = ext4_jbd2_inode_add_write(handle, inode, from,
3823 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3824 * starting from file offset 'from'. The range to be zero'd must
3825 * be contained with in one block. If the specified range exceeds
3826 * the end of the block it will be shortened to end of the block
3827 * that cooresponds to 'from'
3829 static int ext4_block_zero_page_range(handle_t *handle,
3830 struct address_space *mapping, loff_t from, loff_t length)
3832 struct inode *inode = mapping->host;
3833 unsigned offset = from & (PAGE_SIZE-1);
3834 unsigned blocksize = inode->i_sb->s_blocksize;
3835 unsigned max = blocksize - (offset & (blocksize - 1));
3838 * correct length if it does not fall between
3839 * 'from' and the end of the block
3841 if (length > max || length < 0)
3844 if (IS_DAX(inode)) {
3845 return iomap_zero_range(inode, from, length, NULL,
3848 return __ext4_block_zero_page_range(handle, mapping, from, length);
3852 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3853 * up to the end of the block which corresponds to `from'.
3854 * This required during truncate. We need to physically zero the tail end
3855 * of that block so it doesn't yield old data if the file is later grown.
3857 static int ext4_block_truncate_page(handle_t *handle,
3858 struct address_space *mapping, loff_t from)
3860 unsigned offset = from & (PAGE_SIZE-1);
3863 struct inode *inode = mapping->host;
3865 /* If we are processing an encrypted inode during orphan list handling */
3866 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3869 blocksize = inode->i_sb->s_blocksize;
3870 length = blocksize - (offset & (blocksize - 1));
3872 return ext4_block_zero_page_range(handle, mapping, from, length);
3875 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3876 loff_t lstart, loff_t length)
3878 struct super_block *sb = inode->i_sb;
3879 struct address_space *mapping = inode->i_mapping;
3880 unsigned partial_start, partial_end;
3881 ext4_fsblk_t start, end;
3882 loff_t byte_end = (lstart + length - 1);
3885 partial_start = lstart & (sb->s_blocksize - 1);
3886 partial_end = byte_end & (sb->s_blocksize - 1);
3888 start = lstart >> sb->s_blocksize_bits;
3889 end = byte_end >> sb->s_blocksize_bits;
3891 /* Handle partial zero within the single block */
3893 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3894 err = ext4_block_zero_page_range(handle, mapping,
3898 /* Handle partial zero out on the start of the range */
3899 if (partial_start) {
3900 err = ext4_block_zero_page_range(handle, mapping,
3901 lstart, sb->s_blocksize);
3905 /* Handle partial zero out on the end of the range */
3906 if (partial_end != sb->s_blocksize - 1)
3907 err = ext4_block_zero_page_range(handle, mapping,
3908 byte_end - partial_end,
3913 int ext4_can_truncate(struct inode *inode)
3915 if (S_ISREG(inode->i_mode))
3917 if (S_ISDIR(inode->i_mode))
3919 if (S_ISLNK(inode->i_mode))
3920 return !ext4_inode_is_fast_symlink(inode);
3925 * We have to make sure i_disksize gets properly updated before we truncate
3926 * page cache due to hole punching or zero range. Otherwise i_disksize update
3927 * can get lost as it may have been postponed to submission of writeback but
3928 * that will never happen after we truncate page cache.
3930 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3936 loff_t size = i_size_read(inode);
3938 WARN_ON(!inode_is_locked(inode));
3939 if (offset > size || offset + len < size)
3942 if (EXT4_I(inode)->i_disksize >= size)
3945 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3947 return PTR_ERR(handle);
3948 ext4_update_i_disksize(inode, size);
3949 ret = ext4_mark_inode_dirty(handle, inode);
3950 ext4_journal_stop(handle);
3955 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
3957 up_write(&ei->i_mmap_sem);
3959 down_write(&ei->i_mmap_sem);
3962 int ext4_break_layouts(struct inode *inode)
3964 struct ext4_inode_info *ei = EXT4_I(inode);
3968 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
3972 page = dax_layout_busy_page(inode->i_mapping);
3976 error = ___wait_var_event(&page->_refcount,
3977 atomic_read(&page->_refcount) == 1,
3978 TASK_INTERRUPTIBLE, 0, 0,
3979 ext4_wait_dax_page(ei));
3980 } while (error == 0);
3986 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3987 * associated with the given offset and length
3989 * @inode: File inode
3990 * @offset: The offset where the hole will begin
3991 * @len: The length of the hole
3993 * Returns: 0 on success or negative on failure
3996 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3998 struct super_block *sb = inode->i_sb;
3999 ext4_lblk_t first_block, stop_block;
4000 struct address_space *mapping = inode->i_mapping;
4001 loff_t first_block_offset, last_block_offset;
4003 unsigned int credits;
4004 int ret = 0, ret2 = 0;
4006 trace_ext4_punch_hole(inode, offset, length, 0);
4008 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4009 if (ext4_has_inline_data(inode)) {
4010 down_write(&EXT4_I(inode)->i_mmap_sem);
4011 ret = ext4_convert_inline_data(inode);
4012 up_write(&EXT4_I(inode)->i_mmap_sem);
4018 * Write out all dirty pages to avoid race conditions
4019 * Then release them.
4021 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4022 ret = filemap_write_and_wait_range(mapping, offset,
4023 offset + length - 1);
4030 /* No need to punch hole beyond i_size */
4031 if (offset >= inode->i_size)
4035 * If the hole extends beyond i_size, set the hole
4036 * to end after the page that contains i_size
4038 if (offset + length > inode->i_size) {
4039 length = inode->i_size +
4040 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4044 if (offset & (sb->s_blocksize - 1) ||
4045 (offset + length) & (sb->s_blocksize - 1)) {
4047 * Attach jinode to inode for jbd2 if we do any zeroing of
4050 ret = ext4_inode_attach_jinode(inode);
4056 /* Wait all existing dio workers, newcomers will block on i_mutex */
4057 inode_dio_wait(inode);
4060 * Prevent page faults from reinstantiating pages we have released from
4063 down_write(&EXT4_I(inode)->i_mmap_sem);
4065 ret = ext4_break_layouts(inode);
4069 first_block_offset = round_up(offset, sb->s_blocksize);
4070 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4072 /* Now release the pages and zero block aligned part of pages*/
4073 if (last_block_offset > first_block_offset) {
4074 ret = ext4_update_disksize_before_punch(inode, offset, length);
4077 truncate_pagecache_range(inode, first_block_offset,
4081 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4082 credits = ext4_writepage_trans_blocks(inode);
4084 credits = ext4_blocks_for_truncate(inode);
4085 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4086 if (IS_ERR(handle)) {
4087 ret = PTR_ERR(handle);
4088 ext4_std_error(sb, ret);
4092 ret = ext4_zero_partial_blocks(handle, inode, offset,
4097 first_block = (offset + sb->s_blocksize - 1) >>
4098 EXT4_BLOCK_SIZE_BITS(sb);
4099 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4101 /* If there are blocks to remove, do it */
4102 if (stop_block > first_block) {
4104 down_write(&EXT4_I(inode)->i_data_sem);
4105 ext4_discard_preallocations(inode, 0);
4107 ret = ext4_es_remove_extent(inode, first_block,
4108 stop_block - first_block);
4110 up_write(&EXT4_I(inode)->i_data_sem);
4114 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4115 ret = ext4_ext_remove_space(inode, first_block,
4118 ret = ext4_ind_remove_space(handle, inode, first_block,
4121 up_write(&EXT4_I(inode)->i_data_sem);
4123 ext4_fc_track_range(handle, inode, first_block, stop_block);
4125 ext4_handle_sync(handle);
4127 inode->i_mtime = inode->i_ctime = current_time(inode);
4128 ret2 = ext4_mark_inode_dirty(handle, inode);
4132 ext4_update_inode_fsync_trans(handle, inode, 1);
4134 ext4_journal_stop(handle);
4136 up_write(&EXT4_I(inode)->i_mmap_sem);
4138 inode_unlock(inode);
4142 int ext4_inode_attach_jinode(struct inode *inode)
4144 struct ext4_inode_info *ei = EXT4_I(inode);
4145 struct jbd2_inode *jinode;
4147 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4150 jinode = jbd2_alloc_inode(GFP_KERNEL);
4151 spin_lock(&inode->i_lock);
4154 spin_unlock(&inode->i_lock);
4157 ei->jinode = jinode;
4158 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4161 spin_unlock(&inode->i_lock);
4162 if (unlikely(jinode != NULL))
4163 jbd2_free_inode(jinode);
4170 * We block out ext4_get_block() block instantiations across the entire
4171 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4172 * simultaneously on behalf of the same inode.
4174 * As we work through the truncate and commit bits of it to the journal there
4175 * is one core, guiding principle: the file's tree must always be consistent on
4176 * disk. We must be able to restart the truncate after a crash.
4178 * The file's tree may be transiently inconsistent in memory (although it
4179 * probably isn't), but whenever we close off and commit a journal transaction,
4180 * the contents of (the filesystem + the journal) must be consistent and
4181 * restartable. It's pretty simple, really: bottom up, right to left (although
4182 * left-to-right works OK too).
4184 * Note that at recovery time, journal replay occurs *before* the restart of
4185 * truncate against the orphan inode list.
4187 * The committed inode has the new, desired i_size (which is the same as
4188 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4189 * that this inode's truncate did not complete and it will again call
4190 * ext4_truncate() to have another go. So there will be instantiated blocks
4191 * to the right of the truncation point in a crashed ext4 filesystem. But
4192 * that's fine - as long as they are linked from the inode, the post-crash
4193 * ext4_truncate() run will find them and release them.
4195 int ext4_truncate(struct inode *inode)
4197 struct ext4_inode_info *ei = EXT4_I(inode);
4198 unsigned int credits;
4201 struct address_space *mapping = inode->i_mapping;
4204 * There is a possibility that we're either freeing the inode
4205 * or it's a completely new inode. In those cases we might not
4206 * have i_mutex locked because it's not necessary.
4208 if (!(inode->i_state & (I_NEW|I_FREEING)))
4209 WARN_ON(!inode_is_locked(inode));
4210 trace_ext4_truncate_enter(inode);
4212 if (!ext4_can_truncate(inode))
4215 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4216 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4218 if (ext4_has_inline_data(inode)) {
4221 err = ext4_inline_data_truncate(inode, &has_inline);
4222 if (err || has_inline)
4226 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4227 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4228 if (ext4_inode_attach_jinode(inode) < 0)
4232 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4233 credits = ext4_writepage_trans_blocks(inode);
4235 credits = ext4_blocks_for_truncate(inode);
4237 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4238 if (IS_ERR(handle)) {
4239 err = PTR_ERR(handle);
4243 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4244 ext4_block_truncate_page(handle, mapping, inode->i_size);
4247 * We add the inode to the orphan list, so that if this
4248 * truncate spans multiple transactions, and we crash, we will
4249 * resume the truncate when the filesystem recovers. It also
4250 * marks the inode dirty, to catch the new size.
4252 * Implication: the file must always be in a sane, consistent
4253 * truncatable state while each transaction commits.
4255 err = ext4_orphan_add(handle, inode);
4259 down_write(&EXT4_I(inode)->i_data_sem);
4261 ext4_discard_preallocations(inode, 0);
4263 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4264 err = ext4_ext_truncate(handle, inode);
4266 ext4_ind_truncate(handle, inode);
4268 up_write(&ei->i_data_sem);
4273 ext4_handle_sync(handle);
4277 * If this was a simple ftruncate() and the file will remain alive,
4278 * then we need to clear up the orphan record which we created above.
4279 * However, if this was a real unlink then we were called by
4280 * ext4_evict_inode(), and we allow that function to clean up the
4281 * orphan info for us.
4284 ext4_orphan_del(handle, inode);
4286 inode->i_mtime = inode->i_ctime = current_time(inode);
4287 err2 = ext4_mark_inode_dirty(handle, inode);
4288 if (unlikely(err2 && !err))
4290 ext4_journal_stop(handle);
4293 trace_ext4_truncate_exit(inode);
4298 * ext4_get_inode_loc returns with an extra refcount against the inode's
4299 * underlying buffer_head on success. If 'in_mem' is true, we have all
4300 * data in memory that is needed to recreate the on-disk version of this
4303 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4304 struct ext4_iloc *iloc, int in_mem,
4305 ext4_fsblk_t *ret_block)
4307 struct ext4_group_desc *gdp;
4308 struct buffer_head *bh;
4310 struct blk_plug plug;
4311 int inodes_per_block, inode_offset;
4314 if (ino < EXT4_ROOT_INO ||
4315 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4316 return -EFSCORRUPTED;
4318 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4319 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4324 * Figure out the offset within the block group inode table
4326 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4327 inode_offset = ((ino - 1) %
4328 EXT4_INODES_PER_GROUP(sb));
4329 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4330 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4332 bh = sb_getblk(sb, block);
4335 if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO))
4337 if (!buffer_uptodate(bh)) {
4340 if (ext4_buffer_uptodate(bh)) {
4341 /* someone brought it uptodate while we waited */
4347 * If we have all information of the inode in memory and this
4348 * is the only valid inode in the block, we need not read the
4352 struct buffer_head *bitmap_bh;
4355 start = inode_offset & ~(inodes_per_block - 1);
4357 /* Is the inode bitmap in cache? */
4358 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4359 if (unlikely(!bitmap_bh))
4363 * If the inode bitmap isn't in cache then the
4364 * optimisation may end up performing two reads instead
4365 * of one, so skip it.
4367 if (!buffer_uptodate(bitmap_bh)) {
4371 for (i = start; i < start + inodes_per_block; i++) {
4372 if (i == inode_offset)
4374 if (ext4_test_bit(i, bitmap_bh->b_data))
4378 if (i == start + inodes_per_block) {
4379 /* all other inodes are free, so skip I/O */
4380 memset(bh->b_data, 0, bh->b_size);
4381 set_buffer_uptodate(bh);
4389 * If we need to do any I/O, try to pre-readahead extra
4390 * blocks from the inode table.
4392 blk_start_plug(&plug);
4393 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4394 ext4_fsblk_t b, end, table;
4396 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4398 table = ext4_inode_table(sb, gdp);
4399 /* s_inode_readahead_blks is always a power of 2 */
4400 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4404 num = EXT4_INODES_PER_GROUP(sb);
4405 if (ext4_has_group_desc_csum(sb))
4406 num -= ext4_itable_unused_count(sb, gdp);
4407 table += num / inodes_per_block;
4411 ext4_sb_breadahead_unmovable(sb, b++);
4415 * There are other valid inodes in the buffer, this inode
4416 * has in-inode xattrs, or we don't have this inode in memory.
4417 * Read the block from disk.
4419 trace_ext4_load_inode(sb, ino);
4420 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4421 blk_finish_plug(&plug);
4423 if (!buffer_uptodate(bh)) {
4436 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4437 struct ext4_iloc *iloc)
4439 ext4_fsblk_t err_blk;
4442 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, 0,
4446 ext4_error_inode_block(inode, err_blk, EIO,
4447 "unable to read itable block");
4452 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4454 ext4_fsblk_t err_blk;
4457 /* We have all inode data except xattrs in memory here. */
4458 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc,
4459 !ext4_test_inode_state(inode, EXT4_STATE_XATTR), &err_blk);
4462 ext4_error_inode_block(inode, err_blk, EIO,
4463 "unable to read itable block");
4469 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4470 struct ext4_iloc *iloc)
4472 return __ext4_get_inode_loc(sb, ino, iloc, 0, NULL);
4475 static bool ext4_should_enable_dax(struct inode *inode)
4477 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4479 if (test_opt2(inode->i_sb, DAX_NEVER))
4481 if (!S_ISREG(inode->i_mode))
4483 if (ext4_should_journal_data(inode))
4485 if (ext4_has_inline_data(inode))
4487 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4489 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4491 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4493 if (test_opt(inode->i_sb, DAX_ALWAYS))
4496 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4499 void ext4_set_inode_flags(struct inode *inode, bool init)
4501 unsigned int flags = EXT4_I(inode)->i_flags;
4502 unsigned int new_fl = 0;
4504 WARN_ON_ONCE(IS_DAX(inode) && init);
4506 if (flags & EXT4_SYNC_FL)
4508 if (flags & EXT4_APPEND_FL)
4510 if (flags & EXT4_IMMUTABLE_FL)
4511 new_fl |= S_IMMUTABLE;
4512 if (flags & EXT4_NOATIME_FL)
4513 new_fl |= S_NOATIME;
4514 if (flags & EXT4_DIRSYNC_FL)
4515 new_fl |= S_DIRSYNC;
4517 /* Because of the way inode_set_flags() works we must preserve S_DAX
4518 * here if already set. */
4519 new_fl |= (inode->i_flags & S_DAX);
4520 if (init && ext4_should_enable_dax(inode))
4523 if (flags & EXT4_ENCRYPT_FL)
4524 new_fl |= S_ENCRYPTED;
4525 if (flags & EXT4_CASEFOLD_FL)
4526 new_fl |= S_CASEFOLD;
4527 if (flags & EXT4_VERITY_FL)
4529 inode_set_flags(inode, new_fl,
4530 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4531 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4534 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4535 struct ext4_inode_info *ei)
4538 struct inode *inode = &(ei->vfs_inode);
4539 struct super_block *sb = inode->i_sb;
4541 if (ext4_has_feature_huge_file(sb)) {
4542 /* we are using combined 48 bit field */
4543 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4544 le32_to_cpu(raw_inode->i_blocks_lo);
4545 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4546 /* i_blocks represent file system block size */
4547 return i_blocks << (inode->i_blkbits - 9);
4552 return le32_to_cpu(raw_inode->i_blocks_lo);
4556 static inline int ext4_iget_extra_inode(struct inode *inode,
4557 struct ext4_inode *raw_inode,
4558 struct ext4_inode_info *ei)
4560 __le32 *magic = (void *)raw_inode +
4561 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4563 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4564 EXT4_INODE_SIZE(inode->i_sb) &&
4565 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4566 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4567 return ext4_find_inline_data_nolock(inode);
4569 EXT4_I(inode)->i_inline_off = 0;
4573 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4575 if (!ext4_has_feature_project(inode->i_sb))
4577 *projid = EXT4_I(inode)->i_projid;
4582 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4583 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4586 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4588 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4589 inode_set_iversion_raw(inode, val);
4591 inode_set_iversion_queried(inode, val);
4593 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4595 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4596 return inode_peek_iversion_raw(inode);
4598 return inode_peek_iversion(inode);
4601 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4602 ext4_iget_flags flags, const char *function,
4605 struct ext4_iloc iloc;
4606 struct ext4_inode *raw_inode;
4607 struct ext4_inode_info *ei;
4608 struct inode *inode;
4609 journal_t *journal = EXT4_SB(sb)->s_journal;
4617 if ((!(flags & EXT4_IGET_SPECIAL) &&
4618 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4619 (ino < EXT4_ROOT_INO) ||
4620 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4621 if (flags & EXT4_IGET_HANDLE)
4622 return ERR_PTR(-ESTALE);
4623 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4624 "inode #%lu: comm %s: iget: illegal inode #",
4625 ino, current->comm);
4626 return ERR_PTR(-EFSCORRUPTED);
4629 inode = iget_locked(sb, ino);
4631 return ERR_PTR(-ENOMEM);
4632 if (!(inode->i_state & I_NEW))
4638 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4641 raw_inode = ext4_raw_inode(&iloc);
4643 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4644 ext4_error_inode(inode, function, line, 0,
4645 "iget: root inode unallocated");
4646 ret = -EFSCORRUPTED;
4650 if ((flags & EXT4_IGET_HANDLE) &&
4651 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4656 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4657 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4658 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4659 EXT4_INODE_SIZE(inode->i_sb) ||
4660 (ei->i_extra_isize & 3)) {
4661 ext4_error_inode(inode, function, line, 0,
4662 "iget: bad extra_isize %u "
4665 EXT4_INODE_SIZE(inode->i_sb));
4666 ret = -EFSCORRUPTED;
4670 ei->i_extra_isize = 0;
4672 /* Precompute checksum seed for inode metadata */
4673 if (ext4_has_metadata_csum(sb)) {
4674 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4676 __le32 inum = cpu_to_le32(inode->i_ino);
4677 __le32 gen = raw_inode->i_generation;
4678 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4680 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4684 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4685 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4686 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4687 ext4_error_inode_err(inode, function, line, 0,
4688 EFSBADCRC, "iget: checksum invalid");
4693 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4694 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4695 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4696 if (ext4_has_feature_project(sb) &&
4697 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4698 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4699 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4701 i_projid = EXT4_DEF_PROJID;
4703 if (!(test_opt(inode->i_sb, NO_UID32))) {
4704 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4705 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4707 i_uid_write(inode, i_uid);
4708 i_gid_write(inode, i_gid);
4709 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4710 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4712 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4713 ei->i_inline_off = 0;
4714 ei->i_dir_start_lookup = 0;
4715 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4716 /* We now have enough fields to check if the inode was active or not.
4717 * This is needed because nfsd might try to access dead inodes
4718 * the test is that same one that e2fsck uses
4719 * NeilBrown 1999oct15
4721 if (inode->i_nlink == 0) {
4722 if ((inode->i_mode == 0 ||
4723 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4724 ino != EXT4_BOOT_LOADER_INO) {
4725 /* this inode is deleted */
4729 /* The only unlinked inodes we let through here have
4730 * valid i_mode and are being read by the orphan
4731 * recovery code: that's fine, we're about to complete
4732 * the process of deleting those.
4733 * OR it is the EXT4_BOOT_LOADER_INO which is
4734 * not initialized on a new filesystem. */
4736 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4737 ext4_set_inode_flags(inode, true);
4738 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4739 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4740 if (ext4_has_feature_64bit(sb))
4742 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4743 inode->i_size = ext4_isize(sb, raw_inode);
4744 if ((size = i_size_read(inode)) < 0) {
4745 ext4_error_inode(inode, function, line, 0,
4746 "iget: bad i_size value: %lld", size);
4747 ret = -EFSCORRUPTED;
4751 * If dir_index is not enabled but there's dir with INDEX flag set,
4752 * we'd normally treat htree data as empty space. But with metadata
4753 * checksumming that corrupts checksums so forbid that.
4755 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4756 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4757 ext4_error_inode(inode, function, line, 0,
4758 "iget: Dir with htree data on filesystem without dir_index feature.");
4759 ret = -EFSCORRUPTED;
4762 ei->i_disksize = inode->i_size;
4764 ei->i_reserved_quota = 0;
4766 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4767 ei->i_block_group = iloc.block_group;
4768 ei->i_last_alloc_group = ~0;
4770 * NOTE! The in-memory inode i_data array is in little-endian order
4771 * even on big-endian machines: we do NOT byteswap the block numbers!
4773 for (block = 0; block < EXT4_N_BLOCKS; block++)
4774 ei->i_data[block] = raw_inode->i_block[block];
4775 INIT_LIST_HEAD(&ei->i_orphan);
4776 ext4_fc_init_inode(&ei->vfs_inode);
4779 * Set transaction id's of transactions that have to be committed
4780 * to finish f[data]sync. We set them to currently running transaction
4781 * as we cannot be sure that the inode or some of its metadata isn't
4782 * part of the transaction - the inode could have been reclaimed and
4783 * now it is reread from disk.
4786 transaction_t *transaction;
4789 read_lock(&journal->j_state_lock);
4790 if (journal->j_running_transaction)
4791 transaction = journal->j_running_transaction;
4793 transaction = journal->j_committing_transaction;
4795 tid = transaction->t_tid;
4797 tid = journal->j_commit_sequence;
4798 read_unlock(&journal->j_state_lock);
4799 ei->i_sync_tid = tid;
4800 ei->i_datasync_tid = tid;
4803 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4804 if (ei->i_extra_isize == 0) {
4805 /* The extra space is currently unused. Use it. */
4806 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4807 ei->i_extra_isize = sizeof(struct ext4_inode) -
4808 EXT4_GOOD_OLD_INODE_SIZE;
4810 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4816 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4817 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4818 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4819 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4821 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4822 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4824 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4825 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4827 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4829 ext4_inode_set_iversion_queried(inode, ivers);
4833 if (ei->i_file_acl &&
4834 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4835 ext4_error_inode(inode, function, line, 0,
4836 "iget: bad extended attribute block %llu",
4838 ret = -EFSCORRUPTED;
4840 } else if (!ext4_has_inline_data(inode)) {
4841 /* validate the block references in the inode */
4842 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4843 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4844 (S_ISLNK(inode->i_mode) &&
4845 !ext4_inode_is_fast_symlink(inode)))) {
4846 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4847 ret = ext4_ext_check_inode(inode);
4849 ret = ext4_ind_check_inode(inode);
4855 if (S_ISREG(inode->i_mode)) {
4856 inode->i_op = &ext4_file_inode_operations;
4857 inode->i_fop = &ext4_file_operations;
4858 ext4_set_aops(inode);
4859 } else if (S_ISDIR(inode->i_mode)) {
4860 inode->i_op = &ext4_dir_inode_operations;
4861 inode->i_fop = &ext4_dir_operations;
4862 } else if (S_ISLNK(inode->i_mode)) {
4863 /* VFS does not allow setting these so must be corruption */
4864 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4865 ext4_error_inode(inode, function, line, 0,
4866 "iget: immutable or append flags "
4867 "not allowed on symlinks");
4868 ret = -EFSCORRUPTED;
4871 if (IS_ENCRYPTED(inode)) {
4872 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4873 ext4_set_aops(inode);
4874 } else if (ext4_inode_is_fast_symlink(inode)) {
4875 inode->i_link = (char *)ei->i_data;
4876 inode->i_op = &ext4_fast_symlink_inode_operations;
4877 nd_terminate_link(ei->i_data, inode->i_size,
4878 sizeof(ei->i_data) - 1);
4880 inode->i_op = &ext4_symlink_inode_operations;
4881 ext4_set_aops(inode);
4883 inode_nohighmem(inode);
4884 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4885 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4886 inode->i_op = &ext4_special_inode_operations;
4887 if (raw_inode->i_block[0])
4888 init_special_inode(inode, inode->i_mode,
4889 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4891 init_special_inode(inode, inode->i_mode,
4892 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4893 } else if (ino == EXT4_BOOT_LOADER_INO) {
4894 make_bad_inode(inode);
4896 ret = -EFSCORRUPTED;
4897 ext4_error_inode(inode, function, line, 0,
4898 "iget: bogus i_mode (%o)", inode->i_mode);
4901 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4902 ext4_error_inode(inode, function, line, 0,
4903 "casefold flag without casefold feature");
4906 unlock_new_inode(inode);
4912 return ERR_PTR(ret);
4915 static int ext4_inode_blocks_set(handle_t *handle,
4916 struct ext4_inode *raw_inode,
4917 struct ext4_inode_info *ei)
4919 struct inode *inode = &(ei->vfs_inode);
4920 u64 i_blocks = READ_ONCE(inode->i_blocks);
4921 struct super_block *sb = inode->i_sb;
4923 if (i_blocks <= ~0U) {
4925 * i_blocks can be represented in a 32 bit variable
4926 * as multiple of 512 bytes
4928 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4929 raw_inode->i_blocks_high = 0;
4930 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4933 if (!ext4_has_feature_huge_file(sb))
4936 if (i_blocks <= 0xffffffffffffULL) {
4938 * i_blocks can be represented in a 48 bit variable
4939 * as multiple of 512 bytes
4941 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4942 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4943 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4945 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4946 /* i_block is stored in file system block size */
4947 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4948 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4949 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4954 static void __ext4_update_other_inode_time(struct super_block *sb,
4955 unsigned long orig_ino,
4957 struct ext4_inode *raw_inode)
4959 struct inode *inode;
4961 inode = find_inode_by_ino_rcu(sb, ino);
4965 if (!inode_is_dirtytime_only(inode))
4968 spin_lock(&inode->i_lock);
4969 if (inode_is_dirtytime_only(inode)) {
4970 struct ext4_inode_info *ei = EXT4_I(inode);
4972 inode->i_state &= ~I_DIRTY_TIME;
4973 spin_unlock(&inode->i_lock);
4975 spin_lock(&ei->i_raw_lock);
4976 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4977 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4978 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4979 ext4_inode_csum_set(inode, raw_inode, ei);
4980 spin_unlock(&ei->i_raw_lock);
4981 trace_ext4_other_inode_update_time(inode, orig_ino);
4984 spin_unlock(&inode->i_lock);
4988 * Opportunistically update the other time fields for other inodes in
4989 * the same inode table block.
4991 static void ext4_update_other_inodes_time(struct super_block *sb,
4992 unsigned long orig_ino, char *buf)
4995 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4996 int inode_size = EXT4_INODE_SIZE(sb);
4999 * Calculate the first inode in the inode table block. Inode
5000 * numbers are one-based. That is, the first inode in a block
5001 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5003 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5005 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5006 if (ino == orig_ino)
5008 __ext4_update_other_inode_time(sb, orig_ino, ino,
5009 (struct ext4_inode *)buf);
5015 * Post the struct inode info into an on-disk inode location in the
5016 * buffer-cache. This gobbles the caller's reference to the
5017 * buffer_head in the inode location struct.
5019 * The caller must have write access to iloc->bh.
5021 static int ext4_do_update_inode(handle_t *handle,
5022 struct inode *inode,
5023 struct ext4_iloc *iloc)
5025 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5026 struct ext4_inode_info *ei = EXT4_I(inode);
5027 struct buffer_head *bh = iloc->bh;
5028 struct super_block *sb = inode->i_sb;
5029 int err = 0, rc, block;
5030 int need_datasync = 0, set_large_file = 0;
5035 spin_lock(&ei->i_raw_lock);
5037 /* For fields not tracked in the in-memory inode,
5038 * initialise them to zero for new inodes. */
5039 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5040 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5042 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5044 spin_unlock(&ei->i_raw_lock);
5048 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5049 i_uid = i_uid_read(inode);
5050 i_gid = i_gid_read(inode);
5051 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5052 if (!(test_opt(inode->i_sb, NO_UID32))) {
5053 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5054 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5056 * Fix up interoperability with old kernels. Otherwise, old inodes get
5057 * re-used with the upper 16 bits of the uid/gid intact
5059 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5060 raw_inode->i_uid_high = 0;
5061 raw_inode->i_gid_high = 0;
5063 raw_inode->i_uid_high =
5064 cpu_to_le16(high_16_bits(i_uid));
5065 raw_inode->i_gid_high =
5066 cpu_to_le16(high_16_bits(i_gid));
5069 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5070 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5071 raw_inode->i_uid_high = 0;
5072 raw_inode->i_gid_high = 0;
5074 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5076 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5077 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5078 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5079 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5081 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5082 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5083 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5084 raw_inode->i_file_acl_high =
5085 cpu_to_le16(ei->i_file_acl >> 32);
5086 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5087 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5088 ext4_isize_set(raw_inode, ei->i_disksize);
5091 if (ei->i_disksize > 0x7fffffffULL) {
5092 if (!ext4_has_feature_large_file(sb) ||
5093 EXT4_SB(sb)->s_es->s_rev_level ==
5094 cpu_to_le32(EXT4_GOOD_OLD_REV))
5097 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5098 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5099 if (old_valid_dev(inode->i_rdev)) {
5100 raw_inode->i_block[0] =
5101 cpu_to_le32(old_encode_dev(inode->i_rdev));
5102 raw_inode->i_block[1] = 0;
5104 raw_inode->i_block[0] = 0;
5105 raw_inode->i_block[1] =
5106 cpu_to_le32(new_encode_dev(inode->i_rdev));
5107 raw_inode->i_block[2] = 0;
5109 } else if (!ext4_has_inline_data(inode)) {
5110 for (block = 0; block < EXT4_N_BLOCKS; block++)
5111 raw_inode->i_block[block] = ei->i_data[block];
5114 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5115 u64 ivers = ext4_inode_peek_iversion(inode);
5117 raw_inode->i_disk_version = cpu_to_le32(ivers);
5118 if (ei->i_extra_isize) {
5119 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5120 raw_inode->i_version_hi =
5121 cpu_to_le32(ivers >> 32);
5122 raw_inode->i_extra_isize =
5123 cpu_to_le16(ei->i_extra_isize);
5127 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5128 i_projid != EXT4_DEF_PROJID);
5130 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5131 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5132 raw_inode->i_projid = cpu_to_le32(i_projid);
5134 ext4_inode_csum_set(inode, raw_inode, ei);
5135 spin_unlock(&ei->i_raw_lock);
5136 if (inode->i_sb->s_flags & SB_LAZYTIME)
5137 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5140 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5141 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5144 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5145 if (set_large_file) {
5146 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5147 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5150 lock_buffer(EXT4_SB(sb)->s_sbh);
5151 ext4_set_feature_large_file(sb);
5152 ext4_superblock_csum_set(sb);
5153 unlock_buffer(EXT4_SB(sb)->s_sbh);
5154 ext4_handle_sync(handle);
5155 err = ext4_handle_dirty_metadata(handle, NULL,
5156 EXT4_SB(sb)->s_sbh);
5158 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5161 ext4_std_error(inode->i_sb, err);
5166 * ext4_write_inode()
5168 * We are called from a few places:
5170 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5171 * Here, there will be no transaction running. We wait for any running
5172 * transaction to commit.
5174 * - Within flush work (sys_sync(), kupdate and such).
5175 * We wait on commit, if told to.
5177 * - Within iput_final() -> write_inode_now()
5178 * We wait on commit, if told to.
5180 * In all cases it is actually safe for us to return without doing anything,
5181 * because the inode has been copied into a raw inode buffer in
5182 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5185 * Note that we are absolutely dependent upon all inode dirtiers doing the
5186 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5187 * which we are interested.
5189 * It would be a bug for them to not do this. The code:
5191 * mark_inode_dirty(inode)
5193 * inode->i_size = expr;
5195 * is in error because write_inode() could occur while `stuff()' is running,
5196 * and the new i_size will be lost. Plus the inode will no longer be on the
5197 * superblock's dirty inode list.
5199 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5203 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5204 sb_rdonly(inode->i_sb))
5207 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5210 if (EXT4_SB(inode->i_sb)->s_journal) {
5211 if (ext4_journal_current_handle()) {
5212 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5218 * No need to force transaction in WB_SYNC_NONE mode. Also
5219 * ext4_sync_fs() will force the commit after everything is
5222 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5225 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5226 EXT4_I(inode)->i_sync_tid);
5228 struct ext4_iloc iloc;
5230 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5234 * sync(2) will flush the whole buffer cache. No need to do
5235 * it here separately for each inode.
5237 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5238 sync_dirty_buffer(iloc.bh);
5239 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5240 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5241 "IO error syncing inode");
5250 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5251 * buffers that are attached to a page stradding i_size and are undergoing
5252 * commit. In that case we have to wait for commit to finish and try again.
5254 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5258 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5259 tid_t commit_tid = 0;
5262 offset = inode->i_size & (PAGE_SIZE - 1);
5264 * If the page is fully truncated, we don't need to wait for any commit
5265 * (and we even should not as __ext4_journalled_invalidatepage() may
5266 * strip all buffers from the page but keep the page dirty which can then
5267 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5268 * buffers). Also we don't need to wait for any commit if all buffers in
5269 * the page remain valid. This is most beneficial for the common case of
5270 * blocksize == PAGESIZE.
5272 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5275 page = find_lock_page(inode->i_mapping,
5276 inode->i_size >> PAGE_SHIFT);
5279 ret = __ext4_journalled_invalidatepage(page, offset,
5280 PAGE_SIZE - offset);
5286 read_lock(&journal->j_state_lock);
5287 if (journal->j_committing_transaction)
5288 commit_tid = journal->j_committing_transaction->t_tid;
5289 read_unlock(&journal->j_state_lock);
5291 jbd2_log_wait_commit(journal, commit_tid);
5298 * Called from notify_change.
5300 * We want to trap VFS attempts to truncate the file as soon as
5301 * possible. In particular, we want to make sure that when the VFS
5302 * shrinks i_size, we put the inode on the orphan list and modify
5303 * i_disksize immediately, so that during the subsequent flushing of
5304 * dirty pages and freeing of disk blocks, we can guarantee that any
5305 * commit will leave the blocks being flushed in an unused state on
5306 * disk. (On recovery, the inode will get truncated and the blocks will
5307 * be freed, so we have a strong guarantee that no future commit will
5308 * leave these blocks visible to the user.)
5310 * Another thing we have to assure is that if we are in ordered mode
5311 * and inode is still attached to the committing transaction, we must
5312 * we start writeout of all the dirty pages which are being truncated.
5313 * This way we are sure that all the data written in the previous
5314 * transaction are already on disk (truncate waits for pages under
5317 * Called with inode->i_mutex down.
5319 int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5322 struct inode *inode = d_inode(dentry);
5325 const unsigned int ia_valid = attr->ia_valid;
5327 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5330 if (unlikely(IS_IMMUTABLE(inode)))
5333 if (unlikely(IS_APPEND(inode) &&
5334 (ia_valid & (ATTR_MODE | ATTR_UID |
5335 ATTR_GID | ATTR_TIMES_SET))))
5338 error = setattr_prepare(mnt_userns, dentry, attr);
5342 error = fscrypt_prepare_setattr(dentry, attr);
5346 error = fsverity_prepare_setattr(dentry, attr);
5350 if (is_quota_modification(inode, attr)) {
5351 error = dquot_initialize(inode);
5355 ext4_fc_start_update(inode);
5356 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5357 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5360 /* (user+group)*(old+new) structure, inode write (sb,
5361 * inode block, ? - but truncate inode update has it) */
5362 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5363 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5364 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5365 if (IS_ERR(handle)) {
5366 error = PTR_ERR(handle);
5370 /* dquot_transfer() calls back ext4_get_inode_usage() which
5371 * counts xattr inode references.
5373 down_read(&EXT4_I(inode)->xattr_sem);
5374 error = dquot_transfer(inode, attr);
5375 up_read(&EXT4_I(inode)->xattr_sem);
5378 ext4_journal_stop(handle);
5379 ext4_fc_stop_update(inode);
5382 /* Update corresponding info in inode so that everything is in
5383 * one transaction */
5384 if (attr->ia_valid & ATTR_UID)
5385 inode->i_uid = attr->ia_uid;
5386 if (attr->ia_valid & ATTR_GID)
5387 inode->i_gid = attr->ia_gid;
5388 error = ext4_mark_inode_dirty(handle, inode);
5389 ext4_journal_stop(handle);
5390 if (unlikely(error))
5394 if (attr->ia_valid & ATTR_SIZE) {
5396 loff_t oldsize = inode->i_size;
5397 int shrink = (attr->ia_size < inode->i_size);
5399 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5400 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5402 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5403 ext4_fc_stop_update(inode);
5407 if (!S_ISREG(inode->i_mode)) {
5408 ext4_fc_stop_update(inode);
5412 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5413 inode_inc_iversion(inode);
5416 if (ext4_should_order_data(inode)) {
5417 error = ext4_begin_ordered_truncate(inode,
5423 * Blocks are going to be removed from the inode. Wait
5424 * for dio in flight.
5426 inode_dio_wait(inode);
5429 down_write(&EXT4_I(inode)->i_mmap_sem);
5431 rc = ext4_break_layouts(inode);
5433 up_write(&EXT4_I(inode)->i_mmap_sem);
5437 if (attr->ia_size != inode->i_size) {
5438 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5439 if (IS_ERR(handle)) {
5440 error = PTR_ERR(handle);
5443 if (ext4_handle_valid(handle) && shrink) {
5444 error = ext4_orphan_add(handle, inode);
5448 * Update c/mtime on truncate up, ext4_truncate() will
5449 * update c/mtime in shrink case below
5452 inode->i_mtime = current_time(inode);
5453 inode->i_ctime = inode->i_mtime;
5457 ext4_fc_track_range(handle, inode,
5458 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5459 inode->i_sb->s_blocksize_bits,
5460 (oldsize > 0 ? oldsize - 1 : 0) >>
5461 inode->i_sb->s_blocksize_bits);
5463 ext4_fc_track_range(
5465 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5466 inode->i_sb->s_blocksize_bits,
5467 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5468 inode->i_sb->s_blocksize_bits);
5470 down_write(&EXT4_I(inode)->i_data_sem);
5471 EXT4_I(inode)->i_disksize = attr->ia_size;
5472 rc = ext4_mark_inode_dirty(handle, inode);
5476 * We have to update i_size under i_data_sem together
5477 * with i_disksize to avoid races with writeback code
5478 * running ext4_wb_update_i_disksize().
5481 i_size_write(inode, attr->ia_size);
5482 up_write(&EXT4_I(inode)->i_data_sem);
5483 ext4_journal_stop(handle);
5487 pagecache_isize_extended(inode, oldsize,
5489 } else if (ext4_should_journal_data(inode)) {
5490 ext4_wait_for_tail_page_commit(inode);
5495 * Truncate pagecache after we've waited for commit
5496 * in data=journal mode to make pages freeable.
5498 truncate_pagecache(inode, inode->i_size);
5500 * Call ext4_truncate() even if i_size didn't change to
5501 * truncate possible preallocated blocks.
5503 if (attr->ia_size <= oldsize) {
5504 rc = ext4_truncate(inode);
5509 up_write(&EXT4_I(inode)->i_mmap_sem);
5513 setattr_copy(mnt_userns, inode, attr);
5514 mark_inode_dirty(inode);
5518 * If the call to ext4_truncate failed to get a transaction handle at
5519 * all, we need to clean up the in-core orphan list manually.
5521 if (orphan && inode->i_nlink)
5522 ext4_orphan_del(NULL, inode);
5524 if (!error && (ia_valid & ATTR_MODE))
5525 rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5529 ext4_std_error(inode->i_sb, error);
5532 ext4_fc_stop_update(inode);
5536 int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5537 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5539 struct inode *inode = d_inode(path->dentry);
5540 struct ext4_inode *raw_inode;
5541 struct ext4_inode_info *ei = EXT4_I(inode);
5544 if ((request_mask & STATX_BTIME) &&
5545 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5546 stat->result_mask |= STATX_BTIME;
5547 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5548 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5551 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5552 if (flags & EXT4_APPEND_FL)
5553 stat->attributes |= STATX_ATTR_APPEND;
5554 if (flags & EXT4_COMPR_FL)
5555 stat->attributes |= STATX_ATTR_COMPRESSED;
5556 if (flags & EXT4_ENCRYPT_FL)
5557 stat->attributes |= STATX_ATTR_ENCRYPTED;
5558 if (flags & EXT4_IMMUTABLE_FL)
5559 stat->attributes |= STATX_ATTR_IMMUTABLE;
5560 if (flags & EXT4_NODUMP_FL)
5561 stat->attributes |= STATX_ATTR_NODUMP;
5562 if (flags & EXT4_VERITY_FL)
5563 stat->attributes |= STATX_ATTR_VERITY;
5565 stat->attributes_mask |= (STATX_ATTR_APPEND |
5566 STATX_ATTR_COMPRESSED |
5567 STATX_ATTR_ENCRYPTED |
5568 STATX_ATTR_IMMUTABLE |
5572 generic_fillattr(mnt_userns, inode, stat);
5576 int ext4_file_getattr(struct user_namespace *mnt_userns,
5577 const struct path *path, struct kstat *stat,
5578 u32 request_mask, unsigned int query_flags)
5580 struct inode *inode = d_inode(path->dentry);
5581 u64 delalloc_blocks;
5583 ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5586 * If there is inline data in the inode, the inode will normally not
5587 * have data blocks allocated (it may have an external xattr block).
5588 * Report at least one sector for such files, so tools like tar, rsync,
5589 * others don't incorrectly think the file is completely sparse.
5591 if (unlikely(ext4_has_inline_data(inode)))
5592 stat->blocks += (stat->size + 511) >> 9;
5595 * We can't update i_blocks if the block allocation is delayed
5596 * otherwise in the case of system crash before the real block
5597 * allocation is done, we will have i_blocks inconsistent with
5598 * on-disk file blocks.
5599 * We always keep i_blocks updated together with real
5600 * allocation. But to not confuse with user, stat
5601 * will return the blocks that include the delayed allocation
5602 * blocks for this file.
5604 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5605 EXT4_I(inode)->i_reserved_data_blocks);
5606 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5610 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5613 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5614 return ext4_ind_trans_blocks(inode, lblocks);
5615 return ext4_ext_index_trans_blocks(inode, pextents);
5619 * Account for index blocks, block groups bitmaps and block group
5620 * descriptor blocks if modify datablocks and index blocks
5621 * worse case, the indexs blocks spread over different block groups
5623 * If datablocks are discontiguous, they are possible to spread over
5624 * different block groups too. If they are contiguous, with flexbg,
5625 * they could still across block group boundary.
5627 * Also account for superblock, inode, quota and xattr blocks
5629 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5632 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5638 * How many index blocks need to touch to map @lblocks logical blocks
5639 * to @pextents physical extents?
5641 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5646 * Now let's see how many group bitmaps and group descriptors need
5649 groups = idxblocks + pextents;
5651 if (groups > ngroups)
5653 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5654 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5656 /* bitmaps and block group descriptor blocks */
5657 ret += groups + gdpblocks;
5659 /* Blocks for super block, inode, quota and xattr blocks */
5660 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5666 * Calculate the total number of credits to reserve to fit
5667 * the modification of a single pages into a single transaction,
5668 * which may include multiple chunks of block allocations.
5670 * This could be called via ext4_write_begin()
5672 * We need to consider the worse case, when
5673 * one new block per extent.
5675 int ext4_writepage_trans_blocks(struct inode *inode)
5677 int bpp = ext4_journal_blocks_per_page(inode);
5680 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5682 /* Account for data blocks for journalled mode */
5683 if (ext4_should_journal_data(inode))
5689 * Calculate the journal credits for a chunk of data modification.
5691 * This is called from DIO, fallocate or whoever calling
5692 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5694 * journal buffers for data blocks are not included here, as DIO
5695 * and fallocate do no need to journal data buffers.
5697 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5699 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5703 * The caller must have previously called ext4_reserve_inode_write().
5704 * Give this, we know that the caller already has write access to iloc->bh.
5706 int ext4_mark_iloc_dirty(handle_t *handle,
5707 struct inode *inode, struct ext4_iloc *iloc)
5711 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5715 ext4_fc_track_inode(handle, inode);
5717 if (IS_I_VERSION(inode))
5718 inode_inc_iversion(inode);
5720 /* the do_update_inode consumes one bh->b_count */
5723 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5724 err = ext4_do_update_inode(handle, inode, iloc);
5730 * On success, We end up with an outstanding reference count against
5731 * iloc->bh. This _must_ be cleaned up later.
5735 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5736 struct ext4_iloc *iloc)
5740 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5743 err = ext4_get_inode_loc(inode, iloc);
5745 BUFFER_TRACE(iloc->bh, "get_write_access");
5746 err = ext4_journal_get_write_access(handle, iloc->bh);
5752 ext4_std_error(inode->i_sb, err);
5756 static int __ext4_expand_extra_isize(struct inode *inode,
5757 unsigned int new_extra_isize,
5758 struct ext4_iloc *iloc,
5759 handle_t *handle, int *no_expand)
5761 struct ext4_inode *raw_inode;
5762 struct ext4_xattr_ibody_header *header;
5763 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5764 struct ext4_inode_info *ei = EXT4_I(inode);
5767 /* this was checked at iget time, but double check for good measure */
5768 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5769 (ei->i_extra_isize & 3)) {
5770 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5772 EXT4_INODE_SIZE(inode->i_sb));
5773 return -EFSCORRUPTED;
5775 if ((new_extra_isize < ei->i_extra_isize) ||
5776 (new_extra_isize < 4) ||
5777 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5778 return -EINVAL; /* Should never happen */
5780 raw_inode = ext4_raw_inode(iloc);
5782 header = IHDR(inode, raw_inode);
5784 /* No extended attributes present */
5785 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5786 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5787 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5788 EXT4_I(inode)->i_extra_isize, 0,
5789 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5790 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5794 /* try to expand with EAs present */
5795 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5799 * Inode size expansion failed; don't try again
5808 * Expand an inode by new_extra_isize bytes.
5809 * Returns 0 on success or negative error number on failure.
5811 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5812 unsigned int new_extra_isize,
5813 struct ext4_iloc iloc,
5819 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5823 * In nojournal mode, we can immediately attempt to expand
5824 * the inode. When journaled, we first need to obtain extra
5825 * buffer credits since we may write into the EA block
5826 * with this same handle. If journal_extend fails, then it will
5827 * only result in a minor loss of functionality for that inode.
5828 * If this is felt to be critical, then e2fsck should be run to
5829 * force a large enough s_min_extra_isize.
5831 if (ext4_journal_extend(handle,
5832 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5835 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5838 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5839 handle, &no_expand);
5840 ext4_write_unlock_xattr(inode, &no_expand);
5845 int ext4_expand_extra_isize(struct inode *inode,
5846 unsigned int new_extra_isize,
5847 struct ext4_iloc *iloc)
5853 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5858 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5859 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5860 if (IS_ERR(handle)) {
5861 error = PTR_ERR(handle);
5866 ext4_write_lock_xattr(inode, &no_expand);
5868 BUFFER_TRACE(iloc->bh, "get_write_access");
5869 error = ext4_journal_get_write_access(handle, iloc->bh);
5875 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5876 handle, &no_expand);
5878 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5883 ext4_write_unlock_xattr(inode, &no_expand);
5884 ext4_journal_stop(handle);
5889 * What we do here is to mark the in-core inode as clean with respect to inode
5890 * dirtiness (it may still be data-dirty).
5891 * This means that the in-core inode may be reaped by prune_icache
5892 * without having to perform any I/O. This is a very good thing,
5893 * because *any* task may call prune_icache - even ones which
5894 * have a transaction open against a different journal.
5896 * Is this cheating? Not really. Sure, we haven't written the
5897 * inode out, but prune_icache isn't a user-visible syncing function.
5898 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5899 * we start and wait on commits.
5901 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5902 const char *func, unsigned int line)
5904 struct ext4_iloc iloc;
5905 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5909 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5910 err = ext4_reserve_inode_write(handle, inode, &iloc);
5914 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5915 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5918 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5921 ext4_error_inode_err(inode, func, line, 0, err,
5922 "mark_inode_dirty error");
5927 * ext4_dirty_inode() is called from __mark_inode_dirty()
5929 * We're really interested in the case where a file is being extended.
5930 * i_size has been changed by generic_commit_write() and we thus need
5931 * to include the updated inode in the current transaction.
5933 * Also, dquot_alloc_block() will always dirty the inode when blocks
5934 * are allocated to the file.
5936 * If the inode is marked synchronous, we don't honour that here - doing
5937 * so would cause a commit on atime updates, which we don't bother doing.
5938 * We handle synchronous inodes at the highest possible level.
5940 void ext4_dirty_inode(struct inode *inode, int flags)
5944 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5947 ext4_mark_inode_dirty(handle, inode);
5948 ext4_journal_stop(handle);
5951 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5956 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5959 * We have to be very careful here: changing a data block's
5960 * journaling status dynamically is dangerous. If we write a
5961 * data block to the journal, change the status and then delete
5962 * that block, we risk forgetting to revoke the old log record
5963 * from the journal and so a subsequent replay can corrupt data.
5964 * So, first we make sure that the journal is empty and that
5965 * nobody is changing anything.
5968 journal = EXT4_JOURNAL(inode);
5971 if (is_journal_aborted(journal))
5974 /* Wait for all existing dio workers */
5975 inode_dio_wait(inode);
5978 * Before flushing the journal and switching inode's aops, we have
5979 * to flush all dirty data the inode has. There can be outstanding
5980 * delayed allocations, there can be unwritten extents created by
5981 * fallocate or buffered writes in dioread_nolock mode covered by
5982 * dirty data which can be converted only after flushing the dirty
5983 * data (and journalled aops don't know how to handle these cases).
5986 down_write(&EXT4_I(inode)->i_mmap_sem);
5987 err = filemap_write_and_wait(inode->i_mapping);
5989 up_write(&EXT4_I(inode)->i_mmap_sem);
5994 percpu_down_write(&sbi->s_writepages_rwsem);
5995 jbd2_journal_lock_updates(journal);
5998 * OK, there are no updates running now, and all cached data is
5999 * synced to disk. We are now in a completely consistent state
6000 * which doesn't have anything in the journal, and we know that
6001 * no filesystem updates are running, so it is safe to modify
6002 * the inode's in-core data-journaling state flag now.
6006 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6008 err = jbd2_journal_flush(journal);
6010 jbd2_journal_unlock_updates(journal);
6011 percpu_up_write(&sbi->s_writepages_rwsem);
6014 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6016 ext4_set_aops(inode);
6018 jbd2_journal_unlock_updates(journal);
6019 percpu_up_write(&sbi->s_writepages_rwsem);
6022 up_write(&EXT4_I(inode)->i_mmap_sem);
6024 /* Finally we can mark the inode as dirty. */
6026 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6028 return PTR_ERR(handle);
6030 ext4_fc_mark_ineligible(inode->i_sb,
6031 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE);
6032 err = ext4_mark_inode_dirty(handle, inode);
6033 ext4_handle_sync(handle);
6034 ext4_journal_stop(handle);
6035 ext4_std_error(inode->i_sb, err);
6040 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6042 return !buffer_mapped(bh);
6045 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6047 struct vm_area_struct *vma = vmf->vma;
6048 struct page *page = vmf->page;
6053 struct file *file = vma->vm_file;
6054 struct inode *inode = file_inode(file);
6055 struct address_space *mapping = inode->i_mapping;
6057 get_block_t *get_block;
6060 if (unlikely(IS_IMMUTABLE(inode)))
6061 return VM_FAULT_SIGBUS;
6063 sb_start_pagefault(inode->i_sb);
6064 file_update_time(vma->vm_file);
6066 down_read(&EXT4_I(inode)->i_mmap_sem);
6068 err = ext4_convert_inline_data(inode);
6073 * On data journalling we skip straight to the transaction handle:
6074 * there's no delalloc; page truncated will be checked later; the
6075 * early return w/ all buffers mapped (calculates size/len) can't
6076 * be used; and there's no dioread_nolock, so only ext4_get_block.
6078 if (ext4_should_journal_data(inode))
6081 /* Delalloc case is easy... */
6082 if (test_opt(inode->i_sb, DELALLOC) &&
6083 !ext4_nonda_switch(inode->i_sb)) {
6085 err = block_page_mkwrite(vma, vmf,
6086 ext4_da_get_block_prep);
6087 } while (err == -ENOSPC &&
6088 ext4_should_retry_alloc(inode->i_sb, &retries));
6093 size = i_size_read(inode);
6094 /* Page got truncated from under us? */
6095 if (page->mapping != mapping || page_offset(page) > size) {
6097 ret = VM_FAULT_NOPAGE;
6101 if (page->index == size >> PAGE_SHIFT)
6102 len = size & ~PAGE_MASK;
6106 * Return if we have all the buffers mapped. This avoids the need to do
6107 * journal_start/journal_stop which can block and take a long time
6109 * This cannot be done for data journalling, as we have to add the
6110 * inode to the transaction's list to writeprotect pages on commit.
6112 if (page_has_buffers(page)) {
6113 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6115 ext4_bh_unmapped)) {
6116 /* Wait so that we don't change page under IO */
6117 wait_for_stable_page(page);
6118 ret = VM_FAULT_LOCKED;
6123 /* OK, we need to fill the hole... */
6124 if (ext4_should_dioread_nolock(inode))
6125 get_block = ext4_get_block_unwritten;
6127 get_block = ext4_get_block;
6129 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6130 ext4_writepage_trans_blocks(inode));
6131 if (IS_ERR(handle)) {
6132 ret = VM_FAULT_SIGBUS;
6136 * Data journalling can't use block_page_mkwrite() because it
6137 * will set_buffer_dirty() before do_journal_get_write_access()
6138 * thus might hit warning messages for dirty metadata buffers.
6140 if (!ext4_should_journal_data(inode)) {
6141 err = block_page_mkwrite(vma, vmf, get_block);
6144 size = i_size_read(inode);
6145 /* Page got truncated from under us? */
6146 if (page->mapping != mapping || page_offset(page) > size) {
6147 ret = VM_FAULT_NOPAGE;
6151 if (page->index == size >> PAGE_SHIFT)
6152 len = size & ~PAGE_MASK;
6156 err = __block_write_begin(page, 0, len, ext4_get_block);
6158 ret = VM_FAULT_SIGBUS;
6159 if (ext4_walk_page_buffers(handle, page_buffers(page),
6160 0, len, NULL, do_journal_get_write_access))
6162 if (ext4_walk_page_buffers(handle, page_buffers(page),
6163 0, len, NULL, write_end_fn))
6165 if (ext4_jbd2_inode_add_write(handle, inode,
6166 page_offset(page), len))
6168 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6173 ext4_journal_stop(handle);
6174 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6177 ret = block_page_mkwrite_return(err);
6179 up_read(&EXT4_I(inode)->i_mmap_sem);
6180 sb_end_pagefault(inode->i_sb);
6184 ext4_journal_stop(handle);
6188 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6190 struct inode *inode = file_inode(vmf->vma->vm_file);
6193 down_read(&EXT4_I(inode)->i_mmap_sem);
6194 ret = filemap_fault(vmf);
6195 up_read(&EXT4_I(inode)->i_mmap_sem);