1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 struct ext4_inode_info *ei)
54 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 int offset = offsetof(struct ext4_inode, i_checksum_lo);
58 unsigned int csum_size = sizeof(dummy_csum);
60 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
61 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
64 EXT4_GOOD_OLD_INODE_SIZE - offset);
66 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
67 offset = offsetof(struct ext4_inode, i_checksum_hi);
68 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
69 EXT4_GOOD_OLD_INODE_SIZE,
70 offset - EXT4_GOOD_OLD_INODE_SIZE);
71 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
72 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
76 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
77 EXT4_INODE_SIZE(inode->i_sb) - offset);
83 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
84 struct ext4_inode_info *ei)
86 __u32 provided, calculated;
88 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
89 cpu_to_le32(EXT4_OS_LINUX) ||
90 !ext4_has_metadata_csum(inode->i_sb))
93 provided = le16_to_cpu(raw->i_checksum_lo);
94 calculated = ext4_inode_csum(inode, raw, ei);
95 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
96 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
97 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 return provided == calculated;
104 void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
105 struct ext4_inode_info *ei)
109 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
110 cpu_to_le32(EXT4_OS_LINUX) ||
111 !ext4_has_metadata_csum(inode->i_sb))
114 csum = ext4_inode_csum(inode, raw, ei);
115 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
116 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
117 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
118 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode *inode,
124 trace_ext4_begin_ordered_truncate(inode, new_size);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode)->jinode)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
134 EXT4_I(inode)->jinode,
138 static void ext4_invalidatepage(struct page *page, unsigned int offset,
139 unsigned int length);
140 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
141 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
142 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
146 * Test whether an inode is a fast symlink.
147 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
149 int ext4_inode_is_fast_symlink(struct inode *inode)
151 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
152 int ea_blocks = EXT4_I(inode)->i_file_acl ?
153 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
155 if (ext4_has_inline_data(inode))
158 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
160 return S_ISLNK(inode->i_mode) && inode->i_size &&
161 (inode->i_size < EXT4_N_BLOCKS * 4);
165 * Called at the last iput() if i_nlink is zero.
167 void ext4_evict_inode(struct inode *inode)
172 * Credits for final inode cleanup and freeing:
173 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
174 * (xattr block freeing), bitmap, group descriptor (inode freeing)
176 int extra_credits = 6;
177 struct ext4_xattr_inode_array *ea_inode_array = NULL;
178 bool freeze_protected = false;
180 trace_ext4_evict_inode(inode);
182 if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
183 ext4_evict_ea_inode(inode);
184 if (inode->i_nlink) {
186 * When journalling data dirty buffers are tracked only in the
187 * journal. So although mm thinks everything is clean and
188 * ready for reaping the inode might still have some pages to
189 * write in the running transaction or waiting to be
190 * checkpointed. Thus calling jbd2_journal_invalidatepage()
191 * (via truncate_inode_pages()) to discard these buffers can
192 * cause data loss. Also even if we did not discard these
193 * buffers, we would have no way to find them after the inode
194 * is reaped and thus user could see stale data if he tries to
195 * read them before the transaction is checkpointed. So be
196 * careful and force everything to disk here... We use
197 * ei->i_datasync_tid to store the newest transaction
198 * containing inode's data.
200 * Note that directories do not have this problem because they
201 * don't use page cache.
203 if (inode->i_ino != EXT4_JOURNAL_INO &&
204 ext4_should_journal_data(inode) &&
205 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
206 inode->i_data.nrpages) {
207 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
208 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
210 jbd2_complete_transaction(journal, commit_tid);
211 filemap_write_and_wait(&inode->i_data);
213 truncate_inode_pages_final(&inode->i_data);
218 if (is_bad_inode(inode))
220 dquot_initialize(inode);
222 if (ext4_should_order_data(inode))
223 ext4_begin_ordered_truncate(inode, 0);
224 truncate_inode_pages_final(&inode->i_data);
227 * For inodes with journalled data, transaction commit could have
228 * dirtied the inode. And for inodes with dioread_nolock, unwritten
229 * extents converting worker could merge extents and also have dirtied
230 * the inode. Flush worker is ignoring it because of I_FREEING flag but
231 * we still need to remove the inode from the writeback lists.
233 if (!list_empty_careful(&inode->i_io_list))
234 inode_io_list_del(inode);
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it. When we are in a running transaction though,
239 * we are already protected against freezing and we cannot grab further
240 * protection due to lock ordering constraints.
242 if (!ext4_journal_current_handle()) {
243 sb_start_intwrite(inode->i_sb);
244 freeze_protected = true;
247 if (!IS_NOQUOTA(inode))
248 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
251 * Block bitmap, group descriptor, and inode are accounted in both
252 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
254 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
255 ext4_blocks_for_truncate(inode) + extra_credits - 3);
256 if (IS_ERR(handle)) {
257 ext4_std_error(inode->i_sb, PTR_ERR(handle));
259 * If we're going to skip the normal cleanup, we still need to
260 * make sure that the in-core orphan linked list is properly
263 ext4_orphan_del(NULL, inode);
264 if (freeze_protected)
265 sb_end_intwrite(inode->i_sb);
270 ext4_handle_sync(handle);
273 * Set inode->i_size to 0 before calling ext4_truncate(). We need
274 * special handling of symlinks here because i_size is used to
275 * determine whether ext4_inode_info->i_data contains symlink data or
276 * block mappings. Setting i_size to 0 will remove its fast symlink
277 * status. Erase i_data so that it becomes a valid empty block map.
279 if (ext4_inode_is_fast_symlink(inode))
280 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
282 err = ext4_mark_inode_dirty(handle, inode);
284 ext4_warning(inode->i_sb,
285 "couldn't mark inode dirty (err %d)", err);
288 if (inode->i_blocks) {
289 err = ext4_truncate(inode);
291 ext4_error_err(inode->i_sb, -err,
292 "couldn't truncate inode %lu (err %d)",
298 /* Remove xattr references. */
299 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
302 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
304 ext4_journal_stop(handle);
305 ext4_orphan_del(NULL, inode);
306 if (freeze_protected)
307 sb_end_intwrite(inode->i_sb);
308 ext4_xattr_inode_array_free(ea_inode_array);
313 * Kill off the orphan record which ext4_truncate created.
314 * AKPM: I think this can be inside the above `if'.
315 * Note that ext4_orphan_del() has to be able to cope with the
316 * deletion of a non-existent orphan - this is because we don't
317 * know if ext4_truncate() actually created an orphan record.
318 * (Well, we could do this if we need to, but heck - it works)
320 ext4_orphan_del(handle, inode);
321 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
324 * One subtle ordering requirement: if anything has gone wrong
325 * (transaction abort, IO errors, whatever), then we can still
326 * do these next steps (the fs will already have been marked as
327 * having errors), but we can't free the inode if the mark_dirty
330 if (ext4_mark_inode_dirty(handle, inode))
331 /* If that failed, just do the required in-core inode clear. */
332 ext4_clear_inode(inode);
334 ext4_free_inode(handle, inode);
335 ext4_journal_stop(handle);
336 if (freeze_protected)
337 sb_end_intwrite(inode->i_sb);
338 ext4_xattr_inode_array_free(ea_inode_array);
342 * Check out some where else accidentally dirty the evicting inode,
343 * which may probably cause inode use-after-free issues later.
345 WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
347 if (!list_empty(&EXT4_I(inode)->i_fc_list))
348 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM);
349 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
353 qsize_t *ext4_get_reserved_space(struct inode *inode)
355 return &EXT4_I(inode)->i_reserved_quota;
360 * Called with i_data_sem down, which is important since we can call
361 * ext4_discard_preallocations() from here.
363 void ext4_da_update_reserve_space(struct inode *inode,
364 int used, int quota_claim)
366 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
367 struct ext4_inode_info *ei = EXT4_I(inode);
369 spin_lock(&ei->i_block_reservation_lock);
370 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
371 if (unlikely(used > ei->i_reserved_data_blocks)) {
372 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
373 "with only %d reserved data blocks",
374 __func__, inode->i_ino, used,
375 ei->i_reserved_data_blocks);
377 used = ei->i_reserved_data_blocks;
380 /* Update per-inode reservations */
381 ei->i_reserved_data_blocks -= used;
382 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
384 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
386 /* Update quota subsystem for data blocks */
388 dquot_claim_block(inode, EXT4_C2B(sbi, used));
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
393 * not re-claim the quota for fallocated blocks.
395 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
403 if ((ei->i_reserved_data_blocks == 0) &&
404 !inode_is_open_for_write(inode))
405 ext4_discard_preallocations(inode, 0);
408 static int __check_block_validity(struct inode *inode, const char *func,
410 struct ext4_map_blocks *map)
412 if (ext4_has_feature_journal(inode->i_sb) &&
414 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
416 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
417 ext4_error_inode(inode, func, line, map->m_pblk,
418 "lblock %lu mapped to illegal pblock %llu "
419 "(length %d)", (unsigned long) map->m_lblk,
420 map->m_pblk, map->m_len);
421 return -EFSCORRUPTED;
426 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
431 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
432 return fscrypt_zeroout_range(inode, lblk, pblk, len);
434 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
441 #define check_block_validity(inode, map) \
442 __check_block_validity((inode), __func__, __LINE__, (map))
444 #ifdef ES_AGGRESSIVE_TEST
445 static void ext4_map_blocks_es_recheck(handle_t *handle,
447 struct ext4_map_blocks *es_map,
448 struct ext4_map_blocks *map,
455 * There is a race window that the result is not the same.
456 * e.g. xfstests #223 when dioread_nolock enables. The reason
457 * is that we lookup a block mapping in extent status tree with
458 * out taking i_data_sem. So at the time the unwritten extent
459 * could be converted.
461 down_read(&EXT4_I(inode)->i_data_sem);
462 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
463 retval = ext4_ext_map_blocks(handle, inode, map, 0);
465 retval = ext4_ind_map_blocks(handle, inode, map, 0);
467 up_read((&EXT4_I(inode)->i_data_sem));
470 * We don't check m_len because extent will be collpased in status
471 * tree. So the m_len might not equal.
473 if (es_map->m_lblk != map->m_lblk ||
474 es_map->m_flags != map->m_flags ||
475 es_map->m_pblk != map->m_pblk) {
476 printk("ES cache assertion failed for inode: %lu "
477 "es_cached ex [%d/%d/%llu/%x] != "
478 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
479 inode->i_ino, es_map->m_lblk, es_map->m_len,
480 es_map->m_pblk, es_map->m_flags, map->m_lblk,
481 map->m_len, map->m_pblk, map->m_flags,
485 #endif /* ES_AGGRESSIVE_TEST */
488 * The ext4_map_blocks() function tries to look up the requested blocks,
489 * and returns if the blocks are already mapped.
491 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
492 * and store the allocated blocks in the result buffer head and mark it
495 * If file type is extents based, it will call ext4_ext_map_blocks(),
496 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
499 * On success, it returns the number of blocks being mapped or allocated. if
500 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
501 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
503 * It returns 0 if plain look up failed (blocks have not been allocated), in
504 * that case, @map is returned as unmapped but we still do fill map->m_len to
505 * indicate the length of a hole starting at map->m_lblk.
507 * It returns the error in case of allocation failure.
509 int ext4_map_blocks(handle_t *handle, struct inode *inode,
510 struct ext4_map_blocks *map, int flags)
512 struct extent_status es;
515 #ifdef ES_AGGRESSIVE_TEST
516 struct ext4_map_blocks orig_map;
518 memcpy(&orig_map, map, sizeof(*map));
522 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
523 flags, map->m_len, (unsigned long) map->m_lblk);
526 * ext4_map_blocks returns an int, and m_len is an unsigned int
528 if (unlikely(map->m_len > INT_MAX))
529 map->m_len = INT_MAX;
531 /* We can handle the block number less than EXT_MAX_BLOCKS */
532 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
533 return -EFSCORRUPTED;
535 /* Lookup extent status tree firstly */
536 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
537 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
538 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
539 map->m_pblk = ext4_es_pblock(&es) +
540 map->m_lblk - es.es_lblk;
541 map->m_flags |= ext4_es_is_written(&es) ?
542 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
543 retval = es.es_len - (map->m_lblk - es.es_lblk);
544 if (retval > map->m_len)
547 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
549 retval = es.es_len - (map->m_lblk - es.es_lblk);
550 if (retval > map->m_len)
557 #ifdef ES_AGGRESSIVE_TEST
558 ext4_map_blocks_es_recheck(handle, inode, map,
565 * Try to see if we can get the block without requesting a new
568 down_read(&EXT4_I(inode)->i_data_sem);
569 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
570 retval = ext4_ext_map_blocks(handle, inode, map, 0);
572 retval = ext4_ind_map_blocks(handle, inode, map, 0);
577 if (unlikely(retval != map->m_len)) {
578 ext4_warning(inode->i_sb,
579 "ES len assertion failed for inode "
580 "%lu: retval %d != map->m_len %d",
581 inode->i_ino, retval, map->m_len);
585 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
586 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
587 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
588 !(status & EXTENT_STATUS_WRITTEN) &&
589 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
590 map->m_lblk + map->m_len - 1))
591 status |= EXTENT_STATUS_DELAYED;
592 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
593 map->m_pblk, status);
595 up_read((&EXT4_I(inode)->i_data_sem));
598 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
599 ret = check_block_validity(inode, map);
604 /* If it is only a block(s) look up */
605 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
609 * Returns if the blocks have already allocated
611 * Note that if blocks have been preallocated
612 * ext4_ext_get_block() returns the create = 0
613 * with buffer head unmapped.
615 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
617 * If we need to convert extent to unwritten
618 * we continue and do the actual work in
619 * ext4_ext_map_blocks()
621 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
625 * Here we clear m_flags because after allocating an new extent,
626 * it will be set again.
628 map->m_flags &= ~EXT4_MAP_FLAGS;
631 * New blocks allocate and/or writing to unwritten extent
632 * will possibly result in updating i_data, so we take
633 * the write lock of i_data_sem, and call get_block()
634 * with create == 1 flag.
636 down_write(&EXT4_I(inode)->i_data_sem);
639 * We need to check for EXT4 here because migrate
640 * could have changed the inode type in between
642 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
643 retval = ext4_ext_map_blocks(handle, inode, map, flags);
645 retval = ext4_ind_map_blocks(handle, inode, map, flags);
647 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
649 * We allocated new blocks which will result in
650 * i_data's format changing. Force the migrate
651 * to fail by clearing migrate flags
653 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
660 if (unlikely(retval != map->m_len)) {
661 ext4_warning(inode->i_sb,
662 "ES len assertion failed for inode "
663 "%lu: retval %d != map->m_len %d",
664 inode->i_ino, retval, map->m_len);
669 * We have to zeroout blocks before inserting them into extent
670 * status tree. Otherwise someone could look them up there and
671 * use them before they are really zeroed. We also have to
672 * unmap metadata before zeroing as otherwise writeback can
673 * overwrite zeros with stale data from block device.
675 if (flags & EXT4_GET_BLOCKS_ZERO &&
676 map->m_flags & EXT4_MAP_MAPPED &&
677 map->m_flags & EXT4_MAP_NEW) {
678 ret = ext4_issue_zeroout(inode, map->m_lblk,
679 map->m_pblk, map->m_len);
687 * If the extent has been zeroed out, we don't need to update
688 * extent status tree.
690 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
691 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
692 if (ext4_es_is_written(&es))
695 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
696 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
697 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
698 !(status & EXTENT_STATUS_WRITTEN) &&
699 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
700 map->m_lblk + map->m_len - 1))
701 status |= EXTENT_STATUS_DELAYED;
702 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
703 map->m_pblk, status);
707 up_write((&EXT4_I(inode)->i_data_sem));
708 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
709 ret = check_block_validity(inode, map);
714 * Inodes with freshly allocated blocks where contents will be
715 * visible after transaction commit must be on transaction's
718 if (map->m_flags & EXT4_MAP_NEW &&
719 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
720 !(flags & EXT4_GET_BLOCKS_ZERO) &&
721 !ext4_is_quota_file(inode) &&
722 ext4_should_order_data(inode)) {
724 (loff_t)map->m_lblk << inode->i_blkbits;
725 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
727 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
728 ret = ext4_jbd2_inode_add_wait(handle, inode,
731 ret = ext4_jbd2_inode_add_write(handle, inode,
737 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
738 map->m_flags & EXT4_MAP_MAPPED))
739 ext4_fc_track_range(handle, inode, map->m_lblk,
740 map->m_lblk + map->m_len - 1);
742 ext_debug(inode, "failed with err %d\n", retval);
747 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
748 * we have to be careful as someone else may be manipulating b_state as well.
750 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
752 unsigned long old_state;
753 unsigned long new_state;
755 flags &= EXT4_MAP_FLAGS;
757 /* Dummy buffer_head? Set non-atomically. */
759 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
763 * Someone else may be modifying b_state. Be careful! This is ugly but
764 * once we get rid of using bh as a container for mapping information
765 * to pass to / from get_block functions, this can go away.
768 old_state = READ_ONCE(bh->b_state);
769 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
771 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
774 static int _ext4_get_block(struct inode *inode, sector_t iblock,
775 struct buffer_head *bh, int flags)
777 struct ext4_map_blocks map;
780 if (ext4_has_inline_data(inode))
784 map.m_len = bh->b_size >> inode->i_blkbits;
786 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
789 map_bh(bh, inode->i_sb, map.m_pblk);
790 ext4_update_bh_state(bh, map.m_flags);
791 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
793 } else if (ret == 0) {
794 /* hole case, need to fill in bh->b_size */
795 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
800 int ext4_get_block(struct inode *inode, sector_t iblock,
801 struct buffer_head *bh, int create)
803 return _ext4_get_block(inode, iblock, bh,
804 create ? EXT4_GET_BLOCKS_CREATE : 0);
808 * Get block function used when preparing for buffered write if we require
809 * creating an unwritten extent if blocks haven't been allocated. The extent
810 * will be converted to written after the IO is complete.
812 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
813 struct buffer_head *bh_result, int create)
815 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
816 inode->i_ino, create);
817 return _ext4_get_block(inode, iblock, bh_result,
818 EXT4_GET_BLOCKS_IO_CREATE_EXT);
821 /* Maximum number of blocks we map for direct IO at once. */
822 #define DIO_MAX_BLOCKS 4096
825 * `handle' can be NULL if create is zero
827 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
828 ext4_lblk_t block, int map_flags)
830 struct ext4_map_blocks map;
831 struct buffer_head *bh;
832 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
835 J_ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
836 || handle != NULL || create == 0);
840 err = ext4_map_blocks(handle, inode, &map, map_flags);
843 return create ? ERR_PTR(-ENOSPC) : NULL;
847 bh = sb_getblk(inode->i_sb, map.m_pblk);
849 return ERR_PTR(-ENOMEM);
850 if (map.m_flags & EXT4_MAP_NEW) {
851 J_ASSERT(create != 0);
852 J_ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
853 || (handle != NULL));
856 * Now that we do not always journal data, we should
857 * keep in mind whether this should always journal the
858 * new buffer as metadata. For now, regular file
859 * writes use ext4_get_block instead, so it's not a
863 BUFFER_TRACE(bh, "call get_create_access");
864 err = ext4_journal_get_create_access(handle, bh);
869 if (!buffer_uptodate(bh)) {
870 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
871 set_buffer_uptodate(bh);
874 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
875 err = ext4_handle_dirty_metadata(handle, inode, bh);
879 BUFFER_TRACE(bh, "not a new buffer");
886 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
887 ext4_lblk_t block, int map_flags)
889 struct buffer_head *bh;
892 bh = ext4_getblk(handle, inode, block, map_flags);
895 if (!bh || ext4_buffer_uptodate(bh))
898 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
906 /* Read a contiguous batch of blocks. */
907 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
908 bool wait, struct buffer_head **bhs)
912 for (i = 0; i < bh_count; i++) {
913 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
914 if (IS_ERR(bhs[i])) {
915 err = PTR_ERR(bhs[i]);
921 for (i = 0; i < bh_count; i++)
922 /* Note that NULL bhs[i] is valid because of holes. */
923 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
924 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
929 for (i = 0; i < bh_count; i++)
931 wait_on_buffer(bhs[i]);
933 for (i = 0; i < bh_count; i++) {
934 if (bhs[i] && !buffer_uptodate(bhs[i])) {
942 for (i = 0; i < bh_count; i++) {
949 int ext4_walk_page_buffers(handle_t *handle,
950 struct buffer_head *head,
954 int (*fn)(handle_t *handle,
955 struct buffer_head *bh))
957 struct buffer_head *bh;
958 unsigned block_start, block_end;
959 unsigned blocksize = head->b_size;
961 struct buffer_head *next;
963 for (bh = head, block_start = 0;
964 ret == 0 && (bh != head || !block_start);
965 block_start = block_end, bh = next) {
966 next = bh->b_this_page;
967 block_end = block_start + blocksize;
968 if (block_end <= from || block_start >= to) {
969 if (partial && !buffer_uptodate(bh))
973 err = (*fn)(handle, bh);
981 * To preserve ordering, it is essential that the hole instantiation and
982 * the data write be encapsulated in a single transaction. We cannot
983 * close off a transaction and start a new one between the ext4_get_block()
984 * and the commit_write(). So doing the jbd2_journal_start at the start of
985 * prepare_write() is the right place.
987 * Also, this function can nest inside ext4_writepage(). In that case, we
988 * *know* that ext4_writepage() has generated enough buffer credits to do the
989 * whole page. So we won't block on the journal in that case, which is good,
990 * because the caller may be PF_MEMALLOC.
992 * By accident, ext4 can be reentered when a transaction is open via
993 * quota file writes. If we were to commit the transaction while thus
994 * reentered, there can be a deadlock - we would be holding a quota
995 * lock, and the commit would never complete if another thread had a
996 * transaction open and was blocking on the quota lock - a ranking
999 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1000 * will _not_ run commit under these circumstances because handle->h_ref
1001 * is elevated. We'll still have enough credits for the tiny quotafile
1004 int do_journal_get_write_access(handle_t *handle,
1005 struct buffer_head *bh)
1007 int dirty = buffer_dirty(bh);
1010 if (!buffer_mapped(bh) || buffer_freed(bh))
1013 * __block_write_begin() could have dirtied some buffers. Clean
1014 * the dirty bit as jbd2_journal_get_write_access() could complain
1015 * otherwise about fs integrity issues. Setting of the dirty bit
1016 * by __block_write_begin() isn't a real problem here as we clear
1017 * the bit before releasing a page lock and thus writeback cannot
1018 * ever write the buffer.
1021 clear_buffer_dirty(bh);
1022 BUFFER_TRACE(bh, "get write access");
1023 ret = ext4_journal_get_write_access(handle, bh);
1025 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1029 #ifdef CONFIG_FS_ENCRYPTION
1030 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1031 get_block_t *get_block)
1033 unsigned from = pos & (PAGE_SIZE - 1);
1034 unsigned to = from + len;
1035 struct inode *inode = page->mapping->host;
1036 unsigned block_start, block_end;
1039 unsigned blocksize = inode->i_sb->s_blocksize;
1041 struct buffer_head *bh, *head, *wait[2];
1045 BUG_ON(!PageLocked(page));
1046 BUG_ON(from > PAGE_SIZE);
1047 BUG_ON(to > PAGE_SIZE);
1050 if (!page_has_buffers(page))
1051 create_empty_buffers(page, blocksize, 0);
1052 head = page_buffers(page);
1053 bbits = ilog2(blocksize);
1054 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1056 for (bh = head, block_start = 0; bh != head || !block_start;
1057 block++, block_start = block_end, bh = bh->b_this_page) {
1058 block_end = block_start + blocksize;
1059 if (block_end <= from || block_start >= to) {
1060 if (PageUptodate(page)) {
1061 if (!buffer_uptodate(bh))
1062 set_buffer_uptodate(bh);
1067 clear_buffer_new(bh);
1068 if (!buffer_mapped(bh)) {
1069 WARN_ON(bh->b_size != blocksize);
1070 err = get_block(inode, block, bh, 1);
1073 if (buffer_new(bh)) {
1074 if (PageUptodate(page)) {
1075 clear_buffer_new(bh);
1076 set_buffer_uptodate(bh);
1077 mark_buffer_dirty(bh);
1080 if (block_end > to || block_start < from)
1081 zero_user_segments(page, to, block_end,
1086 if (PageUptodate(page)) {
1087 if (!buffer_uptodate(bh))
1088 set_buffer_uptodate(bh);
1091 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1092 !buffer_unwritten(bh) &&
1093 (block_start < from || block_end > to)) {
1094 ext4_read_bh_lock(bh, 0, false);
1095 wait[nr_wait++] = bh;
1099 * If we issued read requests, let them complete.
1101 for (i = 0; i < nr_wait; i++) {
1102 wait_on_buffer(wait[i]);
1103 if (!buffer_uptodate(wait[i]))
1106 if (unlikely(err)) {
1107 page_zero_new_buffers(page, from, to);
1108 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1109 for (i = 0; i < nr_wait; i++) {
1112 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1113 bh_offset(wait[i]));
1115 clear_buffer_uptodate(wait[i]);
1125 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1126 loff_t pos, unsigned len, unsigned flags,
1127 struct page **pagep, void **fsdata)
1129 struct inode *inode = mapping->host;
1130 int ret, needed_blocks;
1137 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1140 trace_ext4_write_begin(inode, pos, len, flags);
1142 * Reserve one block more for addition to orphan list in case
1143 * we allocate blocks but write fails for some reason
1145 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1146 index = pos >> PAGE_SHIFT;
1147 from = pos & (PAGE_SIZE - 1);
1150 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1151 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1160 * grab_cache_page_write_begin() can take a long time if the
1161 * system is thrashing due to memory pressure, or if the page
1162 * is being written back. So grab it first before we start
1163 * the transaction handle. This also allows us to allocate
1164 * the page (if needed) without using GFP_NOFS.
1167 page = grab_cache_page_write_begin(mapping, index, flags);
1171 * The same as page allocation, we prealloc buffer heads before
1172 * starting the handle.
1174 if (!page_has_buffers(page))
1175 create_empty_buffers(page, inode->i_sb->s_blocksize, 0);
1180 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1181 if (IS_ERR(handle)) {
1183 return PTR_ERR(handle);
1187 if (page->mapping != mapping) {
1188 /* The page got truncated from under us */
1191 ext4_journal_stop(handle);
1194 /* In case writeback began while the page was unlocked */
1195 wait_for_stable_page(page);
1197 #ifdef CONFIG_FS_ENCRYPTION
1198 if (ext4_should_dioread_nolock(inode))
1199 ret = ext4_block_write_begin(page, pos, len,
1200 ext4_get_block_unwritten);
1202 ret = ext4_block_write_begin(page, pos, len,
1205 if (ext4_should_dioread_nolock(inode))
1206 ret = __block_write_begin(page, pos, len,
1207 ext4_get_block_unwritten);
1209 ret = __block_write_begin(page, pos, len, ext4_get_block);
1211 if (!ret && ext4_should_journal_data(inode)) {
1212 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1214 do_journal_get_write_access);
1218 bool extended = (pos + len > inode->i_size) &&
1219 !ext4_verity_in_progress(inode);
1223 * __block_write_begin may have instantiated a few blocks
1224 * outside i_size. Trim these off again. Don't need
1225 * i_size_read because we hold i_mutex.
1227 * Add inode to orphan list in case we crash before
1230 if (extended && ext4_can_truncate(inode))
1231 ext4_orphan_add(handle, inode);
1233 ext4_journal_stop(handle);
1235 ext4_truncate_failed_write(inode);
1237 * If truncate failed early the inode might
1238 * still be on the orphan list; we need to
1239 * make sure the inode is removed from the
1240 * orphan list in that case.
1243 ext4_orphan_del(NULL, inode);
1246 if (ret == -ENOSPC &&
1247 ext4_should_retry_alloc(inode->i_sb, &retries))
1256 /* For write_end() in data=journal mode */
1257 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1260 if (!buffer_mapped(bh) || buffer_freed(bh))
1262 set_buffer_uptodate(bh);
1263 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1264 clear_buffer_meta(bh);
1265 clear_buffer_prio(bh);
1270 * We need to pick up the new inode size which generic_commit_write gave us
1271 * `file' can be NULL - eg, when called from page_symlink().
1273 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1274 * buffers are managed internally.
1276 static int ext4_write_end(struct file *file,
1277 struct address_space *mapping,
1278 loff_t pos, unsigned len, unsigned copied,
1279 struct page *page, void *fsdata)
1281 handle_t *handle = ext4_journal_current_handle();
1282 struct inode *inode = mapping->host;
1283 loff_t old_size = inode->i_size;
1285 int i_size_changed = 0;
1286 int inline_data = ext4_has_inline_data(inode);
1287 bool verity = ext4_verity_in_progress(inode);
1289 trace_ext4_write_end(inode, pos, len, copied);
1291 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1292 ret = ext4_write_inline_data_end(inode, pos, len,
1302 copied = block_write_end(file, mapping, pos,
1303 len, copied, page, fsdata);
1305 * it's important to update i_size while still holding page lock:
1306 * page writeout could otherwise come in and zero beyond i_size.
1308 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1309 * blocks are being written past EOF, so skip the i_size update.
1312 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1316 if (old_size < pos && !verity)
1317 pagecache_isize_extended(inode, old_size, pos);
1319 * Don't mark the inode dirty under page lock. First, it unnecessarily
1320 * makes the holding time of page lock longer. Second, it forces lock
1321 * ordering of page lock and transaction start for journaling
1324 if (i_size_changed || inline_data)
1325 ret = ext4_mark_inode_dirty(handle, inode);
1328 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1329 /* if we have allocated more blocks and copied
1330 * less. We will have blocks allocated outside
1331 * inode->i_size. So truncate them
1333 ext4_orphan_add(handle, inode);
1335 ret2 = ext4_journal_stop(handle);
1339 if (pos + len > inode->i_size && !verity) {
1340 ext4_truncate_failed_write(inode);
1342 * If truncate failed early the inode might still be
1343 * on the orphan list; we need to make sure the inode
1344 * is removed from the orphan list in that case.
1347 ext4_orphan_del(NULL, inode);
1350 return ret ? ret : copied;
1354 * This is a private version of page_zero_new_buffers() which doesn't
1355 * set the buffer to be dirty, since in data=journalled mode we need
1356 * to call ext4_handle_dirty_metadata() instead.
1358 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1360 unsigned from, unsigned to)
1362 unsigned int block_start = 0, block_end;
1363 struct buffer_head *head, *bh;
1365 bh = head = page_buffers(page);
1367 block_end = block_start + bh->b_size;
1368 if (buffer_new(bh)) {
1369 if (block_end > from && block_start < to) {
1370 if (!PageUptodate(page)) {
1371 unsigned start, size;
1373 start = max(from, block_start);
1374 size = min(to, block_end) - start;
1376 zero_user(page, start, size);
1377 write_end_fn(handle, bh);
1379 clear_buffer_new(bh);
1382 block_start = block_end;
1383 bh = bh->b_this_page;
1384 } while (bh != head);
1387 static int ext4_journalled_write_end(struct file *file,
1388 struct address_space *mapping,
1389 loff_t pos, unsigned len, unsigned copied,
1390 struct page *page, void *fsdata)
1392 handle_t *handle = ext4_journal_current_handle();
1393 struct inode *inode = mapping->host;
1394 loff_t old_size = inode->i_size;
1398 int size_changed = 0;
1399 int inline_data = ext4_has_inline_data(inode);
1400 bool verity = ext4_verity_in_progress(inode);
1402 trace_ext4_journalled_write_end(inode, pos, len, copied);
1403 from = pos & (PAGE_SIZE - 1);
1406 BUG_ON(!ext4_handle_valid(handle));
1409 ret = ext4_write_inline_data_end(inode, pos, len,
1418 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1420 ext4_journalled_zero_new_buffers(handle, page, from, to);
1422 if (unlikely(copied < len))
1423 ext4_journalled_zero_new_buffers(handle, page,
1425 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1426 from + copied, &partial,
1429 SetPageUptodate(page);
1432 size_changed = ext4_update_inode_size(inode, pos + copied);
1433 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1434 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1438 if (old_size < pos && !verity)
1439 pagecache_isize_extended(inode, old_size, pos);
1441 if (size_changed || inline_data) {
1442 ret2 = ext4_mark_inode_dirty(handle, inode);
1448 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1449 /* if we have allocated more blocks and copied
1450 * less. We will have blocks allocated outside
1451 * inode->i_size. So truncate them
1453 ext4_orphan_add(handle, inode);
1455 ret2 = ext4_journal_stop(handle);
1458 if (pos + len > inode->i_size && !verity) {
1459 ext4_truncate_failed_write(inode);
1461 * If truncate failed early the inode might still be
1462 * on the orphan list; we need to make sure the inode
1463 * is removed from the orphan list in that case.
1466 ext4_orphan_del(NULL, inode);
1469 return ret ? ret : copied;
1473 * Reserve space for a single cluster
1475 static int ext4_da_reserve_space(struct inode *inode)
1477 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1478 struct ext4_inode_info *ei = EXT4_I(inode);
1482 * We will charge metadata quota at writeout time; this saves
1483 * us from metadata over-estimation, though we may go over by
1484 * a small amount in the end. Here we just reserve for data.
1486 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1490 spin_lock(&ei->i_block_reservation_lock);
1491 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1492 spin_unlock(&ei->i_block_reservation_lock);
1493 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1496 ei->i_reserved_data_blocks++;
1497 trace_ext4_da_reserve_space(inode);
1498 spin_unlock(&ei->i_block_reservation_lock);
1500 return 0; /* success */
1503 void ext4_da_release_space(struct inode *inode, int to_free)
1505 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1506 struct ext4_inode_info *ei = EXT4_I(inode);
1509 return; /* Nothing to release, exit */
1511 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1513 trace_ext4_da_release_space(inode, to_free);
1514 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1516 * if there aren't enough reserved blocks, then the
1517 * counter is messed up somewhere. Since this
1518 * function is called from invalidate page, it's
1519 * harmless to return without any action.
1521 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1522 "ino %lu, to_free %d with only %d reserved "
1523 "data blocks", inode->i_ino, to_free,
1524 ei->i_reserved_data_blocks);
1526 to_free = ei->i_reserved_data_blocks;
1528 ei->i_reserved_data_blocks -= to_free;
1530 /* update fs dirty data blocks counter */
1531 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1533 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1535 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1539 * Delayed allocation stuff
1542 struct mpage_da_data {
1543 struct inode *inode;
1544 struct writeback_control *wbc;
1546 pgoff_t first_page; /* The first page to write */
1547 pgoff_t next_page; /* Current page to examine */
1548 pgoff_t last_page; /* Last page to examine */
1550 * Extent to map - this can be after first_page because that can be
1551 * fully mapped. We somewhat abuse m_flags to store whether the extent
1552 * is delalloc or unwritten.
1554 struct ext4_map_blocks map;
1555 struct ext4_io_submit io_submit; /* IO submission data */
1556 unsigned int do_map:1;
1557 unsigned int scanned_until_end:1;
1560 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1565 struct pagevec pvec;
1566 struct inode *inode = mpd->inode;
1567 struct address_space *mapping = inode->i_mapping;
1569 /* This is necessary when next_page == 0. */
1570 if (mpd->first_page >= mpd->next_page)
1573 mpd->scanned_until_end = 0;
1574 index = mpd->first_page;
1575 end = mpd->next_page - 1;
1577 ext4_lblk_t start, last;
1578 start = index << (PAGE_SHIFT - inode->i_blkbits);
1579 last = end << (PAGE_SHIFT - inode->i_blkbits);
1582 * avoid racing with extent status tree scans made by
1583 * ext4_insert_delayed_block()
1585 down_write(&EXT4_I(inode)->i_data_sem);
1586 ext4_es_remove_extent(inode, start, last - start + 1);
1587 up_write(&EXT4_I(inode)->i_data_sem);
1590 pagevec_init(&pvec);
1591 while (index <= end) {
1592 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1595 for (i = 0; i < nr_pages; i++) {
1596 struct page *page = pvec.pages[i];
1598 BUG_ON(!PageLocked(page));
1599 BUG_ON(PageWriteback(page));
1601 if (page_mapped(page))
1602 clear_page_dirty_for_io(page);
1603 block_invalidatepage(page, 0, PAGE_SIZE);
1604 ClearPageUptodate(page);
1608 pagevec_release(&pvec);
1612 static void ext4_print_free_blocks(struct inode *inode)
1614 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1615 struct super_block *sb = inode->i_sb;
1616 struct ext4_inode_info *ei = EXT4_I(inode);
1618 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1619 EXT4_C2B(EXT4_SB(inode->i_sb),
1620 ext4_count_free_clusters(sb)));
1621 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1622 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1623 (long long) EXT4_C2B(EXT4_SB(sb),
1624 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1625 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1626 (long long) EXT4_C2B(EXT4_SB(sb),
1627 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1628 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1629 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1630 ei->i_reserved_data_blocks);
1634 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1636 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1640 * ext4_insert_delayed_block - adds a delayed block to the extents status
1641 * tree, incrementing the reserved cluster/block
1642 * count or making a pending reservation
1645 * @inode - file containing the newly added block
1646 * @lblk - logical block to be added
1648 * Returns 0 on success, negative error code on failure.
1650 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1652 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1654 bool allocated = false;
1655 bool reserved = false;
1658 * If the cluster containing lblk is shared with a delayed,
1659 * written, or unwritten extent in a bigalloc file system, it's
1660 * already been accounted for and does not need to be reserved.
1661 * A pending reservation must be made for the cluster if it's
1662 * shared with a written or unwritten extent and doesn't already
1663 * have one. Written and unwritten extents can be purged from the
1664 * extents status tree if the system is under memory pressure, so
1665 * it's necessary to examine the extent tree if a search of the
1666 * extents status tree doesn't get a match.
1668 if (sbi->s_cluster_ratio == 1) {
1669 ret = ext4_da_reserve_space(inode);
1670 if (ret != 0) /* ENOSPC */
1673 } else { /* bigalloc */
1674 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1675 if (!ext4_es_scan_clu(inode,
1676 &ext4_es_is_mapped, lblk)) {
1677 ret = ext4_clu_mapped(inode,
1678 EXT4_B2C(sbi, lblk));
1682 ret = ext4_da_reserve_space(inode);
1683 if (ret != 0) /* ENOSPC */
1695 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1696 if (ret && reserved)
1697 ext4_da_release_space(inode, 1);
1704 * This function is grabs code from the very beginning of
1705 * ext4_map_blocks, but assumes that the caller is from delayed write
1706 * time. This function looks up the requested blocks and sets the
1707 * buffer delay bit under the protection of i_data_sem.
1709 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1710 struct ext4_map_blocks *map,
1711 struct buffer_head *bh)
1713 struct extent_status es;
1715 sector_t invalid_block = ~((sector_t) 0xffff);
1716 #ifdef ES_AGGRESSIVE_TEST
1717 struct ext4_map_blocks orig_map;
1719 memcpy(&orig_map, map, sizeof(*map));
1722 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1726 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1727 (unsigned long) map->m_lblk);
1729 /* Lookup extent status tree firstly */
1730 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1731 if (ext4_es_is_hole(&es))
1735 * Delayed extent could be allocated by fallocate.
1736 * So we need to check it.
1738 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1739 map_bh(bh, inode->i_sb, invalid_block);
1741 set_buffer_delay(bh);
1745 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1746 retval = es.es_len - (iblock - es.es_lblk);
1747 if (retval > map->m_len)
1748 retval = map->m_len;
1749 map->m_len = retval;
1750 if (ext4_es_is_written(&es))
1751 map->m_flags |= EXT4_MAP_MAPPED;
1752 else if (ext4_es_is_unwritten(&es))
1753 map->m_flags |= EXT4_MAP_UNWRITTEN;
1757 #ifdef ES_AGGRESSIVE_TEST
1758 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1764 * Try to see if we can get the block without requesting a new
1765 * file system block.
1767 down_read(&EXT4_I(inode)->i_data_sem);
1768 if (ext4_has_inline_data(inode))
1770 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1771 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1773 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1775 up_read(&EXT4_I(inode)->i_data_sem);
1779 unsigned int status;
1781 if (unlikely(retval != map->m_len)) {
1782 ext4_warning(inode->i_sb,
1783 "ES len assertion failed for inode "
1784 "%lu: retval %d != map->m_len %d",
1785 inode->i_ino, retval, map->m_len);
1789 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1790 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1791 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1792 map->m_pblk, status);
1793 up_read(&EXT4_I(inode)->i_data_sem);
1796 up_read(&EXT4_I(inode)->i_data_sem);
1799 down_write(&EXT4_I(inode)->i_data_sem);
1800 retval = ext4_insert_delayed_block(inode, map->m_lblk);
1801 up_write(&EXT4_I(inode)->i_data_sem);
1805 map_bh(bh, inode->i_sb, invalid_block);
1807 set_buffer_delay(bh);
1812 * This is a special get_block_t callback which is used by
1813 * ext4_da_write_begin(). It will either return mapped block or
1814 * reserve space for a single block.
1816 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1817 * We also have b_blocknr = -1 and b_bdev initialized properly
1819 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1820 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1821 * initialized properly.
1823 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1824 struct buffer_head *bh, int create)
1826 struct ext4_map_blocks map;
1829 BUG_ON(create == 0);
1830 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1832 map.m_lblk = iblock;
1836 * first, we need to know whether the block is allocated already
1837 * preallocated blocks are unmapped but should treated
1838 * the same as allocated blocks.
1840 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1844 map_bh(bh, inode->i_sb, map.m_pblk);
1845 ext4_update_bh_state(bh, map.m_flags);
1847 if (buffer_unwritten(bh)) {
1848 /* A delayed write to unwritten bh should be marked
1849 * new and mapped. Mapped ensures that we don't do
1850 * get_block multiple times when we write to the same
1851 * offset and new ensures that we do proper zero out
1852 * for partial write.
1855 set_buffer_mapped(bh);
1860 static int bget_one(handle_t *handle, struct buffer_head *bh)
1866 static int bput_one(handle_t *handle, struct buffer_head *bh)
1872 static int __ext4_journalled_writepage(struct page *page,
1875 struct address_space *mapping = page->mapping;
1876 struct inode *inode = mapping->host;
1877 struct buffer_head *page_bufs = NULL;
1878 handle_t *handle = NULL;
1879 int ret = 0, err = 0;
1880 int inline_data = ext4_has_inline_data(inode);
1881 struct buffer_head *inode_bh = NULL;
1883 ClearPageChecked(page);
1886 BUG_ON(page->index != 0);
1887 BUG_ON(len > ext4_get_max_inline_size(inode));
1888 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1889 if (inode_bh == NULL)
1892 page_bufs = page_buffers(page);
1897 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1901 * We need to release the page lock before we start the
1902 * journal, so grab a reference so the page won't disappear
1903 * out from under us.
1908 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1909 ext4_writepage_trans_blocks(inode));
1910 if (IS_ERR(handle)) {
1911 ret = PTR_ERR(handle);
1913 goto out_no_pagelock;
1915 BUG_ON(!ext4_handle_valid(handle));
1919 if (page->mapping != mapping) {
1920 /* The page got truncated from under us */
1921 ext4_journal_stop(handle);
1927 ret = ext4_mark_inode_dirty(handle, inode);
1929 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1930 do_journal_get_write_access);
1932 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1937 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1940 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1941 err = ext4_journal_stop(handle);
1945 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1949 if (!inline_data && page_bufs)
1950 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1957 * Note that we don't need to start a transaction unless we're journaling data
1958 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1959 * need to file the inode to the transaction's list in ordered mode because if
1960 * we are writing back data added by write(), the inode is already there and if
1961 * we are writing back data modified via mmap(), no one guarantees in which
1962 * transaction the data will hit the disk. In case we are journaling data, we
1963 * cannot start transaction directly because transaction start ranks above page
1964 * lock so we have to do some magic.
1966 * This function can get called via...
1967 * - ext4_writepages after taking page lock (have journal handle)
1968 * - journal_submit_inode_data_buffers (no journal handle)
1969 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1970 * - grab_page_cache when doing write_begin (have journal handle)
1972 * We don't do any block allocation in this function. If we have page with
1973 * multiple blocks we need to write those buffer_heads that are mapped. This
1974 * is important for mmaped based write. So if we do with blocksize 1K
1975 * truncate(f, 1024);
1976 * a = mmap(f, 0, 4096);
1978 * truncate(f, 4096);
1979 * we have in the page first buffer_head mapped via page_mkwrite call back
1980 * but other buffer_heads would be unmapped but dirty (dirty done via the
1981 * do_wp_page). So writepage should write the first block. If we modify
1982 * the mmap area beyond 1024 we will again get a page_fault and the
1983 * page_mkwrite callback will do the block allocation and mark the
1984 * buffer_heads mapped.
1986 * We redirty the page if we have any buffer_heads that is either delay or
1987 * unwritten in the page.
1989 * We can get recursively called as show below.
1991 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1994 * But since we don't do any block allocation we should not deadlock.
1995 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1997 static int ext4_writepage(struct page *page,
1998 struct writeback_control *wbc)
2003 struct buffer_head *page_bufs = NULL;
2004 struct inode *inode = page->mapping->host;
2005 struct ext4_io_submit io_submit;
2006 bool keep_towrite = false;
2008 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2009 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
2014 trace_ext4_writepage(page);
2015 size = i_size_read(inode);
2016 if (page->index == size >> PAGE_SHIFT &&
2017 !ext4_verity_in_progress(inode))
2018 len = size & ~PAGE_MASK;
2022 /* Should never happen but for bugs in other kernel subsystems */
2023 if (!page_has_buffers(page)) {
2024 ext4_warning_inode(inode,
2025 "page %lu does not have buffers attached", page->index);
2026 ClearPageDirty(page);
2031 page_bufs = page_buffers(page);
2033 * We cannot do block allocation or other extent handling in this
2034 * function. If there are buffers needing that, we have to redirty
2035 * the page. But we may reach here when we do a journal commit via
2036 * journal_submit_inode_data_buffers() and in that case we must write
2037 * allocated buffers to achieve data=ordered mode guarantees.
2039 * Also, if there is only one buffer per page (the fs block
2040 * size == the page size), if one buffer needs block
2041 * allocation or needs to modify the extent tree to clear the
2042 * unwritten flag, we know that the page can't be written at
2043 * all, so we might as well refuse the write immediately.
2044 * Unfortunately if the block size != page size, we can't as
2045 * easily detect this case using ext4_walk_page_buffers(), but
2046 * for the extremely common case, this is an optimization that
2047 * skips a useless round trip through ext4_bio_write_page().
2049 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2050 ext4_bh_delay_or_unwritten)) {
2051 redirty_page_for_writepage(wbc, page);
2052 if ((current->flags & PF_MEMALLOC) ||
2053 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2055 * For memory cleaning there's no point in writing only
2056 * some buffers. So just bail out. Warn if we came here
2057 * from direct reclaim.
2059 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2064 keep_towrite = true;
2067 if (PageChecked(page) && ext4_should_journal_data(inode))
2069 * It's mmapped pagecache. Add buffers and journal it. There
2070 * doesn't seem much point in redirtying the page here.
2072 return __ext4_journalled_writepage(page, len);
2074 ext4_io_submit_init(&io_submit, wbc);
2075 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2076 if (!io_submit.io_end) {
2077 redirty_page_for_writepage(wbc, page);
2081 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2082 ext4_io_submit(&io_submit);
2083 /* Drop io_end reference we got from init */
2084 ext4_put_io_end_defer(io_submit.io_end);
2088 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2094 BUG_ON(page->index != mpd->first_page);
2095 clear_page_dirty_for_io(page);
2097 * We have to be very careful here! Nothing protects writeback path
2098 * against i_size changes and the page can be writeably mapped into
2099 * page tables. So an application can be growing i_size and writing
2100 * data through mmap while writeback runs. clear_page_dirty_for_io()
2101 * write-protects our page in page tables and the page cannot get
2102 * written to again until we release page lock. So only after
2103 * clear_page_dirty_for_io() we are safe to sample i_size for
2104 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2105 * on the barrier provided by TestClearPageDirty in
2106 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2107 * after page tables are updated.
2109 size = i_size_read(mpd->inode);
2110 if (page->index == size >> PAGE_SHIFT &&
2111 !ext4_verity_in_progress(mpd->inode))
2112 len = size & ~PAGE_MASK;
2115 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2117 mpd->wbc->nr_to_write--;
2123 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2126 * mballoc gives us at most this number of blocks...
2127 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2128 * The rest of mballoc seems to handle chunks up to full group size.
2130 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2133 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2135 * @mpd - extent of blocks
2136 * @lblk - logical number of the block in the file
2137 * @bh - buffer head we want to add to the extent
2139 * The function is used to collect contig. blocks in the same state. If the
2140 * buffer doesn't require mapping for writeback and we haven't started the
2141 * extent of buffers to map yet, the function returns 'true' immediately - the
2142 * caller can write the buffer right away. Otherwise the function returns true
2143 * if the block has been added to the extent, false if the block couldn't be
2146 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2147 struct buffer_head *bh)
2149 struct ext4_map_blocks *map = &mpd->map;
2151 /* Buffer that doesn't need mapping for writeback? */
2152 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2153 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2154 /* So far no extent to map => we write the buffer right away */
2155 if (map->m_len == 0)
2160 /* First block in the extent? */
2161 if (map->m_len == 0) {
2162 /* We cannot map unless handle is started... */
2167 map->m_flags = bh->b_state & BH_FLAGS;
2171 /* Don't go larger than mballoc is willing to allocate */
2172 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2175 /* Can we merge the block to our big extent? */
2176 if (lblk == map->m_lblk + map->m_len &&
2177 (bh->b_state & BH_FLAGS) == map->m_flags) {
2185 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2187 * @mpd - extent of blocks for mapping
2188 * @head - the first buffer in the page
2189 * @bh - buffer we should start processing from
2190 * @lblk - logical number of the block in the file corresponding to @bh
2192 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2193 * the page for IO if all buffers in this page were mapped and there's no
2194 * accumulated extent of buffers to map or add buffers in the page to the
2195 * extent of buffers to map. The function returns 1 if the caller can continue
2196 * by processing the next page, 0 if it should stop adding buffers to the
2197 * extent to map because we cannot extend it anymore. It can also return value
2198 * < 0 in case of error during IO submission.
2200 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2201 struct buffer_head *head,
2202 struct buffer_head *bh,
2205 struct inode *inode = mpd->inode;
2207 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2208 >> inode->i_blkbits;
2210 if (ext4_verity_in_progress(inode))
2211 blocks = EXT_MAX_BLOCKS;
2214 BUG_ON(buffer_locked(bh));
2216 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2217 /* Found extent to map? */
2220 /* Buffer needs mapping and handle is not started? */
2223 /* Everything mapped so far and we hit EOF */
2226 } while (lblk++, (bh = bh->b_this_page) != head);
2227 /* So far everything mapped? Submit the page for IO. */
2228 if (mpd->map.m_len == 0) {
2229 err = mpage_submit_page(mpd, head->b_page);
2233 if (lblk >= blocks) {
2234 mpd->scanned_until_end = 1;
2241 * mpage_process_page - update page buffers corresponding to changed extent and
2242 * may submit fully mapped page for IO
2244 * @mpd - description of extent to map, on return next extent to map
2245 * @m_lblk - logical block mapping.
2246 * @m_pblk - corresponding physical mapping.
2247 * @map_bh - determines on return whether this page requires any further
2249 * Scan given page buffers corresponding to changed extent and update buffer
2250 * state according to new extent state.
2251 * We map delalloc buffers to their physical location, clear unwritten bits.
2252 * If the given page is not fully mapped, we update @map to the next extent in
2253 * the given page that needs mapping & return @map_bh as true.
2255 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2256 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2259 struct buffer_head *head, *bh;
2260 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2261 ext4_lblk_t lblk = *m_lblk;
2262 ext4_fsblk_t pblock = *m_pblk;
2264 int blkbits = mpd->inode->i_blkbits;
2265 ssize_t io_end_size = 0;
2266 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2268 bh = head = page_buffers(page);
2270 if (lblk < mpd->map.m_lblk)
2272 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2274 * Buffer after end of mapped extent.
2275 * Find next buffer in the page to map.
2278 mpd->map.m_flags = 0;
2279 io_end_vec->size += io_end_size;
2282 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2285 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2286 io_end_vec = ext4_alloc_io_end_vec(io_end);
2287 if (IS_ERR(io_end_vec)) {
2288 err = PTR_ERR(io_end_vec);
2291 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2296 if (buffer_delay(bh)) {
2297 clear_buffer_delay(bh);
2298 bh->b_blocknr = pblock++;
2300 clear_buffer_unwritten(bh);
2301 io_end_size += (1 << blkbits);
2302 } while (lblk++, (bh = bh->b_this_page) != head);
2304 io_end_vec->size += io_end_size;
2314 * mpage_map_buffers - update buffers corresponding to changed extent and
2315 * submit fully mapped pages for IO
2317 * @mpd - description of extent to map, on return next extent to map
2319 * Scan buffers corresponding to changed extent (we expect corresponding pages
2320 * to be already locked) and update buffer state according to new extent state.
2321 * We map delalloc buffers to their physical location, clear unwritten bits,
2322 * and mark buffers as uninit when we perform writes to unwritten extents
2323 * and do extent conversion after IO is finished. If the last page is not fully
2324 * mapped, we update @map to the next extent in the last page that needs
2325 * mapping. Otherwise we submit the page for IO.
2327 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2329 struct pagevec pvec;
2331 struct inode *inode = mpd->inode;
2332 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2335 ext4_fsblk_t pblock;
2337 bool map_bh = false;
2339 start = mpd->map.m_lblk >> bpp_bits;
2340 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2341 lblk = start << bpp_bits;
2342 pblock = mpd->map.m_pblk;
2344 pagevec_init(&pvec);
2345 while (start <= end) {
2346 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2350 for (i = 0; i < nr_pages; i++) {
2351 struct page *page = pvec.pages[i];
2353 err = mpage_process_page(mpd, page, &lblk, &pblock,
2356 * If map_bh is true, means page may require further bh
2357 * mapping, or maybe the page was submitted for IO.
2358 * So we return to call further extent mapping.
2360 if (err < 0 || map_bh)
2362 /* Page fully mapped - let IO run! */
2363 err = mpage_submit_page(mpd, page);
2367 pagevec_release(&pvec);
2369 /* Extent fully mapped and matches with page boundary. We are done. */
2371 mpd->map.m_flags = 0;
2374 pagevec_release(&pvec);
2378 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2380 struct inode *inode = mpd->inode;
2381 struct ext4_map_blocks *map = &mpd->map;
2382 int get_blocks_flags;
2383 int err, dioread_nolock;
2385 trace_ext4_da_write_pages_extent(inode, map);
2387 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2388 * to convert an unwritten extent to be initialized (in the case
2389 * where we have written into one or more preallocated blocks). It is
2390 * possible that we're going to need more metadata blocks than
2391 * previously reserved. However we must not fail because we're in
2392 * writeback and there is nothing we can do about it so it might result
2393 * in data loss. So use reserved blocks to allocate metadata if
2396 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2397 * the blocks in question are delalloc blocks. This indicates
2398 * that the blocks and quotas has already been checked when
2399 * the data was copied into the page cache.
2401 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2402 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2403 EXT4_GET_BLOCKS_IO_SUBMIT;
2404 dioread_nolock = ext4_should_dioread_nolock(inode);
2406 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2407 if (map->m_flags & BIT(BH_Delay))
2408 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2410 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2413 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2414 if (!mpd->io_submit.io_end->handle &&
2415 ext4_handle_valid(handle)) {
2416 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2417 handle->h_rsv_handle = NULL;
2419 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2422 BUG_ON(map->m_len == 0);
2427 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2428 * mpd->len and submit pages underlying it for IO
2430 * @handle - handle for journal operations
2431 * @mpd - extent to map
2432 * @give_up_on_write - we set this to true iff there is a fatal error and there
2433 * is no hope of writing the data. The caller should discard
2434 * dirty pages to avoid infinite loops.
2436 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2437 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2438 * them to initialized or split the described range from larger unwritten
2439 * extent. Note that we need not map all the described range since allocation
2440 * can return less blocks or the range is covered by more unwritten extents. We
2441 * cannot map more because we are limited by reserved transaction credits. On
2442 * the other hand we always make sure that the last touched page is fully
2443 * mapped so that it can be written out (and thus forward progress is
2444 * guaranteed). After mapping we submit all mapped pages for IO.
2446 static int mpage_map_and_submit_extent(handle_t *handle,
2447 struct mpage_da_data *mpd,
2448 bool *give_up_on_write)
2450 struct inode *inode = mpd->inode;
2451 struct ext4_map_blocks *map = &mpd->map;
2455 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2456 struct ext4_io_end_vec *io_end_vec;
2458 io_end_vec = ext4_alloc_io_end_vec(io_end);
2459 if (IS_ERR(io_end_vec))
2460 return PTR_ERR(io_end_vec);
2461 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2463 err = mpage_map_one_extent(handle, mpd);
2465 struct super_block *sb = inode->i_sb;
2467 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2468 ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2469 goto invalidate_dirty_pages;
2471 * Let the uper layers retry transient errors.
2472 * In the case of ENOSPC, if ext4_count_free_blocks()
2473 * is non-zero, a commit should free up blocks.
2475 if ((err == -ENOMEM) ||
2476 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2478 goto update_disksize;
2481 ext4_msg(sb, KERN_CRIT,
2482 "Delayed block allocation failed for "
2483 "inode %lu at logical offset %llu with"
2484 " max blocks %u with error %d",
2486 (unsigned long long)map->m_lblk,
2487 (unsigned)map->m_len, -err);
2488 ext4_msg(sb, KERN_CRIT,
2489 "This should not happen!! Data will "
2492 ext4_print_free_blocks(inode);
2493 invalidate_dirty_pages:
2494 *give_up_on_write = true;
2499 * Update buffer state, submit mapped pages, and get us new
2502 err = mpage_map_and_submit_buffers(mpd);
2504 goto update_disksize;
2505 } while (map->m_len);
2509 * Update on-disk size after IO is submitted. Races with
2510 * truncate are avoided by checking i_size under i_data_sem.
2512 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2513 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2517 down_write(&EXT4_I(inode)->i_data_sem);
2518 i_size = i_size_read(inode);
2519 if (disksize > i_size)
2521 if (disksize > EXT4_I(inode)->i_disksize)
2522 EXT4_I(inode)->i_disksize = disksize;
2523 up_write(&EXT4_I(inode)->i_data_sem);
2524 err2 = ext4_mark_inode_dirty(handle, inode);
2526 ext4_error_err(inode->i_sb, -err2,
2527 "Failed to mark inode %lu dirty",
2537 * Calculate the total number of credits to reserve for one writepages
2538 * iteration. This is called from ext4_writepages(). We map an extent of
2539 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2540 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2541 * bpp - 1 blocks in bpp different extents.
2543 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2545 int bpp = ext4_journal_blocks_per_page(inode);
2547 return ext4_meta_trans_blocks(inode,
2548 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2552 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2553 * and underlying extent to map
2555 * @mpd - where to look for pages
2557 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2558 * IO immediately. When we find a page which isn't mapped we start accumulating
2559 * extent of buffers underlying these pages that needs mapping (formed by
2560 * either delayed or unwritten buffers). We also lock the pages containing
2561 * these buffers. The extent found is returned in @mpd structure (starting at
2562 * mpd->lblk with length mpd->len blocks).
2564 * Note that this function can attach bios to one io_end structure which are
2565 * neither logically nor physically contiguous. Although it may seem as an
2566 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2567 * case as we need to track IO to all buffers underlying a page in one io_end.
2569 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2571 struct address_space *mapping = mpd->inode->i_mapping;
2572 struct pagevec pvec;
2573 unsigned int nr_pages;
2574 long left = mpd->wbc->nr_to_write;
2575 pgoff_t index = mpd->first_page;
2576 pgoff_t end = mpd->last_page;
2579 int blkbits = mpd->inode->i_blkbits;
2581 struct buffer_head *head;
2583 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2584 tag = PAGECACHE_TAG_TOWRITE;
2586 tag = PAGECACHE_TAG_DIRTY;
2588 pagevec_init(&pvec);
2590 mpd->next_page = index;
2591 while (index <= end) {
2592 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2597 for (i = 0; i < nr_pages; i++) {
2598 struct page *page = pvec.pages[i];
2601 * Accumulated enough dirty pages? This doesn't apply
2602 * to WB_SYNC_ALL mode. For integrity sync we have to
2603 * keep going because someone may be concurrently
2604 * dirtying pages, and we might have synced a lot of
2605 * newly appeared dirty pages, but have not synced all
2606 * of the old dirty pages.
2608 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2611 /* If we can't merge this page, we are done. */
2612 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2617 * If the page is no longer dirty, or its mapping no
2618 * longer corresponds to inode we are writing (which
2619 * means it has been truncated or invalidated), or the
2620 * page is already under writeback and we are not doing
2621 * a data integrity writeback, skip the page
2623 if (!PageDirty(page) ||
2624 (PageWriteback(page) &&
2625 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2626 unlikely(page->mapping != mapping)) {
2631 wait_on_page_writeback(page);
2632 BUG_ON(PageWriteback(page));
2635 * Should never happen but for buggy code in
2636 * other subsystems that call
2637 * set_page_dirty() without properly warning
2638 * the file system first. See [1] for more
2641 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2643 if (!page_has_buffers(page)) {
2644 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2645 ClearPageDirty(page);
2650 if (mpd->map.m_len == 0)
2651 mpd->first_page = page->index;
2652 mpd->next_page = page->index + 1;
2653 /* Add all dirty buffers to mpd */
2654 lblk = ((ext4_lblk_t)page->index) <<
2655 (PAGE_SHIFT - blkbits);
2656 head = page_buffers(page);
2657 err = mpage_process_page_bufs(mpd, head, head, lblk);
2663 pagevec_release(&pvec);
2666 mpd->scanned_until_end = 1;
2669 pagevec_release(&pvec);
2673 static int ext4_writepages(struct address_space *mapping,
2674 struct writeback_control *wbc)
2676 pgoff_t writeback_index = 0;
2677 long nr_to_write = wbc->nr_to_write;
2678 int range_whole = 0;
2680 handle_t *handle = NULL;
2681 struct mpage_da_data mpd;
2682 struct inode *inode = mapping->host;
2683 int needed_blocks, rsv_blocks = 0, ret = 0;
2684 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2685 struct blk_plug plug;
2686 bool give_up_on_write = false;
2688 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2691 percpu_down_read(&sbi->s_writepages_rwsem);
2692 trace_ext4_writepages(inode, wbc);
2695 * No pages to write? This is mainly a kludge to avoid starting
2696 * a transaction for special inodes like journal inode on last iput()
2697 * because that could violate lock ordering on umount
2699 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2700 goto out_writepages;
2702 if (ext4_should_journal_data(inode)) {
2703 ret = generic_writepages(mapping, wbc);
2704 goto out_writepages;
2708 * If the filesystem has aborted, it is read-only, so return
2709 * right away instead of dumping stack traces later on that
2710 * will obscure the real source of the problem. We test
2711 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2712 * the latter could be true if the filesystem is mounted
2713 * read-only, and in that case, ext4_writepages should
2714 * *never* be called, so if that ever happens, we would want
2717 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2718 ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2720 goto out_writepages;
2724 * If we have inline data and arrive here, it means that
2725 * we will soon create the block for the 1st page, so
2726 * we'd better clear the inline data here.
2728 if (ext4_has_inline_data(inode)) {
2729 /* Just inode will be modified... */
2730 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2731 if (IS_ERR(handle)) {
2732 ret = PTR_ERR(handle);
2733 goto out_writepages;
2735 BUG_ON(ext4_test_inode_state(inode,
2736 EXT4_STATE_MAY_INLINE_DATA));
2737 ext4_destroy_inline_data(handle, inode);
2738 ext4_journal_stop(handle);
2741 if (ext4_should_dioread_nolock(inode)) {
2743 * We may need to convert up to one extent per block in
2744 * the page and we may dirty the inode.
2746 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2747 PAGE_SIZE >> inode->i_blkbits);
2750 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2753 if (wbc->range_cyclic) {
2754 writeback_index = mapping->writeback_index;
2755 if (writeback_index)
2757 mpd.first_page = writeback_index;
2760 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2761 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2766 ext4_io_submit_init(&mpd.io_submit, wbc);
2768 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2769 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2770 blk_start_plug(&plug);
2773 * First writeback pages that don't need mapping - we can avoid
2774 * starting a transaction unnecessarily and also avoid being blocked
2775 * in the block layer on device congestion while having transaction
2779 mpd.scanned_until_end = 0;
2780 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2781 if (!mpd.io_submit.io_end) {
2785 ret = mpage_prepare_extent_to_map(&mpd);
2786 /* Unlock pages we didn't use */
2787 mpage_release_unused_pages(&mpd, false);
2788 /* Submit prepared bio */
2789 ext4_io_submit(&mpd.io_submit);
2790 ext4_put_io_end_defer(mpd.io_submit.io_end);
2791 mpd.io_submit.io_end = NULL;
2795 while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2796 /* For each extent of pages we use new io_end */
2797 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2798 if (!mpd.io_submit.io_end) {
2804 * We have two constraints: We find one extent to map and we
2805 * must always write out whole page (makes a difference when
2806 * blocksize < pagesize) so that we don't block on IO when we
2807 * try to write out the rest of the page. Journalled mode is
2808 * not supported by delalloc.
2810 BUG_ON(ext4_should_journal_data(inode));
2811 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2813 /* start a new transaction */
2814 handle = ext4_journal_start_with_reserve(inode,
2815 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2816 if (IS_ERR(handle)) {
2817 ret = PTR_ERR(handle);
2818 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2819 "%ld pages, ino %lu; err %d", __func__,
2820 wbc->nr_to_write, inode->i_ino, ret);
2821 /* Release allocated io_end */
2822 ext4_put_io_end(mpd.io_submit.io_end);
2823 mpd.io_submit.io_end = NULL;
2828 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2829 ret = mpage_prepare_extent_to_map(&mpd);
2830 if (!ret && mpd.map.m_len)
2831 ret = mpage_map_and_submit_extent(handle, &mpd,
2834 * Caution: If the handle is synchronous,
2835 * ext4_journal_stop() can wait for transaction commit
2836 * to finish which may depend on writeback of pages to
2837 * complete or on page lock to be released. In that
2838 * case, we have to wait until after we have
2839 * submitted all the IO, released page locks we hold,
2840 * and dropped io_end reference (for extent conversion
2841 * to be able to complete) before stopping the handle.
2843 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2844 ext4_journal_stop(handle);
2848 /* Unlock pages we didn't use */
2849 mpage_release_unused_pages(&mpd, give_up_on_write);
2850 /* Submit prepared bio */
2851 ext4_io_submit(&mpd.io_submit);
2854 * Drop our io_end reference we got from init. We have
2855 * to be careful and use deferred io_end finishing if
2856 * we are still holding the transaction as we can
2857 * release the last reference to io_end which may end
2858 * up doing unwritten extent conversion.
2861 ext4_put_io_end_defer(mpd.io_submit.io_end);
2862 ext4_journal_stop(handle);
2864 ext4_put_io_end(mpd.io_submit.io_end);
2865 mpd.io_submit.io_end = NULL;
2867 if (ret == -ENOSPC && sbi->s_journal) {
2869 * Commit the transaction which would
2870 * free blocks released in the transaction
2873 jbd2_journal_force_commit_nested(sbi->s_journal);
2877 /* Fatal error - ENOMEM, EIO... */
2882 blk_finish_plug(&plug);
2883 if (!ret && !cycled && wbc->nr_to_write > 0) {
2885 mpd.last_page = writeback_index - 1;
2891 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2893 * Set the writeback_index so that range_cyclic
2894 * mode will write it back later
2896 mapping->writeback_index = mpd.first_page;
2899 trace_ext4_writepages_result(inode, wbc, ret,
2900 nr_to_write - wbc->nr_to_write);
2901 percpu_up_read(&sbi->s_writepages_rwsem);
2905 static int ext4_dax_writepages(struct address_space *mapping,
2906 struct writeback_control *wbc)
2909 long nr_to_write = wbc->nr_to_write;
2910 struct inode *inode = mapping->host;
2911 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2913 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2916 percpu_down_read(&sbi->s_writepages_rwsem);
2917 trace_ext4_writepages(inode, wbc);
2919 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2920 trace_ext4_writepages_result(inode, wbc, ret,
2921 nr_to_write - wbc->nr_to_write);
2922 percpu_up_read(&sbi->s_writepages_rwsem);
2926 static int ext4_nonda_switch(struct super_block *sb)
2928 s64 free_clusters, dirty_clusters;
2929 struct ext4_sb_info *sbi = EXT4_SB(sb);
2932 * switch to non delalloc mode if we are running low
2933 * on free block. The free block accounting via percpu
2934 * counters can get slightly wrong with percpu_counter_batch getting
2935 * accumulated on each CPU without updating global counters
2936 * Delalloc need an accurate free block accounting. So switch
2937 * to non delalloc when we are near to error range.
2940 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2942 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2944 * Start pushing delalloc when 1/2 of free blocks are dirty.
2946 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2947 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2949 if (2 * free_clusters < 3 * dirty_clusters ||
2950 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2952 * free block count is less than 150% of dirty blocks
2953 * or free blocks is less than watermark
2960 /* We always reserve for an inode update; the superblock could be there too */
2961 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2963 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2966 if (pos + len <= 0x7fffffffULL)
2969 /* We might need to update the superblock to set LARGE_FILE */
2973 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2974 loff_t pos, unsigned len, unsigned flags,
2975 struct page **pagep, void **fsdata)
2977 int ret, retries = 0;
2980 struct inode *inode = mapping->host;
2983 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2986 index = pos >> PAGE_SHIFT;
2988 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2989 ext4_verity_in_progress(inode)) {
2990 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2991 return ext4_write_begin(file, mapping, pos,
2992 len, flags, pagep, fsdata);
2994 *fsdata = (void *)0;
2995 trace_ext4_da_write_begin(inode, pos, len, flags);
2997 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2998 ret = ext4_da_write_inline_data_begin(mapping, inode,
3008 * grab_cache_page_write_begin() can take a long time if the
3009 * system is thrashing due to memory pressure, or if the page
3010 * is being written back. So grab it first before we start
3011 * the transaction handle. This also allows us to allocate
3012 * the page (if needed) without using GFP_NOFS.
3015 page = grab_cache_page_write_begin(mapping, index, flags);
3021 * With delayed allocation, we don't log the i_disksize update
3022 * if there is delayed block allocation. But we still need
3023 * to journalling the i_disksize update if writes to the end
3024 * of file which has an already mapped buffer.
3027 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3028 ext4_da_write_credits(inode, pos, len));
3029 if (IS_ERR(handle)) {
3031 return PTR_ERR(handle);
3035 if (page->mapping != mapping) {
3036 /* The page got truncated from under us */
3039 ext4_journal_stop(handle);
3042 /* In case writeback began while the page was unlocked */
3043 wait_for_stable_page(page);
3045 #ifdef CONFIG_FS_ENCRYPTION
3046 ret = ext4_block_write_begin(page, pos, len,
3047 ext4_da_get_block_prep);
3049 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3053 ext4_journal_stop(handle);
3055 * block_write_begin may have instantiated a few blocks
3056 * outside i_size. Trim these off again. Don't need
3057 * i_size_read because we hold i_mutex.
3059 if (pos + len > inode->i_size)
3060 ext4_truncate_failed_write(inode);
3062 if (ret == -ENOSPC &&
3063 ext4_should_retry_alloc(inode->i_sb, &retries))
3075 * Check if we should update i_disksize
3076 * when write to the end of file but not require block allocation
3078 static int ext4_da_should_update_i_disksize(struct page *page,
3079 unsigned long offset)
3081 struct buffer_head *bh;
3082 struct inode *inode = page->mapping->host;
3086 bh = page_buffers(page);
3087 idx = offset >> inode->i_blkbits;
3089 for (i = 0; i < idx; i++)
3090 bh = bh->b_this_page;
3092 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3097 static int ext4_da_write_end(struct file *file,
3098 struct address_space *mapping,
3099 loff_t pos, unsigned len, unsigned copied,
3100 struct page *page, void *fsdata)
3102 struct inode *inode = mapping->host;
3104 handle_t *handle = ext4_journal_current_handle();
3106 unsigned long start, end;
3107 int write_mode = (int)(unsigned long)fsdata;
3109 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3110 return ext4_write_end(file, mapping, pos,
3111 len, copied, page, fsdata);
3113 trace_ext4_da_write_end(inode, pos, len, copied);
3114 start = pos & (PAGE_SIZE - 1);
3115 end = start + copied - 1;
3118 * Since we are holding inode lock, we are sure i_disksize <=
3119 * i_size. We also know that if i_disksize < i_size, there are
3120 * delalloc writes pending in the range upto i_size. If the end of
3121 * the current write is <= i_size, there's no need to touch
3122 * i_disksize since writeback will push i_disksize upto i_size
3123 * eventually. If the end of the current write is > i_size and
3124 * inside an allocated block (ext4_da_should_update_i_disksize()
3125 * check), we need to update i_disksize here as neither
3126 * ext4_writepage() nor certain ext4_writepages() paths not
3127 * allocating blocks update i_disksize.
3129 * Note that we defer inode dirtying to generic_write_end() /
3130 * ext4_da_write_inline_data_end().
3132 new_i_size = pos + copied;
3133 if (copied && new_i_size > inode->i_size) {
3134 if (ext4_has_inline_data(inode) ||
3135 ext4_da_should_update_i_disksize(page, end))
3136 ext4_update_i_disksize(inode, new_i_size);
3139 if (write_mode != CONVERT_INLINE_DATA &&
3140 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3141 ext4_has_inline_data(inode))
3142 ret = ext4_da_write_inline_data_end(inode, pos, len, copied,
3145 ret = generic_write_end(file, mapping, pos, len, copied,
3149 ret2 = ext4_journal_stop(handle);
3150 if (unlikely(ret2 && !ret))
3153 return ret ? ret : copied;
3157 * Force all delayed allocation blocks to be allocated for a given inode.
3159 int ext4_alloc_da_blocks(struct inode *inode)
3161 trace_ext4_alloc_da_blocks(inode);
3163 if (!EXT4_I(inode)->i_reserved_data_blocks)
3167 * We do something simple for now. The filemap_flush() will
3168 * also start triggering a write of the data blocks, which is
3169 * not strictly speaking necessary (and for users of
3170 * laptop_mode, not even desirable). However, to do otherwise
3171 * would require replicating code paths in:
3173 * ext4_writepages() ->
3174 * write_cache_pages() ---> (via passed in callback function)
3175 * __mpage_da_writepage() -->
3176 * mpage_add_bh_to_extent()
3177 * mpage_da_map_blocks()
3179 * The problem is that write_cache_pages(), located in
3180 * mm/page-writeback.c, marks pages clean in preparation for
3181 * doing I/O, which is not desirable if we're not planning on
3184 * We could call write_cache_pages(), and then redirty all of
3185 * the pages by calling redirty_page_for_writepage() but that
3186 * would be ugly in the extreme. So instead we would need to
3187 * replicate parts of the code in the above functions,
3188 * simplifying them because we wouldn't actually intend to
3189 * write out the pages, but rather only collect contiguous
3190 * logical block extents, call the multi-block allocator, and
3191 * then update the buffer heads with the block allocations.
3193 * For now, though, we'll cheat by calling filemap_flush(),
3194 * which will map the blocks, and start the I/O, but not
3195 * actually wait for the I/O to complete.
3197 return filemap_flush(inode->i_mapping);
3201 * bmap() is special. It gets used by applications such as lilo and by
3202 * the swapper to find the on-disk block of a specific piece of data.
3204 * Naturally, this is dangerous if the block concerned is still in the
3205 * journal. If somebody makes a swapfile on an ext4 data-journaling
3206 * filesystem and enables swap, then they may get a nasty shock when the
3207 * data getting swapped to that swapfile suddenly gets overwritten by
3208 * the original zero's written out previously to the journal and
3209 * awaiting writeback in the kernel's buffer cache.
3211 * So, if we see any bmap calls here on a modified, data-journaled file,
3212 * take extra steps to flush any blocks which might be in the cache.
3214 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3216 struct inode *inode = mapping->host;
3221 inode_lock_shared(inode);
3223 * We can get here for an inline file via the FIBMAP ioctl
3225 if (ext4_has_inline_data(inode))
3228 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3229 test_opt(inode->i_sb, DELALLOC)) {
3231 * With delalloc we want to sync the file
3232 * so that we can make sure we allocate
3235 filemap_write_and_wait(mapping);
3238 if (EXT4_JOURNAL(inode) &&
3239 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3241 * This is a REALLY heavyweight approach, but the use of
3242 * bmap on dirty files is expected to be extremely rare:
3243 * only if we run lilo or swapon on a freshly made file
3244 * do we expect this to happen.
3246 * (bmap requires CAP_SYS_RAWIO so this does not
3247 * represent an unprivileged user DOS attack --- we'd be
3248 * in trouble if mortal users could trigger this path at
3251 * NB. EXT4_STATE_JDATA is not set on files other than
3252 * regular files. If somebody wants to bmap a directory
3253 * or symlink and gets confused because the buffer
3254 * hasn't yet been flushed to disk, they deserve
3255 * everything they get.
3258 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3259 journal = EXT4_JOURNAL(inode);
3260 jbd2_journal_lock_updates(journal);
3261 err = jbd2_journal_flush(journal);
3262 jbd2_journal_unlock_updates(journal);
3268 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3271 inode_unlock_shared(inode);
3275 static int ext4_readpage(struct file *file, struct page *page)
3278 struct inode *inode = page->mapping->host;
3280 trace_ext4_readpage(page);
3282 if (ext4_has_inline_data(inode))
3283 ret = ext4_readpage_inline(inode, page);
3286 return ext4_mpage_readpages(inode, NULL, page);
3291 static void ext4_readahead(struct readahead_control *rac)
3293 struct inode *inode = rac->mapping->host;
3295 /* If the file has inline data, no need to do readahead. */
3296 if (ext4_has_inline_data(inode))
3299 ext4_mpage_readpages(inode, rac, NULL);
3302 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3303 unsigned int length)
3305 trace_ext4_invalidatepage(page, offset, length);
3307 /* No journalling happens on data buffers when this function is used */
3308 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3310 block_invalidatepage(page, offset, length);
3313 static int __ext4_journalled_invalidatepage(struct page *page,
3314 unsigned int offset,
3315 unsigned int length)
3317 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3319 trace_ext4_journalled_invalidatepage(page, offset, length);
3322 * If it's a full truncate we just forget about the pending dirtying
3324 if (offset == 0 && length == PAGE_SIZE)
3325 ClearPageChecked(page);
3327 return jbd2_journal_invalidatepage(journal, page, offset, length);
3330 /* Wrapper for aops... */
3331 static void ext4_journalled_invalidatepage(struct page *page,
3332 unsigned int offset,
3333 unsigned int length)
3335 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3338 static int ext4_releasepage(struct page *page, gfp_t wait)
3340 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3342 trace_ext4_releasepage(page);
3344 /* Page has dirty journalled data -> cannot release */
3345 if (PageChecked(page))
3348 return jbd2_journal_try_to_free_buffers(journal, page);
3350 return try_to_free_buffers(page);
3353 static bool ext4_inode_datasync_dirty(struct inode *inode)
3355 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3358 if (jbd2_transaction_committed(journal,
3359 EXT4_I(inode)->i_datasync_tid))
3361 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3362 return !list_empty(&EXT4_I(inode)->i_fc_list);
3366 /* Any metadata buffers to write? */
3367 if (!list_empty(&inode->i_mapping->private_list))
3369 return inode->i_state & I_DIRTY_DATASYNC;
3372 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3373 struct ext4_map_blocks *map, loff_t offset,
3376 u8 blkbits = inode->i_blkbits;
3379 * Writes that span EOF might trigger an I/O size update on completion,
3380 * so consider them to be dirty for the purpose of O_DSYNC, even if
3381 * there is no other metadata changes being made or are pending.
3384 if (ext4_inode_datasync_dirty(inode) ||
3385 offset + length > i_size_read(inode))
3386 iomap->flags |= IOMAP_F_DIRTY;
3388 if (map->m_flags & EXT4_MAP_NEW)
3389 iomap->flags |= IOMAP_F_NEW;
3391 iomap->bdev = inode->i_sb->s_bdev;
3392 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3393 iomap->offset = (u64) map->m_lblk << blkbits;
3394 iomap->length = (u64) map->m_len << blkbits;
3396 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3397 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3398 iomap->flags |= IOMAP_F_MERGED;
3401 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3402 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3403 * set. In order for any allocated unwritten extents to be converted
3404 * into written extents correctly within the ->end_io() handler, we
3405 * need to ensure that the iomap->type is set appropriately. Hence, the
3406 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3409 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3410 iomap->type = IOMAP_UNWRITTEN;
3411 iomap->addr = (u64) map->m_pblk << blkbits;
3412 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3413 iomap->type = IOMAP_MAPPED;
3414 iomap->addr = (u64) map->m_pblk << blkbits;
3416 iomap->type = IOMAP_HOLE;
3417 iomap->addr = IOMAP_NULL_ADDR;
3421 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3425 u8 blkbits = inode->i_blkbits;
3426 int ret, dio_credits, m_flags = 0, retries = 0;
3429 * Trim the mapping request to the maximum value that we can map at
3430 * once for direct I/O.
3432 if (map->m_len > DIO_MAX_BLOCKS)
3433 map->m_len = DIO_MAX_BLOCKS;
3434 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3438 * Either we allocate blocks and then don't get an unwritten extent, so
3439 * in that case we have reserved enough credits. Or, the blocks are
3440 * already allocated and unwritten. In that case, the extent conversion
3441 * fits into the credits as well.
3443 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3445 return PTR_ERR(handle);
3448 * DAX and direct I/O are the only two operations that are currently
3449 * supported with IOMAP_WRITE.
3451 WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3453 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3455 * We use i_size instead of i_disksize here because delalloc writeback
3456 * can complete at any point during the I/O and subsequently push the
3457 * i_disksize out to i_size. This could be beyond where direct I/O is
3458 * happening and thus expose allocated blocks to direct I/O reads.
3460 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3461 m_flags = EXT4_GET_BLOCKS_CREATE;
3462 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3463 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3465 ret = ext4_map_blocks(handle, inode, map, m_flags);
3468 * We cannot fill holes in indirect tree based inodes as that could
3469 * expose stale data in the case of a crash. Use the magic error code
3470 * to fallback to buffered I/O.
3472 if (!m_flags && !ret)
3475 ext4_journal_stop(handle);
3476 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3483 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3484 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3487 struct ext4_map_blocks map;
3488 u8 blkbits = inode->i_blkbits;
3490 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3493 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3497 * Calculate the first and last logical blocks respectively.
3499 map.m_lblk = offset >> blkbits;
3500 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3501 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3503 if (flags & IOMAP_WRITE) {
3505 * We check here if the blocks are already allocated, then we
3506 * don't need to start a journal txn and we can directly return
3507 * the mapping information. This could boost performance
3508 * especially in multi-threaded overwrite requests.
3510 if (offset + length <= i_size_read(inode)) {
3511 ret = ext4_map_blocks(NULL, inode, &map, 0);
3512 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3515 ret = ext4_iomap_alloc(inode, &map, flags);
3517 ret = ext4_map_blocks(NULL, inode, &map, 0);
3523 ext4_set_iomap(inode, iomap, &map, offset, length);
3528 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3529 loff_t length, unsigned flags, struct iomap *iomap,
3530 struct iomap *srcmap)
3535 * Even for writes we don't need to allocate blocks, so just pretend
3536 * we are reading to save overhead of starting a transaction.
3538 flags &= ~IOMAP_WRITE;
3539 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3540 WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
3544 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3545 ssize_t written, unsigned flags, struct iomap *iomap)
3548 * Check to see whether an error occurred while writing out the data to
3549 * the allocated blocks. If so, return the magic error code so that we
3550 * fallback to buffered I/O and attempt to complete the remainder of
3551 * the I/O. Any blocks that may have been allocated in preparation for
3552 * the direct I/O will be reused during buffered I/O.
3554 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3560 const struct iomap_ops ext4_iomap_ops = {
3561 .iomap_begin = ext4_iomap_begin,
3562 .iomap_end = ext4_iomap_end,
3565 const struct iomap_ops ext4_iomap_overwrite_ops = {
3566 .iomap_begin = ext4_iomap_overwrite_begin,
3567 .iomap_end = ext4_iomap_end,
3570 static bool ext4_iomap_is_delalloc(struct inode *inode,
3571 struct ext4_map_blocks *map)
3573 struct extent_status es;
3574 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3576 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3577 map->m_lblk, end, &es);
3579 if (!es.es_len || es.es_lblk > end)
3582 if (es.es_lblk > map->m_lblk) {
3583 map->m_len = es.es_lblk - map->m_lblk;
3587 offset = map->m_lblk - es.es_lblk;
3588 map->m_len = es.es_len - offset;
3593 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3594 loff_t length, unsigned int flags,
3595 struct iomap *iomap, struct iomap *srcmap)
3598 bool delalloc = false;
3599 struct ext4_map_blocks map;
3600 u8 blkbits = inode->i_blkbits;
3602 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3605 if (ext4_has_inline_data(inode)) {
3606 ret = ext4_inline_data_iomap(inode, iomap);
3607 if (ret != -EAGAIN) {
3608 if (ret == 0 && offset >= iomap->length)
3615 * Calculate the first and last logical block respectively.
3617 map.m_lblk = offset >> blkbits;
3618 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3619 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3622 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3623 * So handle it here itself instead of querying ext4_map_blocks().
3624 * Since ext4_map_blocks() will warn about it and will return
3627 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3628 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3630 if (offset >= sbi->s_bitmap_maxbytes) {
3636 ret = ext4_map_blocks(NULL, inode, &map, 0);
3640 delalloc = ext4_iomap_is_delalloc(inode, &map);
3643 ext4_set_iomap(inode, iomap, &map, offset, length);
3644 if (delalloc && iomap->type == IOMAP_HOLE)
3645 iomap->type = IOMAP_DELALLOC;
3650 const struct iomap_ops ext4_iomap_report_ops = {
3651 .iomap_begin = ext4_iomap_begin_report,
3655 * Pages can be marked dirty completely asynchronously from ext4's journalling
3656 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3657 * much here because ->set_page_dirty is called under VFS locks. The page is
3658 * not necessarily locked.
3660 * We cannot just dirty the page and leave attached buffers clean, because the
3661 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3662 * or jbddirty because all the journalling code will explode.
3664 * So what we do is to mark the page "pending dirty" and next time writepage
3665 * is called, propagate that into the buffers appropriately.
3667 static int ext4_journalled_set_page_dirty(struct page *page)
3669 SetPageChecked(page);
3670 return __set_page_dirty_nobuffers(page);
3673 static int ext4_set_page_dirty(struct page *page)
3675 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3676 WARN_ON_ONCE(!page_has_buffers(page));
3677 return __set_page_dirty_buffers(page);
3680 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3681 struct file *file, sector_t *span)
3683 return iomap_swapfile_activate(sis, file, span,
3684 &ext4_iomap_report_ops);
3687 static const struct address_space_operations ext4_aops = {
3688 .readpage = ext4_readpage,
3689 .readahead = ext4_readahead,
3690 .writepage = ext4_writepage,
3691 .writepages = ext4_writepages,
3692 .write_begin = ext4_write_begin,
3693 .write_end = ext4_write_end,
3694 .set_page_dirty = ext4_set_page_dirty,
3696 .invalidatepage = ext4_invalidatepage,
3697 .releasepage = ext4_releasepage,
3698 .direct_IO = noop_direct_IO,
3699 .migratepage = buffer_migrate_page,
3700 .is_partially_uptodate = block_is_partially_uptodate,
3701 .error_remove_page = generic_error_remove_page,
3702 .swap_activate = ext4_iomap_swap_activate,
3705 static const struct address_space_operations ext4_journalled_aops = {
3706 .readpage = ext4_readpage,
3707 .readahead = ext4_readahead,
3708 .writepage = ext4_writepage,
3709 .writepages = ext4_writepages,
3710 .write_begin = ext4_write_begin,
3711 .write_end = ext4_journalled_write_end,
3712 .set_page_dirty = ext4_journalled_set_page_dirty,
3714 .invalidatepage = ext4_journalled_invalidatepage,
3715 .releasepage = ext4_releasepage,
3716 .direct_IO = noop_direct_IO,
3717 .is_partially_uptodate = block_is_partially_uptodate,
3718 .error_remove_page = generic_error_remove_page,
3719 .swap_activate = ext4_iomap_swap_activate,
3722 static const struct address_space_operations ext4_da_aops = {
3723 .readpage = ext4_readpage,
3724 .readahead = ext4_readahead,
3725 .writepage = ext4_writepage,
3726 .writepages = ext4_writepages,
3727 .write_begin = ext4_da_write_begin,
3728 .write_end = ext4_da_write_end,
3729 .set_page_dirty = ext4_set_page_dirty,
3731 .invalidatepage = ext4_invalidatepage,
3732 .releasepage = ext4_releasepage,
3733 .direct_IO = noop_direct_IO,
3734 .migratepage = buffer_migrate_page,
3735 .is_partially_uptodate = block_is_partially_uptodate,
3736 .error_remove_page = generic_error_remove_page,
3737 .swap_activate = ext4_iomap_swap_activate,
3740 static const struct address_space_operations ext4_dax_aops = {
3741 .writepages = ext4_dax_writepages,
3742 .direct_IO = noop_direct_IO,
3743 .set_page_dirty = noop_set_page_dirty,
3745 .invalidatepage = noop_invalidatepage,
3746 .swap_activate = ext4_iomap_swap_activate,
3749 void ext4_set_aops(struct inode *inode)
3751 switch (ext4_inode_journal_mode(inode)) {
3752 case EXT4_INODE_ORDERED_DATA_MODE:
3753 case EXT4_INODE_WRITEBACK_DATA_MODE:
3755 case EXT4_INODE_JOURNAL_DATA_MODE:
3756 inode->i_mapping->a_ops = &ext4_journalled_aops;
3762 inode->i_mapping->a_ops = &ext4_dax_aops;
3763 else if (test_opt(inode->i_sb, DELALLOC))
3764 inode->i_mapping->a_ops = &ext4_da_aops;
3766 inode->i_mapping->a_ops = &ext4_aops;
3769 static int __ext4_block_zero_page_range(handle_t *handle,
3770 struct address_space *mapping, loff_t from, loff_t length)
3772 ext4_fsblk_t index = from >> PAGE_SHIFT;
3773 unsigned offset = from & (PAGE_SIZE-1);
3774 unsigned blocksize, pos;
3776 struct inode *inode = mapping->host;
3777 struct buffer_head *bh;
3781 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3782 mapping_gfp_constraint(mapping, ~__GFP_FS));
3786 blocksize = inode->i_sb->s_blocksize;
3788 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3790 if (!page_has_buffers(page))
3791 create_empty_buffers(page, blocksize, 0);
3793 /* Find the buffer that contains "offset" */
3794 bh = page_buffers(page);
3796 while (offset >= pos) {
3797 bh = bh->b_this_page;
3801 if (buffer_freed(bh)) {
3802 BUFFER_TRACE(bh, "freed: skip");
3805 if (!buffer_mapped(bh)) {
3806 BUFFER_TRACE(bh, "unmapped");
3807 ext4_get_block(inode, iblock, bh, 0);
3808 /* unmapped? It's a hole - nothing to do */
3809 if (!buffer_mapped(bh)) {
3810 BUFFER_TRACE(bh, "still unmapped");
3815 /* Ok, it's mapped. Make sure it's up-to-date */
3816 if (PageUptodate(page))
3817 set_buffer_uptodate(bh);
3819 if (!buffer_uptodate(bh)) {
3820 err = ext4_read_bh_lock(bh, 0, true);
3823 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3824 /* We expect the key to be set. */
3825 BUG_ON(!fscrypt_has_encryption_key(inode));
3826 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3829 clear_buffer_uptodate(bh);
3834 if (ext4_should_journal_data(inode)) {
3835 BUFFER_TRACE(bh, "get write access");
3836 err = ext4_journal_get_write_access(handle, bh);
3840 zero_user(page, offset, length);
3841 BUFFER_TRACE(bh, "zeroed end of block");
3843 if (ext4_should_journal_data(inode)) {
3844 err = ext4_handle_dirty_metadata(handle, inode, bh);
3847 mark_buffer_dirty(bh);
3848 if (ext4_should_order_data(inode))
3849 err = ext4_jbd2_inode_add_write(handle, inode, from,
3860 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3861 * starting from file offset 'from'. The range to be zero'd must
3862 * be contained with in one block. If the specified range exceeds
3863 * the end of the block it will be shortened to end of the block
3864 * that corresponds to 'from'
3866 static int ext4_block_zero_page_range(handle_t *handle,
3867 struct address_space *mapping, loff_t from, loff_t length)
3869 struct inode *inode = mapping->host;
3870 unsigned offset = from & (PAGE_SIZE-1);
3871 unsigned blocksize = inode->i_sb->s_blocksize;
3872 unsigned max = blocksize - (offset & (blocksize - 1));
3875 * correct length if it does not fall between
3876 * 'from' and the end of the block
3878 if (length > max || length < 0)
3881 if (IS_DAX(inode)) {
3882 return iomap_zero_range(inode, from, length, NULL,
3885 return __ext4_block_zero_page_range(handle, mapping, from, length);
3889 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3890 * up to the end of the block which corresponds to `from'.
3891 * This required during truncate. We need to physically zero the tail end
3892 * of that block so it doesn't yield old data if the file is later grown.
3894 static int ext4_block_truncate_page(handle_t *handle,
3895 struct address_space *mapping, loff_t from)
3897 unsigned offset = from & (PAGE_SIZE-1);
3900 struct inode *inode = mapping->host;
3902 /* If we are processing an encrypted inode during orphan list handling */
3903 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3906 blocksize = inode->i_sb->s_blocksize;
3907 length = blocksize - (offset & (blocksize - 1));
3909 return ext4_block_zero_page_range(handle, mapping, from, length);
3912 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3913 loff_t lstart, loff_t length)
3915 struct super_block *sb = inode->i_sb;
3916 struct address_space *mapping = inode->i_mapping;
3917 unsigned partial_start, partial_end;
3918 ext4_fsblk_t start, end;
3919 loff_t byte_end = (lstart + length - 1);
3922 partial_start = lstart & (sb->s_blocksize - 1);
3923 partial_end = byte_end & (sb->s_blocksize - 1);
3925 start = lstart >> sb->s_blocksize_bits;
3926 end = byte_end >> sb->s_blocksize_bits;
3928 /* Handle partial zero within the single block */
3930 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3931 err = ext4_block_zero_page_range(handle, mapping,
3935 /* Handle partial zero out on the start of the range */
3936 if (partial_start) {
3937 err = ext4_block_zero_page_range(handle, mapping,
3938 lstart, sb->s_blocksize);
3942 /* Handle partial zero out on the end of the range */
3943 if (partial_end != sb->s_blocksize - 1)
3944 err = ext4_block_zero_page_range(handle, mapping,
3945 byte_end - partial_end,
3950 int ext4_can_truncate(struct inode *inode)
3952 if (S_ISREG(inode->i_mode))
3954 if (S_ISDIR(inode->i_mode))
3956 if (S_ISLNK(inode->i_mode))
3957 return !ext4_inode_is_fast_symlink(inode);
3962 * We have to make sure i_disksize gets properly updated before we truncate
3963 * page cache due to hole punching or zero range. Otherwise i_disksize update
3964 * can get lost as it may have been postponed to submission of writeback but
3965 * that will never happen after we truncate page cache.
3967 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3973 loff_t size = i_size_read(inode);
3975 WARN_ON(!inode_is_locked(inode));
3976 if (offset > size || offset + len < size)
3979 if (EXT4_I(inode)->i_disksize >= size)
3982 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3984 return PTR_ERR(handle);
3985 ext4_update_i_disksize(inode, size);
3986 ret = ext4_mark_inode_dirty(handle, inode);
3987 ext4_journal_stop(handle);
3992 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
3994 up_write(&ei->i_mmap_sem);
3996 down_write(&ei->i_mmap_sem);
3999 int ext4_break_layouts(struct inode *inode)
4001 struct ext4_inode_info *ei = EXT4_I(inode);
4005 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4009 page = dax_layout_busy_page(inode->i_mapping);
4013 error = ___wait_var_event(&page->_refcount,
4014 atomic_read(&page->_refcount) == 1,
4015 TASK_INTERRUPTIBLE, 0, 0,
4016 ext4_wait_dax_page(ei));
4017 } while (error == 0);
4023 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4024 * associated with the given offset and length
4026 * @inode: File inode
4027 * @offset: The offset where the hole will begin
4028 * @len: The length of the hole
4030 * Returns: 0 on success or negative on failure
4033 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4035 struct inode *inode = file_inode(file);
4036 struct super_block *sb = inode->i_sb;
4037 ext4_lblk_t first_block, stop_block;
4038 struct address_space *mapping = inode->i_mapping;
4039 loff_t first_block_offset, last_block_offset, max_length;
4040 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4042 unsigned int credits;
4043 int ret = 0, ret2 = 0;
4045 trace_ext4_punch_hole(inode, offset, length, 0);
4048 * Write out all dirty pages to avoid race conditions
4049 * Then release them.
4051 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4052 ret = filemap_write_and_wait_range(mapping, offset,
4053 offset + length - 1);
4060 /* No need to punch hole beyond i_size */
4061 if (offset >= inode->i_size)
4065 * If the hole extends beyond i_size, set the hole
4066 * to end after the page that contains i_size
4068 if (offset + length > inode->i_size) {
4069 length = inode->i_size +
4070 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4075 * For punch hole the length + offset needs to be within one block
4076 * before last range. Adjust the length if it goes beyond that limit.
4078 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4079 if (offset + length > max_length)
4080 length = max_length - offset;
4082 if (offset & (sb->s_blocksize - 1) ||
4083 (offset + length) & (sb->s_blocksize - 1)) {
4085 * Attach jinode to inode for jbd2 if we do any zeroing of
4088 ret = ext4_inode_attach_jinode(inode);
4094 /* Wait all existing dio workers, newcomers will block on i_mutex */
4095 inode_dio_wait(inode);
4097 ret = file_modified(file);
4102 * Prevent page faults from reinstantiating pages we have released from
4105 down_write(&EXT4_I(inode)->i_mmap_sem);
4107 ret = ext4_break_layouts(inode);
4111 first_block_offset = round_up(offset, sb->s_blocksize);
4112 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4114 /* Now release the pages and zero block aligned part of pages*/
4115 if (last_block_offset > first_block_offset) {
4116 ret = ext4_update_disksize_before_punch(inode, offset, length);
4119 truncate_pagecache_range(inode, first_block_offset,
4123 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4124 credits = ext4_writepage_trans_blocks(inode);
4126 credits = ext4_blocks_for_truncate(inode);
4127 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4128 if (IS_ERR(handle)) {
4129 ret = PTR_ERR(handle);
4130 ext4_std_error(sb, ret);
4134 ret = ext4_zero_partial_blocks(handle, inode, offset,
4139 first_block = (offset + sb->s_blocksize - 1) >>
4140 EXT4_BLOCK_SIZE_BITS(sb);
4141 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4143 /* If there are blocks to remove, do it */
4144 if (stop_block > first_block) {
4146 down_write(&EXT4_I(inode)->i_data_sem);
4147 ext4_discard_preallocations(inode, 0);
4149 ret = ext4_es_remove_extent(inode, first_block,
4150 stop_block - first_block);
4152 up_write(&EXT4_I(inode)->i_data_sem);
4156 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4157 ret = ext4_ext_remove_space(inode, first_block,
4160 ret = ext4_ind_remove_space(handle, inode, first_block,
4163 up_write(&EXT4_I(inode)->i_data_sem);
4165 ext4_fc_track_range(handle, inode, first_block, stop_block);
4167 ext4_handle_sync(handle);
4169 inode->i_mtime = inode->i_ctime = current_time(inode);
4170 ret2 = ext4_mark_inode_dirty(handle, inode);
4174 ext4_update_inode_fsync_trans(handle, inode, 1);
4176 ext4_journal_stop(handle);
4178 up_write(&EXT4_I(inode)->i_mmap_sem);
4180 inode_unlock(inode);
4184 int ext4_inode_attach_jinode(struct inode *inode)
4186 struct ext4_inode_info *ei = EXT4_I(inode);
4187 struct jbd2_inode *jinode;
4189 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4192 jinode = jbd2_alloc_inode(GFP_KERNEL);
4193 spin_lock(&inode->i_lock);
4196 spin_unlock(&inode->i_lock);
4199 ei->jinode = jinode;
4200 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4203 spin_unlock(&inode->i_lock);
4204 if (unlikely(jinode != NULL))
4205 jbd2_free_inode(jinode);
4212 * We block out ext4_get_block() block instantiations across the entire
4213 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4214 * simultaneously on behalf of the same inode.
4216 * As we work through the truncate and commit bits of it to the journal there
4217 * is one core, guiding principle: the file's tree must always be consistent on
4218 * disk. We must be able to restart the truncate after a crash.
4220 * The file's tree may be transiently inconsistent in memory (although it
4221 * probably isn't), but whenever we close off and commit a journal transaction,
4222 * the contents of (the filesystem + the journal) must be consistent and
4223 * restartable. It's pretty simple, really: bottom up, right to left (although
4224 * left-to-right works OK too).
4226 * Note that at recovery time, journal replay occurs *before* the restart of
4227 * truncate against the orphan inode list.
4229 * The committed inode has the new, desired i_size (which is the same as
4230 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4231 * that this inode's truncate did not complete and it will again call
4232 * ext4_truncate() to have another go. So there will be instantiated blocks
4233 * to the right of the truncation point in a crashed ext4 filesystem. But
4234 * that's fine - as long as they are linked from the inode, the post-crash
4235 * ext4_truncate() run will find them and release them.
4237 int ext4_truncate(struct inode *inode)
4239 struct ext4_inode_info *ei = EXT4_I(inode);
4240 unsigned int credits;
4243 struct address_space *mapping = inode->i_mapping;
4246 * There is a possibility that we're either freeing the inode
4247 * or it's a completely new inode. In those cases we might not
4248 * have i_mutex locked because it's not necessary.
4250 if (!(inode->i_state & (I_NEW|I_FREEING)))
4251 WARN_ON(!inode_is_locked(inode));
4252 trace_ext4_truncate_enter(inode);
4254 if (!ext4_can_truncate(inode))
4257 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4258 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4260 if (ext4_has_inline_data(inode)) {
4263 err = ext4_inline_data_truncate(inode, &has_inline);
4264 if (err || has_inline)
4268 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4269 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4270 err = ext4_inode_attach_jinode(inode);
4275 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4276 credits = ext4_writepage_trans_blocks(inode);
4278 credits = ext4_blocks_for_truncate(inode);
4280 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4281 if (IS_ERR(handle)) {
4282 err = PTR_ERR(handle);
4286 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4287 ext4_block_truncate_page(handle, mapping, inode->i_size);
4290 * We add the inode to the orphan list, so that if this
4291 * truncate spans multiple transactions, and we crash, we will
4292 * resume the truncate when the filesystem recovers. It also
4293 * marks the inode dirty, to catch the new size.
4295 * Implication: the file must always be in a sane, consistent
4296 * truncatable state while each transaction commits.
4298 err = ext4_orphan_add(handle, inode);
4302 down_write(&EXT4_I(inode)->i_data_sem);
4304 ext4_discard_preallocations(inode, 0);
4306 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4307 err = ext4_ext_truncate(handle, inode);
4309 ext4_ind_truncate(handle, inode);
4311 up_write(&ei->i_data_sem);
4316 ext4_handle_sync(handle);
4320 * If this was a simple ftruncate() and the file will remain alive,
4321 * then we need to clear up the orphan record which we created above.
4322 * However, if this was a real unlink then we were called by
4323 * ext4_evict_inode(), and we allow that function to clean up the
4324 * orphan info for us.
4327 ext4_orphan_del(handle, inode);
4329 inode->i_mtime = inode->i_ctime = current_time(inode);
4330 err2 = ext4_mark_inode_dirty(handle, inode);
4331 if (unlikely(err2 && !err))
4333 ext4_journal_stop(handle);
4336 trace_ext4_truncate_exit(inode);
4341 * ext4_get_inode_loc returns with an extra refcount against the inode's
4342 * underlying buffer_head on success. If 'in_mem' is true, we have all
4343 * data in memory that is needed to recreate the on-disk version of this
4346 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4347 struct ext4_iloc *iloc, int in_mem,
4348 ext4_fsblk_t *ret_block)
4350 struct ext4_group_desc *gdp;
4351 struct buffer_head *bh;
4353 struct blk_plug plug;
4354 int inodes_per_block, inode_offset;
4357 if (ino < EXT4_ROOT_INO ||
4358 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4359 return -EFSCORRUPTED;
4361 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4362 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4367 * Figure out the offset within the block group inode table
4369 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4370 inode_offset = ((ino - 1) %
4371 EXT4_INODES_PER_GROUP(sb));
4372 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4374 block = ext4_inode_table(sb, gdp);
4375 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4376 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4377 ext4_error(sb, "Invalid inode table block %llu in "
4378 "block_group %u", block, iloc->block_group);
4379 return -EFSCORRUPTED;
4381 block += (inode_offset / inodes_per_block);
4383 bh = sb_getblk(sb, block);
4386 if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO))
4388 if (!buffer_uptodate(bh)) {
4391 if (ext4_buffer_uptodate(bh)) {
4392 /* someone brought it uptodate while we waited */
4398 * If we have all information of the inode in memory and this
4399 * is the only valid inode in the block, we need not read the
4403 struct buffer_head *bitmap_bh;
4406 start = inode_offset & ~(inodes_per_block - 1);
4408 /* Is the inode bitmap in cache? */
4409 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4410 if (unlikely(!bitmap_bh))
4414 * If the inode bitmap isn't in cache then the
4415 * optimisation may end up performing two reads instead
4416 * of one, so skip it.
4418 if (!buffer_uptodate(bitmap_bh)) {
4422 for (i = start; i < start + inodes_per_block; i++) {
4423 if (i == inode_offset)
4425 if (ext4_test_bit(i, bitmap_bh->b_data))
4429 if (i == start + inodes_per_block) {
4430 /* all other inodes are free, so skip I/O */
4431 memset(bh->b_data, 0, bh->b_size);
4432 set_buffer_uptodate(bh);
4440 * If we need to do any I/O, try to pre-readahead extra
4441 * blocks from the inode table.
4443 blk_start_plug(&plug);
4444 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4445 ext4_fsblk_t b, end, table;
4447 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4449 table = ext4_inode_table(sb, gdp);
4450 /* s_inode_readahead_blks is always a power of 2 */
4451 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4455 num = EXT4_INODES_PER_GROUP(sb);
4456 if (ext4_has_group_desc_csum(sb))
4457 num -= ext4_itable_unused_count(sb, gdp);
4458 table += num / inodes_per_block;
4462 ext4_sb_breadahead_unmovable(sb, b++);
4466 * There are other valid inodes in the buffer, this inode
4467 * has in-inode xattrs, or we don't have this inode in memory.
4468 * Read the block from disk.
4470 trace_ext4_load_inode(sb, ino);
4471 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4472 blk_finish_plug(&plug);
4474 if (!buffer_uptodate(bh)) {
4487 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4488 struct ext4_iloc *iloc)
4490 ext4_fsblk_t err_blk = 0;
4493 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, 0,
4497 ext4_error_inode_block(inode, err_blk, EIO,
4498 "unable to read itable block");
4503 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4505 ext4_fsblk_t err_blk = 0;
4508 /* We have all inode data except xattrs in memory here. */
4509 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc,
4510 !ext4_test_inode_state(inode, EXT4_STATE_XATTR), &err_blk);
4513 ext4_error_inode_block(inode, err_blk, EIO,
4514 "unable to read itable block");
4520 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4521 struct ext4_iloc *iloc)
4523 return __ext4_get_inode_loc(sb, ino, iloc, 0, NULL);
4526 static bool ext4_should_enable_dax(struct inode *inode)
4528 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4530 if (test_opt2(inode->i_sb, DAX_NEVER))
4532 if (!S_ISREG(inode->i_mode))
4534 if (ext4_should_journal_data(inode))
4536 if (ext4_has_inline_data(inode))
4538 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4540 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4542 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4544 if (test_opt(inode->i_sb, DAX_ALWAYS))
4547 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4550 void ext4_set_inode_flags(struct inode *inode, bool init)
4552 unsigned int flags = EXT4_I(inode)->i_flags;
4553 unsigned int new_fl = 0;
4555 WARN_ON_ONCE(IS_DAX(inode) && init);
4557 if (flags & EXT4_SYNC_FL)
4559 if (flags & EXT4_APPEND_FL)
4561 if (flags & EXT4_IMMUTABLE_FL)
4562 new_fl |= S_IMMUTABLE;
4563 if (flags & EXT4_NOATIME_FL)
4564 new_fl |= S_NOATIME;
4565 if (flags & EXT4_DIRSYNC_FL)
4566 new_fl |= S_DIRSYNC;
4568 /* Because of the way inode_set_flags() works we must preserve S_DAX
4569 * here if already set. */
4570 new_fl |= (inode->i_flags & S_DAX);
4571 if (init && ext4_should_enable_dax(inode))
4574 if (flags & EXT4_ENCRYPT_FL)
4575 new_fl |= S_ENCRYPTED;
4576 if (flags & EXT4_CASEFOLD_FL)
4577 new_fl |= S_CASEFOLD;
4578 if (flags & EXT4_VERITY_FL)
4580 inode_set_flags(inode, new_fl,
4581 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4582 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4585 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4586 struct ext4_inode_info *ei)
4589 struct inode *inode = &(ei->vfs_inode);
4590 struct super_block *sb = inode->i_sb;
4592 if (ext4_has_feature_huge_file(sb)) {
4593 /* we are using combined 48 bit field */
4594 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4595 le32_to_cpu(raw_inode->i_blocks_lo);
4596 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4597 /* i_blocks represent file system block size */
4598 return i_blocks << (inode->i_blkbits - 9);
4603 return le32_to_cpu(raw_inode->i_blocks_lo);
4607 static inline int ext4_iget_extra_inode(struct inode *inode,
4608 struct ext4_inode *raw_inode,
4609 struct ext4_inode_info *ei)
4611 __le32 *magic = (void *)raw_inode +
4612 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4614 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4615 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4618 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4619 err = ext4_find_inline_data_nolock(inode);
4620 if (!err && ext4_has_inline_data(inode))
4621 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4624 EXT4_I(inode)->i_inline_off = 0;
4628 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4630 if (!ext4_has_feature_project(inode->i_sb))
4632 *projid = EXT4_I(inode)->i_projid;
4637 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4638 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4641 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4643 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4644 inode_set_iversion_raw(inode, val);
4646 inode_set_iversion_queried(inode, val);
4648 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4650 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4651 return inode_peek_iversion_raw(inode);
4653 return inode_peek_iversion(inode);
4656 static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags)
4659 if (flags & EXT4_IGET_EA_INODE) {
4660 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4661 return "missing EA_INODE flag";
4662 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4663 EXT4_I(inode)->i_file_acl)
4664 return "ea_inode with extended attributes";
4666 if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4667 return "unexpected EA_INODE flag";
4669 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD))
4670 return "unexpected bad inode w/o EXT4_IGET_BAD";
4674 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4675 ext4_iget_flags flags, const char *function,
4678 struct ext4_iloc iloc;
4679 struct ext4_inode *raw_inode;
4680 struct ext4_inode_info *ei;
4681 struct inode *inode;
4682 const char *err_str;
4683 journal_t *journal = EXT4_SB(sb)->s_journal;
4691 if ((!(flags & EXT4_IGET_SPECIAL) &&
4692 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4693 (ino < EXT4_ROOT_INO) ||
4694 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4695 if (flags & EXT4_IGET_HANDLE)
4696 return ERR_PTR(-ESTALE);
4697 __ext4_error(sb, function, line, EFSCORRUPTED, 0,
4698 "inode #%lu: comm %s: iget: illegal inode #",
4699 ino, current->comm);
4700 return ERR_PTR(-EFSCORRUPTED);
4703 inode = iget_locked(sb, ino);
4705 return ERR_PTR(-ENOMEM);
4706 if (!(inode->i_state & I_NEW)) {
4707 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4708 ext4_error_inode(inode, function, line, 0, err_str);
4710 return ERR_PTR(-EFSCORRUPTED);
4718 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4721 raw_inode = ext4_raw_inode(&iloc);
4723 if ((flags & EXT4_IGET_HANDLE) &&
4724 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4729 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4730 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4731 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4732 EXT4_INODE_SIZE(inode->i_sb) ||
4733 (ei->i_extra_isize & 3)) {
4734 ext4_error_inode(inode, function, line, 0,
4735 "iget: bad extra_isize %u "
4738 EXT4_INODE_SIZE(inode->i_sb));
4739 ret = -EFSCORRUPTED;
4743 ei->i_extra_isize = 0;
4745 /* Precompute checksum seed for inode metadata */
4746 if (ext4_has_metadata_csum(sb)) {
4747 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4749 __le32 inum = cpu_to_le32(inode->i_ino);
4750 __le32 gen = raw_inode->i_generation;
4751 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4753 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4757 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4758 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4759 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4760 ext4_error_inode_err(inode, function, line, 0,
4761 EFSBADCRC, "iget: checksum invalid");
4766 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4767 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4768 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4769 if (ext4_has_feature_project(sb) &&
4770 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4771 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4772 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4774 i_projid = EXT4_DEF_PROJID;
4776 if (!(test_opt(inode->i_sb, NO_UID32))) {
4777 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4778 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4780 i_uid_write(inode, i_uid);
4781 i_gid_write(inode, i_gid);
4782 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4783 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4785 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4786 ei->i_inline_off = 0;
4787 ei->i_dir_start_lookup = 0;
4788 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4789 /* We now have enough fields to check if the inode was active or not.
4790 * This is needed because nfsd might try to access dead inodes
4791 * the test is that same one that e2fsck uses
4792 * NeilBrown 1999oct15
4794 if (inode->i_nlink == 0) {
4795 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4796 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4797 ino != EXT4_BOOT_LOADER_INO) {
4798 /* this inode is deleted or unallocated */
4799 if (flags & EXT4_IGET_SPECIAL) {
4800 ext4_error_inode(inode, function, line, 0,
4801 "iget: special inode unallocated");
4802 ret = -EFSCORRUPTED;
4807 /* The only unlinked inodes we let through here have
4808 * valid i_mode and are being read by the orphan
4809 * recovery code: that's fine, we're about to complete
4810 * the process of deleting those.
4811 * OR it is the EXT4_BOOT_LOADER_INO which is
4812 * not initialized on a new filesystem. */
4814 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4815 ext4_set_inode_flags(inode, true);
4816 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4817 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4818 if (ext4_has_feature_64bit(sb))
4820 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4821 inode->i_size = ext4_isize(sb, raw_inode);
4822 if ((size = i_size_read(inode)) < 0) {
4823 ext4_error_inode(inode, function, line, 0,
4824 "iget: bad i_size value: %lld", size);
4825 ret = -EFSCORRUPTED;
4829 * If dir_index is not enabled but there's dir with INDEX flag set,
4830 * we'd normally treat htree data as empty space. But with metadata
4831 * checksumming that corrupts checksums so forbid that.
4833 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4834 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4835 ext4_error_inode(inode, function, line, 0,
4836 "iget: Dir with htree data on filesystem without dir_index feature.");
4837 ret = -EFSCORRUPTED;
4840 ei->i_disksize = inode->i_size;
4842 ei->i_reserved_quota = 0;
4844 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4845 ei->i_block_group = iloc.block_group;
4846 ei->i_last_alloc_group = ~0;
4848 * NOTE! The in-memory inode i_data array is in little-endian order
4849 * even on big-endian machines: we do NOT byteswap the block numbers!
4851 for (block = 0; block < EXT4_N_BLOCKS; block++)
4852 ei->i_data[block] = raw_inode->i_block[block];
4853 INIT_LIST_HEAD(&ei->i_orphan);
4854 ext4_fc_init_inode(&ei->vfs_inode);
4857 * Set transaction id's of transactions that have to be committed
4858 * to finish f[data]sync. We set them to currently running transaction
4859 * as we cannot be sure that the inode or some of its metadata isn't
4860 * part of the transaction - the inode could have been reclaimed and
4861 * now it is reread from disk.
4864 transaction_t *transaction;
4867 read_lock(&journal->j_state_lock);
4868 if (journal->j_running_transaction)
4869 transaction = journal->j_running_transaction;
4871 transaction = journal->j_committing_transaction;
4873 tid = transaction->t_tid;
4875 tid = journal->j_commit_sequence;
4876 read_unlock(&journal->j_state_lock);
4877 ei->i_sync_tid = tid;
4878 ei->i_datasync_tid = tid;
4881 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4882 if (ei->i_extra_isize == 0) {
4883 /* The extra space is currently unused. Use it. */
4884 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4885 ei->i_extra_isize = sizeof(struct ext4_inode) -
4886 EXT4_GOOD_OLD_INODE_SIZE;
4888 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4894 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4895 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4896 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4897 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4899 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4900 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4902 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4903 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4905 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4907 ext4_inode_set_iversion_queried(inode, ivers);
4911 if (ei->i_file_acl &&
4912 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4913 ext4_error_inode(inode, function, line, 0,
4914 "iget: bad extended attribute block %llu",
4916 ret = -EFSCORRUPTED;
4918 } else if (!ext4_has_inline_data(inode)) {
4919 /* validate the block references in the inode */
4920 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4921 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4922 (S_ISLNK(inode->i_mode) &&
4923 !ext4_inode_is_fast_symlink(inode)))) {
4924 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4925 ret = ext4_ext_check_inode(inode);
4927 ret = ext4_ind_check_inode(inode);
4933 if (S_ISREG(inode->i_mode)) {
4934 inode->i_op = &ext4_file_inode_operations;
4935 inode->i_fop = &ext4_file_operations;
4936 ext4_set_aops(inode);
4937 } else if (S_ISDIR(inode->i_mode)) {
4938 inode->i_op = &ext4_dir_inode_operations;
4939 inode->i_fop = &ext4_dir_operations;
4940 } else if (S_ISLNK(inode->i_mode)) {
4941 /* VFS does not allow setting these so must be corruption */
4942 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4943 ext4_error_inode(inode, function, line, 0,
4944 "iget: immutable or append flags "
4945 "not allowed on symlinks");
4946 ret = -EFSCORRUPTED;
4949 if (IS_ENCRYPTED(inode)) {
4950 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4951 ext4_set_aops(inode);
4952 } else if (ext4_inode_is_fast_symlink(inode)) {
4953 inode->i_link = (char *)ei->i_data;
4954 inode->i_op = &ext4_fast_symlink_inode_operations;
4955 nd_terminate_link(ei->i_data, inode->i_size,
4956 sizeof(ei->i_data) - 1);
4958 inode->i_op = &ext4_symlink_inode_operations;
4959 ext4_set_aops(inode);
4961 inode_nohighmem(inode);
4962 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4963 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4964 inode->i_op = &ext4_special_inode_operations;
4965 if (raw_inode->i_block[0])
4966 init_special_inode(inode, inode->i_mode,
4967 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4969 init_special_inode(inode, inode->i_mode,
4970 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4971 } else if (ino == EXT4_BOOT_LOADER_INO) {
4972 make_bad_inode(inode);
4974 ret = -EFSCORRUPTED;
4975 ext4_error_inode(inode, function, line, 0,
4976 "iget: bogus i_mode (%o)", inode->i_mode);
4979 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4980 ext4_error_inode(inode, function, line, 0,
4981 "casefold flag without casefold feature");
4982 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4983 ext4_error_inode(inode, function, line, 0, err_str);
4984 ret = -EFSCORRUPTED;
4989 unlock_new_inode(inode);
4995 return ERR_PTR(ret);
4998 static int ext4_inode_blocks_set(handle_t *handle,
4999 struct ext4_inode *raw_inode,
5000 struct ext4_inode_info *ei)
5002 struct inode *inode = &(ei->vfs_inode);
5003 u64 i_blocks = READ_ONCE(inode->i_blocks);
5004 struct super_block *sb = inode->i_sb;
5006 if (i_blocks <= ~0U) {
5008 * i_blocks can be represented in a 32 bit variable
5009 * as multiple of 512 bytes
5011 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5012 raw_inode->i_blocks_high = 0;
5013 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5016 if (!ext4_has_feature_huge_file(sb))
5019 if (i_blocks <= 0xffffffffffffULL) {
5021 * i_blocks can be represented in a 48 bit variable
5022 * as multiple of 512 bytes
5024 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5025 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5026 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5028 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5029 /* i_block is stored in file system block size */
5030 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5031 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5032 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5037 static void __ext4_update_other_inode_time(struct super_block *sb,
5038 unsigned long orig_ino,
5040 struct ext4_inode *raw_inode)
5042 struct inode *inode;
5044 inode = find_inode_by_ino_rcu(sb, ino);
5048 if ((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5050 ((inode->i_state & I_DIRTY_TIME) == 0))
5053 spin_lock(&inode->i_lock);
5054 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5055 I_DIRTY_INODE)) == 0) &&
5056 (inode->i_state & I_DIRTY_TIME)) {
5057 struct ext4_inode_info *ei = EXT4_I(inode);
5059 inode->i_state &= ~I_DIRTY_TIME;
5060 spin_unlock(&inode->i_lock);
5062 spin_lock(&ei->i_raw_lock);
5063 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5064 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5065 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5066 ext4_inode_csum_set(inode, raw_inode, ei);
5067 spin_unlock(&ei->i_raw_lock);
5068 trace_ext4_other_inode_update_time(inode, orig_ino);
5071 spin_unlock(&inode->i_lock);
5075 * Opportunistically update the other time fields for other inodes in
5076 * the same inode table block.
5078 static void ext4_update_other_inodes_time(struct super_block *sb,
5079 unsigned long orig_ino, char *buf)
5082 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5083 int inode_size = EXT4_INODE_SIZE(sb);
5086 * Calculate the first inode in the inode table block. Inode
5087 * numbers are one-based. That is, the first inode in a block
5088 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5090 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5092 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5093 if (ino == orig_ino)
5095 __ext4_update_other_inode_time(sb, orig_ino, ino,
5096 (struct ext4_inode *)buf);
5102 * Post the struct inode info into an on-disk inode location in the
5103 * buffer-cache. This gobbles the caller's reference to the
5104 * buffer_head in the inode location struct.
5106 * The caller must have write access to iloc->bh.
5108 static int ext4_do_update_inode(handle_t *handle,
5109 struct inode *inode,
5110 struct ext4_iloc *iloc)
5112 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5113 struct ext4_inode_info *ei = EXT4_I(inode);
5114 struct buffer_head *bh = iloc->bh;
5115 struct super_block *sb = inode->i_sb;
5117 int need_datasync = 0, set_large_file = 0;
5122 spin_lock(&ei->i_raw_lock);
5124 /* For fields not tracked in the in-memory inode,
5125 * initialise them to zero for new inodes. */
5126 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5127 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5129 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5131 spin_unlock(&ei->i_raw_lock);
5135 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5136 i_uid = i_uid_read(inode);
5137 i_gid = i_gid_read(inode);
5138 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5139 if (!(test_opt(inode->i_sb, NO_UID32))) {
5140 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5141 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5143 * Fix up interoperability with old kernels. Otherwise, old inodes get
5144 * re-used with the upper 16 bits of the uid/gid intact
5146 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5147 raw_inode->i_uid_high = 0;
5148 raw_inode->i_gid_high = 0;
5150 raw_inode->i_uid_high =
5151 cpu_to_le16(high_16_bits(i_uid));
5152 raw_inode->i_gid_high =
5153 cpu_to_le16(high_16_bits(i_gid));
5156 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5157 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5158 raw_inode->i_uid_high = 0;
5159 raw_inode->i_gid_high = 0;
5161 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5163 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5164 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5165 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5166 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5168 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5169 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5170 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5171 raw_inode->i_file_acl_high =
5172 cpu_to_le16(ei->i_file_acl >> 32);
5173 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5174 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5175 ext4_isize_set(raw_inode, ei->i_disksize);
5178 if (ei->i_disksize > 0x7fffffffULL) {
5179 if (!ext4_has_feature_large_file(sb) ||
5180 EXT4_SB(sb)->s_es->s_rev_level ==
5181 cpu_to_le32(EXT4_GOOD_OLD_REV))
5184 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5185 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5186 if (old_valid_dev(inode->i_rdev)) {
5187 raw_inode->i_block[0] =
5188 cpu_to_le32(old_encode_dev(inode->i_rdev));
5189 raw_inode->i_block[1] = 0;
5191 raw_inode->i_block[0] = 0;
5192 raw_inode->i_block[1] =
5193 cpu_to_le32(new_encode_dev(inode->i_rdev));
5194 raw_inode->i_block[2] = 0;
5196 } else if (!ext4_has_inline_data(inode)) {
5197 for (block = 0; block < EXT4_N_BLOCKS; block++)
5198 raw_inode->i_block[block] = ei->i_data[block];
5201 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5202 u64 ivers = ext4_inode_peek_iversion(inode);
5204 raw_inode->i_disk_version = cpu_to_le32(ivers);
5205 if (ei->i_extra_isize) {
5206 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5207 raw_inode->i_version_hi =
5208 cpu_to_le32(ivers >> 32);
5209 raw_inode->i_extra_isize =
5210 cpu_to_le16(ei->i_extra_isize);
5214 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5215 i_projid != EXT4_DEF_PROJID);
5217 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5218 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5219 raw_inode->i_projid = cpu_to_le32(i_projid);
5221 ext4_inode_csum_set(inode, raw_inode, ei);
5222 spin_unlock(&ei->i_raw_lock);
5223 if (inode->i_sb->s_flags & SB_LAZYTIME)
5224 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5227 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5228 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5231 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5232 if (set_large_file) {
5233 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5234 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5237 ext4_set_feature_large_file(sb);
5238 ext4_handle_sync(handle);
5239 err = ext4_handle_dirty_super(handle, sb);
5241 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5244 ext4_std_error(inode->i_sb, err);
5249 * ext4_write_inode()
5251 * We are called from a few places:
5253 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5254 * Here, there will be no transaction running. We wait for any running
5255 * transaction to commit.
5257 * - Within flush work (sys_sync(), kupdate and such).
5258 * We wait on commit, if told to.
5260 * - Within iput_final() -> write_inode_now()
5261 * We wait on commit, if told to.
5263 * In all cases it is actually safe for us to return without doing anything,
5264 * because the inode has been copied into a raw inode buffer in
5265 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5268 * Note that we are absolutely dependent upon all inode dirtiers doing the
5269 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5270 * which we are interested.
5272 * It would be a bug for them to not do this. The code:
5274 * mark_inode_dirty(inode)
5276 * inode->i_size = expr;
5278 * is in error because write_inode() could occur while `stuff()' is running,
5279 * and the new i_size will be lost. Plus the inode will no longer be on the
5280 * superblock's dirty inode list.
5282 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5286 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5287 sb_rdonly(inode->i_sb))
5290 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5293 if (EXT4_SB(inode->i_sb)->s_journal) {
5294 if (ext4_journal_current_handle()) {
5295 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5301 * No need to force transaction in WB_SYNC_NONE mode. Also
5302 * ext4_sync_fs() will force the commit after everything is
5305 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5308 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5309 EXT4_I(inode)->i_sync_tid);
5311 struct ext4_iloc iloc;
5313 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5317 * sync(2) will flush the whole buffer cache. No need to do
5318 * it here separately for each inode.
5320 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5321 sync_dirty_buffer(iloc.bh);
5322 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5323 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5324 "IO error syncing inode");
5333 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5334 * buffers that are attached to a page stradding i_size and are undergoing
5335 * commit. In that case we have to wait for commit to finish and try again.
5337 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5341 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5342 tid_t commit_tid = 0;
5345 offset = inode->i_size & (PAGE_SIZE - 1);
5347 * If the page is fully truncated, we don't need to wait for any commit
5348 * (and we even should not as __ext4_journalled_invalidatepage() may
5349 * strip all buffers from the page but keep the page dirty which can then
5350 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5351 * buffers). Also we don't need to wait for any commit if all buffers in
5352 * the page remain valid. This is most beneficial for the common case of
5353 * blocksize == PAGESIZE.
5355 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5358 page = find_lock_page(inode->i_mapping,
5359 inode->i_size >> PAGE_SHIFT);
5362 ret = __ext4_journalled_invalidatepage(page, offset,
5363 PAGE_SIZE - offset);
5369 read_lock(&journal->j_state_lock);
5370 if (journal->j_committing_transaction)
5371 commit_tid = journal->j_committing_transaction->t_tid;
5372 read_unlock(&journal->j_state_lock);
5374 jbd2_log_wait_commit(journal, commit_tid);
5381 * Called from notify_change.
5383 * We want to trap VFS attempts to truncate the file as soon as
5384 * possible. In particular, we want to make sure that when the VFS
5385 * shrinks i_size, we put the inode on the orphan list and modify
5386 * i_disksize immediately, so that during the subsequent flushing of
5387 * dirty pages and freeing of disk blocks, we can guarantee that any
5388 * commit will leave the blocks being flushed in an unused state on
5389 * disk. (On recovery, the inode will get truncated and the blocks will
5390 * be freed, so we have a strong guarantee that no future commit will
5391 * leave these blocks visible to the user.)
5393 * Another thing we have to assure is that if we are in ordered mode
5394 * and inode is still attached to the committing transaction, we must
5395 * we start writeout of all the dirty pages which are being truncated.
5396 * This way we are sure that all the data written in the previous
5397 * transaction are already on disk (truncate waits for pages under
5400 * Called with inode->i_mutex down.
5402 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5404 struct inode *inode = d_inode(dentry);
5407 const unsigned int ia_valid = attr->ia_valid;
5409 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5412 if (unlikely(IS_IMMUTABLE(inode)))
5415 if (unlikely(IS_APPEND(inode) &&
5416 (ia_valid & (ATTR_MODE | ATTR_UID |
5417 ATTR_GID | ATTR_TIMES_SET))))
5420 error = setattr_prepare(dentry, attr);
5424 error = fscrypt_prepare_setattr(dentry, attr);
5428 error = fsverity_prepare_setattr(dentry, attr);
5432 if (is_quota_modification(inode, attr)) {
5433 error = dquot_initialize(inode);
5438 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5439 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5442 /* (user+group)*(old+new) structure, inode write (sb,
5443 * inode block, ? - but truncate inode update has it) */
5444 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5445 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5446 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5447 if (IS_ERR(handle)) {
5448 error = PTR_ERR(handle);
5452 /* dquot_transfer() calls back ext4_get_inode_usage() which
5453 * counts xattr inode references.
5455 down_read(&EXT4_I(inode)->xattr_sem);
5456 error = dquot_transfer(inode, attr);
5457 up_read(&EXT4_I(inode)->xattr_sem);
5460 ext4_journal_stop(handle);
5463 /* Update corresponding info in inode so that everything is in
5464 * one transaction */
5465 if (attr->ia_valid & ATTR_UID)
5466 inode->i_uid = attr->ia_uid;
5467 if (attr->ia_valid & ATTR_GID)
5468 inode->i_gid = attr->ia_gid;
5469 error = ext4_mark_inode_dirty(handle, inode);
5470 ext4_journal_stop(handle);
5471 if (unlikely(error)) {
5476 if (attr->ia_valid & ATTR_SIZE) {
5478 loff_t oldsize = inode->i_size;
5479 loff_t old_disksize;
5480 int shrink = (attr->ia_size < inode->i_size);
5482 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5483 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5485 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5489 if (!S_ISREG(inode->i_mode)) {
5493 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5494 inode_inc_iversion(inode);
5497 if (ext4_should_order_data(inode)) {
5498 error = ext4_begin_ordered_truncate(inode,
5504 * Blocks are going to be removed from the inode. Wait
5505 * for dio in flight.
5507 inode_dio_wait(inode);
5510 down_write(&EXT4_I(inode)->i_mmap_sem);
5512 rc = ext4_break_layouts(inode);
5514 up_write(&EXT4_I(inode)->i_mmap_sem);
5518 if (attr->ia_size != inode->i_size) {
5519 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5520 if (IS_ERR(handle)) {
5521 error = PTR_ERR(handle);
5524 if (ext4_handle_valid(handle) && shrink) {
5525 error = ext4_orphan_add(handle, inode);
5529 * Update c/mtime on truncate up, ext4_truncate() will
5530 * update c/mtime in shrink case below
5533 inode->i_mtime = current_time(inode);
5534 inode->i_ctime = inode->i_mtime;
5538 ext4_fc_track_range(handle, inode,
5539 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5540 inode->i_sb->s_blocksize_bits,
5541 EXT_MAX_BLOCKS - 1);
5543 ext4_fc_track_range(
5545 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5546 inode->i_sb->s_blocksize_bits,
5547 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5548 inode->i_sb->s_blocksize_bits);
5550 down_write(&EXT4_I(inode)->i_data_sem);
5551 old_disksize = EXT4_I(inode)->i_disksize;
5552 EXT4_I(inode)->i_disksize = attr->ia_size;
5553 rc = ext4_mark_inode_dirty(handle, inode);
5557 * We have to update i_size under i_data_sem together
5558 * with i_disksize to avoid races with writeback code
5559 * running ext4_wb_update_i_disksize().
5562 i_size_write(inode, attr->ia_size);
5564 EXT4_I(inode)->i_disksize = old_disksize;
5565 up_write(&EXT4_I(inode)->i_data_sem);
5566 ext4_journal_stop(handle);
5570 pagecache_isize_extended(inode, oldsize,
5572 } else if (ext4_should_journal_data(inode)) {
5573 ext4_wait_for_tail_page_commit(inode);
5578 * Truncate pagecache after we've waited for commit
5579 * in data=journal mode to make pages freeable.
5581 truncate_pagecache(inode, inode->i_size);
5583 * Call ext4_truncate() even if i_size didn't change to
5584 * truncate possible preallocated blocks.
5586 if (attr->ia_size <= oldsize) {
5587 rc = ext4_truncate(inode);
5592 up_write(&EXT4_I(inode)->i_mmap_sem);
5596 setattr_copy(inode, attr);
5597 mark_inode_dirty(inode);
5601 * If the call to ext4_truncate failed to get a transaction handle at
5602 * all, we need to clean up the in-core orphan list manually.
5604 if (orphan && inode->i_nlink)
5605 ext4_orphan_del(NULL, inode);
5607 if (!error && (ia_valid & ATTR_MODE))
5608 rc = posix_acl_chmod(inode, inode->i_mode);
5612 ext4_std_error(inode->i_sb, error);
5618 int ext4_getattr(const struct path *path, struct kstat *stat,
5619 u32 request_mask, unsigned int query_flags)
5621 struct inode *inode = d_inode(path->dentry);
5622 struct ext4_inode *raw_inode;
5623 struct ext4_inode_info *ei = EXT4_I(inode);
5626 if ((request_mask & STATX_BTIME) &&
5627 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5628 stat->result_mask |= STATX_BTIME;
5629 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5630 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5633 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5634 if (flags & EXT4_APPEND_FL)
5635 stat->attributes |= STATX_ATTR_APPEND;
5636 if (flags & EXT4_COMPR_FL)
5637 stat->attributes |= STATX_ATTR_COMPRESSED;
5638 if (flags & EXT4_ENCRYPT_FL)
5639 stat->attributes |= STATX_ATTR_ENCRYPTED;
5640 if (flags & EXT4_IMMUTABLE_FL)
5641 stat->attributes |= STATX_ATTR_IMMUTABLE;
5642 if (flags & EXT4_NODUMP_FL)
5643 stat->attributes |= STATX_ATTR_NODUMP;
5644 if (flags & EXT4_VERITY_FL)
5645 stat->attributes |= STATX_ATTR_VERITY;
5647 stat->attributes_mask |= (STATX_ATTR_APPEND |
5648 STATX_ATTR_COMPRESSED |
5649 STATX_ATTR_ENCRYPTED |
5650 STATX_ATTR_IMMUTABLE |
5654 generic_fillattr(inode, stat);
5658 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5659 u32 request_mask, unsigned int query_flags)
5661 struct inode *inode = d_inode(path->dentry);
5662 u64 delalloc_blocks;
5664 ext4_getattr(path, stat, request_mask, query_flags);
5667 * If there is inline data in the inode, the inode will normally not
5668 * have data blocks allocated (it may have an external xattr block).
5669 * Report at least one sector for such files, so tools like tar, rsync,
5670 * others don't incorrectly think the file is completely sparse.
5672 if (unlikely(ext4_has_inline_data(inode)))
5673 stat->blocks += (stat->size + 511) >> 9;
5676 * We can't update i_blocks if the block allocation is delayed
5677 * otherwise in the case of system crash before the real block
5678 * allocation is done, we will have i_blocks inconsistent with
5679 * on-disk file blocks.
5680 * We always keep i_blocks updated together with real
5681 * allocation. But to not confuse with user, stat
5682 * will return the blocks that include the delayed allocation
5683 * blocks for this file.
5685 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5686 EXT4_I(inode)->i_reserved_data_blocks);
5687 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5691 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5694 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5695 return ext4_ind_trans_blocks(inode, lblocks);
5696 return ext4_ext_index_trans_blocks(inode, pextents);
5700 * Account for index blocks, block groups bitmaps and block group
5701 * descriptor blocks if modify datablocks and index blocks
5702 * worse case, the indexs blocks spread over different block groups
5704 * If datablocks are discontiguous, they are possible to spread over
5705 * different block groups too. If they are contiguous, with flexbg,
5706 * they could still across block group boundary.
5708 * Also account for superblock, inode, quota and xattr blocks
5710 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5713 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5719 * How many index blocks need to touch to map @lblocks logical blocks
5720 * to @pextents physical extents?
5722 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5727 * Now let's see how many group bitmaps and group descriptors need
5730 groups = idxblocks + pextents;
5732 if (groups > ngroups)
5734 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5735 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5737 /* bitmaps and block group descriptor blocks */
5738 ret += groups + gdpblocks;
5740 /* Blocks for super block, inode, quota and xattr blocks */
5741 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5747 * Calculate the total number of credits to reserve to fit
5748 * the modification of a single pages into a single transaction,
5749 * which may include multiple chunks of block allocations.
5751 * This could be called via ext4_write_begin()
5753 * We need to consider the worse case, when
5754 * one new block per extent.
5756 int ext4_writepage_trans_blocks(struct inode *inode)
5758 int bpp = ext4_journal_blocks_per_page(inode);
5761 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5763 /* Account for data blocks for journalled mode */
5764 if (ext4_should_journal_data(inode))
5770 * Calculate the journal credits for a chunk of data modification.
5772 * This is called from DIO, fallocate or whoever calling
5773 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5775 * journal buffers for data blocks are not included here, as DIO
5776 * and fallocate do no need to journal data buffers.
5778 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5780 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5784 * The caller must have previously called ext4_reserve_inode_write().
5785 * Give this, we know that the caller already has write access to iloc->bh.
5787 int ext4_mark_iloc_dirty(handle_t *handle,
5788 struct inode *inode, struct ext4_iloc *iloc)
5792 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5796 ext4_fc_track_inode(handle, inode);
5799 * ea_inodes are using i_version for storing reference count, don't
5802 if (IS_I_VERSION(inode) &&
5803 !(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
5804 inode_inc_iversion(inode);
5806 /* the do_update_inode consumes one bh->b_count */
5809 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5810 err = ext4_do_update_inode(handle, inode, iloc);
5816 * On success, We end up with an outstanding reference count against
5817 * iloc->bh. This _must_ be cleaned up later.
5821 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5822 struct ext4_iloc *iloc)
5826 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5829 err = ext4_get_inode_loc(inode, iloc);
5831 BUFFER_TRACE(iloc->bh, "get_write_access");
5832 err = ext4_journal_get_write_access(handle, iloc->bh);
5838 ext4_std_error(inode->i_sb, err);
5842 static int __ext4_expand_extra_isize(struct inode *inode,
5843 unsigned int new_extra_isize,
5844 struct ext4_iloc *iloc,
5845 handle_t *handle, int *no_expand)
5847 struct ext4_inode *raw_inode;
5848 struct ext4_xattr_ibody_header *header;
5849 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5850 struct ext4_inode_info *ei = EXT4_I(inode);
5853 /* this was checked at iget time, but double check for good measure */
5854 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5855 (ei->i_extra_isize & 3)) {
5856 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5858 EXT4_INODE_SIZE(inode->i_sb));
5859 return -EFSCORRUPTED;
5861 if ((new_extra_isize < ei->i_extra_isize) ||
5862 (new_extra_isize < 4) ||
5863 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5864 return -EINVAL; /* Should never happen */
5866 raw_inode = ext4_raw_inode(iloc);
5868 header = IHDR(inode, raw_inode);
5870 /* No extended attributes present */
5871 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5872 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5873 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5874 EXT4_I(inode)->i_extra_isize, 0,
5875 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5876 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5881 * We may need to allocate external xattr block so we need quotas
5882 * initialized. Here we can be called with various locks held so we
5883 * cannot affort to initialize quotas ourselves. So just bail.
5885 if (dquot_initialize_needed(inode))
5888 /* try to expand with EAs present */
5889 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5893 * Inode size expansion failed; don't try again
5902 * Expand an inode by new_extra_isize bytes.
5903 * Returns 0 on success or negative error number on failure.
5905 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5906 unsigned int new_extra_isize,
5907 struct ext4_iloc iloc,
5913 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5917 * In nojournal mode, we can immediately attempt to expand
5918 * the inode. When journaled, we first need to obtain extra
5919 * buffer credits since we may write into the EA block
5920 * with this same handle. If journal_extend fails, then it will
5921 * only result in a minor loss of functionality for that inode.
5922 * If this is felt to be critical, then e2fsck should be run to
5923 * force a large enough s_min_extra_isize.
5925 if (ext4_journal_extend(handle,
5926 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5929 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5932 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5933 handle, &no_expand);
5934 ext4_write_unlock_xattr(inode, &no_expand);
5939 int ext4_expand_extra_isize(struct inode *inode,
5940 unsigned int new_extra_isize,
5941 struct ext4_iloc *iloc)
5947 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5952 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5953 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5954 if (IS_ERR(handle)) {
5955 error = PTR_ERR(handle);
5960 ext4_write_lock_xattr(inode, &no_expand);
5962 BUFFER_TRACE(iloc->bh, "get_write_access");
5963 error = ext4_journal_get_write_access(handle, iloc->bh);
5969 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5970 handle, &no_expand);
5972 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5977 ext4_write_unlock_xattr(inode, &no_expand);
5978 ext4_journal_stop(handle);
5983 * What we do here is to mark the in-core inode as clean with respect to inode
5984 * dirtiness (it may still be data-dirty).
5985 * This means that the in-core inode may be reaped by prune_icache
5986 * without having to perform any I/O. This is a very good thing,
5987 * because *any* task may call prune_icache - even ones which
5988 * have a transaction open against a different journal.
5990 * Is this cheating? Not really. Sure, we haven't written the
5991 * inode out, but prune_icache isn't a user-visible syncing function.
5992 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5993 * we start and wait on commits.
5995 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5996 const char *func, unsigned int line)
5998 struct ext4_iloc iloc;
5999 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6003 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6004 err = ext4_reserve_inode_write(handle, inode, &iloc);
6008 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6009 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6012 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
6015 ext4_error_inode_err(inode, func, line, 0, err,
6016 "mark_inode_dirty error");
6021 * ext4_dirty_inode() is called from __mark_inode_dirty()
6023 * We're really interested in the case where a file is being extended.
6024 * i_size has been changed by generic_commit_write() and we thus need
6025 * to include the updated inode in the current transaction.
6027 * Also, dquot_alloc_block() will always dirty the inode when blocks
6028 * are allocated to the file.
6030 * If the inode is marked synchronous, we don't honour that here - doing
6031 * so would cause a commit on atime updates, which we don't bother doing.
6032 * We handle synchronous inodes at the highest possible level.
6034 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6035 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6036 * to copy into the on-disk inode structure are the timestamp files.
6038 void ext4_dirty_inode(struct inode *inode, int flags)
6042 if (flags == I_DIRTY_TIME)
6044 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6048 ext4_mark_inode_dirty(handle, inode);
6050 ext4_journal_stop(handle);
6055 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6060 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6063 * We have to be very careful here: changing a data block's
6064 * journaling status dynamically is dangerous. If we write a
6065 * data block to the journal, change the status and then delete
6066 * that block, we risk forgetting to revoke the old log record
6067 * from the journal and so a subsequent replay can corrupt data.
6068 * So, first we make sure that the journal is empty and that
6069 * nobody is changing anything.
6072 journal = EXT4_JOURNAL(inode);
6075 if (is_journal_aborted(journal))
6078 /* Wait for all existing dio workers */
6079 inode_dio_wait(inode);
6082 * Before flushing the journal and switching inode's aops, we have
6083 * to flush all dirty data the inode has. There can be outstanding
6084 * delayed allocations, there can be unwritten extents created by
6085 * fallocate or buffered writes in dioread_nolock mode covered by
6086 * dirty data which can be converted only after flushing the dirty
6087 * data (and journalled aops don't know how to handle these cases).
6090 down_write(&EXT4_I(inode)->i_mmap_sem);
6091 err = filemap_write_and_wait(inode->i_mapping);
6093 up_write(&EXT4_I(inode)->i_mmap_sem);
6098 percpu_down_write(&sbi->s_writepages_rwsem);
6099 jbd2_journal_lock_updates(journal);
6102 * OK, there are no updates running now, and all cached data is
6103 * synced to disk. We are now in a completely consistent state
6104 * which doesn't have anything in the journal, and we know that
6105 * no filesystem updates are running, so it is safe to modify
6106 * the inode's in-core data-journaling state flag now.
6110 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6112 err = jbd2_journal_flush(journal);
6114 jbd2_journal_unlock_updates(journal);
6115 percpu_up_write(&sbi->s_writepages_rwsem);
6118 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6120 ext4_set_aops(inode);
6122 jbd2_journal_unlock_updates(journal);
6123 percpu_up_write(&sbi->s_writepages_rwsem);
6126 up_write(&EXT4_I(inode)->i_mmap_sem);
6128 /* Finally we can mark the inode as dirty. */
6130 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6132 return PTR_ERR(handle);
6134 ext4_fc_mark_ineligible(inode->i_sb,
6135 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE);
6136 err = ext4_mark_inode_dirty(handle, inode);
6137 ext4_handle_sync(handle);
6138 ext4_journal_stop(handle);
6139 ext4_std_error(inode->i_sb, err);
6144 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6146 return !buffer_mapped(bh);
6149 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6151 struct vm_area_struct *vma = vmf->vma;
6152 struct page *page = vmf->page;
6157 struct file *file = vma->vm_file;
6158 struct inode *inode = file_inode(file);
6159 struct address_space *mapping = inode->i_mapping;
6161 get_block_t *get_block;
6164 if (unlikely(IS_IMMUTABLE(inode)))
6165 return VM_FAULT_SIGBUS;
6167 sb_start_pagefault(inode->i_sb);
6168 file_update_time(vma->vm_file);
6170 down_read(&EXT4_I(inode)->i_mmap_sem);
6172 err = ext4_convert_inline_data(inode);
6177 * On data journalling we skip straight to the transaction handle:
6178 * there's no delalloc; page truncated will be checked later; the
6179 * early return w/ all buffers mapped (calculates size/len) can't
6180 * be used; and there's no dioread_nolock, so only ext4_get_block.
6182 if (ext4_should_journal_data(inode))
6185 /* Delalloc case is easy... */
6186 if (test_opt(inode->i_sb, DELALLOC) &&
6187 !ext4_nonda_switch(inode->i_sb)) {
6189 err = block_page_mkwrite(vma, vmf,
6190 ext4_da_get_block_prep);
6191 } while (err == -ENOSPC &&
6192 ext4_should_retry_alloc(inode->i_sb, &retries));
6197 size = i_size_read(inode);
6198 /* Page got truncated from under us? */
6199 if (page->mapping != mapping || page_offset(page) > size) {
6201 ret = VM_FAULT_NOPAGE;
6205 if (page->index == size >> PAGE_SHIFT)
6206 len = size & ~PAGE_MASK;
6210 * Return if we have all the buffers mapped. This avoids the need to do
6211 * journal_start/journal_stop which can block and take a long time
6213 * This cannot be done for data journalling, as we have to add the
6214 * inode to the transaction's list to writeprotect pages on commit.
6216 if (page_has_buffers(page)) {
6217 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6219 ext4_bh_unmapped)) {
6220 /* Wait so that we don't change page under IO */
6221 wait_for_stable_page(page);
6222 ret = VM_FAULT_LOCKED;
6227 /* OK, we need to fill the hole... */
6228 if (ext4_should_dioread_nolock(inode))
6229 get_block = ext4_get_block_unwritten;
6231 get_block = ext4_get_block;
6233 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6234 ext4_writepage_trans_blocks(inode));
6235 if (IS_ERR(handle)) {
6236 ret = VM_FAULT_SIGBUS;
6240 * Data journalling can't use block_page_mkwrite() because it
6241 * will set_buffer_dirty() before do_journal_get_write_access()
6242 * thus might hit warning messages for dirty metadata buffers.
6244 if (!ext4_should_journal_data(inode)) {
6245 err = block_page_mkwrite(vma, vmf, get_block);
6248 size = i_size_read(inode);
6249 /* Page got truncated from under us? */
6250 if (page->mapping != mapping || page_offset(page) > size) {
6251 ret = VM_FAULT_NOPAGE;
6255 if (page->index == size >> PAGE_SHIFT)
6256 len = size & ~PAGE_MASK;
6260 err = __block_write_begin(page, 0, len, ext4_get_block);
6262 ret = VM_FAULT_SIGBUS;
6263 if (ext4_walk_page_buffers(handle, page_buffers(page),
6264 0, len, NULL, do_journal_get_write_access))
6266 if (ext4_walk_page_buffers(handle, page_buffers(page),
6267 0, len, NULL, write_end_fn))
6269 if (ext4_jbd2_inode_add_write(handle, inode,
6270 page_offset(page), len))
6272 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6277 ext4_journal_stop(handle);
6278 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6281 ret = block_page_mkwrite_return(err);
6283 up_read(&EXT4_I(inode)->i_mmap_sem);
6284 sb_end_pagefault(inode->i_sb);
6288 ext4_journal_stop(handle);
6292 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6294 struct inode *inode = file_inode(vmf->vma->vm_file);
6297 down_read(&EXT4_I(inode)->i_mmap_sem);
6298 ret = filemap_fault(vmf);
6299 up_read(&EXT4_I(inode)->i_mmap_sem);