1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/mount.h>
24 #include <linux/time.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/dax.h>
28 #include <linux/quotaops.h>
29 #include <linux/string.h>
30 #include <linux/buffer_head.h>
31 #include <linux/writeback.h>
32 #include <linux/pagevec.h>
33 #include <linux/mpage.h>
34 #include <linux/namei.h>
35 #include <linux/uio.h>
36 #include <linux/bio.h>
37 #include <linux/workqueue.h>
38 #include <linux/kernel.h>
39 #include <linux/printk.h>
40 #include <linux/slab.h>
41 #include <linux/bitops.h>
42 #include <linux/iomap.h>
43 #include <linux/iversion.h>
44 #include <linux/dax.h>
46 #include "ext4_jbd2.h"
51 #include <trace/events/ext4.h>
53 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
54 struct ext4_inode_info *ei)
56 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 int offset = offsetof(struct ext4_inode, i_checksum_lo);
60 unsigned int csum_size = sizeof(dummy_csum);
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
63 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
65 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
66 EXT4_GOOD_OLD_INODE_SIZE - offset);
68 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
69 offset = offsetof(struct ext4_inode, i_checksum_hi);
70 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
71 EXT4_GOOD_OLD_INODE_SIZE,
72 offset - EXT4_GOOD_OLD_INODE_SIZE);
73 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
74 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
78 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
79 EXT4_INODE_SIZE(inode->i_sb) - offset);
85 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
86 struct ext4_inode_info *ei)
88 __u32 provided, calculated;
90 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
91 cpu_to_le32(EXT4_OS_LINUX) ||
92 !ext4_has_metadata_csum(inode->i_sb))
95 provided = le16_to_cpu(raw->i_checksum_lo);
96 calculated = ext4_inode_csum(inode, raw, ei);
97 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
98 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
99 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 calculated &= 0xFFFF;
103 return provided == calculated;
106 void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107 struct ext4_inode_info *ei)
111 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112 cpu_to_le32(EXT4_OS_LINUX) ||
113 !ext4_has_metadata_csum(inode->i_sb))
116 csum = ext4_inode_csum(inode, raw, ei);
117 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
123 static inline int ext4_begin_ordered_truncate(struct inode *inode,
126 trace_ext4_begin_ordered_truncate(inode, new_size);
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
133 if (!EXT4_I(inode)->jinode)
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136 EXT4_I(inode)->jinode,
140 static void ext4_invalidatepage(struct page *page, unsigned int offset,
141 unsigned int length);
142 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
147 * Test whether an inode is a fast symlink.
148 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 int ext4_inode_is_fast_symlink(struct inode *inode)
152 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
153 int ea_blocks = EXT4_I(inode)->i_file_acl ?
154 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
156 if (ext4_has_inline_data(inode))
159 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 return S_ISLNK(inode->i_mode) && inode->i_size &&
162 (inode->i_size < EXT4_N_BLOCKS * 4);
166 * Called at the last iput() if i_nlink is zero.
168 void ext4_evict_inode(struct inode *inode)
173 * Credits for final inode cleanup and freeing:
174 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
175 * (xattr block freeing), bitmap, group descriptor (inode freeing)
177 int extra_credits = 6;
178 struct ext4_xattr_inode_array *ea_inode_array = NULL;
179 bool freeze_protected = false;
181 trace_ext4_evict_inode(inode);
183 if (inode->i_nlink) {
185 * When journalling data dirty buffers are tracked only in the
186 * journal. So although mm thinks everything is clean and
187 * ready for reaping the inode might still have some pages to
188 * write in the running transaction or waiting to be
189 * checkpointed. Thus calling jbd2_journal_invalidatepage()
190 * (via truncate_inode_pages()) to discard these buffers can
191 * cause data loss. Also even if we did not discard these
192 * buffers, we would have no way to find them after the inode
193 * is reaped and thus user could see stale data if he tries to
194 * read them before the transaction is checkpointed. So be
195 * careful and force everything to disk here... We use
196 * ei->i_datasync_tid to store the newest transaction
197 * containing inode's data.
199 * Note that directories do not have this problem because they
200 * don't use page cache.
202 if (inode->i_ino != EXT4_JOURNAL_INO &&
203 ext4_should_journal_data(inode) &&
204 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
205 inode->i_data.nrpages) {
206 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
207 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
209 jbd2_complete_transaction(journal, commit_tid);
210 filemap_write_and_wait(&inode->i_data);
212 truncate_inode_pages_final(&inode->i_data);
217 if (is_bad_inode(inode))
219 dquot_initialize(inode);
221 if (ext4_should_order_data(inode))
222 ext4_begin_ordered_truncate(inode, 0);
223 truncate_inode_pages_final(&inode->i_data);
226 * For inodes with journalled data, transaction commit could have
227 * dirtied the inode. Flush worker is ignoring it because of I_FREEING
228 * flag but we still need to remove the inode from the writeback lists.
230 if (!list_empty_careful(&inode->i_io_list)) {
231 WARN_ON_ONCE(!ext4_should_journal_data(inode));
232 inode_io_list_del(inode);
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it. When we are in a running transaction though,
238 * we are already protected against freezing and we cannot grab further
239 * protection due to lock ordering constraints.
241 if (!ext4_journal_current_handle()) {
242 sb_start_intwrite(inode->i_sb);
243 freeze_protected = true;
246 if (!IS_NOQUOTA(inode))
247 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
250 * Block bitmap, group descriptor, and inode are accounted in both
251 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
253 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
254 ext4_blocks_for_truncate(inode) + extra_credits - 3);
255 if (IS_ERR(handle)) {
256 ext4_std_error(inode->i_sb, PTR_ERR(handle));
258 * If we're going to skip the normal cleanup, we still need to
259 * make sure that the in-core orphan linked list is properly
262 ext4_orphan_del(NULL, inode);
263 if (freeze_protected)
264 sb_end_intwrite(inode->i_sb);
269 ext4_handle_sync(handle);
272 * Set inode->i_size to 0 before calling ext4_truncate(). We need
273 * special handling of symlinks here because i_size is used to
274 * determine whether ext4_inode_info->i_data contains symlink data or
275 * block mappings. Setting i_size to 0 will remove its fast symlink
276 * status. Erase i_data so that it becomes a valid empty block map.
278 if (ext4_inode_is_fast_symlink(inode))
279 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
281 err = ext4_mark_inode_dirty(handle, inode);
283 ext4_warning(inode->i_sb,
284 "couldn't mark inode dirty (err %d)", err);
287 if (inode->i_blocks) {
288 err = ext4_truncate(inode);
290 ext4_error_err(inode->i_sb, -err,
291 "couldn't truncate inode %lu (err %d)",
297 /* Remove xattr references. */
298 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
301 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
303 ext4_journal_stop(handle);
304 ext4_orphan_del(NULL, inode);
305 if (freeze_protected)
306 sb_end_intwrite(inode->i_sb);
307 ext4_xattr_inode_array_free(ea_inode_array);
312 * Kill off the orphan record which ext4_truncate created.
313 * AKPM: I think this can be inside the above `if'.
314 * Note that ext4_orphan_del() has to be able to cope with the
315 * deletion of a non-existent orphan - this is because we don't
316 * know if ext4_truncate() actually created an orphan record.
317 * (Well, we could do this if we need to, but heck - it works)
319 ext4_orphan_del(handle, inode);
320 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
323 * One subtle ordering requirement: if anything has gone wrong
324 * (transaction abort, IO errors, whatever), then we can still
325 * do these next steps (the fs will already have been marked as
326 * having errors), but we can't free the inode if the mark_dirty
329 if (ext4_mark_inode_dirty(handle, inode))
330 /* If that failed, just do the required in-core inode clear. */
331 ext4_clear_inode(inode);
333 ext4_free_inode(handle, inode);
334 ext4_journal_stop(handle);
335 if (freeze_protected)
336 sb_end_intwrite(inode->i_sb);
337 ext4_xattr_inode_array_free(ea_inode_array);
340 if (!list_empty(&EXT4_I(inode)->i_fc_list))
341 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
342 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
346 qsize_t *ext4_get_reserved_space(struct inode *inode)
348 return &EXT4_I(inode)->i_reserved_quota;
353 * Called with i_data_sem down, which is important since we can call
354 * ext4_discard_preallocations() from here.
356 void ext4_da_update_reserve_space(struct inode *inode,
357 int used, int quota_claim)
359 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
360 struct ext4_inode_info *ei = EXT4_I(inode);
362 spin_lock(&ei->i_block_reservation_lock);
363 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
364 if (unlikely(used > ei->i_reserved_data_blocks)) {
365 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
366 "with only %d reserved data blocks",
367 __func__, inode->i_ino, used,
368 ei->i_reserved_data_blocks);
370 used = ei->i_reserved_data_blocks;
373 /* Update per-inode reservations */
374 ei->i_reserved_data_blocks -= used;
375 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
377 spin_unlock(&ei->i_block_reservation_lock);
379 /* Update quota subsystem for data blocks */
381 dquot_claim_block(inode, EXT4_C2B(sbi, used));
384 * We did fallocate with an offset that is already delayed
385 * allocated. So on delayed allocated writeback we should
386 * not re-claim the quota for fallocated blocks.
388 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
392 * If we have done all the pending block allocations and if
393 * there aren't any writers on the inode, we can discard the
394 * inode's preallocations.
396 if ((ei->i_reserved_data_blocks == 0) &&
397 !inode_is_open_for_write(inode))
398 ext4_discard_preallocations(inode, 0);
401 static int __check_block_validity(struct inode *inode, const char *func,
403 struct ext4_map_blocks *map)
405 if (ext4_has_feature_journal(inode->i_sb) &&
407 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
409 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
410 ext4_error_inode(inode, func, line, map->m_pblk,
411 "lblock %lu mapped to illegal pblock %llu "
412 "(length %d)", (unsigned long) map->m_lblk,
413 map->m_pblk, map->m_len);
414 return -EFSCORRUPTED;
419 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
424 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
425 return fscrypt_zeroout_range(inode, lblk, pblk, len);
427 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
434 #define check_block_validity(inode, map) \
435 __check_block_validity((inode), __func__, __LINE__, (map))
437 #ifdef ES_AGGRESSIVE_TEST
438 static void ext4_map_blocks_es_recheck(handle_t *handle,
440 struct ext4_map_blocks *es_map,
441 struct ext4_map_blocks *map,
448 * There is a race window that the result is not the same.
449 * e.g. xfstests #223 when dioread_nolock enables. The reason
450 * is that we lookup a block mapping in extent status tree with
451 * out taking i_data_sem. So at the time the unwritten extent
452 * could be converted.
454 down_read(&EXT4_I(inode)->i_data_sem);
455 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
456 retval = ext4_ext_map_blocks(handle, inode, map, 0);
458 retval = ext4_ind_map_blocks(handle, inode, map, 0);
460 up_read((&EXT4_I(inode)->i_data_sem));
463 * We don't check m_len because extent will be collpased in status
464 * tree. So the m_len might not equal.
466 if (es_map->m_lblk != map->m_lblk ||
467 es_map->m_flags != map->m_flags ||
468 es_map->m_pblk != map->m_pblk) {
469 printk("ES cache assertion failed for inode: %lu "
470 "es_cached ex [%d/%d/%llu/%x] != "
471 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472 inode->i_ino, es_map->m_lblk, es_map->m_len,
473 es_map->m_pblk, es_map->m_flags, map->m_lblk,
474 map->m_len, map->m_pblk, map->m_flags,
478 #endif /* ES_AGGRESSIVE_TEST */
481 * The ext4_map_blocks() function tries to look up the requested blocks,
482 * and returns if the blocks are already mapped.
484 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485 * and store the allocated blocks in the result buffer head and mark it
488 * If file type is extents based, it will call ext4_ext_map_blocks(),
489 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
492 * On success, it returns the number of blocks being mapped or allocated. if
493 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
494 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
496 * It returns 0 if plain look up failed (blocks have not been allocated), in
497 * that case, @map is returned as unmapped but we still do fill map->m_len to
498 * indicate the length of a hole starting at map->m_lblk.
500 * It returns the error in case of allocation failure.
502 int ext4_map_blocks(handle_t *handle, struct inode *inode,
503 struct ext4_map_blocks *map, int flags)
505 struct extent_status es;
508 #ifdef ES_AGGRESSIVE_TEST
509 struct ext4_map_blocks orig_map;
511 memcpy(&orig_map, map, sizeof(*map));
515 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
516 flags, map->m_len, (unsigned long) map->m_lblk);
519 * ext4_map_blocks returns an int, and m_len is an unsigned int
521 if (unlikely(map->m_len > INT_MAX))
522 map->m_len = INT_MAX;
524 /* We can handle the block number less than EXT_MAX_BLOCKS */
525 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
526 return -EFSCORRUPTED;
528 /* Lookup extent status tree firstly */
529 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
530 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
531 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
532 map->m_pblk = ext4_es_pblock(&es) +
533 map->m_lblk - es.es_lblk;
534 map->m_flags |= ext4_es_is_written(&es) ?
535 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
536 retval = es.es_len - (map->m_lblk - es.es_lblk);
537 if (retval > map->m_len)
540 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
542 retval = es.es_len - (map->m_lblk - es.es_lblk);
543 if (retval > map->m_len)
550 #ifdef ES_AGGRESSIVE_TEST
551 ext4_map_blocks_es_recheck(handle, inode, map,
558 * Try to see if we can get the block without requesting a new
561 down_read(&EXT4_I(inode)->i_data_sem);
562 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
563 retval = ext4_ext_map_blocks(handle, inode, map, 0);
565 retval = ext4_ind_map_blocks(handle, inode, map, 0);
570 if (unlikely(retval != map->m_len)) {
571 ext4_warning(inode->i_sb,
572 "ES len assertion failed for inode "
573 "%lu: retval %d != map->m_len %d",
574 inode->i_ino, retval, map->m_len);
578 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
579 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
580 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
581 !(status & EXTENT_STATUS_WRITTEN) &&
582 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
583 map->m_lblk + map->m_len - 1))
584 status |= EXTENT_STATUS_DELAYED;
585 ret = ext4_es_insert_extent(inode, map->m_lblk,
586 map->m_len, map->m_pblk, status);
590 up_read((&EXT4_I(inode)->i_data_sem));
593 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
594 ret = check_block_validity(inode, map);
599 /* If it is only a block(s) look up */
600 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
604 * Returns if the blocks have already allocated
606 * Note that if blocks have been preallocated
607 * ext4_ext_get_block() returns the create = 0
608 * with buffer head unmapped.
610 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
612 * If we need to convert extent to unwritten
613 * we continue and do the actual work in
614 * ext4_ext_map_blocks()
616 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
620 * Here we clear m_flags because after allocating an new extent,
621 * it will be set again.
623 map->m_flags &= ~EXT4_MAP_FLAGS;
626 * New blocks allocate and/or writing to unwritten extent
627 * will possibly result in updating i_data, so we take
628 * the write lock of i_data_sem, and call get_block()
629 * with create == 1 flag.
631 down_write(&EXT4_I(inode)->i_data_sem);
634 * We need to check for EXT4 here because migrate
635 * could have changed the inode type in between
637 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
638 retval = ext4_ext_map_blocks(handle, inode, map, flags);
640 retval = ext4_ind_map_blocks(handle, inode, map, flags);
642 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
644 * We allocated new blocks which will result in
645 * i_data's format changing. Force the migrate
646 * to fail by clearing migrate flags
648 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
652 * Update reserved blocks/metadata blocks after successful
653 * block allocation which had been deferred till now. We don't
654 * support fallocate for non extent files. So we can update
655 * reserve space here.
658 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
659 ext4_da_update_reserve_space(inode, retval, 1);
665 if (unlikely(retval != map->m_len)) {
666 ext4_warning(inode->i_sb,
667 "ES len assertion failed for inode "
668 "%lu: retval %d != map->m_len %d",
669 inode->i_ino, retval, map->m_len);
674 * We have to zeroout blocks before inserting them into extent
675 * status tree. Otherwise someone could look them up there and
676 * use them before they are really zeroed. We also have to
677 * unmap metadata before zeroing as otherwise writeback can
678 * overwrite zeros with stale data from block device.
680 if (flags & EXT4_GET_BLOCKS_ZERO &&
681 map->m_flags & EXT4_MAP_MAPPED &&
682 map->m_flags & EXT4_MAP_NEW) {
683 ret = ext4_issue_zeroout(inode, map->m_lblk,
684 map->m_pblk, map->m_len);
692 * If the extent has been zeroed out, we don't need to update
693 * extent status tree.
695 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
696 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
697 if (ext4_es_is_written(&es))
700 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
701 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
702 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
703 !(status & EXTENT_STATUS_WRITTEN) &&
704 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
705 map->m_lblk + map->m_len - 1))
706 status |= EXTENT_STATUS_DELAYED;
707 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
708 map->m_pblk, status);
716 up_write((&EXT4_I(inode)->i_data_sem));
717 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
718 ret = check_block_validity(inode, map);
723 * Inodes with freshly allocated blocks where contents will be
724 * visible after transaction commit must be on transaction's
727 if (map->m_flags & EXT4_MAP_NEW &&
728 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
729 !(flags & EXT4_GET_BLOCKS_ZERO) &&
730 !ext4_is_quota_file(inode) &&
731 ext4_should_order_data(inode)) {
733 (loff_t)map->m_lblk << inode->i_blkbits;
734 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
736 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
737 ret = ext4_jbd2_inode_add_wait(handle, inode,
740 ret = ext4_jbd2_inode_add_write(handle, inode,
746 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
747 map->m_flags & EXT4_MAP_MAPPED))
748 ext4_fc_track_range(handle, inode, map->m_lblk,
749 map->m_lblk + map->m_len - 1);
751 ext_debug(inode, "failed with err %d\n", retval);
756 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
757 * we have to be careful as someone else may be manipulating b_state as well.
759 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
761 unsigned long old_state;
762 unsigned long new_state;
764 flags &= EXT4_MAP_FLAGS;
766 /* Dummy buffer_head? Set non-atomically. */
768 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
772 * Someone else may be modifying b_state. Be careful! This is ugly but
773 * once we get rid of using bh as a container for mapping information
774 * to pass to / from get_block functions, this can go away.
777 old_state = READ_ONCE(bh->b_state);
778 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
780 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
783 static int _ext4_get_block(struct inode *inode, sector_t iblock,
784 struct buffer_head *bh, int flags)
786 struct ext4_map_blocks map;
789 if (ext4_has_inline_data(inode))
793 map.m_len = bh->b_size >> inode->i_blkbits;
795 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
798 map_bh(bh, inode->i_sb, map.m_pblk);
799 ext4_update_bh_state(bh, map.m_flags);
800 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
802 } else if (ret == 0) {
803 /* hole case, need to fill in bh->b_size */
804 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
809 int ext4_get_block(struct inode *inode, sector_t iblock,
810 struct buffer_head *bh, int create)
812 return _ext4_get_block(inode, iblock, bh,
813 create ? EXT4_GET_BLOCKS_CREATE : 0);
817 * Get block function used when preparing for buffered write if we require
818 * creating an unwritten extent if blocks haven't been allocated. The extent
819 * will be converted to written after the IO is complete.
821 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
822 struct buffer_head *bh_result, int create)
824 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
825 inode->i_ino, create);
826 return _ext4_get_block(inode, iblock, bh_result,
827 EXT4_GET_BLOCKS_IO_CREATE_EXT);
830 /* Maximum number of blocks we map for direct IO at once. */
831 #define DIO_MAX_BLOCKS 4096
834 * `handle' can be NULL if create is zero
836 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
837 ext4_lblk_t block, int map_flags)
839 struct ext4_map_blocks map;
840 struct buffer_head *bh;
841 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
844 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
845 || handle != NULL || create == 0);
849 err = ext4_map_blocks(handle, inode, &map, map_flags);
852 return create ? ERR_PTR(-ENOSPC) : NULL;
856 bh = sb_getblk(inode->i_sb, map.m_pblk);
858 return ERR_PTR(-ENOMEM);
859 if (map.m_flags & EXT4_MAP_NEW) {
861 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
862 || (handle != NULL));
865 * Now that we do not always journal data, we should
866 * keep in mind whether this should always journal the
867 * new buffer as metadata. For now, regular file
868 * writes use ext4_get_block instead, so it's not a
872 BUFFER_TRACE(bh, "call get_create_access");
873 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
879 if (!buffer_uptodate(bh)) {
880 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
881 set_buffer_uptodate(bh);
884 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
885 err = ext4_handle_dirty_metadata(handle, inode, bh);
889 BUFFER_TRACE(bh, "not a new buffer");
896 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
897 ext4_lblk_t block, int map_flags)
899 struct buffer_head *bh;
902 bh = ext4_getblk(handle, inode, block, map_flags);
905 if (!bh || ext4_buffer_uptodate(bh))
908 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
916 /* Read a contiguous batch of blocks. */
917 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
918 bool wait, struct buffer_head **bhs)
922 for (i = 0; i < bh_count; i++) {
923 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
924 if (IS_ERR(bhs[i])) {
925 err = PTR_ERR(bhs[i]);
931 for (i = 0; i < bh_count; i++)
932 /* Note that NULL bhs[i] is valid because of holes. */
933 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
934 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
939 for (i = 0; i < bh_count; i++)
941 wait_on_buffer(bhs[i]);
943 for (i = 0; i < bh_count; i++) {
944 if (bhs[i] && !buffer_uptodate(bhs[i])) {
952 for (i = 0; i < bh_count; i++) {
959 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
960 struct buffer_head *head,
964 int (*fn)(handle_t *handle, struct inode *inode,
965 struct buffer_head *bh))
967 struct buffer_head *bh;
968 unsigned block_start, block_end;
969 unsigned blocksize = head->b_size;
971 struct buffer_head *next;
973 for (bh = head, block_start = 0;
974 ret == 0 && (bh != head || !block_start);
975 block_start = block_end, bh = next) {
976 next = bh->b_this_page;
977 block_end = block_start + blocksize;
978 if (block_end <= from || block_start >= to) {
979 if (partial && !buffer_uptodate(bh))
983 err = (*fn)(handle, inode, bh);
991 * To preserve ordering, it is essential that the hole instantiation and
992 * the data write be encapsulated in a single transaction. We cannot
993 * close off a transaction and start a new one between the ext4_get_block()
994 * and the commit_write(). So doing the jbd2_journal_start at the start of
995 * prepare_write() is the right place.
997 * Also, this function can nest inside ext4_writepage(). In that case, we
998 * *know* that ext4_writepage() has generated enough buffer credits to do the
999 * whole page. So we won't block on the journal in that case, which is good,
1000 * because the caller may be PF_MEMALLOC.
1002 * By accident, ext4 can be reentered when a transaction is open via
1003 * quota file writes. If we were to commit the transaction while thus
1004 * reentered, there can be a deadlock - we would be holding a quota
1005 * lock, and the commit would never complete if another thread had a
1006 * transaction open and was blocking on the quota lock - a ranking
1009 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1010 * will _not_ run commit under these circumstances because handle->h_ref
1011 * is elevated. We'll still have enough credits for the tiny quotafile
1014 int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1015 struct buffer_head *bh)
1017 int dirty = buffer_dirty(bh);
1020 if (!buffer_mapped(bh) || buffer_freed(bh))
1023 * __block_write_begin() could have dirtied some buffers. Clean
1024 * the dirty bit as jbd2_journal_get_write_access() could complain
1025 * otherwise about fs integrity issues. Setting of the dirty bit
1026 * by __block_write_begin() isn't a real problem here as we clear
1027 * the bit before releasing a page lock and thus writeback cannot
1028 * ever write the buffer.
1031 clear_buffer_dirty(bh);
1032 BUFFER_TRACE(bh, "get write access");
1033 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1036 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1040 #ifdef CONFIG_FS_ENCRYPTION
1041 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1042 get_block_t *get_block)
1044 unsigned from = pos & (PAGE_SIZE - 1);
1045 unsigned to = from + len;
1046 struct inode *inode = page->mapping->host;
1047 unsigned block_start, block_end;
1050 unsigned blocksize = inode->i_sb->s_blocksize;
1052 struct buffer_head *bh, *head, *wait[2];
1056 BUG_ON(!PageLocked(page));
1057 BUG_ON(from > PAGE_SIZE);
1058 BUG_ON(to > PAGE_SIZE);
1061 if (!page_has_buffers(page))
1062 create_empty_buffers(page, blocksize, 0);
1063 head = page_buffers(page);
1064 bbits = ilog2(blocksize);
1065 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1067 for (bh = head, block_start = 0; bh != head || !block_start;
1068 block++, block_start = block_end, bh = bh->b_this_page) {
1069 block_end = block_start + blocksize;
1070 if (block_end <= from || block_start >= to) {
1071 if (PageUptodate(page)) {
1072 set_buffer_uptodate(bh);
1077 clear_buffer_new(bh);
1078 if (!buffer_mapped(bh)) {
1079 WARN_ON(bh->b_size != blocksize);
1080 err = get_block(inode, block, bh, 1);
1083 if (buffer_new(bh)) {
1084 if (PageUptodate(page)) {
1085 clear_buffer_new(bh);
1086 set_buffer_uptodate(bh);
1087 mark_buffer_dirty(bh);
1090 if (block_end > to || block_start < from)
1091 zero_user_segments(page, to, block_end,
1096 if (PageUptodate(page)) {
1097 set_buffer_uptodate(bh);
1100 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1101 !buffer_unwritten(bh) &&
1102 (block_start < from || block_end > to)) {
1103 ext4_read_bh_lock(bh, 0, false);
1104 wait[nr_wait++] = bh;
1108 * If we issued read requests, let them complete.
1110 for (i = 0; i < nr_wait; i++) {
1111 wait_on_buffer(wait[i]);
1112 if (!buffer_uptodate(wait[i]))
1115 if (unlikely(err)) {
1116 page_zero_new_buffers(page, from, to);
1117 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1118 for (i = 0; i < nr_wait; i++) {
1121 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1122 bh_offset(wait[i]));
1124 clear_buffer_uptodate(wait[i]);
1134 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1135 loff_t pos, unsigned len, unsigned flags,
1136 struct page **pagep, void **fsdata)
1138 struct inode *inode = mapping->host;
1139 int ret, needed_blocks;
1146 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1149 trace_ext4_write_begin(inode, pos, len, flags);
1151 * Reserve one block more for addition to orphan list in case
1152 * we allocate blocks but write fails for some reason
1154 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1155 index = pos >> PAGE_SHIFT;
1156 from = pos & (PAGE_SIZE - 1);
1159 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1160 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1169 * grab_cache_page_write_begin() can take a long time if the
1170 * system is thrashing due to memory pressure, or if the page
1171 * is being written back. So grab it first before we start
1172 * the transaction handle. This also allows us to allocate
1173 * the page (if needed) without using GFP_NOFS.
1176 page = grab_cache_page_write_begin(mapping, index, flags);
1182 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1183 if (IS_ERR(handle)) {
1185 return PTR_ERR(handle);
1189 if (page->mapping != mapping) {
1190 /* The page got truncated from under us */
1193 ext4_journal_stop(handle);
1196 /* In case writeback began while the page was unlocked */
1197 wait_for_stable_page(page);
1199 #ifdef CONFIG_FS_ENCRYPTION
1200 if (ext4_should_dioread_nolock(inode))
1201 ret = ext4_block_write_begin(page, pos, len,
1202 ext4_get_block_unwritten);
1204 ret = ext4_block_write_begin(page, pos, len,
1207 if (ext4_should_dioread_nolock(inode))
1208 ret = __block_write_begin(page, pos, len,
1209 ext4_get_block_unwritten);
1211 ret = __block_write_begin(page, pos, len, ext4_get_block);
1213 if (!ret && ext4_should_journal_data(inode)) {
1214 ret = ext4_walk_page_buffers(handle, inode,
1215 page_buffers(page), from, to, NULL,
1216 do_journal_get_write_access);
1220 bool extended = (pos + len > inode->i_size) &&
1221 !ext4_verity_in_progress(inode);
1225 * __block_write_begin may have instantiated a few blocks
1226 * outside i_size. Trim these off again. Don't need
1227 * i_size_read because we hold i_rwsem.
1229 * Add inode to orphan list in case we crash before
1232 if (extended && ext4_can_truncate(inode))
1233 ext4_orphan_add(handle, inode);
1235 ext4_journal_stop(handle);
1237 ext4_truncate_failed_write(inode);
1239 * If truncate failed early the inode might
1240 * still be on the orphan list; we need to
1241 * make sure the inode is removed from the
1242 * orphan list in that case.
1245 ext4_orphan_del(NULL, inode);
1248 if (ret == -ENOSPC &&
1249 ext4_should_retry_alloc(inode->i_sb, &retries))
1258 /* For write_end() in data=journal mode */
1259 static int write_end_fn(handle_t *handle, struct inode *inode,
1260 struct buffer_head *bh)
1263 if (!buffer_mapped(bh) || buffer_freed(bh))
1265 set_buffer_uptodate(bh);
1266 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1267 clear_buffer_meta(bh);
1268 clear_buffer_prio(bh);
1273 * We need to pick up the new inode size which generic_commit_write gave us
1274 * `file' can be NULL - eg, when called from page_symlink().
1276 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1277 * buffers are managed internally.
1279 static int ext4_write_end(struct file *file,
1280 struct address_space *mapping,
1281 loff_t pos, unsigned len, unsigned copied,
1282 struct page *page, void *fsdata)
1284 handle_t *handle = ext4_journal_current_handle();
1285 struct inode *inode = mapping->host;
1286 loff_t old_size = inode->i_size;
1288 int i_size_changed = 0;
1289 bool verity = ext4_verity_in_progress(inode);
1291 trace_ext4_write_end(inode, pos, len, copied);
1293 if (ext4_has_inline_data(inode))
1294 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1296 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1298 * it's important to update i_size while still holding page lock:
1299 * page writeout could otherwise come in and zero beyond i_size.
1301 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1302 * blocks are being written past EOF, so skip the i_size update.
1305 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1309 if (old_size < pos && !verity)
1310 pagecache_isize_extended(inode, old_size, pos);
1312 * Don't mark the inode dirty under page lock. First, it unnecessarily
1313 * makes the holding time of page lock longer. Second, it forces lock
1314 * ordering of page lock and transaction start for journaling
1318 ret = ext4_mark_inode_dirty(handle, inode);
1320 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1321 /* if we have allocated more blocks and copied
1322 * less. We will have blocks allocated outside
1323 * inode->i_size. So truncate them
1325 ext4_orphan_add(handle, inode);
1327 ret2 = ext4_journal_stop(handle);
1331 if (pos + len > inode->i_size && !verity) {
1332 ext4_truncate_failed_write(inode);
1334 * If truncate failed early the inode might still be
1335 * on the orphan list; we need to make sure the inode
1336 * is removed from the orphan list in that case.
1339 ext4_orphan_del(NULL, inode);
1342 return ret ? ret : copied;
1346 * This is a private version of page_zero_new_buffers() which doesn't
1347 * set the buffer to be dirty, since in data=journalled mode we need
1348 * to call ext4_handle_dirty_metadata() instead.
1350 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1351 struct inode *inode,
1353 unsigned from, unsigned to)
1355 unsigned int block_start = 0, block_end;
1356 struct buffer_head *head, *bh;
1358 bh = head = page_buffers(page);
1360 block_end = block_start + bh->b_size;
1361 if (buffer_new(bh)) {
1362 if (block_end > from && block_start < to) {
1363 if (!PageUptodate(page)) {
1364 unsigned start, size;
1366 start = max(from, block_start);
1367 size = min(to, block_end) - start;
1369 zero_user(page, start, size);
1370 write_end_fn(handle, inode, bh);
1372 clear_buffer_new(bh);
1375 block_start = block_end;
1376 bh = bh->b_this_page;
1377 } while (bh != head);
1380 static int ext4_journalled_write_end(struct file *file,
1381 struct address_space *mapping,
1382 loff_t pos, unsigned len, unsigned copied,
1383 struct page *page, void *fsdata)
1385 handle_t *handle = ext4_journal_current_handle();
1386 struct inode *inode = mapping->host;
1387 loff_t old_size = inode->i_size;
1391 int size_changed = 0;
1392 bool verity = ext4_verity_in_progress(inode);
1394 trace_ext4_journalled_write_end(inode, pos, len, copied);
1395 from = pos & (PAGE_SIZE - 1);
1398 BUG_ON(!ext4_handle_valid(handle));
1400 if (ext4_has_inline_data(inode))
1401 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1403 if (unlikely(copied < len) && !PageUptodate(page)) {
1405 ext4_journalled_zero_new_buffers(handle, inode, page, from, to);
1407 if (unlikely(copied < len))
1408 ext4_journalled_zero_new_buffers(handle, inode, page,
1410 ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
1411 from, from + copied, &partial,
1414 SetPageUptodate(page);
1417 size_changed = ext4_update_inode_size(inode, pos + copied);
1418 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1419 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1423 if (old_size < pos && !verity)
1424 pagecache_isize_extended(inode, old_size, pos);
1427 ret2 = ext4_mark_inode_dirty(handle, inode);
1432 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1433 /* if we have allocated more blocks and copied
1434 * less. We will have blocks allocated outside
1435 * inode->i_size. So truncate them
1437 ext4_orphan_add(handle, inode);
1439 ret2 = ext4_journal_stop(handle);
1442 if (pos + len > inode->i_size && !verity) {
1443 ext4_truncate_failed_write(inode);
1445 * If truncate failed early the inode might still be
1446 * on the orphan list; we need to make sure the inode
1447 * is removed from the orphan list in that case.
1450 ext4_orphan_del(NULL, inode);
1453 return ret ? ret : copied;
1457 * Reserve space for a single cluster
1459 static int ext4_da_reserve_space(struct inode *inode)
1461 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1462 struct ext4_inode_info *ei = EXT4_I(inode);
1466 * We will charge metadata quota at writeout time; this saves
1467 * us from metadata over-estimation, though we may go over by
1468 * a small amount in the end. Here we just reserve for data.
1470 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1474 spin_lock(&ei->i_block_reservation_lock);
1475 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1476 spin_unlock(&ei->i_block_reservation_lock);
1477 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1480 ei->i_reserved_data_blocks++;
1481 trace_ext4_da_reserve_space(inode);
1482 spin_unlock(&ei->i_block_reservation_lock);
1484 return 0; /* success */
1487 void ext4_da_release_space(struct inode *inode, int to_free)
1489 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1490 struct ext4_inode_info *ei = EXT4_I(inode);
1493 return; /* Nothing to release, exit */
1495 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1497 trace_ext4_da_release_space(inode, to_free);
1498 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1500 * if there aren't enough reserved blocks, then the
1501 * counter is messed up somewhere. Since this
1502 * function is called from invalidate page, it's
1503 * harmless to return without any action.
1505 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1506 "ino %lu, to_free %d with only %d reserved "
1507 "data blocks", inode->i_ino, to_free,
1508 ei->i_reserved_data_blocks);
1510 to_free = ei->i_reserved_data_blocks;
1512 ei->i_reserved_data_blocks -= to_free;
1514 /* update fs dirty data blocks counter */
1515 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1517 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1519 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1523 * Delayed allocation stuff
1526 struct mpage_da_data {
1527 struct inode *inode;
1528 struct writeback_control *wbc;
1530 pgoff_t first_page; /* The first page to write */
1531 pgoff_t next_page; /* Current page to examine */
1532 pgoff_t last_page; /* Last page to examine */
1534 * Extent to map - this can be after first_page because that can be
1535 * fully mapped. We somewhat abuse m_flags to store whether the extent
1536 * is delalloc or unwritten.
1538 struct ext4_map_blocks map;
1539 struct ext4_io_submit io_submit; /* IO submission data */
1540 unsigned int do_map:1;
1541 unsigned int scanned_until_end:1;
1544 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1549 struct pagevec pvec;
1550 struct inode *inode = mpd->inode;
1551 struct address_space *mapping = inode->i_mapping;
1553 /* This is necessary when next_page == 0. */
1554 if (mpd->first_page >= mpd->next_page)
1557 mpd->scanned_until_end = 0;
1558 index = mpd->first_page;
1559 end = mpd->next_page - 1;
1561 ext4_lblk_t start, last;
1562 start = index << (PAGE_SHIFT - inode->i_blkbits);
1563 last = end << (PAGE_SHIFT - inode->i_blkbits);
1564 ext4_es_remove_extent(inode, start, last - start + 1);
1567 pagevec_init(&pvec);
1568 while (index <= end) {
1569 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1572 for (i = 0; i < nr_pages; i++) {
1573 struct page *page = pvec.pages[i];
1575 BUG_ON(!PageLocked(page));
1576 BUG_ON(PageWriteback(page));
1578 if (page_mapped(page))
1579 clear_page_dirty_for_io(page);
1580 block_invalidatepage(page, 0, PAGE_SIZE);
1581 ClearPageUptodate(page);
1585 pagevec_release(&pvec);
1589 static void ext4_print_free_blocks(struct inode *inode)
1591 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1592 struct super_block *sb = inode->i_sb;
1593 struct ext4_inode_info *ei = EXT4_I(inode);
1595 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1596 EXT4_C2B(EXT4_SB(inode->i_sb),
1597 ext4_count_free_clusters(sb)));
1598 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1599 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1600 (long long) EXT4_C2B(EXT4_SB(sb),
1601 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1602 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1603 (long long) EXT4_C2B(EXT4_SB(sb),
1604 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1605 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1606 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1607 ei->i_reserved_data_blocks);
1611 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode,
1612 struct buffer_head *bh)
1614 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1618 * ext4_insert_delayed_block - adds a delayed block to the extents status
1619 * tree, incrementing the reserved cluster/block
1620 * count or making a pending reservation
1623 * @inode - file containing the newly added block
1624 * @lblk - logical block to be added
1626 * Returns 0 on success, negative error code on failure.
1628 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1630 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1632 bool allocated = false;
1633 bool reserved = false;
1636 * If the cluster containing lblk is shared with a delayed,
1637 * written, or unwritten extent in a bigalloc file system, it's
1638 * already been accounted for and does not need to be reserved.
1639 * A pending reservation must be made for the cluster if it's
1640 * shared with a written or unwritten extent and doesn't already
1641 * have one. Written and unwritten extents can be purged from the
1642 * extents status tree if the system is under memory pressure, so
1643 * it's necessary to examine the extent tree if a search of the
1644 * extents status tree doesn't get a match.
1646 if (sbi->s_cluster_ratio == 1) {
1647 ret = ext4_da_reserve_space(inode);
1648 if (ret != 0) /* ENOSPC */
1651 } else { /* bigalloc */
1652 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1653 if (!ext4_es_scan_clu(inode,
1654 &ext4_es_is_mapped, lblk)) {
1655 ret = ext4_clu_mapped(inode,
1656 EXT4_B2C(sbi, lblk));
1660 ret = ext4_da_reserve_space(inode);
1661 if (ret != 0) /* ENOSPC */
1673 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1674 if (ret && reserved)
1675 ext4_da_release_space(inode, 1);
1682 * This function is grabs code from the very beginning of
1683 * ext4_map_blocks, but assumes that the caller is from delayed write
1684 * time. This function looks up the requested blocks and sets the
1685 * buffer delay bit under the protection of i_data_sem.
1687 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1688 struct ext4_map_blocks *map,
1689 struct buffer_head *bh)
1691 struct extent_status es;
1693 sector_t invalid_block = ~((sector_t) 0xffff);
1694 #ifdef ES_AGGRESSIVE_TEST
1695 struct ext4_map_blocks orig_map;
1697 memcpy(&orig_map, map, sizeof(*map));
1700 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1704 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1705 (unsigned long) map->m_lblk);
1707 /* Lookup extent status tree firstly */
1708 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1709 if (ext4_es_is_hole(&es)) {
1711 down_read(&EXT4_I(inode)->i_data_sem);
1716 * Delayed extent could be allocated by fallocate.
1717 * So we need to check it.
1719 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1720 map_bh(bh, inode->i_sb, invalid_block);
1722 set_buffer_delay(bh);
1726 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1727 retval = es.es_len - (iblock - es.es_lblk);
1728 if (retval > map->m_len)
1729 retval = map->m_len;
1730 map->m_len = retval;
1731 if (ext4_es_is_written(&es))
1732 map->m_flags |= EXT4_MAP_MAPPED;
1733 else if (ext4_es_is_unwritten(&es))
1734 map->m_flags |= EXT4_MAP_UNWRITTEN;
1738 #ifdef ES_AGGRESSIVE_TEST
1739 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1745 * Try to see if we can get the block without requesting a new
1746 * file system block.
1748 down_read(&EXT4_I(inode)->i_data_sem);
1749 if (ext4_has_inline_data(inode))
1751 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1752 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1754 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1761 * XXX: __block_prepare_write() unmaps passed block,
1765 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1771 map_bh(bh, inode->i_sb, invalid_block);
1773 set_buffer_delay(bh);
1774 } else if (retval > 0) {
1776 unsigned int status;
1778 if (unlikely(retval != map->m_len)) {
1779 ext4_warning(inode->i_sb,
1780 "ES len assertion failed for inode "
1781 "%lu: retval %d != map->m_len %d",
1782 inode->i_ino, retval, map->m_len);
1786 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1787 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1788 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1789 map->m_pblk, status);
1795 up_read((&EXT4_I(inode)->i_data_sem));
1801 * This is a special get_block_t callback which is used by
1802 * ext4_da_write_begin(). It will either return mapped block or
1803 * reserve space for a single block.
1805 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1806 * We also have b_blocknr = -1 and b_bdev initialized properly
1808 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1809 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1810 * initialized properly.
1812 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1813 struct buffer_head *bh, int create)
1815 struct ext4_map_blocks map;
1818 BUG_ON(create == 0);
1819 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1821 map.m_lblk = iblock;
1825 * first, we need to know whether the block is allocated already
1826 * preallocated blocks are unmapped but should treated
1827 * the same as allocated blocks.
1829 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1833 map_bh(bh, inode->i_sb, map.m_pblk);
1834 ext4_update_bh_state(bh, map.m_flags);
1836 if (buffer_unwritten(bh)) {
1837 /* A delayed write to unwritten bh should be marked
1838 * new and mapped. Mapped ensures that we don't do
1839 * get_block multiple times when we write to the same
1840 * offset and new ensures that we do proper zero out
1841 * for partial write.
1844 set_buffer_mapped(bh);
1849 static int __ext4_journalled_writepage(struct page *page,
1852 struct address_space *mapping = page->mapping;
1853 struct inode *inode = mapping->host;
1854 handle_t *handle = NULL;
1855 int ret = 0, err = 0;
1856 int inline_data = ext4_has_inline_data(inode);
1857 struct buffer_head *inode_bh = NULL;
1860 ClearPageChecked(page);
1863 BUG_ON(page->index != 0);
1864 BUG_ON(len > ext4_get_max_inline_size(inode));
1865 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1866 if (inode_bh == NULL)
1870 * We need to release the page lock before we start the
1871 * journal, so grab a reference so the page won't disappear
1872 * out from under us.
1877 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1878 ext4_writepage_trans_blocks(inode));
1879 if (IS_ERR(handle)) {
1880 ret = PTR_ERR(handle);
1882 goto out_no_pagelock;
1884 BUG_ON(!ext4_handle_valid(handle));
1888 size = i_size_read(inode);
1889 if (page->mapping != mapping || page_offset(page) > size) {
1890 /* The page got truncated from under us */
1891 ext4_journal_stop(handle);
1897 ret = ext4_mark_inode_dirty(handle, inode);
1899 struct buffer_head *page_bufs = page_buffers(page);
1901 if (page->index == size >> PAGE_SHIFT)
1902 len = size & ~PAGE_MASK;
1906 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1907 NULL, do_journal_get_write_access);
1909 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1910 NULL, write_end_fn);
1914 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1917 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1918 err = ext4_journal_stop(handle);
1922 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1931 * Note that we don't need to start a transaction unless we're journaling data
1932 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1933 * need to file the inode to the transaction's list in ordered mode because if
1934 * we are writing back data added by write(), the inode is already there and if
1935 * we are writing back data modified via mmap(), no one guarantees in which
1936 * transaction the data will hit the disk. In case we are journaling data, we
1937 * cannot start transaction directly because transaction start ranks above page
1938 * lock so we have to do some magic.
1940 * This function can get called via...
1941 * - ext4_writepages after taking page lock (have journal handle)
1942 * - journal_submit_inode_data_buffers (no journal handle)
1943 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1944 * - grab_page_cache when doing write_begin (have journal handle)
1946 * We don't do any block allocation in this function. If we have page with
1947 * multiple blocks we need to write those buffer_heads that are mapped. This
1948 * is important for mmaped based write. So if we do with blocksize 1K
1949 * truncate(f, 1024);
1950 * a = mmap(f, 0, 4096);
1952 * truncate(f, 4096);
1953 * we have in the page first buffer_head mapped via page_mkwrite call back
1954 * but other buffer_heads would be unmapped but dirty (dirty done via the
1955 * do_wp_page). So writepage should write the first block. If we modify
1956 * the mmap area beyond 1024 we will again get a page_fault and the
1957 * page_mkwrite callback will do the block allocation and mark the
1958 * buffer_heads mapped.
1960 * We redirty the page if we have any buffer_heads that is either delay or
1961 * unwritten in the page.
1963 * We can get recursively called as show below.
1965 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1968 * But since we don't do any block allocation we should not deadlock.
1969 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1971 static int ext4_writepage(struct page *page,
1972 struct writeback_control *wbc)
1977 struct buffer_head *page_bufs = NULL;
1978 struct inode *inode = page->mapping->host;
1979 struct ext4_io_submit io_submit;
1980 bool keep_towrite = false;
1982 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
1983 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
1988 trace_ext4_writepage(page);
1989 size = i_size_read(inode);
1990 if (page->index == size >> PAGE_SHIFT &&
1991 !ext4_verity_in_progress(inode))
1992 len = size & ~PAGE_MASK;
1996 page_bufs = page_buffers(page);
1998 * We cannot do block allocation or other extent handling in this
1999 * function. If there are buffers needing that, we have to redirty
2000 * the page. But we may reach here when we do a journal commit via
2001 * journal_submit_inode_data_buffers() and in that case we must write
2002 * allocated buffers to achieve data=ordered mode guarantees.
2004 * Also, if there is only one buffer per page (the fs block
2005 * size == the page size), if one buffer needs block
2006 * allocation or needs to modify the extent tree to clear the
2007 * unwritten flag, we know that the page can't be written at
2008 * all, so we might as well refuse the write immediately.
2009 * Unfortunately if the block size != page size, we can't as
2010 * easily detect this case using ext4_walk_page_buffers(), but
2011 * for the extremely common case, this is an optimization that
2012 * skips a useless round trip through ext4_bio_write_page().
2014 if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL,
2015 ext4_bh_delay_or_unwritten)) {
2016 redirty_page_for_writepage(wbc, page);
2017 if ((current->flags & PF_MEMALLOC) ||
2018 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2020 * For memory cleaning there's no point in writing only
2021 * some buffers. So just bail out. Warn if we came here
2022 * from direct reclaim.
2024 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2029 keep_towrite = true;
2032 if (PageChecked(page) && ext4_should_journal_data(inode))
2034 * It's mmapped pagecache. Add buffers and journal it. There
2035 * doesn't seem much point in redirtying the page here.
2037 return __ext4_journalled_writepage(page, len);
2039 ext4_io_submit_init(&io_submit, wbc);
2040 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2041 if (!io_submit.io_end) {
2042 redirty_page_for_writepage(wbc, page);
2046 ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite);
2047 ext4_io_submit(&io_submit);
2048 /* Drop io_end reference we got from init */
2049 ext4_put_io_end_defer(io_submit.io_end);
2053 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2059 BUG_ON(page->index != mpd->first_page);
2060 clear_page_dirty_for_io(page);
2062 * We have to be very careful here! Nothing protects writeback path
2063 * against i_size changes and the page can be writeably mapped into
2064 * page tables. So an application can be growing i_size and writing
2065 * data through mmap while writeback runs. clear_page_dirty_for_io()
2066 * write-protects our page in page tables and the page cannot get
2067 * written to again until we release page lock. So only after
2068 * clear_page_dirty_for_io() we are safe to sample i_size for
2069 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2070 * on the barrier provided by TestClearPageDirty in
2071 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2072 * after page tables are updated.
2074 size = i_size_read(mpd->inode);
2075 if (page->index == size >> PAGE_SHIFT &&
2076 !ext4_verity_in_progress(mpd->inode))
2077 len = size & ~PAGE_MASK;
2080 err = ext4_bio_write_page(&mpd->io_submit, page, len, false);
2082 mpd->wbc->nr_to_write--;
2088 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2091 * mballoc gives us at most this number of blocks...
2092 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2093 * The rest of mballoc seems to handle chunks up to full group size.
2095 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2098 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2100 * @mpd - extent of blocks
2101 * @lblk - logical number of the block in the file
2102 * @bh - buffer head we want to add to the extent
2104 * The function is used to collect contig. blocks in the same state. If the
2105 * buffer doesn't require mapping for writeback and we haven't started the
2106 * extent of buffers to map yet, the function returns 'true' immediately - the
2107 * caller can write the buffer right away. Otherwise the function returns true
2108 * if the block has been added to the extent, false if the block couldn't be
2111 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2112 struct buffer_head *bh)
2114 struct ext4_map_blocks *map = &mpd->map;
2116 /* Buffer that doesn't need mapping for writeback? */
2117 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2118 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2119 /* So far no extent to map => we write the buffer right away */
2120 if (map->m_len == 0)
2125 /* First block in the extent? */
2126 if (map->m_len == 0) {
2127 /* We cannot map unless handle is started... */
2132 map->m_flags = bh->b_state & BH_FLAGS;
2136 /* Don't go larger than mballoc is willing to allocate */
2137 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2140 /* Can we merge the block to our big extent? */
2141 if (lblk == map->m_lblk + map->m_len &&
2142 (bh->b_state & BH_FLAGS) == map->m_flags) {
2150 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2152 * @mpd - extent of blocks for mapping
2153 * @head - the first buffer in the page
2154 * @bh - buffer we should start processing from
2155 * @lblk - logical number of the block in the file corresponding to @bh
2157 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2158 * the page for IO if all buffers in this page were mapped and there's no
2159 * accumulated extent of buffers to map or add buffers in the page to the
2160 * extent of buffers to map. The function returns 1 if the caller can continue
2161 * by processing the next page, 0 if it should stop adding buffers to the
2162 * extent to map because we cannot extend it anymore. It can also return value
2163 * < 0 in case of error during IO submission.
2165 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2166 struct buffer_head *head,
2167 struct buffer_head *bh,
2170 struct inode *inode = mpd->inode;
2172 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2173 >> inode->i_blkbits;
2175 if (ext4_verity_in_progress(inode))
2176 blocks = EXT_MAX_BLOCKS;
2179 BUG_ON(buffer_locked(bh));
2181 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2182 /* Found extent to map? */
2185 /* Buffer needs mapping and handle is not started? */
2188 /* Everything mapped so far and we hit EOF */
2191 } while (lblk++, (bh = bh->b_this_page) != head);
2192 /* So far everything mapped? Submit the page for IO. */
2193 if (mpd->map.m_len == 0) {
2194 err = mpage_submit_page(mpd, head->b_page);
2198 if (lblk >= blocks) {
2199 mpd->scanned_until_end = 1;
2206 * mpage_process_page - update page buffers corresponding to changed extent and
2207 * may submit fully mapped page for IO
2209 * @mpd - description of extent to map, on return next extent to map
2210 * @m_lblk - logical block mapping.
2211 * @m_pblk - corresponding physical mapping.
2212 * @map_bh - determines on return whether this page requires any further
2214 * Scan given page buffers corresponding to changed extent and update buffer
2215 * state according to new extent state.
2216 * We map delalloc buffers to their physical location, clear unwritten bits.
2217 * If the given page is not fully mapped, we update @map to the next extent in
2218 * the given page that needs mapping & return @map_bh as true.
2220 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2221 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2224 struct buffer_head *head, *bh;
2225 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2226 ext4_lblk_t lblk = *m_lblk;
2227 ext4_fsblk_t pblock = *m_pblk;
2229 int blkbits = mpd->inode->i_blkbits;
2230 ssize_t io_end_size = 0;
2231 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2233 bh = head = page_buffers(page);
2235 if (lblk < mpd->map.m_lblk)
2237 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2239 * Buffer after end of mapped extent.
2240 * Find next buffer in the page to map.
2243 mpd->map.m_flags = 0;
2244 io_end_vec->size += io_end_size;
2246 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2249 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2250 io_end_vec = ext4_alloc_io_end_vec(io_end);
2251 if (IS_ERR(io_end_vec)) {
2252 err = PTR_ERR(io_end_vec);
2255 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2260 if (buffer_delay(bh)) {
2261 clear_buffer_delay(bh);
2262 bh->b_blocknr = pblock++;
2264 clear_buffer_unwritten(bh);
2265 io_end_size += (1 << blkbits);
2266 } while (lblk++, (bh = bh->b_this_page) != head);
2268 io_end_vec->size += io_end_size;
2277 * mpage_map_buffers - update buffers corresponding to changed extent and
2278 * submit fully mapped pages for IO
2280 * @mpd - description of extent to map, on return next extent to map
2282 * Scan buffers corresponding to changed extent (we expect corresponding pages
2283 * to be already locked) and update buffer state according to new extent state.
2284 * We map delalloc buffers to their physical location, clear unwritten bits,
2285 * and mark buffers as uninit when we perform writes to unwritten extents
2286 * and do extent conversion after IO is finished. If the last page is not fully
2287 * mapped, we update @map to the next extent in the last page that needs
2288 * mapping. Otherwise we submit the page for IO.
2290 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2292 struct pagevec pvec;
2294 struct inode *inode = mpd->inode;
2295 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2298 ext4_fsblk_t pblock;
2300 bool map_bh = false;
2302 start = mpd->map.m_lblk >> bpp_bits;
2303 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2304 lblk = start << bpp_bits;
2305 pblock = mpd->map.m_pblk;
2307 pagevec_init(&pvec);
2308 while (start <= end) {
2309 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2313 for (i = 0; i < nr_pages; i++) {
2314 struct page *page = pvec.pages[i];
2316 err = mpage_process_page(mpd, page, &lblk, &pblock,
2319 * If map_bh is true, means page may require further bh
2320 * mapping, or maybe the page was submitted for IO.
2321 * So we return to call further extent mapping.
2323 if (err < 0 || map_bh)
2325 /* Page fully mapped - let IO run! */
2326 err = mpage_submit_page(mpd, page);
2330 pagevec_release(&pvec);
2332 /* Extent fully mapped and matches with page boundary. We are done. */
2334 mpd->map.m_flags = 0;
2337 pagevec_release(&pvec);
2341 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2343 struct inode *inode = mpd->inode;
2344 struct ext4_map_blocks *map = &mpd->map;
2345 int get_blocks_flags;
2346 int err, dioread_nolock;
2348 trace_ext4_da_write_pages_extent(inode, map);
2350 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2351 * to convert an unwritten extent to be initialized (in the case
2352 * where we have written into one or more preallocated blocks). It is
2353 * possible that we're going to need more metadata blocks than
2354 * previously reserved. However we must not fail because we're in
2355 * writeback and there is nothing we can do about it so it might result
2356 * in data loss. So use reserved blocks to allocate metadata if
2359 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2360 * the blocks in question are delalloc blocks. This indicates
2361 * that the blocks and quotas has already been checked when
2362 * the data was copied into the page cache.
2364 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2365 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2366 EXT4_GET_BLOCKS_IO_SUBMIT;
2367 dioread_nolock = ext4_should_dioread_nolock(inode);
2369 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2370 if (map->m_flags & BIT(BH_Delay))
2371 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2373 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2376 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2377 if (!mpd->io_submit.io_end->handle &&
2378 ext4_handle_valid(handle)) {
2379 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2380 handle->h_rsv_handle = NULL;
2382 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2385 BUG_ON(map->m_len == 0);
2390 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2391 * mpd->len and submit pages underlying it for IO
2393 * @handle - handle for journal operations
2394 * @mpd - extent to map
2395 * @give_up_on_write - we set this to true iff there is a fatal error and there
2396 * is no hope of writing the data. The caller should discard
2397 * dirty pages to avoid infinite loops.
2399 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2400 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2401 * them to initialized or split the described range from larger unwritten
2402 * extent. Note that we need not map all the described range since allocation
2403 * can return less blocks or the range is covered by more unwritten extents. We
2404 * cannot map more because we are limited by reserved transaction credits. On
2405 * the other hand we always make sure that the last touched page is fully
2406 * mapped so that it can be written out (and thus forward progress is
2407 * guaranteed). After mapping we submit all mapped pages for IO.
2409 static int mpage_map_and_submit_extent(handle_t *handle,
2410 struct mpage_da_data *mpd,
2411 bool *give_up_on_write)
2413 struct inode *inode = mpd->inode;
2414 struct ext4_map_blocks *map = &mpd->map;
2418 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2419 struct ext4_io_end_vec *io_end_vec;
2421 io_end_vec = ext4_alloc_io_end_vec(io_end);
2422 if (IS_ERR(io_end_vec))
2423 return PTR_ERR(io_end_vec);
2424 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2426 err = mpage_map_one_extent(handle, mpd);
2428 struct super_block *sb = inode->i_sb;
2430 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2431 ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2432 goto invalidate_dirty_pages;
2434 * Let the uper layers retry transient errors.
2435 * In the case of ENOSPC, if ext4_count_free_blocks()
2436 * is non-zero, a commit should free up blocks.
2438 if ((err == -ENOMEM) ||
2439 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2441 goto update_disksize;
2444 ext4_msg(sb, KERN_CRIT,
2445 "Delayed block allocation failed for "
2446 "inode %lu at logical offset %llu with"
2447 " max blocks %u with error %d",
2449 (unsigned long long)map->m_lblk,
2450 (unsigned)map->m_len, -err);
2451 ext4_msg(sb, KERN_CRIT,
2452 "This should not happen!! Data will "
2455 ext4_print_free_blocks(inode);
2456 invalidate_dirty_pages:
2457 *give_up_on_write = true;
2462 * Update buffer state, submit mapped pages, and get us new
2465 err = mpage_map_and_submit_buffers(mpd);
2467 goto update_disksize;
2468 } while (map->m_len);
2472 * Update on-disk size after IO is submitted. Races with
2473 * truncate are avoided by checking i_size under i_data_sem.
2475 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2476 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2480 down_write(&EXT4_I(inode)->i_data_sem);
2481 i_size = i_size_read(inode);
2482 if (disksize > i_size)
2484 if (disksize > EXT4_I(inode)->i_disksize)
2485 EXT4_I(inode)->i_disksize = disksize;
2486 up_write(&EXT4_I(inode)->i_data_sem);
2487 err2 = ext4_mark_inode_dirty(handle, inode);
2489 ext4_error_err(inode->i_sb, -err2,
2490 "Failed to mark inode %lu dirty",
2500 * Calculate the total number of credits to reserve for one writepages
2501 * iteration. This is called from ext4_writepages(). We map an extent of
2502 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2503 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2504 * bpp - 1 blocks in bpp different extents.
2506 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2508 int bpp = ext4_journal_blocks_per_page(inode);
2510 return ext4_meta_trans_blocks(inode,
2511 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2515 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2516 * and underlying extent to map
2518 * @mpd - where to look for pages
2520 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2521 * IO immediately. When we find a page which isn't mapped we start accumulating
2522 * extent of buffers underlying these pages that needs mapping (formed by
2523 * either delayed or unwritten buffers). We also lock the pages containing
2524 * these buffers. The extent found is returned in @mpd structure (starting at
2525 * mpd->lblk with length mpd->len blocks).
2527 * Note that this function can attach bios to one io_end structure which are
2528 * neither logically nor physically contiguous. Although it may seem as an
2529 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2530 * case as we need to track IO to all buffers underlying a page in one io_end.
2532 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2534 struct address_space *mapping = mpd->inode->i_mapping;
2535 struct pagevec pvec;
2536 unsigned int nr_pages;
2537 long left = mpd->wbc->nr_to_write;
2538 pgoff_t index = mpd->first_page;
2539 pgoff_t end = mpd->last_page;
2542 int blkbits = mpd->inode->i_blkbits;
2544 struct buffer_head *head;
2546 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2547 tag = PAGECACHE_TAG_TOWRITE;
2549 tag = PAGECACHE_TAG_DIRTY;
2551 pagevec_init(&pvec);
2553 mpd->next_page = index;
2554 while (index <= end) {
2555 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2560 for (i = 0; i < nr_pages; i++) {
2561 struct page *page = pvec.pages[i];
2564 * Accumulated enough dirty pages? This doesn't apply
2565 * to WB_SYNC_ALL mode. For integrity sync we have to
2566 * keep going because someone may be concurrently
2567 * dirtying pages, and we might have synced a lot of
2568 * newly appeared dirty pages, but have not synced all
2569 * of the old dirty pages.
2571 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2574 /* If we can't merge this page, we are done. */
2575 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2580 * If the page is no longer dirty, or its mapping no
2581 * longer corresponds to inode we are writing (which
2582 * means it has been truncated or invalidated), or the
2583 * page is already under writeback and we are not doing
2584 * a data integrity writeback, skip the page
2586 if (!PageDirty(page) ||
2587 (PageWriteback(page) &&
2588 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2589 unlikely(page->mapping != mapping)) {
2594 wait_on_page_writeback(page);
2595 BUG_ON(PageWriteback(page));
2597 if (mpd->map.m_len == 0)
2598 mpd->first_page = page->index;
2599 mpd->next_page = page->index + 1;
2600 /* Add all dirty buffers to mpd */
2601 lblk = ((ext4_lblk_t)page->index) <<
2602 (PAGE_SHIFT - blkbits);
2603 head = page_buffers(page);
2604 err = mpage_process_page_bufs(mpd, head, head, lblk);
2610 pagevec_release(&pvec);
2613 mpd->scanned_until_end = 1;
2616 pagevec_release(&pvec);
2620 static int ext4_writepages(struct address_space *mapping,
2621 struct writeback_control *wbc)
2623 pgoff_t writeback_index = 0;
2624 long nr_to_write = wbc->nr_to_write;
2625 int range_whole = 0;
2627 handle_t *handle = NULL;
2628 struct mpage_da_data mpd;
2629 struct inode *inode = mapping->host;
2630 int needed_blocks, rsv_blocks = 0, ret = 0;
2631 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2632 struct blk_plug plug;
2633 bool give_up_on_write = false;
2635 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2638 percpu_down_read(&sbi->s_writepages_rwsem);
2639 trace_ext4_writepages(inode, wbc);
2642 * No pages to write? This is mainly a kludge to avoid starting
2643 * a transaction for special inodes like journal inode on last iput()
2644 * because that could violate lock ordering on umount
2646 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2647 goto out_writepages;
2649 if (ext4_should_journal_data(inode)) {
2650 ret = generic_writepages(mapping, wbc);
2651 goto out_writepages;
2655 * If the filesystem has aborted, it is read-only, so return
2656 * right away instead of dumping stack traces later on that
2657 * will obscure the real source of the problem. We test
2658 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2659 * the latter could be true if the filesystem is mounted
2660 * read-only, and in that case, ext4_writepages should
2661 * *never* be called, so if that ever happens, we would want
2664 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2665 ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2667 goto out_writepages;
2671 * If we have inline data and arrive here, it means that
2672 * we will soon create the block for the 1st page, so
2673 * we'd better clear the inline data here.
2675 if (ext4_has_inline_data(inode)) {
2676 /* Just inode will be modified... */
2677 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2678 if (IS_ERR(handle)) {
2679 ret = PTR_ERR(handle);
2680 goto out_writepages;
2682 BUG_ON(ext4_test_inode_state(inode,
2683 EXT4_STATE_MAY_INLINE_DATA));
2684 ext4_destroy_inline_data(handle, inode);
2685 ext4_journal_stop(handle);
2688 if (ext4_should_dioread_nolock(inode)) {
2690 * We may need to convert up to one extent per block in
2691 * the page and we may dirty the inode.
2693 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2694 PAGE_SIZE >> inode->i_blkbits);
2697 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2700 if (wbc->range_cyclic) {
2701 writeback_index = mapping->writeback_index;
2702 if (writeback_index)
2704 mpd.first_page = writeback_index;
2707 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2708 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2713 ext4_io_submit_init(&mpd.io_submit, wbc);
2715 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2716 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2717 blk_start_plug(&plug);
2720 * First writeback pages that don't need mapping - we can avoid
2721 * starting a transaction unnecessarily and also avoid being blocked
2722 * in the block layer on device congestion while having transaction
2726 mpd.scanned_until_end = 0;
2727 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2728 if (!mpd.io_submit.io_end) {
2732 ret = mpage_prepare_extent_to_map(&mpd);
2733 /* Unlock pages we didn't use */
2734 mpage_release_unused_pages(&mpd, false);
2735 /* Submit prepared bio */
2736 ext4_io_submit(&mpd.io_submit);
2737 ext4_put_io_end_defer(mpd.io_submit.io_end);
2738 mpd.io_submit.io_end = NULL;
2742 while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2743 /* For each extent of pages we use new io_end */
2744 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2745 if (!mpd.io_submit.io_end) {
2751 * We have two constraints: We find one extent to map and we
2752 * must always write out whole page (makes a difference when
2753 * blocksize < pagesize) so that we don't block on IO when we
2754 * try to write out the rest of the page. Journalled mode is
2755 * not supported by delalloc.
2757 BUG_ON(ext4_should_journal_data(inode));
2758 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2760 /* start a new transaction */
2761 handle = ext4_journal_start_with_reserve(inode,
2762 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2763 if (IS_ERR(handle)) {
2764 ret = PTR_ERR(handle);
2765 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2766 "%ld pages, ino %lu; err %d", __func__,
2767 wbc->nr_to_write, inode->i_ino, ret);
2768 /* Release allocated io_end */
2769 ext4_put_io_end(mpd.io_submit.io_end);
2770 mpd.io_submit.io_end = NULL;
2775 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2776 ret = mpage_prepare_extent_to_map(&mpd);
2777 if (!ret && mpd.map.m_len)
2778 ret = mpage_map_and_submit_extent(handle, &mpd,
2781 * Caution: If the handle is synchronous,
2782 * ext4_journal_stop() can wait for transaction commit
2783 * to finish which may depend on writeback of pages to
2784 * complete or on page lock to be released. In that
2785 * case, we have to wait until after we have
2786 * submitted all the IO, released page locks we hold,
2787 * and dropped io_end reference (for extent conversion
2788 * to be able to complete) before stopping the handle.
2790 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2791 ext4_journal_stop(handle);
2795 /* Unlock pages we didn't use */
2796 mpage_release_unused_pages(&mpd, give_up_on_write);
2797 /* Submit prepared bio */
2798 ext4_io_submit(&mpd.io_submit);
2801 * Drop our io_end reference we got from init. We have
2802 * to be careful and use deferred io_end finishing if
2803 * we are still holding the transaction as we can
2804 * release the last reference to io_end which may end
2805 * up doing unwritten extent conversion.
2808 ext4_put_io_end_defer(mpd.io_submit.io_end);
2809 ext4_journal_stop(handle);
2811 ext4_put_io_end(mpd.io_submit.io_end);
2812 mpd.io_submit.io_end = NULL;
2814 if (ret == -ENOSPC && sbi->s_journal) {
2816 * Commit the transaction which would
2817 * free blocks released in the transaction
2820 jbd2_journal_force_commit_nested(sbi->s_journal);
2824 /* Fatal error - ENOMEM, EIO... */
2829 blk_finish_plug(&plug);
2830 if (!ret && !cycled && wbc->nr_to_write > 0) {
2832 mpd.last_page = writeback_index - 1;
2838 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2840 * Set the writeback_index so that range_cyclic
2841 * mode will write it back later
2843 mapping->writeback_index = mpd.first_page;
2846 trace_ext4_writepages_result(inode, wbc, ret,
2847 nr_to_write - wbc->nr_to_write);
2848 percpu_up_read(&sbi->s_writepages_rwsem);
2852 static int ext4_dax_writepages(struct address_space *mapping,
2853 struct writeback_control *wbc)
2856 long nr_to_write = wbc->nr_to_write;
2857 struct inode *inode = mapping->host;
2858 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2860 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2863 percpu_down_read(&sbi->s_writepages_rwsem);
2864 trace_ext4_writepages(inode, wbc);
2866 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2867 trace_ext4_writepages_result(inode, wbc, ret,
2868 nr_to_write - wbc->nr_to_write);
2869 percpu_up_read(&sbi->s_writepages_rwsem);
2873 static int ext4_nonda_switch(struct super_block *sb)
2875 s64 free_clusters, dirty_clusters;
2876 struct ext4_sb_info *sbi = EXT4_SB(sb);
2879 * switch to non delalloc mode if we are running low
2880 * on free block. The free block accounting via percpu
2881 * counters can get slightly wrong with percpu_counter_batch getting
2882 * accumulated on each CPU without updating global counters
2883 * Delalloc need an accurate free block accounting. So switch
2884 * to non delalloc when we are near to error range.
2887 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2889 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2891 * Start pushing delalloc when 1/2 of free blocks are dirty.
2893 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2894 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2896 if (2 * free_clusters < 3 * dirty_clusters ||
2897 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2899 * free block count is less than 150% of dirty blocks
2900 * or free blocks is less than watermark
2907 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2908 loff_t pos, unsigned len, unsigned flags,
2909 struct page **pagep, void **fsdata)
2911 int ret, retries = 0;
2914 struct inode *inode = mapping->host;
2916 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2919 index = pos >> PAGE_SHIFT;
2921 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2922 ext4_verity_in_progress(inode)) {
2923 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2924 return ext4_write_begin(file, mapping, pos,
2925 len, flags, pagep, fsdata);
2927 *fsdata = (void *)0;
2928 trace_ext4_da_write_begin(inode, pos, len, flags);
2930 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2931 ret = ext4_da_write_inline_data_begin(mapping, inode,
2941 page = grab_cache_page_write_begin(mapping, index, flags);
2945 /* In case writeback began while the page was unlocked */
2946 wait_for_stable_page(page);
2948 #ifdef CONFIG_FS_ENCRYPTION
2949 ret = ext4_block_write_begin(page, pos, len,
2950 ext4_da_get_block_prep);
2952 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2958 * block_write_begin may have instantiated a few blocks
2959 * outside i_size. Trim these off again. Don't need
2960 * i_size_read because we hold inode lock.
2962 if (pos + len > inode->i_size)
2963 ext4_truncate_failed_write(inode);
2965 if (ret == -ENOSPC &&
2966 ext4_should_retry_alloc(inode->i_sb, &retries))
2976 * Check if we should update i_disksize
2977 * when write to the end of file but not require block allocation
2979 static int ext4_da_should_update_i_disksize(struct page *page,
2980 unsigned long offset)
2982 struct buffer_head *bh;
2983 struct inode *inode = page->mapping->host;
2987 bh = page_buffers(page);
2988 idx = offset >> inode->i_blkbits;
2990 for (i = 0; i < idx; i++)
2991 bh = bh->b_this_page;
2993 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2998 static int ext4_da_write_end(struct file *file,
2999 struct address_space *mapping,
3000 loff_t pos, unsigned len, unsigned copied,
3001 struct page *page, void *fsdata)
3003 struct inode *inode = mapping->host;
3005 unsigned long start, end;
3006 int write_mode = (int)(unsigned long)fsdata;
3008 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3009 return ext4_write_end(file, mapping, pos,
3010 len, copied, page, fsdata);
3012 trace_ext4_da_write_end(inode, pos, len, copied);
3014 if (write_mode != CONVERT_INLINE_DATA &&
3015 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3016 ext4_has_inline_data(inode))
3017 return ext4_write_inline_data_end(inode, pos, len, copied, page);
3019 start = pos & (PAGE_SIZE - 1);
3020 end = start + copied - 1;
3023 * Since we are holding inode lock, we are sure i_disksize <=
3024 * i_size. We also know that if i_disksize < i_size, there are
3025 * delalloc writes pending in the range upto i_size. If the end of
3026 * the current write is <= i_size, there's no need to touch
3027 * i_disksize since writeback will push i_disksize upto i_size
3028 * eventually. If the end of the current write is > i_size and
3029 * inside an allocated block (ext4_da_should_update_i_disksize()
3030 * check), we need to update i_disksize here as neither
3031 * ext4_writepage() nor certain ext4_writepages() paths not
3032 * allocating blocks update i_disksize.
3034 * Note that we defer inode dirtying to generic_write_end() /
3035 * ext4_da_write_inline_data_end().
3037 new_i_size = pos + copied;
3038 if (copied && new_i_size > inode->i_size &&
3039 ext4_da_should_update_i_disksize(page, end))
3040 ext4_update_i_disksize(inode, new_i_size);
3042 return generic_write_end(file, mapping, pos, len, copied, page, fsdata);
3046 * Force all delayed allocation blocks to be allocated for a given inode.
3048 int ext4_alloc_da_blocks(struct inode *inode)
3050 trace_ext4_alloc_da_blocks(inode);
3052 if (!EXT4_I(inode)->i_reserved_data_blocks)
3056 * We do something simple for now. The filemap_flush() will
3057 * also start triggering a write of the data blocks, which is
3058 * not strictly speaking necessary (and for users of
3059 * laptop_mode, not even desirable). However, to do otherwise
3060 * would require replicating code paths in:
3062 * ext4_writepages() ->
3063 * write_cache_pages() ---> (via passed in callback function)
3064 * __mpage_da_writepage() -->
3065 * mpage_add_bh_to_extent()
3066 * mpage_da_map_blocks()
3068 * The problem is that write_cache_pages(), located in
3069 * mm/page-writeback.c, marks pages clean in preparation for
3070 * doing I/O, which is not desirable if we're not planning on
3073 * We could call write_cache_pages(), and then redirty all of
3074 * the pages by calling redirty_page_for_writepage() but that
3075 * would be ugly in the extreme. So instead we would need to
3076 * replicate parts of the code in the above functions,
3077 * simplifying them because we wouldn't actually intend to
3078 * write out the pages, but rather only collect contiguous
3079 * logical block extents, call the multi-block allocator, and
3080 * then update the buffer heads with the block allocations.
3082 * For now, though, we'll cheat by calling filemap_flush(),
3083 * which will map the blocks, and start the I/O, but not
3084 * actually wait for the I/O to complete.
3086 return filemap_flush(inode->i_mapping);
3090 * bmap() is special. It gets used by applications such as lilo and by
3091 * the swapper to find the on-disk block of a specific piece of data.
3093 * Naturally, this is dangerous if the block concerned is still in the
3094 * journal. If somebody makes a swapfile on an ext4 data-journaling
3095 * filesystem and enables swap, then they may get a nasty shock when the
3096 * data getting swapped to that swapfile suddenly gets overwritten by
3097 * the original zero's written out previously to the journal and
3098 * awaiting writeback in the kernel's buffer cache.
3100 * So, if we see any bmap calls here on a modified, data-journaled file,
3101 * take extra steps to flush any blocks which might be in the cache.
3103 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3105 struct inode *inode = mapping->host;
3110 * We can get here for an inline file via the FIBMAP ioctl
3112 if (ext4_has_inline_data(inode))
3115 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3116 test_opt(inode->i_sb, DELALLOC)) {
3118 * With delalloc we want to sync the file
3119 * so that we can make sure we allocate
3122 filemap_write_and_wait(mapping);
3125 if (EXT4_JOURNAL(inode) &&
3126 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3128 * This is a REALLY heavyweight approach, but the use of
3129 * bmap on dirty files is expected to be extremely rare:
3130 * only if we run lilo or swapon on a freshly made file
3131 * do we expect this to happen.
3133 * (bmap requires CAP_SYS_RAWIO so this does not
3134 * represent an unprivileged user DOS attack --- we'd be
3135 * in trouble if mortal users could trigger this path at
3138 * NB. EXT4_STATE_JDATA is not set on files other than
3139 * regular files. If somebody wants to bmap a directory
3140 * or symlink and gets confused because the buffer
3141 * hasn't yet been flushed to disk, they deserve
3142 * everything they get.
3145 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3146 journal = EXT4_JOURNAL(inode);
3147 jbd2_journal_lock_updates(journal);
3148 err = jbd2_journal_flush(journal, 0);
3149 jbd2_journal_unlock_updates(journal);
3155 return iomap_bmap(mapping, block, &ext4_iomap_ops);
3158 static int ext4_readpage(struct file *file, struct page *page)
3161 struct inode *inode = page->mapping->host;
3163 trace_ext4_readpage(page);
3165 if (ext4_has_inline_data(inode))
3166 ret = ext4_readpage_inline(inode, page);
3169 return ext4_mpage_readpages(inode, NULL, page);
3174 static void ext4_readahead(struct readahead_control *rac)
3176 struct inode *inode = rac->mapping->host;
3178 /* If the file has inline data, no need to do readahead. */
3179 if (ext4_has_inline_data(inode))
3182 ext4_mpage_readpages(inode, rac, NULL);
3185 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3186 unsigned int length)
3188 trace_ext4_invalidatepage(page, offset, length);
3190 /* No journalling happens on data buffers when this function is used */
3191 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3193 block_invalidatepage(page, offset, length);
3196 static int __ext4_journalled_invalidatepage(struct page *page,
3197 unsigned int offset,
3198 unsigned int length)
3200 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3202 trace_ext4_journalled_invalidatepage(page, offset, length);
3205 * If it's a full truncate we just forget about the pending dirtying
3207 if (offset == 0 && length == PAGE_SIZE)
3208 ClearPageChecked(page);
3210 return jbd2_journal_invalidatepage(journal, page, offset, length);
3213 /* Wrapper for aops... */
3214 static void ext4_journalled_invalidatepage(struct page *page,
3215 unsigned int offset,
3216 unsigned int length)
3218 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3221 static int ext4_releasepage(struct page *page, gfp_t wait)
3223 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3225 trace_ext4_releasepage(page);
3227 /* Page has dirty journalled data -> cannot release */
3228 if (PageChecked(page))
3231 return jbd2_journal_try_to_free_buffers(journal, page);
3233 return try_to_free_buffers(page);
3236 static bool ext4_inode_datasync_dirty(struct inode *inode)
3238 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3241 if (jbd2_transaction_committed(journal,
3242 EXT4_I(inode)->i_datasync_tid))
3244 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3245 return !list_empty(&EXT4_I(inode)->i_fc_list);
3249 /* Any metadata buffers to write? */
3250 if (!list_empty(&inode->i_mapping->private_list))
3252 return inode->i_state & I_DIRTY_DATASYNC;
3255 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3256 struct ext4_map_blocks *map, loff_t offset,
3257 loff_t length, unsigned int flags)
3259 u8 blkbits = inode->i_blkbits;
3262 * Writes that span EOF might trigger an I/O size update on completion,
3263 * so consider them to be dirty for the purpose of O_DSYNC, even if
3264 * there is no other metadata changes being made or are pending.
3267 if (ext4_inode_datasync_dirty(inode) ||
3268 offset + length > i_size_read(inode))
3269 iomap->flags |= IOMAP_F_DIRTY;
3271 if (map->m_flags & EXT4_MAP_NEW)
3272 iomap->flags |= IOMAP_F_NEW;
3274 if (flags & IOMAP_DAX)
3275 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3277 iomap->bdev = inode->i_sb->s_bdev;
3278 iomap->offset = (u64) map->m_lblk << blkbits;
3279 iomap->length = (u64) map->m_len << blkbits;
3281 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3282 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3283 iomap->flags |= IOMAP_F_MERGED;
3286 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3287 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3288 * set. In order for any allocated unwritten extents to be converted
3289 * into written extents correctly within the ->end_io() handler, we
3290 * need to ensure that the iomap->type is set appropriately. Hence, the
3291 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3294 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3295 iomap->type = IOMAP_UNWRITTEN;
3296 iomap->addr = (u64) map->m_pblk << blkbits;
3297 if (flags & IOMAP_DAX)
3298 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3299 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3300 iomap->type = IOMAP_MAPPED;
3301 iomap->addr = (u64) map->m_pblk << blkbits;
3302 if (flags & IOMAP_DAX)
3303 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3305 iomap->type = IOMAP_HOLE;
3306 iomap->addr = IOMAP_NULL_ADDR;
3310 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3314 u8 blkbits = inode->i_blkbits;
3315 int ret, dio_credits, m_flags = 0, retries = 0;
3318 * Trim the mapping request to the maximum value that we can map at
3319 * once for direct I/O.
3321 if (map->m_len > DIO_MAX_BLOCKS)
3322 map->m_len = DIO_MAX_BLOCKS;
3323 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3327 * Either we allocate blocks and then don't get an unwritten extent, so
3328 * in that case we have reserved enough credits. Or, the blocks are
3329 * already allocated and unwritten. In that case, the extent conversion
3330 * fits into the credits as well.
3332 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3334 return PTR_ERR(handle);
3337 * DAX and direct I/O are the only two operations that are currently
3338 * supported with IOMAP_WRITE.
3340 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3341 if (flags & IOMAP_DAX)
3342 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3344 * We use i_size instead of i_disksize here because delalloc writeback
3345 * can complete at any point during the I/O and subsequently push the
3346 * i_disksize out to i_size. This could be beyond where direct I/O is
3347 * happening and thus expose allocated blocks to direct I/O reads.
3349 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3350 m_flags = EXT4_GET_BLOCKS_CREATE;
3351 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3352 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3354 ret = ext4_map_blocks(handle, inode, map, m_flags);
3357 * We cannot fill holes in indirect tree based inodes as that could
3358 * expose stale data in the case of a crash. Use the magic error code
3359 * to fallback to buffered I/O.
3361 if (!m_flags && !ret)
3364 ext4_journal_stop(handle);
3365 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3372 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3373 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3376 struct ext4_map_blocks map;
3377 u8 blkbits = inode->i_blkbits;
3379 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3382 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3386 * Calculate the first and last logical blocks respectively.
3388 map.m_lblk = offset >> blkbits;
3389 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3390 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3392 if (flags & IOMAP_WRITE) {
3394 * We check here if the blocks are already allocated, then we
3395 * don't need to start a journal txn and we can directly return
3396 * the mapping information. This could boost performance
3397 * especially in multi-threaded overwrite requests.
3399 if (offset + length <= i_size_read(inode)) {
3400 ret = ext4_map_blocks(NULL, inode, &map, 0);
3401 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3404 ret = ext4_iomap_alloc(inode, &map, flags);
3406 ret = ext4_map_blocks(NULL, inode, &map, 0);
3413 * When inline encryption is enabled, sometimes I/O to an encrypted file
3414 * has to be broken up to guarantee DUN contiguity. Handle this by
3415 * limiting the length of the mapping returned.
3417 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3419 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3424 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3425 loff_t length, unsigned flags, struct iomap *iomap,
3426 struct iomap *srcmap)
3431 * Even for writes we don't need to allocate blocks, so just pretend
3432 * we are reading to save overhead of starting a transaction.
3434 flags &= ~IOMAP_WRITE;
3435 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3436 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3440 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3441 ssize_t written, unsigned flags, struct iomap *iomap)
3444 * Check to see whether an error occurred while writing out the data to
3445 * the allocated blocks. If so, return the magic error code so that we
3446 * fallback to buffered I/O and attempt to complete the remainder of
3447 * the I/O. Any blocks that may have been allocated in preparation for
3448 * the direct I/O will be reused during buffered I/O.
3450 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3456 const struct iomap_ops ext4_iomap_ops = {
3457 .iomap_begin = ext4_iomap_begin,
3458 .iomap_end = ext4_iomap_end,
3461 const struct iomap_ops ext4_iomap_overwrite_ops = {
3462 .iomap_begin = ext4_iomap_overwrite_begin,
3463 .iomap_end = ext4_iomap_end,
3466 static bool ext4_iomap_is_delalloc(struct inode *inode,
3467 struct ext4_map_blocks *map)
3469 struct extent_status es;
3470 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3472 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3473 map->m_lblk, end, &es);
3475 if (!es.es_len || es.es_lblk > end)
3478 if (es.es_lblk > map->m_lblk) {
3479 map->m_len = es.es_lblk - map->m_lblk;
3483 offset = map->m_lblk - es.es_lblk;
3484 map->m_len = es.es_len - offset;
3489 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3490 loff_t length, unsigned int flags,
3491 struct iomap *iomap, struct iomap *srcmap)
3494 bool delalloc = false;
3495 struct ext4_map_blocks map;
3496 u8 blkbits = inode->i_blkbits;
3498 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3501 if (ext4_has_inline_data(inode)) {
3502 ret = ext4_inline_data_iomap(inode, iomap);
3503 if (ret != -EAGAIN) {
3504 if (ret == 0 && offset >= iomap->length)
3511 * Calculate the first and last logical block respectively.
3513 map.m_lblk = offset >> blkbits;
3514 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3515 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3518 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3519 * So handle it here itself instead of querying ext4_map_blocks().
3520 * Since ext4_map_blocks() will warn about it and will return
3523 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3524 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3526 if (offset >= sbi->s_bitmap_maxbytes) {
3532 ret = ext4_map_blocks(NULL, inode, &map, 0);
3536 delalloc = ext4_iomap_is_delalloc(inode, &map);
3539 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3540 if (delalloc && iomap->type == IOMAP_HOLE)
3541 iomap->type = IOMAP_DELALLOC;
3546 const struct iomap_ops ext4_iomap_report_ops = {
3547 .iomap_begin = ext4_iomap_begin_report,
3551 * Pages can be marked dirty completely asynchronously from ext4's journalling
3552 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3553 * much here because ->set_page_dirty is called under VFS locks. The page is
3554 * not necessarily locked.
3556 * We cannot just dirty the page and leave attached buffers clean, because the
3557 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3558 * or jbddirty because all the journalling code will explode.
3560 * So what we do is to mark the page "pending dirty" and next time writepage
3561 * is called, propagate that into the buffers appropriately.
3563 static int ext4_journalled_set_page_dirty(struct page *page)
3565 SetPageChecked(page);
3566 return __set_page_dirty_nobuffers(page);
3569 static int ext4_set_page_dirty(struct page *page)
3571 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3572 WARN_ON_ONCE(!page_has_buffers(page));
3573 return __set_page_dirty_buffers(page);
3576 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3577 struct file *file, sector_t *span)
3579 return iomap_swapfile_activate(sis, file, span,
3580 &ext4_iomap_report_ops);
3583 static const struct address_space_operations ext4_aops = {
3584 .readpage = ext4_readpage,
3585 .readahead = ext4_readahead,
3586 .writepage = ext4_writepage,
3587 .writepages = ext4_writepages,
3588 .write_begin = ext4_write_begin,
3589 .write_end = ext4_write_end,
3590 .set_page_dirty = ext4_set_page_dirty,
3592 .invalidatepage = ext4_invalidatepage,
3593 .releasepage = ext4_releasepage,
3594 .direct_IO = noop_direct_IO,
3595 .migratepage = buffer_migrate_page,
3596 .is_partially_uptodate = block_is_partially_uptodate,
3597 .error_remove_page = generic_error_remove_page,
3598 .swap_activate = ext4_iomap_swap_activate,
3601 static const struct address_space_operations ext4_journalled_aops = {
3602 .readpage = ext4_readpage,
3603 .readahead = ext4_readahead,
3604 .writepage = ext4_writepage,
3605 .writepages = ext4_writepages,
3606 .write_begin = ext4_write_begin,
3607 .write_end = ext4_journalled_write_end,
3608 .set_page_dirty = ext4_journalled_set_page_dirty,
3610 .invalidatepage = ext4_journalled_invalidatepage,
3611 .releasepage = ext4_releasepage,
3612 .direct_IO = noop_direct_IO,
3613 .is_partially_uptodate = block_is_partially_uptodate,
3614 .error_remove_page = generic_error_remove_page,
3615 .swap_activate = ext4_iomap_swap_activate,
3618 static const struct address_space_operations ext4_da_aops = {
3619 .readpage = ext4_readpage,
3620 .readahead = ext4_readahead,
3621 .writepage = ext4_writepage,
3622 .writepages = ext4_writepages,
3623 .write_begin = ext4_da_write_begin,
3624 .write_end = ext4_da_write_end,
3625 .set_page_dirty = ext4_set_page_dirty,
3627 .invalidatepage = ext4_invalidatepage,
3628 .releasepage = ext4_releasepage,
3629 .direct_IO = noop_direct_IO,
3630 .migratepage = buffer_migrate_page,
3631 .is_partially_uptodate = block_is_partially_uptodate,
3632 .error_remove_page = generic_error_remove_page,
3633 .swap_activate = ext4_iomap_swap_activate,
3636 static const struct address_space_operations ext4_dax_aops = {
3637 .writepages = ext4_dax_writepages,
3638 .direct_IO = noop_direct_IO,
3639 .set_page_dirty = __set_page_dirty_no_writeback,
3641 .invalidatepage = noop_invalidatepage,
3642 .swap_activate = ext4_iomap_swap_activate,
3645 void ext4_set_aops(struct inode *inode)
3647 switch (ext4_inode_journal_mode(inode)) {
3648 case EXT4_INODE_ORDERED_DATA_MODE:
3649 case EXT4_INODE_WRITEBACK_DATA_MODE:
3651 case EXT4_INODE_JOURNAL_DATA_MODE:
3652 inode->i_mapping->a_ops = &ext4_journalled_aops;
3658 inode->i_mapping->a_ops = &ext4_dax_aops;
3659 else if (test_opt(inode->i_sb, DELALLOC))
3660 inode->i_mapping->a_ops = &ext4_da_aops;
3662 inode->i_mapping->a_ops = &ext4_aops;
3665 static int __ext4_block_zero_page_range(handle_t *handle,
3666 struct address_space *mapping, loff_t from, loff_t length)
3668 ext4_fsblk_t index = from >> PAGE_SHIFT;
3669 unsigned offset = from & (PAGE_SIZE-1);
3670 unsigned blocksize, pos;
3672 struct inode *inode = mapping->host;
3673 struct buffer_head *bh;
3677 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3678 mapping_gfp_constraint(mapping, ~__GFP_FS));
3682 blocksize = inode->i_sb->s_blocksize;
3684 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3686 if (!page_has_buffers(page))
3687 create_empty_buffers(page, blocksize, 0);
3689 /* Find the buffer that contains "offset" */
3690 bh = page_buffers(page);
3692 while (offset >= pos) {
3693 bh = bh->b_this_page;
3697 if (buffer_freed(bh)) {
3698 BUFFER_TRACE(bh, "freed: skip");
3701 if (!buffer_mapped(bh)) {
3702 BUFFER_TRACE(bh, "unmapped");
3703 ext4_get_block(inode, iblock, bh, 0);
3704 /* unmapped? It's a hole - nothing to do */
3705 if (!buffer_mapped(bh)) {
3706 BUFFER_TRACE(bh, "still unmapped");
3711 /* Ok, it's mapped. Make sure it's up-to-date */
3712 if (PageUptodate(page))
3713 set_buffer_uptodate(bh);
3715 if (!buffer_uptodate(bh)) {
3716 err = ext4_read_bh_lock(bh, 0, true);
3719 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3720 /* We expect the key to be set. */
3721 BUG_ON(!fscrypt_has_encryption_key(inode));
3722 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3725 clear_buffer_uptodate(bh);
3730 if (ext4_should_journal_data(inode)) {
3731 BUFFER_TRACE(bh, "get write access");
3732 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3737 zero_user(page, offset, length);
3738 BUFFER_TRACE(bh, "zeroed end of block");
3740 if (ext4_should_journal_data(inode)) {
3741 err = ext4_handle_dirty_metadata(handle, inode, bh);
3744 mark_buffer_dirty(bh);
3745 if (ext4_should_order_data(inode))
3746 err = ext4_jbd2_inode_add_write(handle, inode, from,
3757 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3758 * starting from file offset 'from'. The range to be zero'd must
3759 * be contained with in one block. If the specified range exceeds
3760 * the end of the block it will be shortened to end of the block
3761 * that corresponds to 'from'
3763 static int ext4_block_zero_page_range(handle_t *handle,
3764 struct address_space *mapping, loff_t from, loff_t length)
3766 struct inode *inode = mapping->host;
3767 unsigned offset = from & (PAGE_SIZE-1);
3768 unsigned blocksize = inode->i_sb->s_blocksize;
3769 unsigned max = blocksize - (offset & (blocksize - 1));
3772 * correct length if it does not fall between
3773 * 'from' and the end of the block
3775 if (length > max || length < 0)
3778 if (IS_DAX(inode)) {
3779 return dax_zero_range(inode, from, length, NULL,
3782 return __ext4_block_zero_page_range(handle, mapping, from, length);
3786 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3787 * up to the end of the block which corresponds to `from'.
3788 * This required during truncate. We need to physically zero the tail end
3789 * of that block so it doesn't yield old data if the file is later grown.
3791 static int ext4_block_truncate_page(handle_t *handle,
3792 struct address_space *mapping, loff_t from)
3794 unsigned offset = from & (PAGE_SIZE-1);
3797 struct inode *inode = mapping->host;
3799 /* If we are processing an encrypted inode during orphan list handling */
3800 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3803 blocksize = inode->i_sb->s_blocksize;
3804 length = blocksize - (offset & (blocksize - 1));
3806 return ext4_block_zero_page_range(handle, mapping, from, length);
3809 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3810 loff_t lstart, loff_t length)
3812 struct super_block *sb = inode->i_sb;
3813 struct address_space *mapping = inode->i_mapping;
3814 unsigned partial_start, partial_end;
3815 ext4_fsblk_t start, end;
3816 loff_t byte_end = (lstart + length - 1);
3819 partial_start = lstart & (sb->s_blocksize - 1);
3820 partial_end = byte_end & (sb->s_blocksize - 1);
3822 start = lstart >> sb->s_blocksize_bits;
3823 end = byte_end >> sb->s_blocksize_bits;
3825 /* Handle partial zero within the single block */
3827 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3828 err = ext4_block_zero_page_range(handle, mapping,
3832 /* Handle partial zero out on the start of the range */
3833 if (partial_start) {
3834 err = ext4_block_zero_page_range(handle, mapping,
3835 lstart, sb->s_blocksize);
3839 /* Handle partial zero out on the end of the range */
3840 if (partial_end != sb->s_blocksize - 1)
3841 err = ext4_block_zero_page_range(handle, mapping,
3842 byte_end - partial_end,
3847 int ext4_can_truncate(struct inode *inode)
3849 if (S_ISREG(inode->i_mode))
3851 if (S_ISDIR(inode->i_mode))
3853 if (S_ISLNK(inode->i_mode))
3854 return !ext4_inode_is_fast_symlink(inode);
3859 * We have to make sure i_disksize gets properly updated before we truncate
3860 * page cache due to hole punching or zero range. Otherwise i_disksize update
3861 * can get lost as it may have been postponed to submission of writeback but
3862 * that will never happen after we truncate page cache.
3864 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3870 loff_t size = i_size_read(inode);
3872 WARN_ON(!inode_is_locked(inode));
3873 if (offset > size || offset + len < size)
3876 if (EXT4_I(inode)->i_disksize >= size)
3879 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3881 return PTR_ERR(handle);
3882 ext4_update_i_disksize(inode, size);
3883 ret = ext4_mark_inode_dirty(handle, inode);
3884 ext4_journal_stop(handle);
3889 static void ext4_wait_dax_page(struct inode *inode)
3891 filemap_invalidate_unlock(inode->i_mapping);
3893 filemap_invalidate_lock(inode->i_mapping);
3896 int ext4_break_layouts(struct inode *inode)
3901 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3905 page = dax_layout_busy_page(inode->i_mapping);
3909 error = ___wait_var_event(&page->_refcount,
3910 atomic_read(&page->_refcount) == 1,
3911 TASK_INTERRUPTIBLE, 0, 0,
3912 ext4_wait_dax_page(inode));
3913 } while (error == 0);
3919 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3920 * associated with the given offset and length
3922 * @inode: File inode
3923 * @offset: The offset where the hole will begin
3924 * @len: The length of the hole
3926 * Returns: 0 on success or negative on failure
3929 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3931 struct super_block *sb = inode->i_sb;
3932 ext4_lblk_t first_block, stop_block;
3933 struct address_space *mapping = inode->i_mapping;
3934 loff_t first_block_offset, last_block_offset;
3936 unsigned int credits;
3937 int ret = 0, ret2 = 0;
3939 trace_ext4_punch_hole(inode, offset, length, 0);
3941 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
3942 if (ext4_has_inline_data(inode)) {
3943 filemap_invalidate_lock(mapping);
3944 ret = ext4_convert_inline_data(inode);
3945 filemap_invalidate_unlock(mapping);
3951 * Write out all dirty pages to avoid race conditions
3952 * Then release them.
3954 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3955 ret = filemap_write_and_wait_range(mapping, offset,
3956 offset + length - 1);
3963 /* No need to punch hole beyond i_size */
3964 if (offset >= inode->i_size)
3968 * If the hole extends beyond i_size, set the hole
3969 * to end after the page that contains i_size
3971 if (offset + length > inode->i_size) {
3972 length = inode->i_size +
3973 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3977 if (offset & (sb->s_blocksize - 1) ||
3978 (offset + length) & (sb->s_blocksize - 1)) {
3980 * Attach jinode to inode for jbd2 if we do any zeroing of
3983 ret = ext4_inode_attach_jinode(inode);
3989 /* Wait all existing dio workers, newcomers will block on i_rwsem */
3990 inode_dio_wait(inode);
3993 * Prevent page faults from reinstantiating pages we have released from
3996 filemap_invalidate_lock(mapping);
3998 ret = ext4_break_layouts(inode);
4002 first_block_offset = round_up(offset, sb->s_blocksize);
4003 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4005 /* Now release the pages and zero block aligned part of pages*/
4006 if (last_block_offset > first_block_offset) {
4007 ret = ext4_update_disksize_before_punch(inode, offset, length);
4010 truncate_pagecache_range(inode, first_block_offset,
4014 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4015 credits = ext4_writepage_trans_blocks(inode);
4017 credits = ext4_blocks_for_truncate(inode);
4018 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4019 if (IS_ERR(handle)) {
4020 ret = PTR_ERR(handle);
4021 ext4_std_error(sb, ret);
4025 ret = ext4_zero_partial_blocks(handle, inode, offset,
4030 first_block = (offset + sb->s_blocksize - 1) >>
4031 EXT4_BLOCK_SIZE_BITS(sb);
4032 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4034 /* If there are blocks to remove, do it */
4035 if (stop_block > first_block) {
4037 down_write(&EXT4_I(inode)->i_data_sem);
4038 ext4_discard_preallocations(inode, 0);
4040 ret = ext4_es_remove_extent(inode, first_block,
4041 stop_block - first_block);
4043 up_write(&EXT4_I(inode)->i_data_sem);
4047 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4048 ret = ext4_ext_remove_space(inode, first_block,
4051 ret = ext4_ind_remove_space(handle, inode, first_block,
4054 up_write(&EXT4_I(inode)->i_data_sem);
4056 ext4_fc_track_range(handle, inode, first_block, stop_block);
4058 ext4_handle_sync(handle);
4060 inode->i_mtime = inode->i_ctime = current_time(inode);
4061 ret2 = ext4_mark_inode_dirty(handle, inode);
4065 ext4_update_inode_fsync_trans(handle, inode, 1);
4067 ext4_journal_stop(handle);
4069 filemap_invalidate_unlock(mapping);
4071 inode_unlock(inode);
4075 int ext4_inode_attach_jinode(struct inode *inode)
4077 struct ext4_inode_info *ei = EXT4_I(inode);
4078 struct jbd2_inode *jinode;
4080 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4083 jinode = jbd2_alloc_inode(GFP_KERNEL);
4084 spin_lock(&inode->i_lock);
4087 spin_unlock(&inode->i_lock);
4090 ei->jinode = jinode;
4091 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4094 spin_unlock(&inode->i_lock);
4095 if (unlikely(jinode != NULL))
4096 jbd2_free_inode(jinode);
4103 * We block out ext4_get_block() block instantiations across the entire
4104 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4105 * simultaneously on behalf of the same inode.
4107 * As we work through the truncate and commit bits of it to the journal there
4108 * is one core, guiding principle: the file's tree must always be consistent on
4109 * disk. We must be able to restart the truncate after a crash.
4111 * The file's tree may be transiently inconsistent in memory (although it
4112 * probably isn't), but whenever we close off and commit a journal transaction,
4113 * the contents of (the filesystem + the journal) must be consistent and
4114 * restartable. It's pretty simple, really: bottom up, right to left (although
4115 * left-to-right works OK too).
4117 * Note that at recovery time, journal replay occurs *before* the restart of
4118 * truncate against the orphan inode list.
4120 * The committed inode has the new, desired i_size (which is the same as
4121 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4122 * that this inode's truncate did not complete and it will again call
4123 * ext4_truncate() to have another go. So there will be instantiated blocks
4124 * to the right of the truncation point in a crashed ext4 filesystem. But
4125 * that's fine - as long as they are linked from the inode, the post-crash
4126 * ext4_truncate() run will find them and release them.
4128 int ext4_truncate(struct inode *inode)
4130 struct ext4_inode_info *ei = EXT4_I(inode);
4131 unsigned int credits;
4134 struct address_space *mapping = inode->i_mapping;
4137 * There is a possibility that we're either freeing the inode
4138 * or it's a completely new inode. In those cases we might not
4139 * have i_rwsem locked because it's not necessary.
4141 if (!(inode->i_state & (I_NEW|I_FREEING)))
4142 WARN_ON(!inode_is_locked(inode));
4143 trace_ext4_truncate_enter(inode);
4145 if (!ext4_can_truncate(inode))
4148 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4149 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4151 if (ext4_has_inline_data(inode)) {
4154 err = ext4_inline_data_truncate(inode, &has_inline);
4155 if (err || has_inline)
4159 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4160 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4161 if (ext4_inode_attach_jinode(inode) < 0)
4165 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4166 credits = ext4_writepage_trans_blocks(inode);
4168 credits = ext4_blocks_for_truncate(inode);
4170 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4171 if (IS_ERR(handle)) {
4172 err = PTR_ERR(handle);
4176 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4177 ext4_block_truncate_page(handle, mapping, inode->i_size);
4180 * We add the inode to the orphan list, so that if this
4181 * truncate spans multiple transactions, and we crash, we will
4182 * resume the truncate when the filesystem recovers. It also
4183 * marks the inode dirty, to catch the new size.
4185 * Implication: the file must always be in a sane, consistent
4186 * truncatable state while each transaction commits.
4188 err = ext4_orphan_add(handle, inode);
4192 down_write(&EXT4_I(inode)->i_data_sem);
4194 ext4_discard_preallocations(inode, 0);
4196 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4197 err = ext4_ext_truncate(handle, inode);
4199 ext4_ind_truncate(handle, inode);
4201 up_write(&ei->i_data_sem);
4206 ext4_handle_sync(handle);
4210 * If this was a simple ftruncate() and the file will remain alive,
4211 * then we need to clear up the orphan record which we created above.
4212 * However, if this was a real unlink then we were called by
4213 * ext4_evict_inode(), and we allow that function to clean up the
4214 * orphan info for us.
4217 ext4_orphan_del(handle, inode);
4219 inode->i_mtime = inode->i_ctime = current_time(inode);
4220 err2 = ext4_mark_inode_dirty(handle, inode);
4221 if (unlikely(err2 && !err))
4223 ext4_journal_stop(handle);
4226 trace_ext4_truncate_exit(inode);
4230 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4232 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4233 return inode_peek_iversion_raw(inode);
4235 return inode_peek_iversion(inode);
4238 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4239 struct ext4_inode_info *ei)
4241 struct inode *inode = &(ei->vfs_inode);
4242 u64 i_blocks = READ_ONCE(inode->i_blocks);
4243 struct super_block *sb = inode->i_sb;
4245 if (i_blocks <= ~0U) {
4247 * i_blocks can be represented in a 32 bit variable
4248 * as multiple of 512 bytes
4250 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4251 raw_inode->i_blocks_high = 0;
4252 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4257 * This should never happen since sb->s_maxbytes should not have
4258 * allowed this, sb->s_maxbytes was set according to the huge_file
4259 * feature in ext4_fill_super().
4261 if (!ext4_has_feature_huge_file(sb))
4262 return -EFSCORRUPTED;
4264 if (i_blocks <= 0xffffffffffffULL) {
4266 * i_blocks can be represented in a 48 bit variable
4267 * as multiple of 512 bytes
4269 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4270 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4271 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4273 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4274 /* i_block is stored in file system block size */
4275 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4276 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4277 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4282 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4284 struct ext4_inode_info *ei = EXT4_I(inode);
4291 err = ext4_inode_blocks_set(raw_inode, ei);
4293 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4294 i_uid = i_uid_read(inode);
4295 i_gid = i_gid_read(inode);
4296 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4297 if (!(test_opt(inode->i_sb, NO_UID32))) {
4298 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4299 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4301 * Fix up interoperability with old kernels. Otherwise,
4302 * old inodes get re-used with the upper 16 bits of the
4305 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4306 raw_inode->i_uid_high = 0;
4307 raw_inode->i_gid_high = 0;
4309 raw_inode->i_uid_high =
4310 cpu_to_le16(high_16_bits(i_uid));
4311 raw_inode->i_gid_high =
4312 cpu_to_le16(high_16_bits(i_gid));
4315 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4316 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4317 raw_inode->i_uid_high = 0;
4318 raw_inode->i_gid_high = 0;
4320 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4322 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4323 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4324 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4325 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4327 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4328 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4329 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4330 raw_inode->i_file_acl_high =
4331 cpu_to_le16(ei->i_file_acl >> 32);
4332 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4333 ext4_isize_set(raw_inode, ei->i_disksize);
4335 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4336 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4337 if (old_valid_dev(inode->i_rdev)) {
4338 raw_inode->i_block[0] =
4339 cpu_to_le32(old_encode_dev(inode->i_rdev));
4340 raw_inode->i_block[1] = 0;
4342 raw_inode->i_block[0] = 0;
4343 raw_inode->i_block[1] =
4344 cpu_to_le32(new_encode_dev(inode->i_rdev));
4345 raw_inode->i_block[2] = 0;
4347 } else if (!ext4_has_inline_data(inode)) {
4348 for (block = 0; block < EXT4_N_BLOCKS; block++)
4349 raw_inode->i_block[block] = ei->i_data[block];
4352 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4353 u64 ivers = ext4_inode_peek_iversion(inode);
4355 raw_inode->i_disk_version = cpu_to_le32(ivers);
4356 if (ei->i_extra_isize) {
4357 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4358 raw_inode->i_version_hi =
4359 cpu_to_le32(ivers >> 32);
4360 raw_inode->i_extra_isize =
4361 cpu_to_le16(ei->i_extra_isize);
4365 if (i_projid != EXT4_DEF_PROJID &&
4366 !ext4_has_feature_project(inode->i_sb))
4367 err = err ?: -EFSCORRUPTED;
4369 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4370 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4371 raw_inode->i_projid = cpu_to_le32(i_projid);
4373 ext4_inode_csum_set(inode, raw_inode, ei);
4378 * ext4_get_inode_loc returns with an extra refcount against the inode's
4379 * underlying buffer_head on success. If we pass 'inode' and it does not
4380 * have in-inode xattr, we have all inode data in memory that is needed
4381 * to recreate the on-disk version of this inode.
4383 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4384 struct inode *inode, struct ext4_iloc *iloc,
4385 ext4_fsblk_t *ret_block)
4387 struct ext4_group_desc *gdp;
4388 struct buffer_head *bh;
4390 struct blk_plug plug;
4391 int inodes_per_block, inode_offset;
4394 if (ino < EXT4_ROOT_INO ||
4395 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4396 return -EFSCORRUPTED;
4398 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4399 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4404 * Figure out the offset within the block group inode table
4406 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4407 inode_offset = ((ino - 1) %
4408 EXT4_INODES_PER_GROUP(sb));
4409 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4410 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4412 bh = sb_getblk(sb, block);
4415 if (ext4_buffer_uptodate(bh))
4419 if (ext4_buffer_uptodate(bh)) {
4420 /* Someone brought it uptodate while we waited */
4426 * If we have all information of the inode in memory and this
4427 * is the only valid inode in the block, we need not read the
4430 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4431 struct buffer_head *bitmap_bh;
4434 start = inode_offset & ~(inodes_per_block - 1);
4436 /* Is the inode bitmap in cache? */
4437 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4438 if (unlikely(!bitmap_bh))
4442 * If the inode bitmap isn't in cache then the
4443 * optimisation may end up performing two reads instead
4444 * of one, so skip it.
4446 if (!buffer_uptodate(bitmap_bh)) {
4450 for (i = start; i < start + inodes_per_block; i++) {
4451 if (i == inode_offset)
4453 if (ext4_test_bit(i, bitmap_bh->b_data))
4457 if (i == start + inodes_per_block) {
4458 struct ext4_inode *raw_inode =
4459 (struct ext4_inode *) (bh->b_data + iloc->offset);
4461 /* all other inodes are free, so skip I/O */
4462 memset(bh->b_data, 0, bh->b_size);
4463 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4464 ext4_fill_raw_inode(inode, raw_inode);
4465 set_buffer_uptodate(bh);
4473 * If we need to do any I/O, try to pre-readahead extra
4474 * blocks from the inode table.
4476 blk_start_plug(&plug);
4477 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4478 ext4_fsblk_t b, end, table;
4480 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4482 table = ext4_inode_table(sb, gdp);
4483 /* s_inode_readahead_blks is always a power of 2 */
4484 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4488 num = EXT4_INODES_PER_GROUP(sb);
4489 if (ext4_has_group_desc_csum(sb))
4490 num -= ext4_itable_unused_count(sb, gdp);
4491 table += num / inodes_per_block;
4495 ext4_sb_breadahead_unmovable(sb, b++);
4499 * There are other valid inodes in the buffer, this inode
4500 * has in-inode xattrs, or we don't have this inode in memory.
4501 * Read the block from disk.
4503 trace_ext4_load_inode(sb, ino);
4504 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4505 blk_finish_plug(&plug);
4507 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4508 if (!buffer_uptodate(bh)) {
4519 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4520 struct ext4_iloc *iloc)
4522 ext4_fsblk_t err_blk = 0;
4525 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4529 ext4_error_inode_block(inode, err_blk, EIO,
4530 "unable to read itable block");
4535 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4537 ext4_fsblk_t err_blk = 0;
4540 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4544 ext4_error_inode_block(inode, err_blk, EIO,
4545 "unable to read itable block");
4551 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4552 struct ext4_iloc *iloc)
4554 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4557 static bool ext4_should_enable_dax(struct inode *inode)
4559 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4561 if (test_opt2(inode->i_sb, DAX_NEVER))
4563 if (!S_ISREG(inode->i_mode))
4565 if (ext4_should_journal_data(inode))
4567 if (ext4_has_inline_data(inode))
4569 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4571 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4573 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4575 if (test_opt(inode->i_sb, DAX_ALWAYS))
4578 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4581 void ext4_set_inode_flags(struct inode *inode, bool init)
4583 unsigned int flags = EXT4_I(inode)->i_flags;
4584 unsigned int new_fl = 0;
4586 WARN_ON_ONCE(IS_DAX(inode) && init);
4588 if (flags & EXT4_SYNC_FL)
4590 if (flags & EXT4_APPEND_FL)
4592 if (flags & EXT4_IMMUTABLE_FL)
4593 new_fl |= S_IMMUTABLE;
4594 if (flags & EXT4_NOATIME_FL)
4595 new_fl |= S_NOATIME;
4596 if (flags & EXT4_DIRSYNC_FL)
4597 new_fl |= S_DIRSYNC;
4599 /* Because of the way inode_set_flags() works we must preserve S_DAX
4600 * here if already set. */
4601 new_fl |= (inode->i_flags & S_DAX);
4602 if (init && ext4_should_enable_dax(inode))
4605 if (flags & EXT4_ENCRYPT_FL)
4606 new_fl |= S_ENCRYPTED;
4607 if (flags & EXT4_CASEFOLD_FL)
4608 new_fl |= S_CASEFOLD;
4609 if (flags & EXT4_VERITY_FL)
4611 inode_set_flags(inode, new_fl,
4612 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4613 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4616 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4617 struct ext4_inode_info *ei)
4620 struct inode *inode = &(ei->vfs_inode);
4621 struct super_block *sb = inode->i_sb;
4623 if (ext4_has_feature_huge_file(sb)) {
4624 /* we are using combined 48 bit field */
4625 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4626 le32_to_cpu(raw_inode->i_blocks_lo);
4627 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4628 /* i_blocks represent file system block size */
4629 return i_blocks << (inode->i_blkbits - 9);
4634 return le32_to_cpu(raw_inode->i_blocks_lo);
4638 static inline int ext4_iget_extra_inode(struct inode *inode,
4639 struct ext4_inode *raw_inode,
4640 struct ext4_inode_info *ei)
4642 __le32 *magic = (void *)raw_inode +
4643 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4645 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4646 EXT4_INODE_SIZE(inode->i_sb) &&
4647 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4648 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4649 return ext4_find_inline_data_nolock(inode);
4651 EXT4_I(inode)->i_inline_off = 0;
4655 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4657 if (!ext4_has_feature_project(inode->i_sb))
4659 *projid = EXT4_I(inode)->i_projid;
4664 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4665 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4668 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4670 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4671 inode_set_iversion_raw(inode, val);
4673 inode_set_iversion_queried(inode, val);
4676 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4677 ext4_iget_flags flags, const char *function,
4680 struct ext4_iloc iloc;
4681 struct ext4_inode *raw_inode;
4682 struct ext4_inode_info *ei;
4683 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4684 struct inode *inode;
4685 journal_t *journal = EXT4_SB(sb)->s_journal;
4693 if ((!(flags & EXT4_IGET_SPECIAL) &&
4694 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4695 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4696 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4697 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4698 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4699 (ino < EXT4_ROOT_INO) ||
4700 (ino > le32_to_cpu(es->s_inodes_count))) {
4701 if (flags & EXT4_IGET_HANDLE)
4702 return ERR_PTR(-ESTALE);
4703 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4704 "inode #%lu: comm %s: iget: illegal inode #",
4705 ino, current->comm);
4706 return ERR_PTR(-EFSCORRUPTED);
4709 inode = iget_locked(sb, ino);
4711 return ERR_PTR(-ENOMEM);
4712 if (!(inode->i_state & I_NEW))
4718 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4721 raw_inode = ext4_raw_inode(&iloc);
4723 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4724 ext4_error_inode(inode, function, line, 0,
4725 "iget: root inode unallocated");
4726 ret = -EFSCORRUPTED;
4730 if ((flags & EXT4_IGET_HANDLE) &&
4731 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4736 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4737 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4738 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4739 EXT4_INODE_SIZE(inode->i_sb) ||
4740 (ei->i_extra_isize & 3)) {
4741 ext4_error_inode(inode, function, line, 0,
4742 "iget: bad extra_isize %u "
4745 EXT4_INODE_SIZE(inode->i_sb));
4746 ret = -EFSCORRUPTED;
4750 ei->i_extra_isize = 0;
4752 /* Precompute checksum seed for inode metadata */
4753 if (ext4_has_metadata_csum(sb)) {
4754 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4756 __le32 inum = cpu_to_le32(inode->i_ino);
4757 __le32 gen = raw_inode->i_generation;
4758 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4760 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4764 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4765 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4766 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4767 ext4_error_inode_err(inode, function, line, 0,
4768 EFSBADCRC, "iget: checksum invalid");
4773 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4774 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4775 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4776 if (ext4_has_feature_project(sb) &&
4777 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4778 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4779 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4781 i_projid = EXT4_DEF_PROJID;
4783 if (!(test_opt(inode->i_sb, NO_UID32))) {
4784 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4785 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4787 i_uid_write(inode, i_uid);
4788 i_gid_write(inode, i_gid);
4789 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4790 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4792 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4793 ei->i_inline_off = 0;
4794 ei->i_dir_start_lookup = 0;
4795 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4796 /* We now have enough fields to check if the inode was active or not.
4797 * This is needed because nfsd might try to access dead inodes
4798 * the test is that same one that e2fsck uses
4799 * NeilBrown 1999oct15
4801 if (inode->i_nlink == 0) {
4802 if ((inode->i_mode == 0 ||
4803 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4804 ino != EXT4_BOOT_LOADER_INO) {
4805 /* this inode is deleted */
4809 /* The only unlinked inodes we let through here have
4810 * valid i_mode and are being read by the orphan
4811 * recovery code: that's fine, we're about to complete
4812 * the process of deleting those.
4813 * OR it is the EXT4_BOOT_LOADER_INO which is
4814 * not initialized on a new filesystem. */
4816 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4817 ext4_set_inode_flags(inode, true);
4818 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4819 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4820 if (ext4_has_feature_64bit(sb))
4822 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4823 inode->i_size = ext4_isize(sb, raw_inode);
4824 if ((size = i_size_read(inode)) < 0) {
4825 ext4_error_inode(inode, function, line, 0,
4826 "iget: bad i_size value: %lld", size);
4827 ret = -EFSCORRUPTED;
4831 * If dir_index is not enabled but there's dir with INDEX flag set,
4832 * we'd normally treat htree data as empty space. But with metadata
4833 * checksumming that corrupts checksums so forbid that.
4835 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4836 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4837 ext4_error_inode(inode, function, line, 0,
4838 "iget: Dir with htree data on filesystem without dir_index feature.");
4839 ret = -EFSCORRUPTED;
4842 ei->i_disksize = inode->i_size;
4844 ei->i_reserved_quota = 0;
4846 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4847 ei->i_block_group = iloc.block_group;
4848 ei->i_last_alloc_group = ~0;
4850 * NOTE! The in-memory inode i_data array is in little-endian order
4851 * even on big-endian machines: we do NOT byteswap the block numbers!
4853 for (block = 0; block < EXT4_N_BLOCKS; block++)
4854 ei->i_data[block] = raw_inode->i_block[block];
4855 INIT_LIST_HEAD(&ei->i_orphan);
4856 ext4_fc_init_inode(&ei->vfs_inode);
4859 * Set transaction id's of transactions that have to be committed
4860 * to finish f[data]sync. We set them to currently running transaction
4861 * as we cannot be sure that the inode or some of its metadata isn't
4862 * part of the transaction - the inode could have been reclaimed and
4863 * now it is reread from disk.
4866 transaction_t *transaction;
4869 read_lock(&journal->j_state_lock);
4870 if (journal->j_running_transaction)
4871 transaction = journal->j_running_transaction;
4873 transaction = journal->j_committing_transaction;
4875 tid = transaction->t_tid;
4877 tid = journal->j_commit_sequence;
4878 read_unlock(&journal->j_state_lock);
4879 ei->i_sync_tid = tid;
4880 ei->i_datasync_tid = tid;
4883 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4884 if (ei->i_extra_isize == 0) {
4885 /* The extra space is currently unused. Use it. */
4886 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4887 ei->i_extra_isize = sizeof(struct ext4_inode) -
4888 EXT4_GOOD_OLD_INODE_SIZE;
4890 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4896 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4897 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4898 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4899 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4901 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4902 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4904 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4905 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4907 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4909 ext4_inode_set_iversion_queried(inode, ivers);
4913 if (ei->i_file_acl &&
4914 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4915 ext4_error_inode(inode, function, line, 0,
4916 "iget: bad extended attribute block %llu",
4918 ret = -EFSCORRUPTED;
4920 } else if (!ext4_has_inline_data(inode)) {
4921 /* validate the block references in the inode */
4922 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4923 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4924 (S_ISLNK(inode->i_mode) &&
4925 !ext4_inode_is_fast_symlink(inode)))) {
4926 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4927 ret = ext4_ext_check_inode(inode);
4929 ret = ext4_ind_check_inode(inode);
4935 if (S_ISREG(inode->i_mode)) {
4936 inode->i_op = &ext4_file_inode_operations;
4937 inode->i_fop = &ext4_file_operations;
4938 ext4_set_aops(inode);
4939 } else if (S_ISDIR(inode->i_mode)) {
4940 inode->i_op = &ext4_dir_inode_operations;
4941 inode->i_fop = &ext4_dir_operations;
4942 } else if (S_ISLNK(inode->i_mode)) {
4943 /* VFS does not allow setting these so must be corruption */
4944 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4945 ext4_error_inode(inode, function, line, 0,
4946 "iget: immutable or append flags "
4947 "not allowed on symlinks");
4948 ret = -EFSCORRUPTED;
4951 if (IS_ENCRYPTED(inode)) {
4952 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4953 ext4_set_aops(inode);
4954 } else if (ext4_inode_is_fast_symlink(inode)) {
4955 inode->i_link = (char *)ei->i_data;
4956 inode->i_op = &ext4_fast_symlink_inode_operations;
4957 nd_terminate_link(ei->i_data, inode->i_size,
4958 sizeof(ei->i_data) - 1);
4960 inode->i_op = &ext4_symlink_inode_operations;
4961 ext4_set_aops(inode);
4963 inode_nohighmem(inode);
4964 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4965 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4966 inode->i_op = &ext4_special_inode_operations;
4967 if (raw_inode->i_block[0])
4968 init_special_inode(inode, inode->i_mode,
4969 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4971 init_special_inode(inode, inode->i_mode,
4972 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4973 } else if (ino == EXT4_BOOT_LOADER_INO) {
4974 make_bad_inode(inode);
4976 ret = -EFSCORRUPTED;
4977 ext4_error_inode(inode, function, line, 0,
4978 "iget: bogus i_mode (%o)", inode->i_mode);
4981 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4982 ext4_error_inode(inode, function, line, 0,
4983 "casefold flag without casefold feature");
4986 unlock_new_inode(inode);
4992 return ERR_PTR(ret);
4995 static void __ext4_update_other_inode_time(struct super_block *sb,
4996 unsigned long orig_ino,
4998 struct ext4_inode *raw_inode)
5000 struct inode *inode;
5002 inode = find_inode_by_ino_rcu(sb, ino);
5006 if (!inode_is_dirtytime_only(inode))
5009 spin_lock(&inode->i_lock);
5010 if (inode_is_dirtytime_only(inode)) {
5011 struct ext4_inode_info *ei = EXT4_I(inode);
5013 inode->i_state &= ~I_DIRTY_TIME;
5014 spin_unlock(&inode->i_lock);
5016 spin_lock(&ei->i_raw_lock);
5017 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5018 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5019 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5020 ext4_inode_csum_set(inode, raw_inode, ei);
5021 spin_unlock(&ei->i_raw_lock);
5022 trace_ext4_other_inode_update_time(inode, orig_ino);
5025 spin_unlock(&inode->i_lock);
5029 * Opportunistically update the other time fields for other inodes in
5030 * the same inode table block.
5032 static void ext4_update_other_inodes_time(struct super_block *sb,
5033 unsigned long orig_ino, char *buf)
5036 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5037 int inode_size = EXT4_INODE_SIZE(sb);
5040 * Calculate the first inode in the inode table block. Inode
5041 * numbers are one-based. That is, the first inode in a block
5042 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5044 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5046 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5047 if (ino == orig_ino)
5049 __ext4_update_other_inode_time(sb, orig_ino, ino,
5050 (struct ext4_inode *)buf);
5056 * Post the struct inode info into an on-disk inode location in the
5057 * buffer-cache. This gobbles the caller's reference to the
5058 * buffer_head in the inode location struct.
5060 * The caller must have write access to iloc->bh.
5062 static int ext4_do_update_inode(handle_t *handle,
5063 struct inode *inode,
5064 struct ext4_iloc *iloc)
5066 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5067 struct ext4_inode_info *ei = EXT4_I(inode);
5068 struct buffer_head *bh = iloc->bh;
5069 struct super_block *sb = inode->i_sb;
5071 int need_datasync = 0, set_large_file = 0;
5073 spin_lock(&ei->i_raw_lock);
5076 * For fields not tracked in the in-memory inode, initialise them
5077 * to zero for new inodes.
5079 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5080 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5082 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5084 if (ei->i_disksize > 0x7fffffffULL) {
5085 if (!ext4_has_feature_large_file(sb) ||
5086 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5090 err = ext4_fill_raw_inode(inode, raw_inode);
5091 spin_unlock(&ei->i_raw_lock);
5093 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5097 if (inode->i_sb->s_flags & SB_LAZYTIME)
5098 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5101 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5102 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5105 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5106 if (set_large_file) {
5107 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5108 err = ext4_journal_get_write_access(handle, sb,
5113 lock_buffer(EXT4_SB(sb)->s_sbh);
5114 ext4_set_feature_large_file(sb);
5115 ext4_superblock_csum_set(sb);
5116 unlock_buffer(EXT4_SB(sb)->s_sbh);
5117 ext4_handle_sync(handle);
5118 err = ext4_handle_dirty_metadata(handle, NULL,
5119 EXT4_SB(sb)->s_sbh);
5121 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5123 ext4_std_error(inode->i_sb, err);
5130 * ext4_write_inode()
5132 * We are called from a few places:
5134 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5135 * Here, there will be no transaction running. We wait for any running
5136 * transaction to commit.
5138 * - Within flush work (sys_sync(), kupdate and such).
5139 * We wait on commit, if told to.
5141 * - Within iput_final() -> write_inode_now()
5142 * We wait on commit, if told to.
5144 * In all cases it is actually safe for us to return without doing anything,
5145 * because the inode has been copied into a raw inode buffer in
5146 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5149 * Note that we are absolutely dependent upon all inode dirtiers doing the
5150 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5151 * which we are interested.
5153 * It would be a bug for them to not do this. The code:
5155 * mark_inode_dirty(inode)
5157 * inode->i_size = expr;
5159 * is in error because write_inode() could occur while `stuff()' is running,
5160 * and the new i_size will be lost. Plus the inode will no longer be on the
5161 * superblock's dirty inode list.
5163 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5167 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5168 sb_rdonly(inode->i_sb))
5171 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5174 if (EXT4_SB(inode->i_sb)->s_journal) {
5175 if (ext4_journal_current_handle()) {
5176 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5182 * No need to force transaction in WB_SYNC_NONE mode. Also
5183 * ext4_sync_fs() will force the commit after everything is
5186 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5189 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5190 EXT4_I(inode)->i_sync_tid);
5192 struct ext4_iloc iloc;
5194 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5198 * sync(2) will flush the whole buffer cache. No need to do
5199 * it here separately for each inode.
5201 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5202 sync_dirty_buffer(iloc.bh);
5203 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5204 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5205 "IO error syncing inode");
5214 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5215 * buffers that are attached to a page stradding i_size and are undergoing
5216 * commit. In that case we have to wait for commit to finish and try again.
5218 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5222 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5223 tid_t commit_tid = 0;
5226 offset = inode->i_size & (PAGE_SIZE - 1);
5228 * If the page is fully truncated, we don't need to wait for any commit
5229 * (and we even should not as __ext4_journalled_invalidatepage() may
5230 * strip all buffers from the page but keep the page dirty which can then
5231 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5232 * buffers). Also we don't need to wait for any commit if all buffers in
5233 * the page remain valid. This is most beneficial for the common case of
5234 * blocksize == PAGESIZE.
5236 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5239 page = find_lock_page(inode->i_mapping,
5240 inode->i_size >> PAGE_SHIFT);
5243 ret = __ext4_journalled_invalidatepage(page, offset,
5244 PAGE_SIZE - offset);
5250 read_lock(&journal->j_state_lock);
5251 if (journal->j_committing_transaction)
5252 commit_tid = journal->j_committing_transaction->t_tid;
5253 read_unlock(&journal->j_state_lock);
5255 jbd2_log_wait_commit(journal, commit_tid);
5262 * Called from notify_change.
5264 * We want to trap VFS attempts to truncate the file as soon as
5265 * possible. In particular, we want to make sure that when the VFS
5266 * shrinks i_size, we put the inode on the orphan list and modify
5267 * i_disksize immediately, so that during the subsequent flushing of
5268 * dirty pages and freeing of disk blocks, we can guarantee that any
5269 * commit will leave the blocks being flushed in an unused state on
5270 * disk. (On recovery, the inode will get truncated and the blocks will
5271 * be freed, so we have a strong guarantee that no future commit will
5272 * leave these blocks visible to the user.)
5274 * Another thing we have to assure is that if we are in ordered mode
5275 * and inode is still attached to the committing transaction, we must
5276 * we start writeout of all the dirty pages which are being truncated.
5277 * This way we are sure that all the data written in the previous
5278 * transaction are already on disk (truncate waits for pages under
5281 * Called with inode->i_rwsem down.
5283 int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5286 struct inode *inode = d_inode(dentry);
5289 const unsigned int ia_valid = attr->ia_valid;
5291 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5294 if (unlikely(IS_IMMUTABLE(inode)))
5297 if (unlikely(IS_APPEND(inode) &&
5298 (ia_valid & (ATTR_MODE | ATTR_UID |
5299 ATTR_GID | ATTR_TIMES_SET))))
5302 error = setattr_prepare(mnt_userns, dentry, attr);
5306 error = fscrypt_prepare_setattr(dentry, attr);
5310 error = fsverity_prepare_setattr(dentry, attr);
5314 if (is_quota_modification(inode, attr)) {
5315 error = dquot_initialize(inode);
5320 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5321 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5324 /* (user+group)*(old+new) structure, inode write (sb,
5325 * inode block, ? - but truncate inode update has it) */
5326 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5327 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5328 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5329 if (IS_ERR(handle)) {
5330 error = PTR_ERR(handle);
5334 /* dquot_transfer() calls back ext4_get_inode_usage() which
5335 * counts xattr inode references.
5337 down_read(&EXT4_I(inode)->xattr_sem);
5338 error = dquot_transfer(inode, attr);
5339 up_read(&EXT4_I(inode)->xattr_sem);
5342 ext4_journal_stop(handle);
5345 /* Update corresponding info in inode so that everything is in
5346 * one transaction */
5347 if (attr->ia_valid & ATTR_UID)
5348 inode->i_uid = attr->ia_uid;
5349 if (attr->ia_valid & ATTR_GID)
5350 inode->i_gid = attr->ia_gid;
5351 error = ext4_mark_inode_dirty(handle, inode);
5352 ext4_journal_stop(handle);
5353 if (unlikely(error)) {
5358 if (attr->ia_valid & ATTR_SIZE) {
5360 loff_t oldsize = inode->i_size;
5361 int shrink = (attr->ia_size < inode->i_size);
5363 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5364 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5366 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5370 if (!S_ISREG(inode->i_mode)) {
5374 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5375 inode_inc_iversion(inode);
5378 if (ext4_should_order_data(inode)) {
5379 error = ext4_begin_ordered_truncate(inode,
5385 * Blocks are going to be removed from the inode. Wait
5386 * for dio in flight.
5388 inode_dio_wait(inode);
5391 filemap_invalidate_lock(inode->i_mapping);
5393 rc = ext4_break_layouts(inode);
5395 filemap_invalidate_unlock(inode->i_mapping);
5399 if (attr->ia_size != inode->i_size) {
5400 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5401 if (IS_ERR(handle)) {
5402 error = PTR_ERR(handle);
5405 if (ext4_handle_valid(handle) && shrink) {
5406 error = ext4_orphan_add(handle, inode);
5410 * Update c/mtime on truncate up, ext4_truncate() will
5411 * update c/mtime in shrink case below
5414 inode->i_mtime = current_time(inode);
5415 inode->i_ctime = inode->i_mtime;
5419 ext4_fc_track_range(handle, inode,
5420 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5421 inode->i_sb->s_blocksize_bits,
5422 EXT_MAX_BLOCKS - 1);
5424 ext4_fc_track_range(
5426 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5427 inode->i_sb->s_blocksize_bits,
5428 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5429 inode->i_sb->s_blocksize_bits);
5431 down_write(&EXT4_I(inode)->i_data_sem);
5432 EXT4_I(inode)->i_disksize = attr->ia_size;
5433 rc = ext4_mark_inode_dirty(handle, inode);
5437 * We have to update i_size under i_data_sem together
5438 * with i_disksize to avoid races with writeback code
5439 * running ext4_wb_update_i_disksize().
5442 i_size_write(inode, attr->ia_size);
5443 up_write(&EXT4_I(inode)->i_data_sem);
5444 ext4_journal_stop(handle);
5448 pagecache_isize_extended(inode, oldsize,
5450 } else if (ext4_should_journal_data(inode)) {
5451 ext4_wait_for_tail_page_commit(inode);
5456 * Truncate pagecache after we've waited for commit
5457 * in data=journal mode to make pages freeable.
5459 truncate_pagecache(inode, inode->i_size);
5461 * Call ext4_truncate() even if i_size didn't change to
5462 * truncate possible preallocated blocks.
5464 if (attr->ia_size <= oldsize) {
5465 rc = ext4_truncate(inode);
5470 filemap_invalidate_unlock(inode->i_mapping);
5474 setattr_copy(mnt_userns, inode, attr);
5475 mark_inode_dirty(inode);
5479 * If the call to ext4_truncate failed to get a transaction handle at
5480 * all, we need to clean up the in-core orphan list manually.
5482 if (orphan && inode->i_nlink)
5483 ext4_orphan_del(NULL, inode);
5485 if (!error && (ia_valid & ATTR_MODE))
5486 rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5490 ext4_std_error(inode->i_sb, error);
5496 int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5497 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5499 struct inode *inode = d_inode(path->dentry);
5500 struct ext4_inode *raw_inode;
5501 struct ext4_inode_info *ei = EXT4_I(inode);
5504 if ((request_mask & STATX_BTIME) &&
5505 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5506 stat->result_mask |= STATX_BTIME;
5507 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5508 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5511 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5512 if (flags & EXT4_APPEND_FL)
5513 stat->attributes |= STATX_ATTR_APPEND;
5514 if (flags & EXT4_COMPR_FL)
5515 stat->attributes |= STATX_ATTR_COMPRESSED;
5516 if (flags & EXT4_ENCRYPT_FL)
5517 stat->attributes |= STATX_ATTR_ENCRYPTED;
5518 if (flags & EXT4_IMMUTABLE_FL)
5519 stat->attributes |= STATX_ATTR_IMMUTABLE;
5520 if (flags & EXT4_NODUMP_FL)
5521 stat->attributes |= STATX_ATTR_NODUMP;
5522 if (flags & EXT4_VERITY_FL)
5523 stat->attributes |= STATX_ATTR_VERITY;
5525 stat->attributes_mask |= (STATX_ATTR_APPEND |
5526 STATX_ATTR_COMPRESSED |
5527 STATX_ATTR_ENCRYPTED |
5528 STATX_ATTR_IMMUTABLE |
5532 generic_fillattr(mnt_userns, inode, stat);
5536 int ext4_file_getattr(struct user_namespace *mnt_userns,
5537 const struct path *path, struct kstat *stat,
5538 u32 request_mask, unsigned int query_flags)
5540 struct inode *inode = d_inode(path->dentry);
5541 u64 delalloc_blocks;
5543 ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5546 * If there is inline data in the inode, the inode will normally not
5547 * have data blocks allocated (it may have an external xattr block).
5548 * Report at least one sector for such files, so tools like tar, rsync,
5549 * others don't incorrectly think the file is completely sparse.
5551 if (unlikely(ext4_has_inline_data(inode)))
5552 stat->blocks += (stat->size + 511) >> 9;
5555 * We can't update i_blocks if the block allocation is delayed
5556 * otherwise in the case of system crash before the real block
5557 * allocation is done, we will have i_blocks inconsistent with
5558 * on-disk file blocks.
5559 * We always keep i_blocks updated together with real
5560 * allocation. But to not confuse with user, stat
5561 * will return the blocks that include the delayed allocation
5562 * blocks for this file.
5564 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5565 EXT4_I(inode)->i_reserved_data_blocks);
5566 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5570 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5573 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5574 return ext4_ind_trans_blocks(inode, lblocks);
5575 return ext4_ext_index_trans_blocks(inode, pextents);
5579 * Account for index blocks, block groups bitmaps and block group
5580 * descriptor blocks if modify datablocks and index blocks
5581 * worse case, the indexs blocks spread over different block groups
5583 * If datablocks are discontiguous, they are possible to spread over
5584 * different block groups too. If they are contiguous, with flexbg,
5585 * they could still across block group boundary.
5587 * Also account for superblock, inode, quota and xattr blocks
5589 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5592 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5598 * How many index blocks need to touch to map @lblocks logical blocks
5599 * to @pextents physical extents?
5601 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5606 * Now let's see how many group bitmaps and group descriptors need
5609 groups = idxblocks + pextents;
5611 if (groups > ngroups)
5613 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5614 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5616 /* bitmaps and block group descriptor blocks */
5617 ret += groups + gdpblocks;
5619 /* Blocks for super block, inode, quota and xattr blocks */
5620 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5626 * Calculate the total number of credits to reserve to fit
5627 * the modification of a single pages into a single transaction,
5628 * which may include multiple chunks of block allocations.
5630 * This could be called via ext4_write_begin()
5632 * We need to consider the worse case, when
5633 * one new block per extent.
5635 int ext4_writepage_trans_blocks(struct inode *inode)
5637 int bpp = ext4_journal_blocks_per_page(inode);
5640 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5642 /* Account for data blocks for journalled mode */
5643 if (ext4_should_journal_data(inode))
5649 * Calculate the journal credits for a chunk of data modification.
5651 * This is called from DIO, fallocate or whoever calling
5652 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5654 * journal buffers for data blocks are not included here, as DIO
5655 * and fallocate do no need to journal data buffers.
5657 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5659 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5663 * The caller must have previously called ext4_reserve_inode_write().
5664 * Give this, we know that the caller already has write access to iloc->bh.
5666 int ext4_mark_iloc_dirty(handle_t *handle,
5667 struct inode *inode, struct ext4_iloc *iloc)
5671 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5675 ext4_fc_track_inode(handle, inode);
5677 if (IS_I_VERSION(inode))
5678 inode_inc_iversion(inode);
5680 /* the do_update_inode consumes one bh->b_count */
5683 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5684 err = ext4_do_update_inode(handle, inode, iloc);
5690 * On success, We end up with an outstanding reference count against
5691 * iloc->bh. This _must_ be cleaned up later.
5695 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5696 struct ext4_iloc *iloc)
5700 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5703 err = ext4_get_inode_loc(inode, iloc);
5705 BUFFER_TRACE(iloc->bh, "get_write_access");
5706 err = ext4_journal_get_write_access(handle, inode->i_sb,
5707 iloc->bh, EXT4_JTR_NONE);
5713 ext4_std_error(inode->i_sb, err);
5717 static int __ext4_expand_extra_isize(struct inode *inode,
5718 unsigned int new_extra_isize,
5719 struct ext4_iloc *iloc,
5720 handle_t *handle, int *no_expand)
5722 struct ext4_inode *raw_inode;
5723 struct ext4_xattr_ibody_header *header;
5724 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5725 struct ext4_inode_info *ei = EXT4_I(inode);
5728 /* this was checked at iget time, but double check for good measure */
5729 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5730 (ei->i_extra_isize & 3)) {
5731 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5733 EXT4_INODE_SIZE(inode->i_sb));
5734 return -EFSCORRUPTED;
5736 if ((new_extra_isize < ei->i_extra_isize) ||
5737 (new_extra_isize < 4) ||
5738 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5739 return -EINVAL; /* Should never happen */
5741 raw_inode = ext4_raw_inode(iloc);
5743 header = IHDR(inode, raw_inode);
5745 /* No extended attributes present */
5746 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5747 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5748 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5749 EXT4_I(inode)->i_extra_isize, 0,
5750 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5751 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5755 /* try to expand with EAs present */
5756 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5760 * Inode size expansion failed; don't try again
5769 * Expand an inode by new_extra_isize bytes.
5770 * Returns 0 on success or negative error number on failure.
5772 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5773 unsigned int new_extra_isize,
5774 struct ext4_iloc iloc,
5780 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5784 * In nojournal mode, we can immediately attempt to expand
5785 * the inode. When journaled, we first need to obtain extra
5786 * buffer credits since we may write into the EA block
5787 * with this same handle. If journal_extend fails, then it will
5788 * only result in a minor loss of functionality for that inode.
5789 * If this is felt to be critical, then e2fsck should be run to
5790 * force a large enough s_min_extra_isize.
5792 if (ext4_journal_extend(handle,
5793 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5796 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5799 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5800 handle, &no_expand);
5801 ext4_write_unlock_xattr(inode, &no_expand);
5806 int ext4_expand_extra_isize(struct inode *inode,
5807 unsigned int new_extra_isize,
5808 struct ext4_iloc *iloc)
5814 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5819 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5820 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5821 if (IS_ERR(handle)) {
5822 error = PTR_ERR(handle);
5827 ext4_write_lock_xattr(inode, &no_expand);
5829 BUFFER_TRACE(iloc->bh, "get_write_access");
5830 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5837 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5838 handle, &no_expand);
5840 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5845 ext4_write_unlock_xattr(inode, &no_expand);
5846 ext4_journal_stop(handle);
5851 * What we do here is to mark the in-core inode as clean with respect to inode
5852 * dirtiness (it may still be data-dirty).
5853 * This means that the in-core inode may be reaped by prune_icache
5854 * without having to perform any I/O. This is a very good thing,
5855 * because *any* task may call prune_icache - even ones which
5856 * have a transaction open against a different journal.
5858 * Is this cheating? Not really. Sure, we haven't written the
5859 * inode out, but prune_icache isn't a user-visible syncing function.
5860 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5861 * we start and wait on commits.
5863 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5864 const char *func, unsigned int line)
5866 struct ext4_iloc iloc;
5867 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5871 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5872 err = ext4_reserve_inode_write(handle, inode, &iloc);
5876 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5877 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5880 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5883 ext4_error_inode_err(inode, func, line, 0, err,
5884 "mark_inode_dirty error");
5889 * ext4_dirty_inode() is called from __mark_inode_dirty()
5891 * We're really interested in the case where a file is being extended.
5892 * i_size has been changed by generic_commit_write() and we thus need
5893 * to include the updated inode in the current transaction.
5895 * Also, dquot_alloc_block() will always dirty the inode when blocks
5896 * are allocated to the file.
5898 * If the inode is marked synchronous, we don't honour that here - doing
5899 * so would cause a commit on atime updates, which we don't bother doing.
5900 * We handle synchronous inodes at the highest possible level.
5902 void ext4_dirty_inode(struct inode *inode, int flags)
5906 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5909 ext4_mark_inode_dirty(handle, inode);
5910 ext4_journal_stop(handle);
5913 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5918 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5921 * We have to be very careful here: changing a data block's
5922 * journaling status dynamically is dangerous. If we write a
5923 * data block to the journal, change the status and then delete
5924 * that block, we risk forgetting to revoke the old log record
5925 * from the journal and so a subsequent replay can corrupt data.
5926 * So, first we make sure that the journal is empty and that
5927 * nobody is changing anything.
5930 journal = EXT4_JOURNAL(inode);
5933 if (is_journal_aborted(journal))
5936 /* Wait for all existing dio workers */
5937 inode_dio_wait(inode);
5940 * Before flushing the journal and switching inode's aops, we have
5941 * to flush all dirty data the inode has. There can be outstanding
5942 * delayed allocations, there can be unwritten extents created by
5943 * fallocate or buffered writes in dioread_nolock mode covered by
5944 * dirty data which can be converted only after flushing the dirty
5945 * data (and journalled aops don't know how to handle these cases).
5948 filemap_invalidate_lock(inode->i_mapping);
5949 err = filemap_write_and_wait(inode->i_mapping);
5951 filemap_invalidate_unlock(inode->i_mapping);
5956 percpu_down_write(&sbi->s_writepages_rwsem);
5957 jbd2_journal_lock_updates(journal);
5960 * OK, there are no updates running now, and all cached data is
5961 * synced to disk. We are now in a completely consistent state
5962 * which doesn't have anything in the journal, and we know that
5963 * no filesystem updates are running, so it is safe to modify
5964 * the inode's in-core data-journaling state flag now.
5968 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5970 err = jbd2_journal_flush(journal, 0);
5972 jbd2_journal_unlock_updates(journal);
5973 percpu_up_write(&sbi->s_writepages_rwsem);
5976 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5978 ext4_set_aops(inode);
5980 jbd2_journal_unlock_updates(journal);
5981 percpu_up_write(&sbi->s_writepages_rwsem);
5984 filemap_invalidate_unlock(inode->i_mapping);
5986 /* Finally we can mark the inode as dirty. */
5988 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5990 return PTR_ERR(handle);
5992 ext4_fc_mark_ineligible(inode->i_sb,
5993 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
5994 err = ext4_mark_inode_dirty(handle, inode);
5995 ext4_handle_sync(handle);
5996 ext4_journal_stop(handle);
5997 ext4_std_error(inode->i_sb, err);
6002 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6003 struct buffer_head *bh)
6005 return !buffer_mapped(bh);
6008 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6010 struct vm_area_struct *vma = vmf->vma;
6011 struct page *page = vmf->page;
6016 struct file *file = vma->vm_file;
6017 struct inode *inode = file_inode(file);
6018 struct address_space *mapping = inode->i_mapping;
6020 get_block_t *get_block;
6023 if (unlikely(IS_IMMUTABLE(inode)))
6024 return VM_FAULT_SIGBUS;
6026 sb_start_pagefault(inode->i_sb);
6027 file_update_time(vma->vm_file);
6029 filemap_invalidate_lock_shared(mapping);
6031 err = ext4_convert_inline_data(inode);
6036 * On data journalling we skip straight to the transaction handle:
6037 * there's no delalloc; page truncated will be checked later; the
6038 * early return w/ all buffers mapped (calculates size/len) can't
6039 * be used; and there's no dioread_nolock, so only ext4_get_block.
6041 if (ext4_should_journal_data(inode))
6044 /* Delalloc case is easy... */
6045 if (test_opt(inode->i_sb, DELALLOC) &&
6046 !ext4_nonda_switch(inode->i_sb)) {
6048 err = block_page_mkwrite(vma, vmf,
6049 ext4_da_get_block_prep);
6050 } while (err == -ENOSPC &&
6051 ext4_should_retry_alloc(inode->i_sb, &retries));
6056 size = i_size_read(inode);
6057 /* Page got truncated from under us? */
6058 if (page->mapping != mapping || page_offset(page) > size) {
6060 ret = VM_FAULT_NOPAGE;
6064 if (page->index == size >> PAGE_SHIFT)
6065 len = size & ~PAGE_MASK;
6069 * Return if we have all the buffers mapped. This avoids the need to do
6070 * journal_start/journal_stop which can block and take a long time
6072 * This cannot be done for data journalling, as we have to add the
6073 * inode to the transaction's list to writeprotect pages on commit.
6075 if (page_has_buffers(page)) {
6076 if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page),
6078 ext4_bh_unmapped)) {
6079 /* Wait so that we don't change page under IO */
6080 wait_for_stable_page(page);
6081 ret = VM_FAULT_LOCKED;
6086 /* OK, we need to fill the hole... */
6087 if (ext4_should_dioread_nolock(inode))
6088 get_block = ext4_get_block_unwritten;
6090 get_block = ext4_get_block;
6092 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6093 ext4_writepage_trans_blocks(inode));
6094 if (IS_ERR(handle)) {
6095 ret = VM_FAULT_SIGBUS;
6099 * Data journalling can't use block_page_mkwrite() because it
6100 * will set_buffer_dirty() before do_journal_get_write_access()
6101 * thus might hit warning messages for dirty metadata buffers.
6103 if (!ext4_should_journal_data(inode)) {
6104 err = block_page_mkwrite(vma, vmf, get_block);
6107 size = i_size_read(inode);
6108 /* Page got truncated from under us? */
6109 if (page->mapping != mapping || page_offset(page) > size) {
6110 ret = VM_FAULT_NOPAGE;
6114 if (page->index == size >> PAGE_SHIFT)
6115 len = size & ~PAGE_MASK;
6119 err = __block_write_begin(page, 0, len, ext4_get_block);
6121 ret = VM_FAULT_SIGBUS;
6122 if (ext4_walk_page_buffers(handle, inode,
6123 page_buffers(page), 0, len, NULL,
6124 do_journal_get_write_access))
6126 if (ext4_walk_page_buffers(handle, inode,
6127 page_buffers(page), 0, len, NULL,
6130 if (ext4_jbd2_inode_add_write(handle, inode,
6131 page_offset(page), len))
6133 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6138 ext4_journal_stop(handle);
6139 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6142 ret = block_page_mkwrite_return(err);
6144 filemap_invalidate_unlock_shared(mapping);
6145 sb_end_pagefault(inode->i_sb);
6149 ext4_journal_stop(handle);