1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 struct ext4_inode_info *ei)
54 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 int offset = offsetof(struct ext4_inode, i_checksum_lo);
58 unsigned int csum_size = sizeof(dummy_csum);
60 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
61 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
64 EXT4_GOOD_OLD_INODE_SIZE - offset);
66 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
67 offset = offsetof(struct ext4_inode, i_checksum_hi);
68 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
69 EXT4_GOOD_OLD_INODE_SIZE,
70 offset - EXT4_GOOD_OLD_INODE_SIZE);
71 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
72 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
76 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
77 EXT4_INODE_SIZE(inode->i_sb) - offset);
83 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
84 struct ext4_inode_info *ei)
86 __u32 provided, calculated;
88 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
89 cpu_to_le32(EXT4_OS_LINUX) ||
90 !ext4_has_metadata_csum(inode->i_sb))
93 provided = le16_to_cpu(raw->i_checksum_lo);
94 calculated = ext4_inode_csum(inode, raw, ei);
95 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
96 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
97 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 return provided == calculated;
104 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
105 struct ext4_inode_info *ei)
109 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
110 cpu_to_le32(EXT4_OS_LINUX) ||
111 !ext4_has_metadata_csum(inode->i_sb))
114 csum = ext4_inode_csum(inode, raw, ei);
115 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
116 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
117 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
118 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode *inode,
124 trace_ext4_begin_ordered_truncate(inode, new_size);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode)->jinode)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
134 EXT4_I(inode)->jinode,
138 static void ext4_invalidatepage(struct page *page, unsigned int offset,
139 unsigned int length);
140 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
141 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
142 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
146 * Test whether an inode is a fast symlink.
147 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
149 int ext4_inode_is_fast_symlink(struct inode *inode)
151 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
152 int ea_blocks = EXT4_I(inode)->i_file_acl ?
153 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
155 if (ext4_has_inline_data(inode))
158 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
160 return S_ISLNK(inode->i_mode) && inode->i_size &&
161 (inode->i_size < EXT4_N_BLOCKS * 4);
165 * Called at the last iput() if i_nlink is zero.
167 void ext4_evict_inode(struct inode *inode)
172 * Credits for final inode cleanup and freeing:
173 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
174 * (xattr block freeing), bitmap, group descriptor (inode freeing)
176 int extra_credits = 6;
177 struct ext4_xattr_inode_array *ea_inode_array = NULL;
179 trace_ext4_evict_inode(inode);
181 if (inode->i_nlink) {
183 * When journalling data dirty buffers are tracked only in the
184 * journal. So although mm thinks everything is clean and
185 * ready for reaping the inode might still have some pages to
186 * write in the running transaction or waiting to be
187 * checkpointed. Thus calling jbd2_journal_invalidatepage()
188 * (via truncate_inode_pages()) to discard these buffers can
189 * cause data loss. Also even if we did not discard these
190 * buffers, we would have no way to find them after the inode
191 * is reaped and thus user could see stale data if he tries to
192 * read them before the transaction is checkpointed. So be
193 * careful and force everything to disk here... We use
194 * ei->i_datasync_tid to store the newest transaction
195 * containing inode's data.
197 * Note that directories do not have this problem because they
198 * don't use page cache.
200 if (inode->i_ino != EXT4_JOURNAL_INO &&
201 ext4_should_journal_data(inode) &&
202 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
203 inode->i_data.nrpages) {
204 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
205 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
207 jbd2_complete_transaction(journal, commit_tid);
208 filemap_write_and_wait(&inode->i_data);
210 truncate_inode_pages_final(&inode->i_data);
215 if (is_bad_inode(inode))
217 dquot_initialize(inode);
219 if (ext4_should_order_data(inode))
220 ext4_begin_ordered_truncate(inode, 0);
221 truncate_inode_pages_final(&inode->i_data);
224 * Protect us against freezing - iput() caller didn't have to have any
225 * protection against it
227 sb_start_intwrite(inode->i_sb);
229 if (!IS_NOQUOTA(inode))
230 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
233 * Block bitmap, group descriptor, and inode are accounted in both
234 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
236 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
237 ext4_blocks_for_truncate(inode) + extra_credits - 3);
238 if (IS_ERR(handle)) {
239 ext4_std_error(inode->i_sb, PTR_ERR(handle));
241 * If we're going to skip the normal cleanup, we still need to
242 * make sure that the in-core orphan linked list is properly
245 ext4_orphan_del(NULL, inode);
246 sb_end_intwrite(inode->i_sb);
251 ext4_handle_sync(handle);
254 * Set inode->i_size to 0 before calling ext4_truncate(). We need
255 * special handling of symlinks here because i_size is used to
256 * determine whether ext4_inode_info->i_data contains symlink data or
257 * block mappings. Setting i_size to 0 will remove its fast symlink
258 * status. Erase i_data so that it becomes a valid empty block map.
260 if (ext4_inode_is_fast_symlink(inode))
261 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
263 err = ext4_mark_inode_dirty(handle, inode);
265 ext4_warning(inode->i_sb,
266 "couldn't mark inode dirty (err %d)", err);
269 if (inode->i_blocks) {
270 err = ext4_truncate(inode);
272 ext4_set_errno(inode->i_sb, -err);
273 ext4_error(inode->i_sb,
274 "couldn't truncate inode %lu (err %d)",
280 /* Remove xattr references. */
281 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
284 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
286 ext4_journal_stop(handle);
287 ext4_orphan_del(NULL, inode);
288 sb_end_intwrite(inode->i_sb);
289 ext4_xattr_inode_array_free(ea_inode_array);
294 * Kill off the orphan record which ext4_truncate created.
295 * AKPM: I think this can be inside the above `if'.
296 * Note that ext4_orphan_del() has to be able to cope with the
297 * deletion of a non-existent orphan - this is because we don't
298 * know if ext4_truncate() actually created an orphan record.
299 * (Well, we could do this if we need to, but heck - it works)
301 ext4_orphan_del(handle, inode);
302 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
305 * One subtle ordering requirement: if anything has gone wrong
306 * (transaction abort, IO errors, whatever), then we can still
307 * do these next steps (the fs will already have been marked as
308 * having errors), but we can't free the inode if the mark_dirty
311 if (ext4_mark_inode_dirty(handle, inode))
312 /* If that failed, just do the required in-core inode clear. */
313 ext4_clear_inode(inode);
315 ext4_free_inode(handle, inode);
316 ext4_journal_stop(handle);
317 sb_end_intwrite(inode->i_sb);
318 ext4_xattr_inode_array_free(ea_inode_array);
321 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
325 qsize_t *ext4_get_reserved_space(struct inode *inode)
327 return &EXT4_I(inode)->i_reserved_quota;
332 * Called with i_data_sem down, which is important since we can call
333 * ext4_discard_preallocations() from here.
335 void ext4_da_update_reserve_space(struct inode *inode,
336 int used, int quota_claim)
338 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
339 struct ext4_inode_info *ei = EXT4_I(inode);
341 spin_lock(&ei->i_block_reservation_lock);
342 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
343 if (unlikely(used > ei->i_reserved_data_blocks)) {
344 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
345 "with only %d reserved data blocks",
346 __func__, inode->i_ino, used,
347 ei->i_reserved_data_blocks);
349 used = ei->i_reserved_data_blocks;
352 /* Update per-inode reservations */
353 ei->i_reserved_data_blocks -= used;
354 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
356 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
358 /* Update quota subsystem for data blocks */
360 dquot_claim_block(inode, EXT4_C2B(sbi, used));
363 * We did fallocate with an offset that is already delayed
364 * allocated. So on delayed allocated writeback we should
365 * not re-claim the quota for fallocated blocks.
367 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
371 * If we have done all the pending block allocations and if
372 * there aren't any writers on the inode, we can discard the
373 * inode's preallocations.
375 if ((ei->i_reserved_data_blocks == 0) &&
376 !inode_is_open_for_write(inode))
377 ext4_discard_preallocations(inode);
380 static int __check_block_validity(struct inode *inode, const char *func,
382 struct ext4_map_blocks *map)
384 if (ext4_has_feature_journal(inode->i_sb) &&
386 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
388 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
390 ext4_error_inode(inode, func, line, map->m_pblk,
391 "lblock %lu mapped to illegal pblock %llu "
392 "(length %d)", (unsigned long) map->m_lblk,
393 map->m_pblk, map->m_len);
394 return -EFSCORRUPTED;
399 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
404 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
405 return fscrypt_zeroout_range(inode, lblk, pblk, len);
407 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
414 #define check_block_validity(inode, map) \
415 __check_block_validity((inode), __func__, __LINE__, (map))
417 #ifdef ES_AGGRESSIVE_TEST
418 static void ext4_map_blocks_es_recheck(handle_t *handle,
420 struct ext4_map_blocks *es_map,
421 struct ext4_map_blocks *map,
428 * There is a race window that the result is not the same.
429 * e.g. xfstests #223 when dioread_nolock enables. The reason
430 * is that we lookup a block mapping in extent status tree with
431 * out taking i_data_sem. So at the time the unwritten extent
432 * could be converted.
434 down_read(&EXT4_I(inode)->i_data_sem);
435 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
436 retval = ext4_ext_map_blocks(handle, inode, map, flags &
437 EXT4_GET_BLOCKS_KEEP_SIZE);
439 retval = ext4_ind_map_blocks(handle, inode, map, flags &
440 EXT4_GET_BLOCKS_KEEP_SIZE);
442 up_read((&EXT4_I(inode)->i_data_sem));
445 * We don't check m_len because extent will be collpased in status
446 * tree. So the m_len might not equal.
448 if (es_map->m_lblk != map->m_lblk ||
449 es_map->m_flags != map->m_flags ||
450 es_map->m_pblk != map->m_pblk) {
451 printk("ES cache assertion failed for inode: %lu "
452 "es_cached ex [%d/%d/%llu/%x] != "
453 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
454 inode->i_ino, es_map->m_lblk, es_map->m_len,
455 es_map->m_pblk, es_map->m_flags, map->m_lblk,
456 map->m_len, map->m_pblk, map->m_flags,
460 #endif /* ES_AGGRESSIVE_TEST */
463 * The ext4_map_blocks() function tries to look up the requested blocks,
464 * and returns if the blocks are already mapped.
466 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
467 * and store the allocated blocks in the result buffer head and mark it
470 * If file type is extents based, it will call ext4_ext_map_blocks(),
471 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
474 * On success, it returns the number of blocks being mapped or allocated. if
475 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
476 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
478 * It returns 0 if plain look up failed (blocks have not been allocated), in
479 * that case, @map is returned as unmapped but we still do fill map->m_len to
480 * indicate the length of a hole starting at map->m_lblk.
482 * It returns the error in case of allocation failure.
484 int ext4_map_blocks(handle_t *handle, struct inode *inode,
485 struct ext4_map_blocks *map, int flags)
487 struct extent_status es;
490 #ifdef ES_AGGRESSIVE_TEST
491 struct ext4_map_blocks orig_map;
493 memcpy(&orig_map, map, sizeof(*map));
497 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
498 "logical block %lu\n", inode->i_ino, flags, map->m_len,
499 (unsigned long) map->m_lblk);
502 * ext4_map_blocks returns an int, and m_len is an unsigned int
504 if (unlikely(map->m_len > INT_MAX))
505 map->m_len = INT_MAX;
507 /* We can handle the block number less than EXT_MAX_BLOCKS */
508 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
509 return -EFSCORRUPTED;
511 /* Lookup extent status tree firstly */
512 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
513 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
514 map->m_pblk = ext4_es_pblock(&es) +
515 map->m_lblk - es.es_lblk;
516 map->m_flags |= ext4_es_is_written(&es) ?
517 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
518 retval = es.es_len - (map->m_lblk - es.es_lblk);
519 if (retval > map->m_len)
522 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
524 retval = es.es_len - (map->m_lblk - es.es_lblk);
525 if (retval > map->m_len)
532 #ifdef ES_AGGRESSIVE_TEST
533 ext4_map_blocks_es_recheck(handle, inode, map,
540 * Try to see if we can get the block without requesting a new
543 down_read(&EXT4_I(inode)->i_data_sem);
544 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
545 retval = ext4_ext_map_blocks(handle, inode, map, flags &
546 EXT4_GET_BLOCKS_KEEP_SIZE);
548 retval = ext4_ind_map_blocks(handle, inode, map, flags &
549 EXT4_GET_BLOCKS_KEEP_SIZE);
554 if (unlikely(retval != map->m_len)) {
555 ext4_warning(inode->i_sb,
556 "ES len assertion failed for inode "
557 "%lu: retval %d != map->m_len %d",
558 inode->i_ino, retval, map->m_len);
562 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
563 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
564 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
565 !(status & EXTENT_STATUS_WRITTEN) &&
566 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
567 map->m_lblk + map->m_len - 1))
568 status |= EXTENT_STATUS_DELAYED;
569 ret = ext4_es_insert_extent(inode, map->m_lblk,
570 map->m_len, map->m_pblk, status);
574 up_read((&EXT4_I(inode)->i_data_sem));
577 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
578 ret = check_block_validity(inode, map);
583 /* If it is only a block(s) look up */
584 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
588 * Returns if the blocks have already allocated
590 * Note that if blocks have been preallocated
591 * ext4_ext_get_block() returns the create = 0
592 * with buffer head unmapped.
594 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
596 * If we need to convert extent to unwritten
597 * we continue and do the actual work in
598 * ext4_ext_map_blocks()
600 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
604 * Here we clear m_flags because after allocating an new extent,
605 * it will be set again.
607 map->m_flags &= ~EXT4_MAP_FLAGS;
610 * New blocks allocate and/or writing to unwritten extent
611 * will possibly result in updating i_data, so we take
612 * the write lock of i_data_sem, and call get_block()
613 * with create == 1 flag.
615 down_write(&EXT4_I(inode)->i_data_sem);
618 * We need to check for EXT4 here because migrate
619 * could have changed the inode type in between
621 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
622 retval = ext4_ext_map_blocks(handle, inode, map, flags);
624 retval = ext4_ind_map_blocks(handle, inode, map, flags);
626 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
628 * We allocated new blocks which will result in
629 * i_data's format changing. Force the migrate
630 * to fail by clearing migrate flags
632 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
636 * Update reserved blocks/metadata blocks after successful
637 * block allocation which had been deferred till now. We don't
638 * support fallocate for non extent files. So we can update
639 * reserve space here.
642 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
643 ext4_da_update_reserve_space(inode, retval, 1);
649 if (unlikely(retval != map->m_len)) {
650 ext4_warning(inode->i_sb,
651 "ES len assertion failed for inode "
652 "%lu: retval %d != map->m_len %d",
653 inode->i_ino, retval, map->m_len);
658 * We have to zeroout blocks before inserting them into extent
659 * status tree. Otherwise someone could look them up there and
660 * use them before they are really zeroed. We also have to
661 * unmap metadata before zeroing as otherwise writeback can
662 * overwrite zeros with stale data from block device.
664 if (flags & EXT4_GET_BLOCKS_ZERO &&
665 map->m_flags & EXT4_MAP_MAPPED &&
666 map->m_flags & EXT4_MAP_NEW) {
667 ret = ext4_issue_zeroout(inode, map->m_lblk,
668 map->m_pblk, map->m_len);
676 * If the extent has been zeroed out, we don't need to update
677 * extent status tree.
679 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
680 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
681 if (ext4_es_is_written(&es))
684 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
685 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
686 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
687 !(status & EXTENT_STATUS_WRITTEN) &&
688 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
689 map->m_lblk + map->m_len - 1))
690 status |= EXTENT_STATUS_DELAYED;
691 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
692 map->m_pblk, status);
700 up_write((&EXT4_I(inode)->i_data_sem));
701 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
702 ret = check_block_validity(inode, map);
707 * Inodes with freshly allocated blocks where contents will be
708 * visible after transaction commit must be on transaction's
711 if (map->m_flags & EXT4_MAP_NEW &&
712 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
713 !(flags & EXT4_GET_BLOCKS_ZERO) &&
714 !ext4_is_quota_file(inode) &&
715 ext4_should_order_data(inode)) {
717 (loff_t)map->m_lblk << inode->i_blkbits;
718 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
720 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
721 ret = ext4_jbd2_inode_add_wait(handle, inode,
724 ret = ext4_jbd2_inode_add_write(handle, inode,
734 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
735 * we have to be careful as someone else may be manipulating b_state as well.
737 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
739 unsigned long old_state;
740 unsigned long new_state;
742 flags &= EXT4_MAP_FLAGS;
744 /* Dummy buffer_head? Set non-atomically. */
746 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
750 * Someone else may be modifying b_state. Be careful! This is ugly but
751 * once we get rid of using bh as a container for mapping information
752 * to pass to / from get_block functions, this can go away.
755 old_state = READ_ONCE(bh->b_state);
756 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
758 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
761 static int _ext4_get_block(struct inode *inode, sector_t iblock,
762 struct buffer_head *bh, int flags)
764 struct ext4_map_blocks map;
767 if (ext4_has_inline_data(inode))
771 map.m_len = bh->b_size >> inode->i_blkbits;
773 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
776 map_bh(bh, inode->i_sb, map.m_pblk);
777 ext4_update_bh_state(bh, map.m_flags);
778 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
780 } else if (ret == 0) {
781 /* hole case, need to fill in bh->b_size */
782 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
787 int ext4_get_block(struct inode *inode, sector_t iblock,
788 struct buffer_head *bh, int create)
790 return _ext4_get_block(inode, iblock, bh,
791 create ? EXT4_GET_BLOCKS_CREATE : 0);
795 * Get block function used when preparing for buffered write if we require
796 * creating an unwritten extent if blocks haven't been allocated. The extent
797 * will be converted to written after the IO is complete.
799 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
800 struct buffer_head *bh_result, int create)
802 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
803 inode->i_ino, create);
804 return _ext4_get_block(inode, iblock, bh_result,
805 EXT4_GET_BLOCKS_IO_CREATE_EXT);
808 /* Maximum number of blocks we map for direct IO at once. */
809 #define DIO_MAX_BLOCKS 4096
812 * `handle' can be NULL if create is zero
814 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
815 ext4_lblk_t block, int map_flags)
817 struct ext4_map_blocks map;
818 struct buffer_head *bh;
819 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
822 J_ASSERT(handle != NULL || create == 0);
826 err = ext4_map_blocks(handle, inode, &map, map_flags);
829 return create ? ERR_PTR(-ENOSPC) : NULL;
833 bh = sb_getblk(inode->i_sb, map.m_pblk);
835 return ERR_PTR(-ENOMEM);
836 if (map.m_flags & EXT4_MAP_NEW) {
837 J_ASSERT(create != 0);
838 J_ASSERT(handle != NULL);
841 * Now that we do not always journal data, we should
842 * keep in mind whether this should always journal the
843 * new buffer as metadata. For now, regular file
844 * writes use ext4_get_block instead, so it's not a
848 BUFFER_TRACE(bh, "call get_create_access");
849 err = ext4_journal_get_create_access(handle, bh);
854 if (!buffer_uptodate(bh)) {
855 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
856 set_buffer_uptodate(bh);
859 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
860 err = ext4_handle_dirty_metadata(handle, inode, bh);
864 BUFFER_TRACE(bh, "not a new buffer");
871 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
872 ext4_lblk_t block, int map_flags)
874 struct buffer_head *bh;
876 bh = ext4_getblk(handle, inode, block, map_flags);
879 if (!bh || ext4_buffer_uptodate(bh))
881 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
883 if (buffer_uptodate(bh))
886 return ERR_PTR(-EIO);
889 /* Read a contiguous batch of blocks. */
890 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
891 bool wait, struct buffer_head **bhs)
895 for (i = 0; i < bh_count; i++) {
896 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
897 if (IS_ERR(bhs[i])) {
898 err = PTR_ERR(bhs[i]);
904 for (i = 0; i < bh_count; i++)
905 /* Note that NULL bhs[i] is valid because of holes. */
906 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
907 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
913 for (i = 0; i < bh_count; i++)
915 wait_on_buffer(bhs[i]);
917 for (i = 0; i < bh_count; i++) {
918 if (bhs[i] && !buffer_uptodate(bhs[i])) {
926 for (i = 0; i < bh_count; i++) {
933 int ext4_walk_page_buffers(handle_t *handle,
934 struct buffer_head *head,
938 int (*fn)(handle_t *handle,
939 struct buffer_head *bh))
941 struct buffer_head *bh;
942 unsigned block_start, block_end;
943 unsigned blocksize = head->b_size;
945 struct buffer_head *next;
947 for (bh = head, block_start = 0;
948 ret == 0 && (bh != head || !block_start);
949 block_start = block_end, bh = next) {
950 next = bh->b_this_page;
951 block_end = block_start + blocksize;
952 if (block_end <= from || block_start >= to) {
953 if (partial && !buffer_uptodate(bh))
957 err = (*fn)(handle, bh);
965 * To preserve ordering, it is essential that the hole instantiation and
966 * the data write be encapsulated in a single transaction. We cannot
967 * close off a transaction and start a new one between the ext4_get_block()
968 * and the commit_write(). So doing the jbd2_journal_start at the start of
969 * prepare_write() is the right place.
971 * Also, this function can nest inside ext4_writepage(). In that case, we
972 * *know* that ext4_writepage() has generated enough buffer credits to do the
973 * whole page. So we won't block on the journal in that case, which is good,
974 * because the caller may be PF_MEMALLOC.
976 * By accident, ext4 can be reentered when a transaction is open via
977 * quota file writes. If we were to commit the transaction while thus
978 * reentered, there can be a deadlock - we would be holding a quota
979 * lock, and the commit would never complete if another thread had a
980 * transaction open and was blocking on the quota lock - a ranking
983 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
984 * will _not_ run commit under these circumstances because handle->h_ref
985 * is elevated. We'll still have enough credits for the tiny quotafile
988 int do_journal_get_write_access(handle_t *handle,
989 struct buffer_head *bh)
991 int dirty = buffer_dirty(bh);
994 if (!buffer_mapped(bh) || buffer_freed(bh))
997 * __block_write_begin() could have dirtied some buffers. Clean
998 * the dirty bit as jbd2_journal_get_write_access() could complain
999 * otherwise about fs integrity issues. Setting of the dirty bit
1000 * by __block_write_begin() isn't a real problem here as we clear
1001 * the bit before releasing a page lock and thus writeback cannot
1002 * ever write the buffer.
1005 clear_buffer_dirty(bh);
1006 BUFFER_TRACE(bh, "get write access");
1007 ret = ext4_journal_get_write_access(handle, bh);
1009 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1013 #ifdef CONFIG_FS_ENCRYPTION
1014 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1015 get_block_t *get_block)
1017 unsigned from = pos & (PAGE_SIZE - 1);
1018 unsigned to = from + len;
1019 struct inode *inode = page->mapping->host;
1020 unsigned block_start, block_end;
1023 unsigned blocksize = inode->i_sb->s_blocksize;
1025 struct buffer_head *bh, *head, *wait[2];
1029 BUG_ON(!PageLocked(page));
1030 BUG_ON(from > PAGE_SIZE);
1031 BUG_ON(to > PAGE_SIZE);
1034 if (!page_has_buffers(page))
1035 create_empty_buffers(page, blocksize, 0);
1036 head = page_buffers(page);
1037 bbits = ilog2(blocksize);
1038 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1040 for (bh = head, block_start = 0; bh != head || !block_start;
1041 block++, block_start = block_end, bh = bh->b_this_page) {
1042 block_end = block_start + blocksize;
1043 if (block_end <= from || block_start >= to) {
1044 if (PageUptodate(page)) {
1045 if (!buffer_uptodate(bh))
1046 set_buffer_uptodate(bh);
1051 clear_buffer_new(bh);
1052 if (!buffer_mapped(bh)) {
1053 WARN_ON(bh->b_size != blocksize);
1054 err = get_block(inode, block, bh, 1);
1057 if (buffer_new(bh)) {
1058 if (PageUptodate(page)) {
1059 clear_buffer_new(bh);
1060 set_buffer_uptodate(bh);
1061 mark_buffer_dirty(bh);
1064 if (block_end > to || block_start < from)
1065 zero_user_segments(page, to, block_end,
1070 if (PageUptodate(page)) {
1071 if (!buffer_uptodate(bh))
1072 set_buffer_uptodate(bh);
1075 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1076 !buffer_unwritten(bh) &&
1077 (block_start < from || block_end > to)) {
1078 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1079 wait[nr_wait++] = bh;
1083 * If we issued read requests, let them complete.
1085 for (i = 0; i < nr_wait; i++) {
1086 wait_on_buffer(wait[i]);
1087 if (!buffer_uptodate(wait[i]))
1090 if (unlikely(err)) {
1091 page_zero_new_buffers(page, from, to);
1092 } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
1093 for (i = 0; i < nr_wait; i++) {
1096 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1097 bh_offset(wait[i]));
1099 clear_buffer_uptodate(wait[i]);
1109 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1110 loff_t pos, unsigned len, unsigned flags,
1111 struct page **pagep, void **fsdata)
1113 struct inode *inode = mapping->host;
1114 int ret, needed_blocks;
1121 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1124 trace_ext4_write_begin(inode, pos, len, flags);
1126 * Reserve one block more for addition to orphan list in case
1127 * we allocate blocks but write fails for some reason
1129 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1130 index = pos >> PAGE_SHIFT;
1131 from = pos & (PAGE_SIZE - 1);
1134 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1135 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1144 * grab_cache_page_write_begin() can take a long time if the
1145 * system is thrashing due to memory pressure, or if the page
1146 * is being written back. So grab it first before we start
1147 * the transaction handle. This also allows us to allocate
1148 * the page (if needed) without using GFP_NOFS.
1151 page = grab_cache_page_write_begin(mapping, index, flags);
1157 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1158 if (IS_ERR(handle)) {
1160 return PTR_ERR(handle);
1164 if (page->mapping != mapping) {
1165 /* The page got truncated from under us */
1168 ext4_journal_stop(handle);
1171 /* In case writeback began while the page was unlocked */
1172 wait_for_stable_page(page);
1174 #ifdef CONFIG_FS_ENCRYPTION
1175 if (ext4_should_dioread_nolock(inode))
1176 ret = ext4_block_write_begin(page, pos, len,
1177 ext4_get_block_unwritten);
1179 ret = ext4_block_write_begin(page, pos, len,
1182 if (ext4_should_dioread_nolock(inode))
1183 ret = __block_write_begin(page, pos, len,
1184 ext4_get_block_unwritten);
1186 ret = __block_write_begin(page, pos, len, ext4_get_block);
1188 if (!ret && ext4_should_journal_data(inode)) {
1189 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1191 do_journal_get_write_access);
1195 bool extended = (pos + len > inode->i_size) &&
1196 !ext4_verity_in_progress(inode);
1200 * __block_write_begin may have instantiated a few blocks
1201 * outside i_size. Trim these off again. Don't need
1202 * i_size_read because we hold i_mutex.
1204 * Add inode to orphan list in case we crash before
1207 if (extended && ext4_can_truncate(inode))
1208 ext4_orphan_add(handle, inode);
1210 ext4_journal_stop(handle);
1212 ext4_truncate_failed_write(inode);
1214 * If truncate failed early the inode might
1215 * still be on the orphan list; we need to
1216 * make sure the inode is removed from the
1217 * orphan list in that case.
1220 ext4_orphan_del(NULL, inode);
1223 if (ret == -ENOSPC &&
1224 ext4_should_retry_alloc(inode->i_sb, &retries))
1233 /* For write_end() in data=journal mode */
1234 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1237 if (!buffer_mapped(bh) || buffer_freed(bh))
1239 set_buffer_uptodate(bh);
1240 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1241 clear_buffer_meta(bh);
1242 clear_buffer_prio(bh);
1247 * We need to pick up the new inode size which generic_commit_write gave us
1248 * `file' can be NULL - eg, when called from page_symlink().
1250 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1251 * buffers are managed internally.
1253 static int ext4_write_end(struct file *file,
1254 struct address_space *mapping,
1255 loff_t pos, unsigned len, unsigned copied,
1256 struct page *page, void *fsdata)
1258 handle_t *handle = ext4_journal_current_handle();
1259 struct inode *inode = mapping->host;
1260 loff_t old_size = inode->i_size;
1262 int i_size_changed = 0;
1263 int inline_data = ext4_has_inline_data(inode);
1264 bool verity = ext4_verity_in_progress(inode);
1266 trace_ext4_write_end(inode, pos, len, copied);
1268 ret = ext4_write_inline_data_end(inode, pos, len,
1277 copied = block_write_end(file, mapping, pos,
1278 len, copied, page, fsdata);
1280 * it's important to update i_size while still holding page lock:
1281 * page writeout could otherwise come in and zero beyond i_size.
1283 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1284 * blocks are being written past EOF, so skip the i_size update.
1287 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1291 if (old_size < pos && !verity)
1292 pagecache_isize_extended(inode, old_size, pos);
1294 * Don't mark the inode dirty under page lock. First, it unnecessarily
1295 * makes the holding time of page lock longer. Second, it forces lock
1296 * ordering of page lock and transaction start for journaling
1299 if (i_size_changed || inline_data)
1300 ext4_mark_inode_dirty(handle, inode);
1302 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1303 /* if we have allocated more blocks and copied
1304 * less. We will have blocks allocated outside
1305 * inode->i_size. So truncate them
1307 ext4_orphan_add(handle, inode);
1309 ret2 = ext4_journal_stop(handle);
1313 if (pos + len > inode->i_size && !verity) {
1314 ext4_truncate_failed_write(inode);
1316 * If truncate failed early the inode might still be
1317 * on the orphan list; we need to make sure the inode
1318 * is removed from the orphan list in that case.
1321 ext4_orphan_del(NULL, inode);
1324 return ret ? ret : copied;
1328 * This is a private version of page_zero_new_buffers() which doesn't
1329 * set the buffer to be dirty, since in data=journalled mode we need
1330 * to call ext4_handle_dirty_metadata() instead.
1332 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1334 unsigned from, unsigned to)
1336 unsigned int block_start = 0, block_end;
1337 struct buffer_head *head, *bh;
1339 bh = head = page_buffers(page);
1341 block_end = block_start + bh->b_size;
1342 if (buffer_new(bh)) {
1343 if (block_end > from && block_start < to) {
1344 if (!PageUptodate(page)) {
1345 unsigned start, size;
1347 start = max(from, block_start);
1348 size = min(to, block_end) - start;
1350 zero_user(page, start, size);
1351 write_end_fn(handle, bh);
1353 clear_buffer_new(bh);
1356 block_start = block_end;
1357 bh = bh->b_this_page;
1358 } while (bh != head);
1361 static int ext4_journalled_write_end(struct file *file,
1362 struct address_space *mapping,
1363 loff_t pos, unsigned len, unsigned copied,
1364 struct page *page, void *fsdata)
1366 handle_t *handle = ext4_journal_current_handle();
1367 struct inode *inode = mapping->host;
1368 loff_t old_size = inode->i_size;
1372 int size_changed = 0;
1373 int inline_data = ext4_has_inline_data(inode);
1374 bool verity = ext4_verity_in_progress(inode);
1376 trace_ext4_journalled_write_end(inode, pos, len, copied);
1377 from = pos & (PAGE_SIZE - 1);
1380 BUG_ON(!ext4_handle_valid(handle));
1383 ret = ext4_write_inline_data_end(inode, pos, len,
1391 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1393 ext4_journalled_zero_new_buffers(handle, page, from, to);
1395 if (unlikely(copied < len))
1396 ext4_journalled_zero_new_buffers(handle, page,
1398 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1399 from + copied, &partial,
1402 SetPageUptodate(page);
1405 size_changed = ext4_update_inode_size(inode, pos + copied);
1406 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1407 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1411 if (old_size < pos && !verity)
1412 pagecache_isize_extended(inode, old_size, pos);
1414 if (size_changed || inline_data) {
1415 ret2 = ext4_mark_inode_dirty(handle, inode);
1420 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1421 /* if we have allocated more blocks and copied
1422 * less. We will have blocks allocated outside
1423 * inode->i_size. So truncate them
1425 ext4_orphan_add(handle, inode);
1428 ret2 = ext4_journal_stop(handle);
1431 if (pos + len > inode->i_size && !verity) {
1432 ext4_truncate_failed_write(inode);
1434 * If truncate failed early the inode might still be
1435 * on the orphan list; we need to make sure the inode
1436 * is removed from the orphan list in that case.
1439 ext4_orphan_del(NULL, inode);
1442 return ret ? ret : copied;
1446 * Reserve space for a single cluster
1448 static int ext4_da_reserve_space(struct inode *inode)
1450 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1451 struct ext4_inode_info *ei = EXT4_I(inode);
1455 * We will charge metadata quota at writeout time; this saves
1456 * us from metadata over-estimation, though we may go over by
1457 * a small amount in the end. Here we just reserve for data.
1459 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1463 spin_lock(&ei->i_block_reservation_lock);
1464 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1465 spin_unlock(&ei->i_block_reservation_lock);
1466 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1469 ei->i_reserved_data_blocks++;
1470 trace_ext4_da_reserve_space(inode);
1471 spin_unlock(&ei->i_block_reservation_lock);
1473 return 0; /* success */
1476 void ext4_da_release_space(struct inode *inode, int to_free)
1478 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1479 struct ext4_inode_info *ei = EXT4_I(inode);
1482 return; /* Nothing to release, exit */
1484 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1486 trace_ext4_da_release_space(inode, to_free);
1487 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1489 * if there aren't enough reserved blocks, then the
1490 * counter is messed up somewhere. Since this
1491 * function is called from invalidate page, it's
1492 * harmless to return without any action.
1494 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1495 "ino %lu, to_free %d with only %d reserved "
1496 "data blocks", inode->i_ino, to_free,
1497 ei->i_reserved_data_blocks);
1499 to_free = ei->i_reserved_data_blocks;
1501 ei->i_reserved_data_blocks -= to_free;
1503 /* update fs dirty data blocks counter */
1504 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1506 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1508 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1512 * Delayed allocation stuff
1515 struct mpage_da_data {
1516 struct inode *inode;
1517 struct writeback_control *wbc;
1519 pgoff_t first_page; /* The first page to write */
1520 pgoff_t next_page; /* Current page to examine */
1521 pgoff_t last_page; /* Last page to examine */
1523 * Extent to map - this can be after first_page because that can be
1524 * fully mapped. We somewhat abuse m_flags to store whether the extent
1525 * is delalloc or unwritten.
1527 struct ext4_map_blocks map;
1528 struct ext4_io_submit io_submit; /* IO submission data */
1529 unsigned int do_map:1;
1532 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1537 struct pagevec pvec;
1538 struct inode *inode = mpd->inode;
1539 struct address_space *mapping = inode->i_mapping;
1541 /* This is necessary when next_page == 0. */
1542 if (mpd->first_page >= mpd->next_page)
1545 index = mpd->first_page;
1546 end = mpd->next_page - 1;
1548 ext4_lblk_t start, last;
1549 start = index << (PAGE_SHIFT - inode->i_blkbits);
1550 last = end << (PAGE_SHIFT - inode->i_blkbits);
1551 ext4_es_remove_extent(inode, start, last - start + 1);
1554 pagevec_init(&pvec);
1555 while (index <= end) {
1556 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1559 for (i = 0; i < nr_pages; i++) {
1560 struct page *page = pvec.pages[i];
1562 BUG_ON(!PageLocked(page));
1563 BUG_ON(PageWriteback(page));
1565 if (page_mapped(page))
1566 clear_page_dirty_for_io(page);
1567 block_invalidatepage(page, 0, PAGE_SIZE);
1568 ClearPageUptodate(page);
1572 pagevec_release(&pvec);
1576 static void ext4_print_free_blocks(struct inode *inode)
1578 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1579 struct super_block *sb = inode->i_sb;
1580 struct ext4_inode_info *ei = EXT4_I(inode);
1582 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1583 EXT4_C2B(EXT4_SB(inode->i_sb),
1584 ext4_count_free_clusters(sb)));
1585 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1586 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1587 (long long) EXT4_C2B(EXT4_SB(sb),
1588 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1589 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1590 (long long) EXT4_C2B(EXT4_SB(sb),
1591 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1592 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1593 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1594 ei->i_reserved_data_blocks);
1598 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1600 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1604 * ext4_insert_delayed_block - adds a delayed block to the extents status
1605 * tree, incrementing the reserved cluster/block
1606 * count or making a pending reservation
1609 * @inode - file containing the newly added block
1610 * @lblk - logical block to be added
1612 * Returns 0 on success, negative error code on failure.
1614 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1616 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1618 bool allocated = false;
1621 * If the cluster containing lblk is shared with a delayed,
1622 * written, or unwritten extent in a bigalloc file system, it's
1623 * already been accounted for and does not need to be reserved.
1624 * A pending reservation must be made for the cluster if it's
1625 * shared with a written or unwritten extent and doesn't already
1626 * have one. Written and unwritten extents can be purged from the
1627 * extents status tree if the system is under memory pressure, so
1628 * it's necessary to examine the extent tree if a search of the
1629 * extents status tree doesn't get a match.
1631 if (sbi->s_cluster_ratio == 1) {
1632 ret = ext4_da_reserve_space(inode);
1633 if (ret != 0) /* ENOSPC */
1635 } else { /* bigalloc */
1636 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1637 if (!ext4_es_scan_clu(inode,
1638 &ext4_es_is_mapped, lblk)) {
1639 ret = ext4_clu_mapped(inode,
1640 EXT4_B2C(sbi, lblk));
1644 ret = ext4_da_reserve_space(inode);
1645 if (ret != 0) /* ENOSPC */
1656 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1663 * This function is grabs code from the very beginning of
1664 * ext4_map_blocks, but assumes that the caller is from delayed write
1665 * time. This function looks up the requested blocks and sets the
1666 * buffer delay bit under the protection of i_data_sem.
1668 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1669 struct ext4_map_blocks *map,
1670 struct buffer_head *bh)
1672 struct extent_status es;
1674 sector_t invalid_block = ~((sector_t) 0xffff);
1675 #ifdef ES_AGGRESSIVE_TEST
1676 struct ext4_map_blocks orig_map;
1678 memcpy(&orig_map, map, sizeof(*map));
1681 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1685 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1686 "logical block %lu\n", inode->i_ino, map->m_len,
1687 (unsigned long) map->m_lblk);
1689 /* Lookup extent status tree firstly */
1690 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1691 if (ext4_es_is_hole(&es)) {
1693 down_read(&EXT4_I(inode)->i_data_sem);
1698 * Delayed extent could be allocated by fallocate.
1699 * So we need to check it.
1701 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1702 map_bh(bh, inode->i_sb, invalid_block);
1704 set_buffer_delay(bh);
1708 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1709 retval = es.es_len - (iblock - es.es_lblk);
1710 if (retval > map->m_len)
1711 retval = map->m_len;
1712 map->m_len = retval;
1713 if (ext4_es_is_written(&es))
1714 map->m_flags |= EXT4_MAP_MAPPED;
1715 else if (ext4_es_is_unwritten(&es))
1716 map->m_flags |= EXT4_MAP_UNWRITTEN;
1720 #ifdef ES_AGGRESSIVE_TEST
1721 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1727 * Try to see if we can get the block without requesting a new
1728 * file system block.
1730 down_read(&EXT4_I(inode)->i_data_sem);
1731 if (ext4_has_inline_data(inode))
1733 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1734 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1736 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1743 * XXX: __block_prepare_write() unmaps passed block,
1747 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1753 map_bh(bh, inode->i_sb, invalid_block);
1755 set_buffer_delay(bh);
1756 } else if (retval > 0) {
1758 unsigned int status;
1760 if (unlikely(retval != map->m_len)) {
1761 ext4_warning(inode->i_sb,
1762 "ES len assertion failed for inode "
1763 "%lu: retval %d != map->m_len %d",
1764 inode->i_ino, retval, map->m_len);
1768 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1769 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1770 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1771 map->m_pblk, status);
1777 up_read((&EXT4_I(inode)->i_data_sem));
1783 * This is a special get_block_t callback which is used by
1784 * ext4_da_write_begin(). It will either return mapped block or
1785 * reserve space for a single block.
1787 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1788 * We also have b_blocknr = -1 and b_bdev initialized properly
1790 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1791 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1792 * initialized properly.
1794 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1795 struct buffer_head *bh, int create)
1797 struct ext4_map_blocks map;
1800 BUG_ON(create == 0);
1801 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1803 map.m_lblk = iblock;
1807 * first, we need to know whether the block is allocated already
1808 * preallocated blocks are unmapped but should treated
1809 * the same as allocated blocks.
1811 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1815 map_bh(bh, inode->i_sb, map.m_pblk);
1816 ext4_update_bh_state(bh, map.m_flags);
1818 if (buffer_unwritten(bh)) {
1819 /* A delayed write to unwritten bh should be marked
1820 * new and mapped. Mapped ensures that we don't do
1821 * get_block multiple times when we write to the same
1822 * offset and new ensures that we do proper zero out
1823 * for partial write.
1826 set_buffer_mapped(bh);
1831 static int bget_one(handle_t *handle, struct buffer_head *bh)
1837 static int bput_one(handle_t *handle, struct buffer_head *bh)
1843 static int __ext4_journalled_writepage(struct page *page,
1846 struct address_space *mapping = page->mapping;
1847 struct inode *inode = mapping->host;
1848 struct buffer_head *page_bufs = NULL;
1849 handle_t *handle = NULL;
1850 int ret = 0, err = 0;
1851 int inline_data = ext4_has_inline_data(inode);
1852 struct buffer_head *inode_bh = NULL;
1854 ClearPageChecked(page);
1857 BUG_ON(page->index != 0);
1858 BUG_ON(len > ext4_get_max_inline_size(inode));
1859 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1860 if (inode_bh == NULL)
1863 page_bufs = page_buffers(page);
1868 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1872 * We need to release the page lock before we start the
1873 * journal, so grab a reference so the page won't disappear
1874 * out from under us.
1879 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1880 ext4_writepage_trans_blocks(inode));
1881 if (IS_ERR(handle)) {
1882 ret = PTR_ERR(handle);
1884 goto out_no_pagelock;
1886 BUG_ON(!ext4_handle_valid(handle));
1890 if (page->mapping != mapping) {
1891 /* The page got truncated from under us */
1892 ext4_journal_stop(handle);
1898 ret = ext4_mark_inode_dirty(handle, inode);
1900 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1901 do_journal_get_write_access);
1903 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1908 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1909 err = ext4_journal_stop(handle);
1913 if (!ext4_has_inline_data(inode))
1914 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1916 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1925 * Note that we don't need to start a transaction unless we're journaling data
1926 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1927 * need to file the inode to the transaction's list in ordered mode because if
1928 * we are writing back data added by write(), the inode is already there and if
1929 * we are writing back data modified via mmap(), no one guarantees in which
1930 * transaction the data will hit the disk. In case we are journaling data, we
1931 * cannot start transaction directly because transaction start ranks above page
1932 * lock so we have to do some magic.
1934 * This function can get called via...
1935 * - ext4_writepages after taking page lock (have journal handle)
1936 * - journal_submit_inode_data_buffers (no journal handle)
1937 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1938 * - grab_page_cache when doing write_begin (have journal handle)
1940 * We don't do any block allocation in this function. If we have page with
1941 * multiple blocks we need to write those buffer_heads that are mapped. This
1942 * is important for mmaped based write. So if we do with blocksize 1K
1943 * truncate(f, 1024);
1944 * a = mmap(f, 0, 4096);
1946 * truncate(f, 4096);
1947 * we have in the page first buffer_head mapped via page_mkwrite call back
1948 * but other buffer_heads would be unmapped but dirty (dirty done via the
1949 * do_wp_page). So writepage should write the first block. If we modify
1950 * the mmap area beyond 1024 we will again get a page_fault and the
1951 * page_mkwrite callback will do the block allocation and mark the
1952 * buffer_heads mapped.
1954 * We redirty the page if we have any buffer_heads that is either delay or
1955 * unwritten in the page.
1957 * We can get recursively called as show below.
1959 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1962 * But since we don't do any block allocation we should not deadlock.
1963 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1965 static int ext4_writepage(struct page *page,
1966 struct writeback_control *wbc)
1971 struct buffer_head *page_bufs = NULL;
1972 struct inode *inode = page->mapping->host;
1973 struct ext4_io_submit io_submit;
1974 bool keep_towrite = false;
1976 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
1977 ext4_invalidatepage(page, 0, PAGE_SIZE);
1982 trace_ext4_writepage(page);
1983 size = i_size_read(inode);
1984 if (page->index == size >> PAGE_SHIFT &&
1985 !ext4_verity_in_progress(inode))
1986 len = size & ~PAGE_MASK;
1990 page_bufs = page_buffers(page);
1992 * We cannot do block allocation or other extent handling in this
1993 * function. If there are buffers needing that, we have to redirty
1994 * the page. But we may reach here when we do a journal commit via
1995 * journal_submit_inode_data_buffers() and in that case we must write
1996 * allocated buffers to achieve data=ordered mode guarantees.
1998 * Also, if there is only one buffer per page (the fs block
1999 * size == the page size), if one buffer needs block
2000 * allocation or needs to modify the extent tree to clear the
2001 * unwritten flag, we know that the page can't be written at
2002 * all, so we might as well refuse the write immediately.
2003 * Unfortunately if the block size != page size, we can't as
2004 * easily detect this case using ext4_walk_page_buffers(), but
2005 * for the extremely common case, this is an optimization that
2006 * skips a useless round trip through ext4_bio_write_page().
2008 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2009 ext4_bh_delay_or_unwritten)) {
2010 redirty_page_for_writepage(wbc, page);
2011 if ((current->flags & PF_MEMALLOC) ||
2012 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2014 * For memory cleaning there's no point in writing only
2015 * some buffers. So just bail out. Warn if we came here
2016 * from direct reclaim.
2018 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2023 keep_towrite = true;
2026 if (PageChecked(page) && ext4_should_journal_data(inode))
2028 * It's mmapped pagecache. Add buffers and journal it. There
2029 * doesn't seem much point in redirtying the page here.
2031 return __ext4_journalled_writepage(page, len);
2033 ext4_io_submit_init(&io_submit, wbc);
2034 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2035 if (!io_submit.io_end) {
2036 redirty_page_for_writepage(wbc, page);
2040 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2041 ext4_io_submit(&io_submit);
2042 /* Drop io_end reference we got from init */
2043 ext4_put_io_end_defer(io_submit.io_end);
2047 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2053 BUG_ON(page->index != mpd->first_page);
2054 clear_page_dirty_for_io(page);
2056 * We have to be very careful here! Nothing protects writeback path
2057 * against i_size changes and the page can be writeably mapped into
2058 * page tables. So an application can be growing i_size and writing
2059 * data through mmap while writeback runs. clear_page_dirty_for_io()
2060 * write-protects our page in page tables and the page cannot get
2061 * written to again until we release page lock. So only after
2062 * clear_page_dirty_for_io() we are safe to sample i_size for
2063 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2064 * on the barrier provided by TestClearPageDirty in
2065 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2066 * after page tables are updated.
2068 size = i_size_read(mpd->inode);
2069 if (page->index == size >> PAGE_SHIFT &&
2070 !ext4_verity_in_progress(mpd->inode))
2071 len = size & ~PAGE_MASK;
2074 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2076 mpd->wbc->nr_to_write--;
2082 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2085 * mballoc gives us at most this number of blocks...
2086 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2087 * The rest of mballoc seems to handle chunks up to full group size.
2089 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2092 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2094 * @mpd - extent of blocks
2095 * @lblk - logical number of the block in the file
2096 * @bh - buffer head we want to add to the extent
2098 * The function is used to collect contig. blocks in the same state. If the
2099 * buffer doesn't require mapping for writeback and we haven't started the
2100 * extent of buffers to map yet, the function returns 'true' immediately - the
2101 * caller can write the buffer right away. Otherwise the function returns true
2102 * if the block has been added to the extent, false if the block couldn't be
2105 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2106 struct buffer_head *bh)
2108 struct ext4_map_blocks *map = &mpd->map;
2110 /* Buffer that doesn't need mapping for writeback? */
2111 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2112 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2113 /* So far no extent to map => we write the buffer right away */
2114 if (map->m_len == 0)
2119 /* First block in the extent? */
2120 if (map->m_len == 0) {
2121 /* We cannot map unless handle is started... */
2126 map->m_flags = bh->b_state & BH_FLAGS;
2130 /* Don't go larger than mballoc is willing to allocate */
2131 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2134 /* Can we merge the block to our big extent? */
2135 if (lblk == map->m_lblk + map->m_len &&
2136 (bh->b_state & BH_FLAGS) == map->m_flags) {
2144 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2146 * @mpd - extent of blocks for mapping
2147 * @head - the first buffer in the page
2148 * @bh - buffer we should start processing from
2149 * @lblk - logical number of the block in the file corresponding to @bh
2151 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2152 * the page for IO if all buffers in this page were mapped and there's no
2153 * accumulated extent of buffers to map or add buffers in the page to the
2154 * extent of buffers to map. The function returns 1 if the caller can continue
2155 * by processing the next page, 0 if it should stop adding buffers to the
2156 * extent to map because we cannot extend it anymore. It can also return value
2157 * < 0 in case of error during IO submission.
2159 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2160 struct buffer_head *head,
2161 struct buffer_head *bh,
2164 struct inode *inode = mpd->inode;
2166 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2167 >> inode->i_blkbits;
2169 if (ext4_verity_in_progress(inode))
2170 blocks = EXT_MAX_BLOCKS;
2173 BUG_ON(buffer_locked(bh));
2175 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2176 /* Found extent to map? */
2179 /* Buffer needs mapping and handle is not started? */
2182 /* Everything mapped so far and we hit EOF */
2185 } while (lblk++, (bh = bh->b_this_page) != head);
2186 /* So far everything mapped? Submit the page for IO. */
2187 if (mpd->map.m_len == 0) {
2188 err = mpage_submit_page(mpd, head->b_page);
2192 return lblk < blocks;
2196 * mpage_process_page - update page buffers corresponding to changed extent and
2197 * may submit fully mapped page for IO
2199 * @mpd - description of extent to map, on return next extent to map
2200 * @m_lblk - logical block mapping.
2201 * @m_pblk - corresponding physical mapping.
2202 * @map_bh - determines on return whether this page requires any further
2204 * Scan given page buffers corresponding to changed extent and update buffer
2205 * state according to new extent state.
2206 * We map delalloc buffers to their physical location, clear unwritten bits.
2207 * If the given page is not fully mapped, we update @map to the next extent in
2208 * the given page that needs mapping & return @map_bh as true.
2210 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2211 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2214 struct buffer_head *head, *bh;
2215 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2216 ext4_lblk_t lblk = *m_lblk;
2217 ext4_fsblk_t pblock = *m_pblk;
2219 int blkbits = mpd->inode->i_blkbits;
2220 ssize_t io_end_size = 0;
2221 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2223 bh = head = page_buffers(page);
2225 if (lblk < mpd->map.m_lblk)
2227 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2229 * Buffer after end of mapped extent.
2230 * Find next buffer in the page to map.
2233 mpd->map.m_flags = 0;
2234 io_end_vec->size += io_end_size;
2237 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2240 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2241 io_end_vec = ext4_alloc_io_end_vec(io_end);
2242 if (IS_ERR(io_end_vec)) {
2243 err = PTR_ERR(io_end_vec);
2246 io_end_vec->offset = mpd->map.m_lblk << blkbits;
2251 if (buffer_delay(bh)) {
2252 clear_buffer_delay(bh);
2253 bh->b_blocknr = pblock++;
2255 clear_buffer_unwritten(bh);
2256 io_end_size += (1 << blkbits);
2257 } while (lblk++, (bh = bh->b_this_page) != head);
2259 io_end_vec->size += io_end_size;
2269 * mpage_map_buffers - update buffers corresponding to changed extent and
2270 * submit fully mapped pages for IO
2272 * @mpd - description of extent to map, on return next extent to map
2274 * Scan buffers corresponding to changed extent (we expect corresponding pages
2275 * to be already locked) and update buffer state according to new extent state.
2276 * We map delalloc buffers to their physical location, clear unwritten bits,
2277 * and mark buffers as uninit when we perform writes to unwritten extents
2278 * and do extent conversion after IO is finished. If the last page is not fully
2279 * mapped, we update @map to the next extent in the last page that needs
2280 * mapping. Otherwise we submit the page for IO.
2282 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2284 struct pagevec pvec;
2286 struct inode *inode = mpd->inode;
2287 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2290 ext4_fsblk_t pblock;
2292 bool map_bh = false;
2294 start = mpd->map.m_lblk >> bpp_bits;
2295 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2296 lblk = start << bpp_bits;
2297 pblock = mpd->map.m_pblk;
2299 pagevec_init(&pvec);
2300 while (start <= end) {
2301 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2305 for (i = 0; i < nr_pages; i++) {
2306 struct page *page = pvec.pages[i];
2308 err = mpage_process_page(mpd, page, &lblk, &pblock,
2311 * If map_bh is true, means page may require further bh
2312 * mapping, or maybe the page was submitted for IO.
2313 * So we return to call further extent mapping.
2315 if (err < 0 || map_bh == true)
2317 /* Page fully mapped - let IO run! */
2318 err = mpage_submit_page(mpd, page);
2322 pagevec_release(&pvec);
2324 /* Extent fully mapped and matches with page boundary. We are done. */
2326 mpd->map.m_flags = 0;
2329 pagevec_release(&pvec);
2333 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2335 struct inode *inode = mpd->inode;
2336 struct ext4_map_blocks *map = &mpd->map;
2337 int get_blocks_flags;
2338 int err, dioread_nolock;
2340 trace_ext4_da_write_pages_extent(inode, map);
2342 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2343 * to convert an unwritten extent to be initialized (in the case
2344 * where we have written into one or more preallocated blocks). It is
2345 * possible that we're going to need more metadata blocks than
2346 * previously reserved. However we must not fail because we're in
2347 * writeback and there is nothing we can do about it so it might result
2348 * in data loss. So use reserved blocks to allocate metadata if
2351 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2352 * the blocks in question are delalloc blocks. This indicates
2353 * that the blocks and quotas has already been checked when
2354 * the data was copied into the page cache.
2356 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2357 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2358 EXT4_GET_BLOCKS_IO_SUBMIT;
2359 dioread_nolock = ext4_should_dioread_nolock(inode);
2361 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2362 if (map->m_flags & (1 << BH_Delay))
2363 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2365 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2368 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2369 if (!mpd->io_submit.io_end->handle &&
2370 ext4_handle_valid(handle)) {
2371 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2372 handle->h_rsv_handle = NULL;
2374 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2377 BUG_ON(map->m_len == 0);
2382 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2383 * mpd->len and submit pages underlying it for IO
2385 * @handle - handle for journal operations
2386 * @mpd - extent to map
2387 * @give_up_on_write - we set this to true iff there is a fatal error and there
2388 * is no hope of writing the data. The caller should discard
2389 * dirty pages to avoid infinite loops.
2391 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2392 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2393 * them to initialized or split the described range from larger unwritten
2394 * extent. Note that we need not map all the described range since allocation
2395 * can return less blocks or the range is covered by more unwritten extents. We
2396 * cannot map more because we are limited by reserved transaction credits. On
2397 * the other hand we always make sure that the last touched page is fully
2398 * mapped so that it can be written out (and thus forward progress is
2399 * guaranteed). After mapping we submit all mapped pages for IO.
2401 static int mpage_map_and_submit_extent(handle_t *handle,
2402 struct mpage_da_data *mpd,
2403 bool *give_up_on_write)
2405 struct inode *inode = mpd->inode;
2406 struct ext4_map_blocks *map = &mpd->map;
2410 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2411 struct ext4_io_end_vec *io_end_vec;
2413 io_end_vec = ext4_alloc_io_end_vec(io_end);
2414 if (IS_ERR(io_end_vec))
2415 return PTR_ERR(io_end_vec);
2416 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2418 err = mpage_map_one_extent(handle, mpd);
2420 struct super_block *sb = inode->i_sb;
2422 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2423 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2424 goto invalidate_dirty_pages;
2426 * Let the uper layers retry transient errors.
2427 * In the case of ENOSPC, if ext4_count_free_blocks()
2428 * is non-zero, a commit should free up blocks.
2430 if ((err == -ENOMEM) ||
2431 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2433 goto update_disksize;
2436 ext4_msg(sb, KERN_CRIT,
2437 "Delayed block allocation failed for "
2438 "inode %lu at logical offset %llu with"
2439 " max blocks %u with error %d",
2441 (unsigned long long)map->m_lblk,
2442 (unsigned)map->m_len, -err);
2443 ext4_msg(sb, KERN_CRIT,
2444 "This should not happen!! Data will "
2447 ext4_print_free_blocks(inode);
2448 invalidate_dirty_pages:
2449 *give_up_on_write = true;
2454 * Update buffer state, submit mapped pages, and get us new
2457 err = mpage_map_and_submit_buffers(mpd);
2459 goto update_disksize;
2460 } while (map->m_len);
2464 * Update on-disk size after IO is submitted. Races with
2465 * truncate are avoided by checking i_size under i_data_sem.
2467 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2468 if (disksize > EXT4_I(inode)->i_disksize) {
2472 down_write(&EXT4_I(inode)->i_data_sem);
2473 i_size = i_size_read(inode);
2474 if (disksize > i_size)
2476 if (disksize > EXT4_I(inode)->i_disksize)
2477 EXT4_I(inode)->i_disksize = disksize;
2478 up_write(&EXT4_I(inode)->i_data_sem);
2479 err2 = ext4_mark_inode_dirty(handle, inode);
2481 ext4_set_errno(inode->i_sb, -err2);
2482 ext4_error(inode->i_sb,
2483 "Failed to mark inode %lu dirty",
2493 * Calculate the total number of credits to reserve for one writepages
2494 * iteration. This is called from ext4_writepages(). We map an extent of
2495 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2496 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2497 * bpp - 1 blocks in bpp different extents.
2499 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2501 int bpp = ext4_journal_blocks_per_page(inode);
2503 return ext4_meta_trans_blocks(inode,
2504 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2508 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2509 * and underlying extent to map
2511 * @mpd - where to look for pages
2513 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2514 * IO immediately. When we find a page which isn't mapped we start accumulating
2515 * extent of buffers underlying these pages that needs mapping (formed by
2516 * either delayed or unwritten buffers). We also lock the pages containing
2517 * these buffers. The extent found is returned in @mpd structure (starting at
2518 * mpd->lblk with length mpd->len blocks).
2520 * Note that this function can attach bios to one io_end structure which are
2521 * neither logically nor physically contiguous. Although it may seem as an
2522 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2523 * case as we need to track IO to all buffers underlying a page in one io_end.
2525 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2527 struct address_space *mapping = mpd->inode->i_mapping;
2528 struct pagevec pvec;
2529 unsigned int nr_pages;
2530 long left = mpd->wbc->nr_to_write;
2531 pgoff_t index = mpd->first_page;
2532 pgoff_t end = mpd->last_page;
2535 int blkbits = mpd->inode->i_blkbits;
2537 struct buffer_head *head;
2539 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2540 tag = PAGECACHE_TAG_TOWRITE;
2542 tag = PAGECACHE_TAG_DIRTY;
2544 pagevec_init(&pvec);
2546 mpd->next_page = index;
2547 while (index <= end) {
2548 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2553 for (i = 0; i < nr_pages; i++) {
2554 struct page *page = pvec.pages[i];
2557 * Accumulated enough dirty pages? This doesn't apply
2558 * to WB_SYNC_ALL mode. For integrity sync we have to
2559 * keep going because someone may be concurrently
2560 * dirtying pages, and we might have synced a lot of
2561 * newly appeared dirty pages, but have not synced all
2562 * of the old dirty pages.
2564 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2567 /* If we can't merge this page, we are done. */
2568 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2573 * If the page is no longer dirty, or its mapping no
2574 * longer corresponds to inode we are writing (which
2575 * means it has been truncated or invalidated), or the
2576 * page is already under writeback and we are not doing
2577 * a data integrity writeback, skip the page
2579 if (!PageDirty(page) ||
2580 (PageWriteback(page) &&
2581 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2582 unlikely(page->mapping != mapping)) {
2587 wait_on_page_writeback(page);
2588 BUG_ON(PageWriteback(page));
2590 if (mpd->map.m_len == 0)
2591 mpd->first_page = page->index;
2592 mpd->next_page = page->index + 1;
2593 /* Add all dirty buffers to mpd */
2594 lblk = ((ext4_lblk_t)page->index) <<
2595 (PAGE_SHIFT - blkbits);
2596 head = page_buffers(page);
2597 err = mpage_process_page_bufs(mpd, head, head, lblk);
2603 pagevec_release(&pvec);
2608 pagevec_release(&pvec);
2612 static int ext4_writepages(struct address_space *mapping,
2613 struct writeback_control *wbc)
2615 pgoff_t writeback_index = 0;
2616 long nr_to_write = wbc->nr_to_write;
2617 int range_whole = 0;
2619 handle_t *handle = NULL;
2620 struct mpage_da_data mpd;
2621 struct inode *inode = mapping->host;
2622 int needed_blocks, rsv_blocks = 0, ret = 0;
2623 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2625 struct blk_plug plug;
2626 bool give_up_on_write = false;
2628 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2631 percpu_down_read(&sbi->s_journal_flag_rwsem);
2632 trace_ext4_writepages(inode, wbc);
2635 * No pages to write? This is mainly a kludge to avoid starting
2636 * a transaction for special inodes like journal inode on last iput()
2637 * because that could violate lock ordering on umount
2639 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2640 goto out_writepages;
2642 if (ext4_should_journal_data(inode)) {
2643 ret = generic_writepages(mapping, wbc);
2644 goto out_writepages;
2648 * If the filesystem has aborted, it is read-only, so return
2649 * right away instead of dumping stack traces later on that
2650 * will obscure the real source of the problem. We test
2651 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2652 * the latter could be true if the filesystem is mounted
2653 * read-only, and in that case, ext4_writepages should
2654 * *never* be called, so if that ever happens, we would want
2657 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2658 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2660 goto out_writepages;
2664 * If we have inline data and arrive here, it means that
2665 * we will soon create the block for the 1st page, so
2666 * we'd better clear the inline data here.
2668 if (ext4_has_inline_data(inode)) {
2669 /* Just inode will be modified... */
2670 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2671 if (IS_ERR(handle)) {
2672 ret = PTR_ERR(handle);
2673 goto out_writepages;
2675 BUG_ON(ext4_test_inode_state(inode,
2676 EXT4_STATE_MAY_INLINE_DATA));
2677 ext4_destroy_inline_data(handle, inode);
2678 ext4_journal_stop(handle);
2681 if (ext4_should_dioread_nolock(inode)) {
2683 * We may need to convert up to one extent per block in
2684 * the page and we may dirty the inode.
2686 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2687 PAGE_SIZE >> inode->i_blkbits);
2690 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2693 if (wbc->range_cyclic) {
2694 writeback_index = mapping->writeback_index;
2695 if (writeback_index)
2697 mpd.first_page = writeback_index;
2700 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2701 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2706 ext4_io_submit_init(&mpd.io_submit, wbc);
2708 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2709 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2711 blk_start_plug(&plug);
2714 * First writeback pages that don't need mapping - we can avoid
2715 * starting a transaction unnecessarily and also avoid being blocked
2716 * in the block layer on device congestion while having transaction
2720 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2721 if (!mpd.io_submit.io_end) {
2725 ret = mpage_prepare_extent_to_map(&mpd);
2726 /* Unlock pages we didn't use */
2727 mpage_release_unused_pages(&mpd, false);
2728 /* Submit prepared bio */
2729 ext4_io_submit(&mpd.io_submit);
2730 ext4_put_io_end_defer(mpd.io_submit.io_end);
2731 mpd.io_submit.io_end = NULL;
2735 while (!done && mpd.first_page <= mpd.last_page) {
2736 /* For each extent of pages we use new io_end */
2737 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2738 if (!mpd.io_submit.io_end) {
2744 * We have two constraints: We find one extent to map and we
2745 * must always write out whole page (makes a difference when
2746 * blocksize < pagesize) so that we don't block on IO when we
2747 * try to write out the rest of the page. Journalled mode is
2748 * not supported by delalloc.
2750 BUG_ON(ext4_should_journal_data(inode));
2751 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2753 /* start a new transaction */
2754 handle = ext4_journal_start_with_reserve(inode,
2755 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2756 if (IS_ERR(handle)) {
2757 ret = PTR_ERR(handle);
2758 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2759 "%ld pages, ino %lu; err %d", __func__,
2760 wbc->nr_to_write, inode->i_ino, ret);
2761 /* Release allocated io_end */
2762 ext4_put_io_end(mpd.io_submit.io_end);
2763 mpd.io_submit.io_end = NULL;
2768 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2769 ret = mpage_prepare_extent_to_map(&mpd);
2772 ret = mpage_map_and_submit_extent(handle, &mpd,
2776 * We scanned the whole range (or exhausted
2777 * nr_to_write), submitted what was mapped and
2778 * didn't find anything needing mapping. We are
2785 * Caution: If the handle is synchronous,
2786 * ext4_journal_stop() can wait for transaction commit
2787 * to finish which may depend on writeback of pages to
2788 * complete or on page lock to be released. In that
2789 * case, we have to wait until after after we have
2790 * submitted all the IO, released page locks we hold,
2791 * and dropped io_end reference (for extent conversion
2792 * to be able to complete) before stopping the handle.
2794 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2795 ext4_journal_stop(handle);
2799 /* Unlock pages we didn't use */
2800 mpage_release_unused_pages(&mpd, give_up_on_write);
2801 /* Submit prepared bio */
2802 ext4_io_submit(&mpd.io_submit);
2805 * Drop our io_end reference we got from init. We have
2806 * to be careful and use deferred io_end finishing if
2807 * we are still holding the transaction as we can
2808 * release the last reference to io_end which may end
2809 * up doing unwritten extent conversion.
2812 ext4_put_io_end_defer(mpd.io_submit.io_end);
2813 ext4_journal_stop(handle);
2815 ext4_put_io_end(mpd.io_submit.io_end);
2816 mpd.io_submit.io_end = NULL;
2818 if (ret == -ENOSPC && sbi->s_journal) {
2820 * Commit the transaction which would
2821 * free blocks released in the transaction
2824 jbd2_journal_force_commit_nested(sbi->s_journal);
2828 /* Fatal error - ENOMEM, EIO... */
2833 blk_finish_plug(&plug);
2834 if (!ret && !cycled && wbc->nr_to_write > 0) {
2836 mpd.last_page = writeback_index - 1;
2842 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2844 * Set the writeback_index so that range_cyclic
2845 * mode will write it back later
2847 mapping->writeback_index = mpd.first_page;
2850 trace_ext4_writepages_result(inode, wbc, ret,
2851 nr_to_write - wbc->nr_to_write);
2852 percpu_up_read(&sbi->s_journal_flag_rwsem);
2856 static int ext4_dax_writepages(struct address_space *mapping,
2857 struct writeback_control *wbc)
2860 long nr_to_write = wbc->nr_to_write;
2861 struct inode *inode = mapping->host;
2862 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2864 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2867 percpu_down_read(&sbi->s_journal_flag_rwsem);
2868 trace_ext4_writepages(inode, wbc);
2870 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2871 trace_ext4_writepages_result(inode, wbc, ret,
2872 nr_to_write - wbc->nr_to_write);
2873 percpu_up_read(&sbi->s_journal_flag_rwsem);
2877 static int ext4_nonda_switch(struct super_block *sb)
2879 s64 free_clusters, dirty_clusters;
2880 struct ext4_sb_info *sbi = EXT4_SB(sb);
2883 * switch to non delalloc mode if we are running low
2884 * on free block. The free block accounting via percpu
2885 * counters can get slightly wrong with percpu_counter_batch getting
2886 * accumulated on each CPU without updating global counters
2887 * Delalloc need an accurate free block accounting. So switch
2888 * to non delalloc when we are near to error range.
2891 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2893 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2895 * Start pushing delalloc when 1/2 of free blocks are dirty.
2897 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2898 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2900 if (2 * free_clusters < 3 * dirty_clusters ||
2901 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2903 * free block count is less than 150% of dirty blocks
2904 * or free blocks is less than watermark
2911 /* We always reserve for an inode update; the superblock could be there too */
2912 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2914 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2917 if (pos + len <= 0x7fffffffULL)
2920 /* We might need to update the superblock to set LARGE_FILE */
2924 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2925 loff_t pos, unsigned len, unsigned flags,
2926 struct page **pagep, void **fsdata)
2928 int ret, retries = 0;
2931 struct inode *inode = mapping->host;
2934 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2937 index = pos >> PAGE_SHIFT;
2939 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2940 ext4_verity_in_progress(inode)) {
2941 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2942 return ext4_write_begin(file, mapping, pos,
2943 len, flags, pagep, fsdata);
2945 *fsdata = (void *)0;
2946 trace_ext4_da_write_begin(inode, pos, len, flags);
2948 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2949 ret = ext4_da_write_inline_data_begin(mapping, inode,
2959 * grab_cache_page_write_begin() can take a long time if the
2960 * system is thrashing due to memory pressure, or if the page
2961 * is being written back. So grab it first before we start
2962 * the transaction handle. This also allows us to allocate
2963 * the page (if needed) without using GFP_NOFS.
2966 page = grab_cache_page_write_begin(mapping, index, flags);
2972 * With delayed allocation, we don't log the i_disksize update
2973 * if there is delayed block allocation. But we still need
2974 * to journalling the i_disksize update if writes to the end
2975 * of file which has an already mapped buffer.
2978 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2979 ext4_da_write_credits(inode, pos, len));
2980 if (IS_ERR(handle)) {
2982 return PTR_ERR(handle);
2986 if (page->mapping != mapping) {
2987 /* The page got truncated from under us */
2990 ext4_journal_stop(handle);
2993 /* In case writeback began while the page was unlocked */
2994 wait_for_stable_page(page);
2996 #ifdef CONFIG_FS_ENCRYPTION
2997 ret = ext4_block_write_begin(page, pos, len,
2998 ext4_da_get_block_prep);
3000 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3004 ext4_journal_stop(handle);
3006 * block_write_begin may have instantiated a few blocks
3007 * outside i_size. Trim these off again. Don't need
3008 * i_size_read because we hold i_mutex.
3010 if (pos + len > inode->i_size)
3011 ext4_truncate_failed_write(inode);
3013 if (ret == -ENOSPC &&
3014 ext4_should_retry_alloc(inode->i_sb, &retries))
3026 * Check if we should update i_disksize
3027 * when write to the end of file but not require block allocation
3029 static int ext4_da_should_update_i_disksize(struct page *page,
3030 unsigned long offset)
3032 struct buffer_head *bh;
3033 struct inode *inode = page->mapping->host;
3037 bh = page_buffers(page);
3038 idx = offset >> inode->i_blkbits;
3040 for (i = 0; i < idx; i++)
3041 bh = bh->b_this_page;
3043 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3048 static int ext4_da_write_end(struct file *file,
3049 struct address_space *mapping,
3050 loff_t pos, unsigned len, unsigned copied,
3051 struct page *page, void *fsdata)
3053 struct inode *inode = mapping->host;
3055 handle_t *handle = ext4_journal_current_handle();
3057 unsigned long start, end;
3058 int write_mode = (int)(unsigned long)fsdata;
3060 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3061 return ext4_write_end(file, mapping, pos,
3062 len, copied, page, fsdata);
3064 trace_ext4_da_write_end(inode, pos, len, copied);
3065 start = pos & (PAGE_SIZE - 1);
3066 end = start + copied - 1;
3069 * generic_write_end() will run mark_inode_dirty() if i_size
3070 * changes. So let's piggyback the i_disksize mark_inode_dirty
3073 new_i_size = pos + copied;
3074 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3075 if (ext4_has_inline_data(inode) ||
3076 ext4_da_should_update_i_disksize(page, end)) {
3077 ext4_update_i_disksize(inode, new_i_size);
3078 /* We need to mark inode dirty even if
3079 * new_i_size is less that inode->i_size
3080 * bu greater than i_disksize.(hint delalloc)
3082 ext4_mark_inode_dirty(handle, inode);
3086 if (write_mode != CONVERT_INLINE_DATA &&
3087 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3088 ext4_has_inline_data(inode))
3089 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3092 ret2 = generic_write_end(file, mapping, pos, len, copied,
3098 ret2 = ext4_journal_stop(handle);
3102 return ret ? ret : copied;
3106 * Force all delayed allocation blocks to be allocated for a given inode.
3108 int ext4_alloc_da_blocks(struct inode *inode)
3110 trace_ext4_alloc_da_blocks(inode);
3112 if (!EXT4_I(inode)->i_reserved_data_blocks)
3116 * We do something simple for now. The filemap_flush() will
3117 * also start triggering a write of the data blocks, which is
3118 * not strictly speaking necessary (and for users of
3119 * laptop_mode, not even desirable). However, to do otherwise
3120 * would require replicating code paths in:
3122 * ext4_writepages() ->
3123 * write_cache_pages() ---> (via passed in callback function)
3124 * __mpage_da_writepage() -->
3125 * mpage_add_bh_to_extent()
3126 * mpage_da_map_blocks()
3128 * The problem is that write_cache_pages(), located in
3129 * mm/page-writeback.c, marks pages clean in preparation for
3130 * doing I/O, which is not desirable if we're not planning on
3133 * We could call write_cache_pages(), and then redirty all of
3134 * the pages by calling redirty_page_for_writepage() but that
3135 * would be ugly in the extreme. So instead we would need to
3136 * replicate parts of the code in the above functions,
3137 * simplifying them because we wouldn't actually intend to
3138 * write out the pages, but rather only collect contiguous
3139 * logical block extents, call the multi-block allocator, and
3140 * then update the buffer heads with the block allocations.
3142 * For now, though, we'll cheat by calling filemap_flush(),
3143 * which will map the blocks, and start the I/O, but not
3144 * actually wait for the I/O to complete.
3146 return filemap_flush(inode->i_mapping);
3150 * bmap() is special. It gets used by applications such as lilo and by
3151 * the swapper to find the on-disk block of a specific piece of data.
3153 * Naturally, this is dangerous if the block concerned is still in the
3154 * journal. If somebody makes a swapfile on an ext4 data-journaling
3155 * filesystem and enables swap, then they may get a nasty shock when the
3156 * data getting swapped to that swapfile suddenly gets overwritten by
3157 * the original zero's written out previously to the journal and
3158 * awaiting writeback in the kernel's buffer cache.
3160 * So, if we see any bmap calls here on a modified, data-journaled file,
3161 * take extra steps to flush any blocks which might be in the cache.
3163 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3165 struct inode *inode = mapping->host;
3170 * We can get here for an inline file via the FIBMAP ioctl
3172 if (ext4_has_inline_data(inode))
3175 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3176 test_opt(inode->i_sb, DELALLOC)) {
3178 * With delalloc we want to sync the file
3179 * so that we can make sure we allocate
3182 filemap_write_and_wait(mapping);
3185 if (EXT4_JOURNAL(inode) &&
3186 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3188 * This is a REALLY heavyweight approach, but the use of
3189 * bmap on dirty files is expected to be extremely rare:
3190 * only if we run lilo or swapon on a freshly made file
3191 * do we expect this to happen.
3193 * (bmap requires CAP_SYS_RAWIO so this does not
3194 * represent an unprivileged user DOS attack --- we'd be
3195 * in trouble if mortal users could trigger this path at
3198 * NB. EXT4_STATE_JDATA is not set on files other than
3199 * regular files. If somebody wants to bmap a directory
3200 * or symlink and gets confused because the buffer
3201 * hasn't yet been flushed to disk, they deserve
3202 * everything they get.
3205 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3206 journal = EXT4_JOURNAL(inode);
3207 jbd2_journal_lock_updates(journal);
3208 err = jbd2_journal_flush(journal);
3209 jbd2_journal_unlock_updates(journal);
3215 return generic_block_bmap(mapping, block, ext4_get_block);
3218 static int ext4_readpage(struct file *file, struct page *page)
3221 struct inode *inode = page->mapping->host;
3223 trace_ext4_readpage(page);
3225 if (ext4_has_inline_data(inode))
3226 ret = ext4_readpage_inline(inode, page);
3229 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3236 ext4_readpages(struct file *file, struct address_space *mapping,
3237 struct list_head *pages, unsigned nr_pages)
3239 struct inode *inode = mapping->host;
3241 /* If the file has inline data, no need to do readpages. */
3242 if (ext4_has_inline_data(inode))
3245 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3248 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3249 unsigned int length)
3251 trace_ext4_invalidatepage(page, offset, length);
3253 /* No journalling happens on data buffers when this function is used */
3254 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3256 block_invalidatepage(page, offset, length);
3259 static int __ext4_journalled_invalidatepage(struct page *page,
3260 unsigned int offset,
3261 unsigned int length)
3263 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3265 trace_ext4_journalled_invalidatepage(page, offset, length);
3268 * If it's a full truncate we just forget about the pending dirtying
3270 if (offset == 0 && length == PAGE_SIZE)
3271 ClearPageChecked(page);
3273 return jbd2_journal_invalidatepage(journal, page, offset, length);
3276 /* Wrapper for aops... */
3277 static void ext4_journalled_invalidatepage(struct page *page,
3278 unsigned int offset,
3279 unsigned int length)
3281 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3284 static int ext4_releasepage(struct page *page, gfp_t wait)
3286 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3288 trace_ext4_releasepage(page);
3290 /* Page has dirty journalled data -> cannot release */
3291 if (PageChecked(page))
3294 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3296 return try_to_free_buffers(page);
3299 static bool ext4_inode_datasync_dirty(struct inode *inode)
3301 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3304 return !jbd2_transaction_committed(journal,
3305 EXT4_I(inode)->i_datasync_tid);
3306 /* Any metadata buffers to write? */
3307 if (!list_empty(&inode->i_mapping->private_list))
3309 return inode->i_state & I_DIRTY_DATASYNC;
3312 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3313 struct ext4_map_blocks *map, loff_t offset,
3316 u8 blkbits = inode->i_blkbits;
3319 * Writes that span EOF might trigger an I/O size update on completion,
3320 * so consider them to be dirty for the purpose of O_DSYNC, even if
3321 * there is no other metadata changes being made or are pending.
3324 if (ext4_inode_datasync_dirty(inode) ||
3325 offset + length > i_size_read(inode))
3326 iomap->flags |= IOMAP_F_DIRTY;
3328 if (map->m_flags & EXT4_MAP_NEW)
3329 iomap->flags |= IOMAP_F_NEW;
3331 iomap->bdev = inode->i_sb->s_bdev;
3332 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3333 iomap->offset = (u64) map->m_lblk << blkbits;
3334 iomap->length = (u64) map->m_len << blkbits;
3337 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3338 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3339 * set. In order for any allocated unwritten extents to be converted
3340 * into written extents correctly within the ->end_io() handler, we
3341 * need to ensure that the iomap->type is set appropriately. Hence, the
3342 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3345 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3346 iomap->type = IOMAP_UNWRITTEN;
3347 iomap->addr = (u64) map->m_pblk << blkbits;
3348 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3349 iomap->type = IOMAP_MAPPED;
3350 iomap->addr = (u64) map->m_pblk << blkbits;
3352 iomap->type = IOMAP_HOLE;
3353 iomap->addr = IOMAP_NULL_ADDR;
3357 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3361 u8 blkbits = inode->i_blkbits;
3362 int ret, dio_credits, m_flags = 0, retries = 0;
3365 * Trim the mapping request to the maximum value that we can map at
3366 * once for direct I/O.
3368 if (map->m_len > DIO_MAX_BLOCKS)
3369 map->m_len = DIO_MAX_BLOCKS;
3370 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3374 * Either we allocate blocks and then don't get an unwritten extent, so
3375 * in that case we have reserved enough credits. Or, the blocks are
3376 * already allocated and unwritten. In that case, the extent conversion
3377 * fits into the credits as well.
3379 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3381 return PTR_ERR(handle);
3384 * DAX and direct I/O are the only two operations that are currently
3385 * supported with IOMAP_WRITE.
3387 WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3389 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3391 * We use i_size instead of i_disksize here because delalloc writeback
3392 * can complete at any point during the I/O and subsequently push the
3393 * i_disksize out to i_size. This could be beyond where direct I/O is
3394 * happening and thus expose allocated blocks to direct I/O reads.
3396 else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode))
3397 m_flags = EXT4_GET_BLOCKS_CREATE;
3398 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3399 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3401 ret = ext4_map_blocks(handle, inode, map, m_flags);
3404 * We cannot fill holes in indirect tree based inodes as that could
3405 * expose stale data in the case of a crash. Use the magic error code
3406 * to fallback to buffered I/O.
3408 if (!m_flags && !ret)
3411 ext4_journal_stop(handle);
3412 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3419 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3420 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3423 struct ext4_map_blocks map;
3424 u8 blkbits = inode->i_blkbits;
3426 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3429 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3433 * Calculate the first and last logical blocks respectively.
3435 map.m_lblk = offset >> blkbits;
3436 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3437 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3439 if (flags & IOMAP_WRITE)
3440 ret = ext4_iomap_alloc(inode, &map, flags);
3442 ret = ext4_map_blocks(NULL, inode, &map, 0);
3447 ext4_set_iomap(inode, iomap, &map, offset, length);
3452 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3453 loff_t length, unsigned flags, struct iomap *iomap,
3454 struct iomap *srcmap)
3459 * Even for writes we don't need to allocate blocks, so just pretend
3460 * we are reading to save overhead of starting a transaction.
3462 flags &= ~IOMAP_WRITE;
3463 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3464 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3468 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3469 ssize_t written, unsigned flags, struct iomap *iomap)
3472 * Check to see whether an error occurred while writing out the data to
3473 * the allocated blocks. If so, return the magic error code so that we
3474 * fallback to buffered I/O and attempt to complete the remainder of
3475 * the I/O. Any blocks that may have been allocated in preparation for
3476 * the direct I/O will be reused during buffered I/O.
3478 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3484 const struct iomap_ops ext4_iomap_ops = {
3485 .iomap_begin = ext4_iomap_begin,
3486 .iomap_end = ext4_iomap_end,
3489 const struct iomap_ops ext4_iomap_overwrite_ops = {
3490 .iomap_begin = ext4_iomap_overwrite_begin,
3491 .iomap_end = ext4_iomap_end,
3494 static bool ext4_iomap_is_delalloc(struct inode *inode,
3495 struct ext4_map_blocks *map)
3497 struct extent_status es;
3498 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3500 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3501 map->m_lblk, end, &es);
3503 if (!es.es_len || es.es_lblk > end)
3506 if (es.es_lblk > map->m_lblk) {
3507 map->m_len = es.es_lblk - map->m_lblk;
3511 offset = map->m_lblk - es.es_lblk;
3512 map->m_len = es.es_len - offset;
3517 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3518 loff_t length, unsigned int flags,
3519 struct iomap *iomap, struct iomap *srcmap)
3522 bool delalloc = false;
3523 struct ext4_map_blocks map;
3524 u8 blkbits = inode->i_blkbits;
3526 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3529 if (ext4_has_inline_data(inode)) {
3530 ret = ext4_inline_data_iomap(inode, iomap);
3531 if (ret != -EAGAIN) {
3532 if (ret == 0 && offset >= iomap->length)
3539 * Calculate the first and last logical block respectively.
3541 map.m_lblk = offset >> blkbits;
3542 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3543 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3545 ret = ext4_map_blocks(NULL, inode, &map, 0);
3549 delalloc = ext4_iomap_is_delalloc(inode, &map);
3551 ext4_set_iomap(inode, iomap, &map, offset, length);
3552 if (delalloc && iomap->type == IOMAP_HOLE)
3553 iomap->type = IOMAP_DELALLOC;
3558 const struct iomap_ops ext4_iomap_report_ops = {
3559 .iomap_begin = ext4_iomap_begin_report,
3563 * Pages can be marked dirty completely asynchronously from ext4's journalling
3564 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3565 * much here because ->set_page_dirty is called under VFS locks. The page is
3566 * not necessarily locked.
3568 * We cannot just dirty the page and leave attached buffers clean, because the
3569 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3570 * or jbddirty because all the journalling code will explode.
3572 * So what we do is to mark the page "pending dirty" and next time writepage
3573 * is called, propagate that into the buffers appropriately.
3575 static int ext4_journalled_set_page_dirty(struct page *page)
3577 SetPageChecked(page);
3578 return __set_page_dirty_nobuffers(page);
3581 static int ext4_set_page_dirty(struct page *page)
3583 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3584 WARN_ON_ONCE(!page_has_buffers(page));
3585 return __set_page_dirty_buffers(page);
3588 static const struct address_space_operations ext4_aops = {
3589 .readpage = ext4_readpage,
3590 .readpages = ext4_readpages,
3591 .writepage = ext4_writepage,
3592 .writepages = ext4_writepages,
3593 .write_begin = ext4_write_begin,
3594 .write_end = ext4_write_end,
3595 .set_page_dirty = ext4_set_page_dirty,
3597 .invalidatepage = ext4_invalidatepage,
3598 .releasepage = ext4_releasepage,
3599 .direct_IO = noop_direct_IO,
3600 .migratepage = buffer_migrate_page,
3601 .is_partially_uptodate = block_is_partially_uptodate,
3602 .error_remove_page = generic_error_remove_page,
3605 static const struct address_space_operations ext4_journalled_aops = {
3606 .readpage = ext4_readpage,
3607 .readpages = ext4_readpages,
3608 .writepage = ext4_writepage,
3609 .writepages = ext4_writepages,
3610 .write_begin = ext4_write_begin,
3611 .write_end = ext4_journalled_write_end,
3612 .set_page_dirty = ext4_journalled_set_page_dirty,
3614 .invalidatepage = ext4_journalled_invalidatepage,
3615 .releasepage = ext4_releasepage,
3616 .direct_IO = noop_direct_IO,
3617 .is_partially_uptodate = block_is_partially_uptodate,
3618 .error_remove_page = generic_error_remove_page,
3621 static const struct address_space_operations ext4_da_aops = {
3622 .readpage = ext4_readpage,
3623 .readpages = ext4_readpages,
3624 .writepage = ext4_writepage,
3625 .writepages = ext4_writepages,
3626 .write_begin = ext4_da_write_begin,
3627 .write_end = ext4_da_write_end,
3628 .set_page_dirty = ext4_set_page_dirty,
3630 .invalidatepage = ext4_invalidatepage,
3631 .releasepage = ext4_releasepage,
3632 .direct_IO = noop_direct_IO,
3633 .migratepage = buffer_migrate_page,
3634 .is_partially_uptodate = block_is_partially_uptodate,
3635 .error_remove_page = generic_error_remove_page,
3638 static const struct address_space_operations ext4_dax_aops = {
3639 .writepages = ext4_dax_writepages,
3640 .direct_IO = noop_direct_IO,
3641 .set_page_dirty = noop_set_page_dirty,
3643 .invalidatepage = noop_invalidatepage,
3646 void ext4_set_aops(struct inode *inode)
3648 switch (ext4_inode_journal_mode(inode)) {
3649 case EXT4_INODE_ORDERED_DATA_MODE:
3650 case EXT4_INODE_WRITEBACK_DATA_MODE:
3652 case EXT4_INODE_JOURNAL_DATA_MODE:
3653 inode->i_mapping->a_ops = &ext4_journalled_aops;
3659 inode->i_mapping->a_ops = &ext4_dax_aops;
3660 else if (test_opt(inode->i_sb, DELALLOC))
3661 inode->i_mapping->a_ops = &ext4_da_aops;
3663 inode->i_mapping->a_ops = &ext4_aops;
3666 static int __ext4_block_zero_page_range(handle_t *handle,
3667 struct address_space *mapping, loff_t from, loff_t length)
3669 ext4_fsblk_t index = from >> PAGE_SHIFT;
3670 unsigned offset = from & (PAGE_SIZE-1);
3671 unsigned blocksize, pos;
3673 struct inode *inode = mapping->host;
3674 struct buffer_head *bh;
3678 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3679 mapping_gfp_constraint(mapping, ~__GFP_FS));
3683 blocksize = inode->i_sb->s_blocksize;
3685 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3687 if (!page_has_buffers(page))
3688 create_empty_buffers(page, blocksize, 0);
3690 /* Find the buffer that contains "offset" */
3691 bh = page_buffers(page);
3693 while (offset >= pos) {
3694 bh = bh->b_this_page;
3698 if (buffer_freed(bh)) {
3699 BUFFER_TRACE(bh, "freed: skip");
3702 if (!buffer_mapped(bh)) {
3703 BUFFER_TRACE(bh, "unmapped");
3704 ext4_get_block(inode, iblock, bh, 0);
3705 /* unmapped? It's a hole - nothing to do */
3706 if (!buffer_mapped(bh)) {
3707 BUFFER_TRACE(bh, "still unmapped");
3712 /* Ok, it's mapped. Make sure it's up-to-date */
3713 if (PageUptodate(page))
3714 set_buffer_uptodate(bh);
3716 if (!buffer_uptodate(bh)) {
3718 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3720 /* Uhhuh. Read error. Complain and punt. */
3721 if (!buffer_uptodate(bh))
3723 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
3724 /* We expect the key to be set. */
3725 BUG_ON(!fscrypt_has_encryption_key(inode));
3726 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3729 clear_buffer_uptodate(bh);
3734 if (ext4_should_journal_data(inode)) {
3735 BUFFER_TRACE(bh, "get write access");
3736 err = ext4_journal_get_write_access(handle, bh);
3740 zero_user(page, offset, length);
3741 BUFFER_TRACE(bh, "zeroed end of block");
3743 if (ext4_should_journal_data(inode)) {
3744 err = ext4_handle_dirty_metadata(handle, inode, bh);
3747 mark_buffer_dirty(bh);
3748 if (ext4_should_order_data(inode))
3749 err = ext4_jbd2_inode_add_write(handle, inode, from,
3760 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3761 * starting from file offset 'from'. The range to be zero'd must
3762 * be contained with in one block. If the specified range exceeds
3763 * the end of the block it will be shortened to end of the block
3764 * that cooresponds to 'from'
3766 static int ext4_block_zero_page_range(handle_t *handle,
3767 struct address_space *mapping, loff_t from, loff_t length)
3769 struct inode *inode = mapping->host;
3770 unsigned offset = from & (PAGE_SIZE-1);
3771 unsigned blocksize = inode->i_sb->s_blocksize;
3772 unsigned max = blocksize - (offset & (blocksize - 1));
3775 * correct length if it does not fall between
3776 * 'from' and the end of the block
3778 if (length > max || length < 0)
3781 if (IS_DAX(inode)) {
3782 return iomap_zero_range(inode, from, length, NULL,
3785 return __ext4_block_zero_page_range(handle, mapping, from, length);
3789 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3790 * up to the end of the block which corresponds to `from'.
3791 * This required during truncate. We need to physically zero the tail end
3792 * of that block so it doesn't yield old data if the file is later grown.
3794 static int ext4_block_truncate_page(handle_t *handle,
3795 struct address_space *mapping, loff_t from)
3797 unsigned offset = from & (PAGE_SIZE-1);
3800 struct inode *inode = mapping->host;
3802 /* If we are processing an encrypted inode during orphan list handling */
3803 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3806 blocksize = inode->i_sb->s_blocksize;
3807 length = blocksize - (offset & (blocksize - 1));
3809 return ext4_block_zero_page_range(handle, mapping, from, length);
3812 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3813 loff_t lstart, loff_t length)
3815 struct super_block *sb = inode->i_sb;
3816 struct address_space *mapping = inode->i_mapping;
3817 unsigned partial_start, partial_end;
3818 ext4_fsblk_t start, end;
3819 loff_t byte_end = (lstart + length - 1);
3822 partial_start = lstart & (sb->s_blocksize - 1);
3823 partial_end = byte_end & (sb->s_blocksize - 1);
3825 start = lstart >> sb->s_blocksize_bits;
3826 end = byte_end >> sb->s_blocksize_bits;
3828 /* Handle partial zero within the single block */
3830 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3831 err = ext4_block_zero_page_range(handle, mapping,
3835 /* Handle partial zero out on the start of the range */
3836 if (partial_start) {
3837 err = ext4_block_zero_page_range(handle, mapping,
3838 lstart, sb->s_blocksize);
3842 /* Handle partial zero out on the end of the range */
3843 if (partial_end != sb->s_blocksize - 1)
3844 err = ext4_block_zero_page_range(handle, mapping,
3845 byte_end - partial_end,
3850 int ext4_can_truncate(struct inode *inode)
3852 if (S_ISREG(inode->i_mode))
3854 if (S_ISDIR(inode->i_mode))
3856 if (S_ISLNK(inode->i_mode))
3857 return !ext4_inode_is_fast_symlink(inode);
3862 * We have to make sure i_disksize gets properly updated before we truncate
3863 * page cache due to hole punching or zero range. Otherwise i_disksize update
3864 * can get lost as it may have been postponed to submission of writeback but
3865 * that will never happen after we truncate page cache.
3867 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3871 loff_t size = i_size_read(inode);
3873 WARN_ON(!inode_is_locked(inode));
3874 if (offset > size || offset + len < size)
3877 if (EXT4_I(inode)->i_disksize >= size)
3880 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3882 return PTR_ERR(handle);
3883 ext4_update_i_disksize(inode, size);
3884 ext4_mark_inode_dirty(handle, inode);
3885 ext4_journal_stop(handle);
3890 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
3892 up_write(&ei->i_mmap_sem);
3894 down_write(&ei->i_mmap_sem);
3897 int ext4_break_layouts(struct inode *inode)
3899 struct ext4_inode_info *ei = EXT4_I(inode);
3903 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
3907 page = dax_layout_busy_page(inode->i_mapping);
3911 error = ___wait_var_event(&page->_refcount,
3912 atomic_read(&page->_refcount) == 1,
3913 TASK_INTERRUPTIBLE, 0, 0,
3914 ext4_wait_dax_page(ei));
3915 } while (error == 0);
3921 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3922 * associated with the given offset and length
3924 * @inode: File inode
3925 * @offset: The offset where the hole will begin
3926 * @len: The length of the hole
3928 * Returns: 0 on success or negative on failure
3931 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3933 struct super_block *sb = inode->i_sb;
3934 ext4_lblk_t first_block, stop_block;
3935 struct address_space *mapping = inode->i_mapping;
3936 loff_t first_block_offset, last_block_offset;
3938 unsigned int credits;
3941 trace_ext4_punch_hole(inode, offset, length, 0);
3943 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
3944 if (ext4_has_inline_data(inode)) {
3945 down_write(&EXT4_I(inode)->i_mmap_sem);
3946 ret = ext4_convert_inline_data(inode);
3947 up_write(&EXT4_I(inode)->i_mmap_sem);
3953 * Write out all dirty pages to avoid race conditions
3954 * Then release them.
3956 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3957 ret = filemap_write_and_wait_range(mapping, offset,
3958 offset + length - 1);
3965 /* No need to punch hole beyond i_size */
3966 if (offset >= inode->i_size)
3970 * If the hole extends beyond i_size, set the hole
3971 * to end after the page that contains i_size
3973 if (offset + length > inode->i_size) {
3974 length = inode->i_size +
3975 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3979 if (offset & (sb->s_blocksize - 1) ||
3980 (offset + length) & (sb->s_blocksize - 1)) {
3982 * Attach jinode to inode for jbd2 if we do any zeroing of
3985 ret = ext4_inode_attach_jinode(inode);
3991 /* Wait all existing dio workers, newcomers will block on i_mutex */
3992 inode_dio_wait(inode);
3995 * Prevent page faults from reinstantiating pages we have released from
3998 down_write(&EXT4_I(inode)->i_mmap_sem);
4000 ret = ext4_break_layouts(inode);
4004 first_block_offset = round_up(offset, sb->s_blocksize);
4005 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4007 /* Now release the pages and zero block aligned part of pages*/
4008 if (last_block_offset > first_block_offset) {
4009 ret = ext4_update_disksize_before_punch(inode, offset, length);
4012 truncate_pagecache_range(inode, first_block_offset,
4016 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4017 credits = ext4_writepage_trans_blocks(inode);
4019 credits = ext4_blocks_for_truncate(inode);
4020 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4021 if (IS_ERR(handle)) {
4022 ret = PTR_ERR(handle);
4023 ext4_std_error(sb, ret);
4027 ret = ext4_zero_partial_blocks(handle, inode, offset,
4032 first_block = (offset + sb->s_blocksize - 1) >>
4033 EXT4_BLOCK_SIZE_BITS(sb);
4034 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4036 /* If there are blocks to remove, do it */
4037 if (stop_block > first_block) {
4039 down_write(&EXT4_I(inode)->i_data_sem);
4040 ext4_discard_preallocations(inode);
4042 ret = ext4_es_remove_extent(inode, first_block,
4043 stop_block - first_block);
4045 up_write(&EXT4_I(inode)->i_data_sem);
4049 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4050 ret = ext4_ext_remove_space(inode, first_block,
4053 ret = ext4_ind_remove_space(handle, inode, first_block,
4056 up_write(&EXT4_I(inode)->i_data_sem);
4059 ext4_handle_sync(handle);
4061 inode->i_mtime = inode->i_ctime = current_time(inode);
4062 ext4_mark_inode_dirty(handle, inode);
4064 ext4_update_inode_fsync_trans(handle, inode, 1);
4066 ext4_journal_stop(handle);
4068 up_write(&EXT4_I(inode)->i_mmap_sem);
4070 inode_unlock(inode);
4074 int ext4_inode_attach_jinode(struct inode *inode)
4076 struct ext4_inode_info *ei = EXT4_I(inode);
4077 struct jbd2_inode *jinode;
4079 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4082 jinode = jbd2_alloc_inode(GFP_KERNEL);
4083 spin_lock(&inode->i_lock);
4086 spin_unlock(&inode->i_lock);
4089 ei->jinode = jinode;
4090 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4093 spin_unlock(&inode->i_lock);
4094 if (unlikely(jinode != NULL))
4095 jbd2_free_inode(jinode);
4102 * We block out ext4_get_block() block instantiations across the entire
4103 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4104 * simultaneously on behalf of the same inode.
4106 * As we work through the truncate and commit bits of it to the journal there
4107 * is one core, guiding principle: the file's tree must always be consistent on
4108 * disk. We must be able to restart the truncate after a crash.
4110 * The file's tree may be transiently inconsistent in memory (although it
4111 * probably isn't), but whenever we close off and commit a journal transaction,
4112 * the contents of (the filesystem + the journal) must be consistent and
4113 * restartable. It's pretty simple, really: bottom up, right to left (although
4114 * left-to-right works OK too).
4116 * Note that at recovery time, journal replay occurs *before* the restart of
4117 * truncate against the orphan inode list.
4119 * The committed inode has the new, desired i_size (which is the same as
4120 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4121 * that this inode's truncate did not complete and it will again call
4122 * ext4_truncate() to have another go. So there will be instantiated blocks
4123 * to the right of the truncation point in a crashed ext4 filesystem. But
4124 * that's fine - as long as they are linked from the inode, the post-crash
4125 * ext4_truncate() run will find them and release them.
4127 int ext4_truncate(struct inode *inode)
4129 struct ext4_inode_info *ei = EXT4_I(inode);
4130 unsigned int credits;
4133 struct address_space *mapping = inode->i_mapping;
4136 * There is a possibility that we're either freeing the inode
4137 * or it's a completely new inode. In those cases we might not
4138 * have i_mutex locked because it's not necessary.
4140 if (!(inode->i_state & (I_NEW|I_FREEING)))
4141 WARN_ON(!inode_is_locked(inode));
4142 trace_ext4_truncate_enter(inode);
4144 if (!ext4_can_truncate(inode))
4147 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4149 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4150 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4152 if (ext4_has_inline_data(inode)) {
4155 err = ext4_inline_data_truncate(inode, &has_inline);
4162 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4163 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4164 if (ext4_inode_attach_jinode(inode) < 0)
4168 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4169 credits = ext4_writepage_trans_blocks(inode);
4171 credits = ext4_blocks_for_truncate(inode);
4173 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4175 return PTR_ERR(handle);
4177 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4178 ext4_block_truncate_page(handle, mapping, inode->i_size);
4181 * We add the inode to the orphan list, so that if this
4182 * truncate spans multiple transactions, and we crash, we will
4183 * resume the truncate when the filesystem recovers. It also
4184 * marks the inode dirty, to catch the new size.
4186 * Implication: the file must always be in a sane, consistent
4187 * truncatable state while each transaction commits.
4189 err = ext4_orphan_add(handle, inode);
4193 down_write(&EXT4_I(inode)->i_data_sem);
4195 ext4_discard_preallocations(inode);
4197 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4198 err = ext4_ext_truncate(handle, inode);
4200 ext4_ind_truncate(handle, inode);
4202 up_write(&ei->i_data_sem);
4207 ext4_handle_sync(handle);
4211 * If this was a simple ftruncate() and the file will remain alive,
4212 * then we need to clear up the orphan record which we created above.
4213 * However, if this was a real unlink then we were called by
4214 * ext4_evict_inode(), and we allow that function to clean up the
4215 * orphan info for us.
4218 ext4_orphan_del(handle, inode);
4220 inode->i_mtime = inode->i_ctime = current_time(inode);
4221 ext4_mark_inode_dirty(handle, inode);
4222 ext4_journal_stop(handle);
4224 trace_ext4_truncate_exit(inode);
4229 * ext4_get_inode_loc returns with an extra refcount against the inode's
4230 * underlying buffer_head on success. If 'in_mem' is true, we have all
4231 * data in memory that is needed to recreate the on-disk version of this
4234 static int __ext4_get_inode_loc(struct inode *inode,
4235 struct ext4_iloc *iloc, int in_mem)
4237 struct ext4_group_desc *gdp;
4238 struct buffer_head *bh;
4239 struct super_block *sb = inode->i_sb;
4241 struct blk_plug plug;
4242 int inodes_per_block, inode_offset;
4245 if (inode->i_ino < EXT4_ROOT_INO ||
4246 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4247 return -EFSCORRUPTED;
4249 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4250 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4255 * Figure out the offset within the block group inode table
4257 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4258 inode_offset = ((inode->i_ino - 1) %
4259 EXT4_INODES_PER_GROUP(sb));
4260 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4261 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4263 bh = sb_getblk(sb, block);
4266 if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO))
4268 if (!buffer_uptodate(bh)) {
4272 * If the buffer has the write error flag, we have failed
4273 * to write out another inode in the same block. In this
4274 * case, we don't have to read the block because we may
4275 * read the old inode data successfully.
4277 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4278 set_buffer_uptodate(bh);
4280 if (buffer_uptodate(bh)) {
4281 /* someone brought it uptodate while we waited */
4287 * If we have all information of the inode in memory and this
4288 * is the only valid inode in the block, we need not read the
4292 struct buffer_head *bitmap_bh;
4295 start = inode_offset & ~(inodes_per_block - 1);
4297 /* Is the inode bitmap in cache? */
4298 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4299 if (unlikely(!bitmap_bh))
4303 * If the inode bitmap isn't in cache then the
4304 * optimisation may end up performing two reads instead
4305 * of one, so skip it.
4307 if (!buffer_uptodate(bitmap_bh)) {
4311 for (i = start; i < start + inodes_per_block; i++) {
4312 if (i == inode_offset)
4314 if (ext4_test_bit(i, bitmap_bh->b_data))
4318 if (i == start + inodes_per_block) {
4319 /* all other inodes are free, so skip I/O */
4320 memset(bh->b_data, 0, bh->b_size);
4321 set_buffer_uptodate(bh);
4329 * If we need to do any I/O, try to pre-readahead extra
4330 * blocks from the inode table.
4332 blk_start_plug(&plug);
4333 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4334 ext4_fsblk_t b, end, table;
4336 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4338 table = ext4_inode_table(sb, gdp);
4339 /* s_inode_readahead_blks is always a power of 2 */
4340 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4344 num = EXT4_INODES_PER_GROUP(sb);
4345 if (ext4_has_group_desc_csum(sb))
4346 num -= ext4_itable_unused_count(sb, gdp);
4347 table += num / inodes_per_block;
4351 sb_breadahead(sb, b++);
4355 * There are other valid inodes in the buffer, this inode
4356 * has in-inode xattrs, or we don't have this inode in memory.
4357 * Read the block from disk.
4359 trace_ext4_load_inode(inode);
4361 bh->b_end_io = end_buffer_read_sync;
4362 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4363 blk_finish_plug(&plug);
4365 if (!buffer_uptodate(bh)) {
4367 ext4_set_errno(inode->i_sb, EIO);
4368 EXT4_ERROR_INODE_BLOCK(inode, block,
4369 "unable to read itable block");
4379 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4381 /* We have all inode data except xattrs in memory here. */
4382 return __ext4_get_inode_loc(inode, iloc,
4383 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4386 static bool ext4_should_use_dax(struct inode *inode)
4388 if (!test_opt(inode->i_sb, DAX))
4390 if (!S_ISREG(inode->i_mode))
4392 if (ext4_should_journal_data(inode))
4394 if (ext4_has_inline_data(inode))
4396 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4398 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4403 void ext4_set_inode_flags(struct inode *inode)
4405 unsigned int flags = EXT4_I(inode)->i_flags;
4406 unsigned int new_fl = 0;
4408 if (flags & EXT4_SYNC_FL)
4410 if (flags & EXT4_APPEND_FL)
4412 if (flags & EXT4_IMMUTABLE_FL)
4413 new_fl |= S_IMMUTABLE;
4414 if (flags & EXT4_NOATIME_FL)
4415 new_fl |= S_NOATIME;
4416 if (flags & EXT4_DIRSYNC_FL)
4417 new_fl |= S_DIRSYNC;
4418 if (ext4_should_use_dax(inode))
4420 if (flags & EXT4_ENCRYPT_FL)
4421 new_fl |= S_ENCRYPTED;
4422 if (flags & EXT4_CASEFOLD_FL)
4423 new_fl |= S_CASEFOLD;
4424 if (flags & EXT4_VERITY_FL)
4426 inode_set_flags(inode, new_fl,
4427 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4428 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4431 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4432 struct ext4_inode_info *ei)
4435 struct inode *inode = &(ei->vfs_inode);
4436 struct super_block *sb = inode->i_sb;
4438 if (ext4_has_feature_huge_file(sb)) {
4439 /* we are using combined 48 bit field */
4440 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4441 le32_to_cpu(raw_inode->i_blocks_lo);
4442 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4443 /* i_blocks represent file system block size */
4444 return i_blocks << (inode->i_blkbits - 9);
4449 return le32_to_cpu(raw_inode->i_blocks_lo);
4453 static inline int ext4_iget_extra_inode(struct inode *inode,
4454 struct ext4_inode *raw_inode,
4455 struct ext4_inode_info *ei)
4457 __le32 *magic = (void *)raw_inode +
4458 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4460 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4461 EXT4_INODE_SIZE(inode->i_sb) &&
4462 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4463 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4464 return ext4_find_inline_data_nolock(inode);
4466 EXT4_I(inode)->i_inline_off = 0;
4470 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4472 if (!ext4_has_feature_project(inode->i_sb))
4474 *projid = EXT4_I(inode)->i_projid;
4479 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4480 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4483 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4485 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4486 inode_set_iversion_raw(inode, val);
4488 inode_set_iversion_queried(inode, val);
4490 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4492 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4493 return inode_peek_iversion_raw(inode);
4495 return inode_peek_iversion(inode);
4498 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4499 ext4_iget_flags flags, const char *function,
4502 struct ext4_iloc iloc;
4503 struct ext4_inode *raw_inode;
4504 struct ext4_inode_info *ei;
4505 struct inode *inode;
4506 journal_t *journal = EXT4_SB(sb)->s_journal;
4514 if ((!(flags & EXT4_IGET_SPECIAL) &&
4515 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4516 (ino < EXT4_ROOT_INO) ||
4517 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4518 if (flags & EXT4_IGET_HANDLE)
4519 return ERR_PTR(-ESTALE);
4520 __ext4_error(sb, function, line,
4521 "inode #%lu: comm %s: iget: illegal inode #",
4522 ino, current->comm);
4523 return ERR_PTR(-EFSCORRUPTED);
4526 inode = iget_locked(sb, ino);
4528 return ERR_PTR(-ENOMEM);
4529 if (!(inode->i_state & I_NEW))
4535 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4538 raw_inode = ext4_raw_inode(&iloc);
4540 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4541 ext4_error_inode(inode, function, line, 0,
4542 "iget: root inode unallocated");
4543 ret = -EFSCORRUPTED;
4547 if ((flags & EXT4_IGET_HANDLE) &&
4548 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4553 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4554 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4555 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4556 EXT4_INODE_SIZE(inode->i_sb) ||
4557 (ei->i_extra_isize & 3)) {
4558 ext4_error_inode(inode, function, line, 0,
4559 "iget: bad extra_isize %u "
4562 EXT4_INODE_SIZE(inode->i_sb));
4563 ret = -EFSCORRUPTED;
4567 ei->i_extra_isize = 0;
4569 /* Precompute checksum seed for inode metadata */
4570 if (ext4_has_metadata_csum(sb)) {
4571 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4573 __le32 inum = cpu_to_le32(inode->i_ino);
4574 __le32 gen = raw_inode->i_generation;
4575 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4577 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4581 if (!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4582 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) {
4583 ext4_set_errno(inode->i_sb, EFSBADCRC);
4584 ext4_error_inode(inode, function, line, 0,
4585 "iget: checksum invalid");
4590 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4591 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4592 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4593 if (ext4_has_feature_project(sb) &&
4594 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4595 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4596 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4598 i_projid = EXT4_DEF_PROJID;
4600 if (!(test_opt(inode->i_sb, NO_UID32))) {
4601 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4602 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4604 i_uid_write(inode, i_uid);
4605 i_gid_write(inode, i_gid);
4606 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4607 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4609 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4610 ei->i_inline_off = 0;
4611 ei->i_dir_start_lookup = 0;
4612 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4613 /* We now have enough fields to check if the inode was active or not.
4614 * This is needed because nfsd might try to access dead inodes
4615 * the test is that same one that e2fsck uses
4616 * NeilBrown 1999oct15
4618 if (inode->i_nlink == 0) {
4619 if ((inode->i_mode == 0 ||
4620 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4621 ino != EXT4_BOOT_LOADER_INO) {
4622 /* this inode is deleted */
4626 /* The only unlinked inodes we let through here have
4627 * valid i_mode and are being read by the orphan
4628 * recovery code: that's fine, we're about to complete
4629 * the process of deleting those.
4630 * OR it is the EXT4_BOOT_LOADER_INO which is
4631 * not initialized on a new filesystem. */
4633 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4634 ext4_set_inode_flags(inode);
4635 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4636 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4637 if (ext4_has_feature_64bit(sb))
4639 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4640 inode->i_size = ext4_isize(sb, raw_inode);
4641 if ((size = i_size_read(inode)) < 0) {
4642 ext4_error_inode(inode, function, line, 0,
4643 "iget: bad i_size value: %lld", size);
4644 ret = -EFSCORRUPTED;
4647 ei->i_disksize = inode->i_size;
4649 ei->i_reserved_quota = 0;
4651 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4652 ei->i_block_group = iloc.block_group;
4653 ei->i_last_alloc_group = ~0;
4655 * NOTE! The in-memory inode i_data array is in little-endian order
4656 * even on big-endian machines: we do NOT byteswap the block numbers!
4658 for (block = 0; block < EXT4_N_BLOCKS; block++)
4659 ei->i_data[block] = raw_inode->i_block[block];
4660 INIT_LIST_HEAD(&ei->i_orphan);
4663 * Set transaction id's of transactions that have to be committed
4664 * to finish f[data]sync. We set them to currently running transaction
4665 * as we cannot be sure that the inode or some of its metadata isn't
4666 * part of the transaction - the inode could have been reclaimed and
4667 * now it is reread from disk.
4670 transaction_t *transaction;
4673 read_lock(&journal->j_state_lock);
4674 if (journal->j_running_transaction)
4675 transaction = journal->j_running_transaction;
4677 transaction = journal->j_committing_transaction;
4679 tid = transaction->t_tid;
4681 tid = journal->j_commit_sequence;
4682 read_unlock(&journal->j_state_lock);
4683 ei->i_sync_tid = tid;
4684 ei->i_datasync_tid = tid;
4687 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4688 if (ei->i_extra_isize == 0) {
4689 /* The extra space is currently unused. Use it. */
4690 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4691 ei->i_extra_isize = sizeof(struct ext4_inode) -
4692 EXT4_GOOD_OLD_INODE_SIZE;
4694 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4700 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4701 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4702 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4703 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4705 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4706 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4708 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4709 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4711 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4713 ext4_inode_set_iversion_queried(inode, ivers);
4717 if (ei->i_file_acl &&
4718 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4719 ext4_error_inode(inode, function, line, 0,
4720 "iget: bad extended attribute block %llu",
4722 ret = -EFSCORRUPTED;
4724 } else if (!ext4_has_inline_data(inode)) {
4725 /* validate the block references in the inode */
4726 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4727 (S_ISLNK(inode->i_mode) &&
4728 !ext4_inode_is_fast_symlink(inode))) {
4729 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4730 ret = ext4_ext_check_inode(inode);
4732 ret = ext4_ind_check_inode(inode);
4738 if (S_ISREG(inode->i_mode)) {
4739 inode->i_op = &ext4_file_inode_operations;
4740 inode->i_fop = &ext4_file_operations;
4741 ext4_set_aops(inode);
4742 } else if (S_ISDIR(inode->i_mode)) {
4743 inode->i_op = &ext4_dir_inode_operations;
4744 inode->i_fop = &ext4_dir_operations;
4745 } else if (S_ISLNK(inode->i_mode)) {
4746 /* VFS does not allow setting these so must be corruption */
4747 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4748 ext4_error_inode(inode, function, line, 0,
4749 "iget: immutable or append flags "
4750 "not allowed on symlinks");
4751 ret = -EFSCORRUPTED;
4754 if (IS_ENCRYPTED(inode)) {
4755 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4756 ext4_set_aops(inode);
4757 } else if (ext4_inode_is_fast_symlink(inode)) {
4758 inode->i_link = (char *)ei->i_data;
4759 inode->i_op = &ext4_fast_symlink_inode_operations;
4760 nd_terminate_link(ei->i_data, inode->i_size,
4761 sizeof(ei->i_data) - 1);
4763 inode->i_op = &ext4_symlink_inode_operations;
4764 ext4_set_aops(inode);
4766 inode_nohighmem(inode);
4767 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4768 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4769 inode->i_op = &ext4_special_inode_operations;
4770 if (raw_inode->i_block[0])
4771 init_special_inode(inode, inode->i_mode,
4772 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4774 init_special_inode(inode, inode->i_mode,
4775 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4776 } else if (ino == EXT4_BOOT_LOADER_INO) {
4777 make_bad_inode(inode);
4779 ret = -EFSCORRUPTED;
4780 ext4_error_inode(inode, function, line, 0,
4781 "iget: bogus i_mode (%o)", inode->i_mode);
4784 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4785 ext4_error_inode(inode, function, line, 0,
4786 "casefold flag without casefold feature");
4789 unlock_new_inode(inode);
4795 return ERR_PTR(ret);
4798 static int ext4_inode_blocks_set(handle_t *handle,
4799 struct ext4_inode *raw_inode,
4800 struct ext4_inode_info *ei)
4802 struct inode *inode = &(ei->vfs_inode);
4803 u64 i_blocks = inode->i_blocks;
4804 struct super_block *sb = inode->i_sb;
4806 if (i_blocks <= ~0U) {
4808 * i_blocks can be represented in a 32 bit variable
4809 * as multiple of 512 bytes
4811 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4812 raw_inode->i_blocks_high = 0;
4813 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4816 if (!ext4_has_feature_huge_file(sb))
4819 if (i_blocks <= 0xffffffffffffULL) {
4821 * i_blocks can be represented in a 48 bit variable
4822 * as multiple of 512 bytes
4824 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4825 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4826 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4828 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4829 /* i_block is stored in file system block size */
4830 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4831 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4832 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4837 struct other_inode {
4838 unsigned long orig_ino;
4839 struct ext4_inode *raw_inode;
4842 static int other_inode_match(struct inode * inode, unsigned long ino,
4845 struct other_inode *oi = (struct other_inode *) data;
4847 if ((inode->i_ino != ino) ||
4848 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4850 ((inode->i_state & I_DIRTY_TIME) == 0))
4852 spin_lock(&inode->i_lock);
4853 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4854 I_DIRTY_INODE)) == 0) &&
4855 (inode->i_state & I_DIRTY_TIME)) {
4856 struct ext4_inode_info *ei = EXT4_I(inode);
4858 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4859 spin_unlock(&inode->i_lock);
4861 spin_lock(&ei->i_raw_lock);
4862 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4863 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4864 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4865 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4866 spin_unlock(&ei->i_raw_lock);
4867 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4870 spin_unlock(&inode->i_lock);
4875 * Opportunistically update the other time fields for other inodes in
4876 * the same inode table block.
4878 static void ext4_update_other_inodes_time(struct super_block *sb,
4879 unsigned long orig_ino, char *buf)
4881 struct other_inode oi;
4883 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4884 int inode_size = EXT4_INODE_SIZE(sb);
4886 oi.orig_ino = orig_ino;
4888 * Calculate the first inode in the inode table block. Inode
4889 * numbers are one-based. That is, the first inode in a block
4890 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4892 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4893 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4894 if (ino == orig_ino)
4896 oi.raw_inode = (struct ext4_inode *) buf;
4897 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4902 * Post the struct inode info into an on-disk inode location in the
4903 * buffer-cache. This gobbles the caller's reference to the
4904 * buffer_head in the inode location struct.
4906 * The caller must have write access to iloc->bh.
4908 static int ext4_do_update_inode(handle_t *handle,
4909 struct inode *inode,
4910 struct ext4_iloc *iloc)
4912 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4913 struct ext4_inode_info *ei = EXT4_I(inode);
4914 struct buffer_head *bh = iloc->bh;
4915 struct super_block *sb = inode->i_sb;
4916 int err = 0, rc, block;
4917 int need_datasync = 0, set_large_file = 0;
4922 spin_lock(&ei->i_raw_lock);
4924 /* For fields not tracked in the in-memory inode,
4925 * initialise them to zero for new inodes. */
4926 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4927 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4929 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4930 i_uid = i_uid_read(inode);
4931 i_gid = i_gid_read(inode);
4932 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4933 if (!(test_opt(inode->i_sb, NO_UID32))) {
4934 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4935 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4937 * Fix up interoperability with old kernels. Otherwise, old inodes get
4938 * re-used with the upper 16 bits of the uid/gid intact
4940 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4941 raw_inode->i_uid_high = 0;
4942 raw_inode->i_gid_high = 0;
4944 raw_inode->i_uid_high =
4945 cpu_to_le16(high_16_bits(i_uid));
4946 raw_inode->i_gid_high =
4947 cpu_to_le16(high_16_bits(i_gid));
4950 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4951 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4952 raw_inode->i_uid_high = 0;
4953 raw_inode->i_gid_high = 0;
4955 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4957 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4958 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4959 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4960 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4962 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4964 spin_unlock(&ei->i_raw_lock);
4967 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4968 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4969 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4970 raw_inode->i_file_acl_high =
4971 cpu_to_le16(ei->i_file_acl >> 32);
4972 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4973 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
4974 ext4_isize_set(raw_inode, ei->i_disksize);
4977 if (ei->i_disksize > 0x7fffffffULL) {
4978 if (!ext4_has_feature_large_file(sb) ||
4979 EXT4_SB(sb)->s_es->s_rev_level ==
4980 cpu_to_le32(EXT4_GOOD_OLD_REV))
4983 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4984 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4985 if (old_valid_dev(inode->i_rdev)) {
4986 raw_inode->i_block[0] =
4987 cpu_to_le32(old_encode_dev(inode->i_rdev));
4988 raw_inode->i_block[1] = 0;
4990 raw_inode->i_block[0] = 0;
4991 raw_inode->i_block[1] =
4992 cpu_to_le32(new_encode_dev(inode->i_rdev));
4993 raw_inode->i_block[2] = 0;
4995 } else if (!ext4_has_inline_data(inode)) {
4996 for (block = 0; block < EXT4_N_BLOCKS; block++)
4997 raw_inode->i_block[block] = ei->i_data[block];
5000 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5001 u64 ivers = ext4_inode_peek_iversion(inode);
5003 raw_inode->i_disk_version = cpu_to_le32(ivers);
5004 if (ei->i_extra_isize) {
5005 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5006 raw_inode->i_version_hi =
5007 cpu_to_le32(ivers >> 32);
5008 raw_inode->i_extra_isize =
5009 cpu_to_le16(ei->i_extra_isize);
5013 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5014 i_projid != EXT4_DEF_PROJID);
5016 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5017 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5018 raw_inode->i_projid = cpu_to_le32(i_projid);
5020 ext4_inode_csum_set(inode, raw_inode, ei);
5021 spin_unlock(&ei->i_raw_lock);
5022 if (inode->i_sb->s_flags & SB_LAZYTIME)
5023 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5026 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5027 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5030 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5031 if (set_large_file) {
5032 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5033 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5036 ext4_set_feature_large_file(sb);
5037 ext4_handle_sync(handle);
5038 err = ext4_handle_dirty_super(handle, sb);
5040 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5043 ext4_std_error(inode->i_sb, err);
5048 * ext4_write_inode()
5050 * We are called from a few places:
5052 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5053 * Here, there will be no transaction running. We wait for any running
5054 * transaction to commit.
5056 * - Within flush work (sys_sync(), kupdate and such).
5057 * We wait on commit, if told to.
5059 * - Within iput_final() -> write_inode_now()
5060 * We wait on commit, if told to.
5062 * In all cases it is actually safe for us to return without doing anything,
5063 * because the inode has been copied into a raw inode buffer in
5064 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5067 * Note that we are absolutely dependent upon all inode dirtiers doing the
5068 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5069 * which we are interested.
5071 * It would be a bug for them to not do this. The code:
5073 * mark_inode_dirty(inode)
5075 * inode->i_size = expr;
5077 * is in error because write_inode() could occur while `stuff()' is running,
5078 * and the new i_size will be lost. Plus the inode will no longer be on the
5079 * superblock's dirty inode list.
5081 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5085 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5086 sb_rdonly(inode->i_sb))
5089 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5092 if (EXT4_SB(inode->i_sb)->s_journal) {
5093 if (ext4_journal_current_handle()) {
5094 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5100 * No need to force transaction in WB_SYNC_NONE mode. Also
5101 * ext4_sync_fs() will force the commit after everything is
5104 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5107 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5108 EXT4_I(inode)->i_sync_tid);
5110 struct ext4_iloc iloc;
5112 err = __ext4_get_inode_loc(inode, &iloc, 0);
5116 * sync(2) will flush the whole buffer cache. No need to do
5117 * it here separately for each inode.
5119 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5120 sync_dirty_buffer(iloc.bh);
5121 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5122 ext4_set_errno(inode->i_sb, EIO);
5123 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5124 "IO error syncing inode");
5133 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5134 * buffers that are attached to a page stradding i_size and are undergoing
5135 * commit. In that case we have to wait for commit to finish and try again.
5137 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5141 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5142 tid_t commit_tid = 0;
5145 offset = inode->i_size & (PAGE_SIZE - 1);
5147 * If the page is fully truncated, we don't need to wait for any commit
5148 * (and we even should not as __ext4_journalled_invalidatepage() may
5149 * strip all buffers from the page but keep the page dirty which can then
5150 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5151 * buffers). Also we don't need to wait for any commit if all buffers in
5152 * the page remain valid. This is most beneficial for the common case of
5153 * blocksize == PAGESIZE.
5155 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5158 page = find_lock_page(inode->i_mapping,
5159 inode->i_size >> PAGE_SHIFT);
5162 ret = __ext4_journalled_invalidatepage(page, offset,
5163 PAGE_SIZE - offset);
5169 read_lock(&journal->j_state_lock);
5170 if (journal->j_committing_transaction)
5171 commit_tid = journal->j_committing_transaction->t_tid;
5172 read_unlock(&journal->j_state_lock);
5174 jbd2_log_wait_commit(journal, commit_tid);
5181 * Called from notify_change.
5183 * We want to trap VFS attempts to truncate the file as soon as
5184 * possible. In particular, we want to make sure that when the VFS
5185 * shrinks i_size, we put the inode on the orphan list and modify
5186 * i_disksize immediately, so that during the subsequent flushing of
5187 * dirty pages and freeing of disk blocks, we can guarantee that any
5188 * commit will leave the blocks being flushed in an unused state on
5189 * disk. (On recovery, the inode will get truncated and the blocks will
5190 * be freed, so we have a strong guarantee that no future commit will
5191 * leave these blocks visible to the user.)
5193 * Another thing we have to assure is that if we are in ordered mode
5194 * and inode is still attached to the committing transaction, we must
5195 * we start writeout of all the dirty pages which are being truncated.
5196 * This way we are sure that all the data written in the previous
5197 * transaction are already on disk (truncate waits for pages under
5200 * Called with inode->i_mutex down.
5202 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5204 struct inode *inode = d_inode(dentry);
5207 const unsigned int ia_valid = attr->ia_valid;
5209 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5212 if (unlikely(IS_IMMUTABLE(inode)))
5215 if (unlikely(IS_APPEND(inode) &&
5216 (ia_valid & (ATTR_MODE | ATTR_UID |
5217 ATTR_GID | ATTR_TIMES_SET))))
5220 error = setattr_prepare(dentry, attr);
5224 error = fscrypt_prepare_setattr(dentry, attr);
5228 error = fsverity_prepare_setattr(dentry, attr);
5232 if (is_quota_modification(inode, attr)) {
5233 error = dquot_initialize(inode);
5237 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5238 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5241 /* (user+group)*(old+new) structure, inode write (sb,
5242 * inode block, ? - but truncate inode update has it) */
5243 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5244 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5245 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5246 if (IS_ERR(handle)) {
5247 error = PTR_ERR(handle);
5251 /* dquot_transfer() calls back ext4_get_inode_usage() which
5252 * counts xattr inode references.
5254 down_read(&EXT4_I(inode)->xattr_sem);
5255 error = dquot_transfer(inode, attr);
5256 up_read(&EXT4_I(inode)->xattr_sem);
5259 ext4_journal_stop(handle);
5262 /* Update corresponding info in inode so that everything is in
5263 * one transaction */
5264 if (attr->ia_valid & ATTR_UID)
5265 inode->i_uid = attr->ia_uid;
5266 if (attr->ia_valid & ATTR_GID)
5267 inode->i_gid = attr->ia_gid;
5268 error = ext4_mark_inode_dirty(handle, inode);
5269 ext4_journal_stop(handle);
5272 if (attr->ia_valid & ATTR_SIZE) {
5274 loff_t oldsize = inode->i_size;
5275 int shrink = (attr->ia_size < inode->i_size);
5277 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5278 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5280 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5283 if (!S_ISREG(inode->i_mode))
5286 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5287 inode_inc_iversion(inode);
5290 if (ext4_should_order_data(inode)) {
5291 error = ext4_begin_ordered_truncate(inode,
5297 * Blocks are going to be removed from the inode. Wait
5298 * for dio in flight.
5300 inode_dio_wait(inode);
5303 down_write(&EXT4_I(inode)->i_mmap_sem);
5305 rc = ext4_break_layouts(inode);
5307 up_write(&EXT4_I(inode)->i_mmap_sem);
5311 if (attr->ia_size != inode->i_size) {
5312 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5313 if (IS_ERR(handle)) {
5314 error = PTR_ERR(handle);
5317 if (ext4_handle_valid(handle) && shrink) {
5318 error = ext4_orphan_add(handle, inode);
5322 * Update c/mtime on truncate up, ext4_truncate() will
5323 * update c/mtime in shrink case below
5326 inode->i_mtime = current_time(inode);
5327 inode->i_ctime = inode->i_mtime;
5329 down_write(&EXT4_I(inode)->i_data_sem);
5330 EXT4_I(inode)->i_disksize = attr->ia_size;
5331 rc = ext4_mark_inode_dirty(handle, inode);
5335 * We have to update i_size under i_data_sem together
5336 * with i_disksize to avoid races with writeback code
5337 * running ext4_wb_update_i_disksize().
5340 i_size_write(inode, attr->ia_size);
5341 up_write(&EXT4_I(inode)->i_data_sem);
5342 ext4_journal_stop(handle);
5346 pagecache_isize_extended(inode, oldsize,
5348 } else if (ext4_should_journal_data(inode)) {
5349 ext4_wait_for_tail_page_commit(inode);
5354 * Truncate pagecache after we've waited for commit
5355 * in data=journal mode to make pages freeable.
5357 truncate_pagecache(inode, inode->i_size);
5359 * Call ext4_truncate() even if i_size didn't change to
5360 * truncate possible preallocated blocks.
5362 if (attr->ia_size <= oldsize) {
5363 rc = ext4_truncate(inode);
5368 up_write(&EXT4_I(inode)->i_mmap_sem);
5372 setattr_copy(inode, attr);
5373 mark_inode_dirty(inode);
5377 * If the call to ext4_truncate failed to get a transaction handle at
5378 * all, we need to clean up the in-core orphan list manually.
5380 if (orphan && inode->i_nlink)
5381 ext4_orphan_del(NULL, inode);
5383 if (!error && (ia_valid & ATTR_MODE))
5384 rc = posix_acl_chmod(inode, inode->i_mode);
5387 ext4_std_error(inode->i_sb, error);
5393 int ext4_getattr(const struct path *path, struct kstat *stat,
5394 u32 request_mask, unsigned int query_flags)
5396 struct inode *inode = d_inode(path->dentry);
5397 struct ext4_inode *raw_inode;
5398 struct ext4_inode_info *ei = EXT4_I(inode);
5401 if ((request_mask & STATX_BTIME) &&
5402 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5403 stat->result_mask |= STATX_BTIME;
5404 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5405 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5408 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5409 if (flags & EXT4_APPEND_FL)
5410 stat->attributes |= STATX_ATTR_APPEND;
5411 if (flags & EXT4_COMPR_FL)
5412 stat->attributes |= STATX_ATTR_COMPRESSED;
5413 if (flags & EXT4_ENCRYPT_FL)
5414 stat->attributes |= STATX_ATTR_ENCRYPTED;
5415 if (flags & EXT4_IMMUTABLE_FL)
5416 stat->attributes |= STATX_ATTR_IMMUTABLE;
5417 if (flags & EXT4_NODUMP_FL)
5418 stat->attributes |= STATX_ATTR_NODUMP;
5419 if (flags & EXT4_VERITY_FL)
5420 stat->attributes |= STATX_ATTR_VERITY;
5422 stat->attributes_mask |= (STATX_ATTR_APPEND |
5423 STATX_ATTR_COMPRESSED |
5424 STATX_ATTR_ENCRYPTED |
5425 STATX_ATTR_IMMUTABLE |
5429 generic_fillattr(inode, stat);
5433 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5434 u32 request_mask, unsigned int query_flags)
5436 struct inode *inode = d_inode(path->dentry);
5437 u64 delalloc_blocks;
5439 ext4_getattr(path, stat, request_mask, query_flags);
5442 * If there is inline data in the inode, the inode will normally not
5443 * have data blocks allocated (it may have an external xattr block).
5444 * Report at least one sector for such files, so tools like tar, rsync,
5445 * others don't incorrectly think the file is completely sparse.
5447 if (unlikely(ext4_has_inline_data(inode)))
5448 stat->blocks += (stat->size + 511) >> 9;
5451 * We can't update i_blocks if the block allocation is delayed
5452 * otherwise in the case of system crash before the real block
5453 * allocation is done, we will have i_blocks inconsistent with
5454 * on-disk file blocks.
5455 * We always keep i_blocks updated together with real
5456 * allocation. But to not confuse with user, stat
5457 * will return the blocks that include the delayed allocation
5458 * blocks for this file.
5460 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5461 EXT4_I(inode)->i_reserved_data_blocks);
5462 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5466 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5469 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5470 return ext4_ind_trans_blocks(inode, lblocks);
5471 return ext4_ext_index_trans_blocks(inode, pextents);
5475 * Account for index blocks, block groups bitmaps and block group
5476 * descriptor blocks if modify datablocks and index blocks
5477 * worse case, the indexs blocks spread over different block groups
5479 * If datablocks are discontiguous, they are possible to spread over
5480 * different block groups too. If they are contiguous, with flexbg,
5481 * they could still across block group boundary.
5483 * Also account for superblock, inode, quota and xattr blocks
5485 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5488 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5494 * How many index blocks need to touch to map @lblocks logical blocks
5495 * to @pextents physical extents?
5497 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5502 * Now let's see how many group bitmaps and group descriptors need
5505 groups = idxblocks + pextents;
5507 if (groups > ngroups)
5509 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5510 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5512 /* bitmaps and block group descriptor blocks */
5513 ret += groups + gdpblocks;
5515 /* Blocks for super block, inode, quota and xattr blocks */
5516 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5522 * Calculate the total number of credits to reserve to fit
5523 * the modification of a single pages into a single transaction,
5524 * which may include multiple chunks of block allocations.
5526 * This could be called via ext4_write_begin()
5528 * We need to consider the worse case, when
5529 * one new block per extent.
5531 int ext4_writepage_trans_blocks(struct inode *inode)
5533 int bpp = ext4_journal_blocks_per_page(inode);
5536 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5538 /* Account for data blocks for journalled mode */
5539 if (ext4_should_journal_data(inode))
5545 * Calculate the journal credits for a chunk of data modification.
5547 * This is called from DIO, fallocate or whoever calling
5548 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5550 * journal buffers for data blocks are not included here, as DIO
5551 * and fallocate do no need to journal data buffers.
5553 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5555 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5559 * The caller must have previously called ext4_reserve_inode_write().
5560 * Give this, we know that the caller already has write access to iloc->bh.
5562 int ext4_mark_iloc_dirty(handle_t *handle,
5563 struct inode *inode, struct ext4_iloc *iloc)
5567 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5571 if (IS_I_VERSION(inode))
5572 inode_inc_iversion(inode);
5574 /* the do_update_inode consumes one bh->b_count */
5577 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5578 err = ext4_do_update_inode(handle, inode, iloc);
5584 * On success, We end up with an outstanding reference count against
5585 * iloc->bh. This _must_ be cleaned up later.
5589 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5590 struct ext4_iloc *iloc)
5594 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5597 err = ext4_get_inode_loc(inode, iloc);
5599 BUFFER_TRACE(iloc->bh, "get_write_access");
5600 err = ext4_journal_get_write_access(handle, iloc->bh);
5606 ext4_std_error(inode->i_sb, err);
5610 static int __ext4_expand_extra_isize(struct inode *inode,
5611 unsigned int new_extra_isize,
5612 struct ext4_iloc *iloc,
5613 handle_t *handle, int *no_expand)
5615 struct ext4_inode *raw_inode;
5616 struct ext4_xattr_ibody_header *header;
5617 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5618 struct ext4_inode_info *ei = EXT4_I(inode);
5621 /* this was checked at iget time, but double check for good measure */
5622 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5623 (ei->i_extra_isize & 3)) {
5624 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5626 EXT4_INODE_SIZE(inode->i_sb));
5627 return -EFSCORRUPTED;
5629 if ((new_extra_isize < ei->i_extra_isize) ||
5630 (new_extra_isize < 4) ||
5631 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5632 return -EINVAL; /* Should never happen */
5634 raw_inode = ext4_raw_inode(iloc);
5636 header = IHDR(inode, raw_inode);
5638 /* No extended attributes present */
5639 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5640 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5641 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5642 EXT4_I(inode)->i_extra_isize, 0,
5643 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5644 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5648 /* try to expand with EAs present */
5649 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5653 * Inode size expansion failed; don't try again
5662 * Expand an inode by new_extra_isize bytes.
5663 * Returns 0 on success or negative error number on failure.
5665 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5666 unsigned int new_extra_isize,
5667 struct ext4_iloc iloc,
5673 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5677 * In nojournal mode, we can immediately attempt to expand
5678 * the inode. When journaled, we first need to obtain extra
5679 * buffer credits since we may write into the EA block
5680 * with this same handle. If journal_extend fails, then it will
5681 * only result in a minor loss of functionality for that inode.
5682 * If this is felt to be critical, then e2fsck should be run to
5683 * force a large enough s_min_extra_isize.
5685 if (ext4_journal_extend(handle,
5686 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5689 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5692 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5693 handle, &no_expand);
5694 ext4_write_unlock_xattr(inode, &no_expand);
5699 int ext4_expand_extra_isize(struct inode *inode,
5700 unsigned int new_extra_isize,
5701 struct ext4_iloc *iloc)
5707 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5712 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5713 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5714 if (IS_ERR(handle)) {
5715 error = PTR_ERR(handle);
5720 ext4_write_lock_xattr(inode, &no_expand);
5722 BUFFER_TRACE(iloc->bh, "get_write_access");
5723 error = ext4_journal_get_write_access(handle, iloc->bh);
5729 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5730 handle, &no_expand);
5732 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5737 ext4_write_unlock_xattr(inode, &no_expand);
5738 ext4_journal_stop(handle);
5743 * What we do here is to mark the in-core inode as clean with respect to inode
5744 * dirtiness (it may still be data-dirty).
5745 * This means that the in-core inode may be reaped by prune_icache
5746 * without having to perform any I/O. This is a very good thing,
5747 * because *any* task may call prune_icache - even ones which
5748 * have a transaction open against a different journal.
5750 * Is this cheating? Not really. Sure, we haven't written the
5751 * inode out, but prune_icache isn't a user-visible syncing function.
5752 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5753 * we start and wait on commits.
5755 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5757 struct ext4_iloc iloc;
5758 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5762 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5763 err = ext4_reserve_inode_write(handle, inode, &iloc);
5767 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5768 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5771 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5775 * ext4_dirty_inode() is called from __mark_inode_dirty()
5777 * We're really interested in the case where a file is being extended.
5778 * i_size has been changed by generic_commit_write() and we thus need
5779 * to include the updated inode in the current transaction.
5781 * Also, dquot_alloc_block() will always dirty the inode when blocks
5782 * are allocated to the file.
5784 * If the inode is marked synchronous, we don't honour that here - doing
5785 * so would cause a commit on atime updates, which we don't bother doing.
5786 * We handle synchronous inodes at the highest possible level.
5788 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5789 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5790 * to copy into the on-disk inode structure are the timestamp files.
5792 void ext4_dirty_inode(struct inode *inode, int flags)
5796 if (flags == I_DIRTY_TIME)
5798 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5802 ext4_mark_inode_dirty(handle, inode);
5804 ext4_journal_stop(handle);
5809 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5814 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5817 * We have to be very careful here: changing a data block's
5818 * journaling status dynamically is dangerous. If we write a
5819 * data block to the journal, change the status and then delete
5820 * that block, we risk forgetting to revoke the old log record
5821 * from the journal and so a subsequent replay can corrupt data.
5822 * So, first we make sure that the journal is empty and that
5823 * nobody is changing anything.
5826 journal = EXT4_JOURNAL(inode);
5829 if (is_journal_aborted(journal))
5832 /* Wait for all existing dio workers */
5833 inode_dio_wait(inode);
5836 * Before flushing the journal and switching inode's aops, we have
5837 * to flush all dirty data the inode has. There can be outstanding
5838 * delayed allocations, there can be unwritten extents created by
5839 * fallocate or buffered writes in dioread_nolock mode covered by
5840 * dirty data which can be converted only after flushing the dirty
5841 * data (and journalled aops don't know how to handle these cases).
5844 down_write(&EXT4_I(inode)->i_mmap_sem);
5845 err = filemap_write_and_wait(inode->i_mapping);
5847 up_write(&EXT4_I(inode)->i_mmap_sem);
5852 percpu_down_write(&sbi->s_journal_flag_rwsem);
5853 jbd2_journal_lock_updates(journal);
5856 * OK, there are no updates running now, and all cached data is
5857 * synced to disk. We are now in a completely consistent state
5858 * which doesn't have anything in the journal, and we know that
5859 * no filesystem updates are running, so it is safe to modify
5860 * the inode's in-core data-journaling state flag now.
5864 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5866 err = jbd2_journal_flush(journal);
5868 jbd2_journal_unlock_updates(journal);
5869 percpu_up_write(&sbi->s_journal_flag_rwsem);
5872 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5874 ext4_set_aops(inode);
5876 jbd2_journal_unlock_updates(journal);
5877 percpu_up_write(&sbi->s_journal_flag_rwsem);
5880 up_write(&EXT4_I(inode)->i_mmap_sem);
5882 /* Finally we can mark the inode as dirty. */
5884 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5886 return PTR_ERR(handle);
5888 err = ext4_mark_inode_dirty(handle, inode);
5889 ext4_handle_sync(handle);
5890 ext4_journal_stop(handle);
5891 ext4_std_error(inode->i_sb, err);
5896 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5898 return !buffer_mapped(bh);
5901 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
5903 struct vm_area_struct *vma = vmf->vma;
5904 struct page *page = vmf->page;
5909 struct file *file = vma->vm_file;
5910 struct inode *inode = file_inode(file);
5911 struct address_space *mapping = inode->i_mapping;
5913 get_block_t *get_block;
5916 if (unlikely(IS_IMMUTABLE(inode)))
5917 return VM_FAULT_SIGBUS;
5919 sb_start_pagefault(inode->i_sb);
5920 file_update_time(vma->vm_file);
5922 down_read(&EXT4_I(inode)->i_mmap_sem);
5924 err = ext4_convert_inline_data(inode);
5928 /* Delalloc case is easy... */
5929 if (test_opt(inode->i_sb, DELALLOC) &&
5930 !ext4_should_journal_data(inode) &&
5931 !ext4_nonda_switch(inode->i_sb)) {
5933 err = block_page_mkwrite(vma, vmf,
5934 ext4_da_get_block_prep);
5935 } while (err == -ENOSPC &&
5936 ext4_should_retry_alloc(inode->i_sb, &retries));
5941 size = i_size_read(inode);
5942 /* Page got truncated from under us? */
5943 if (page->mapping != mapping || page_offset(page) > size) {
5945 ret = VM_FAULT_NOPAGE;
5949 if (page->index == size >> PAGE_SHIFT)
5950 len = size & ~PAGE_MASK;
5954 * Return if we have all the buffers mapped. This avoids the need to do
5955 * journal_start/journal_stop which can block and take a long time
5957 if (page_has_buffers(page)) {
5958 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5960 ext4_bh_unmapped)) {
5961 /* Wait so that we don't change page under IO */
5962 wait_for_stable_page(page);
5963 ret = VM_FAULT_LOCKED;
5968 /* OK, we need to fill the hole... */
5969 if (ext4_should_dioread_nolock(inode))
5970 get_block = ext4_get_block_unwritten;
5972 get_block = ext4_get_block;
5974 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5975 ext4_writepage_trans_blocks(inode));
5976 if (IS_ERR(handle)) {
5977 ret = VM_FAULT_SIGBUS;
5980 err = block_page_mkwrite(vma, vmf, get_block);
5981 if (!err && ext4_should_journal_data(inode)) {
5982 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5983 PAGE_SIZE, NULL, do_journal_get_write_access)) {
5985 ret = VM_FAULT_SIGBUS;
5986 ext4_journal_stop(handle);
5989 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5991 ext4_journal_stop(handle);
5992 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5995 ret = block_page_mkwrite_return(err);
5997 up_read(&EXT4_I(inode)->i_mmap_sem);
5998 sb_end_pagefault(inode->i_sb);
6002 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6004 struct inode *inode = file_inode(vmf->vma->vm_file);
6007 down_read(&EXT4_I(inode)->i_mmap_sem);
6008 ret = filemap_fault(vmf);
6009 up_read(&EXT4_I(inode)->i_mmap_sem);