1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
54 struct ext4_inode_info *ei)
56 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 int offset = offsetof(struct ext4_inode, i_checksum_lo);
60 unsigned int csum_size = sizeof(dummy_csum);
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
63 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
65 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
66 EXT4_GOOD_OLD_INODE_SIZE - offset);
68 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
69 offset = offsetof(struct ext4_inode, i_checksum_hi);
70 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
71 EXT4_GOOD_OLD_INODE_SIZE,
72 offset - EXT4_GOOD_OLD_INODE_SIZE);
73 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
74 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
78 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
79 EXT4_INODE_SIZE(inode->i_sb) - offset);
85 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
86 struct ext4_inode_info *ei)
88 __u32 provided, calculated;
90 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
91 cpu_to_le32(EXT4_OS_LINUX) ||
92 !ext4_has_metadata_csum(inode->i_sb))
95 provided = le16_to_cpu(raw->i_checksum_lo);
96 calculated = ext4_inode_csum(inode, raw, ei);
97 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
98 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
99 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 calculated &= 0xFFFF;
103 return provided == calculated;
106 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107 struct ext4_inode_info *ei)
111 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112 cpu_to_le32(EXT4_OS_LINUX) ||
113 !ext4_has_metadata_csum(inode->i_sb))
116 csum = ext4_inode_csum(inode, raw, ei);
117 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
123 static inline int ext4_begin_ordered_truncate(struct inode *inode,
126 trace_ext4_begin_ordered_truncate(inode, new_size);
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
133 if (!EXT4_I(inode)->jinode)
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136 EXT4_I(inode)->jinode,
140 static void ext4_invalidatepage(struct page *page, unsigned int offset,
141 unsigned int length);
142 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
144 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
151 int ext4_inode_is_fast_symlink(struct inode *inode)
153 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
157 if (ext4_has_inline_data(inode))
160 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
162 return S_ISLNK(inode->i_mode) && inode->i_size &&
163 (inode->i_size < EXT4_N_BLOCKS * 4);
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
177 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
178 * moment, get_block can be called only for blocks inside i_size since
179 * page cache has been already dropped and writes are blocked by
180 * i_mutex. So we can safely drop the i_data_sem here.
182 BUG_ON(EXT4_JOURNAL(inode) == NULL);
183 jbd_debug(2, "restarting handle %p\n", handle);
184 up_write(&EXT4_I(inode)->i_data_sem);
185 ret = ext4_journal_restart(handle, nblocks);
186 down_write(&EXT4_I(inode)->i_data_sem);
187 ext4_discard_preallocations(inode);
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_evict_inode(struct inode *inode)
200 * Credits for final inode cleanup and freeing:
201 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
202 * (xattr block freeing), bitmap, group descriptor (inode freeing)
204 int extra_credits = 6;
205 struct ext4_xattr_inode_array *ea_inode_array = NULL;
206 bool freeze_protected = false;
208 trace_ext4_evict_inode(inode);
210 if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
211 ext4_evict_ea_inode(inode);
212 if (inode->i_nlink) {
214 * When journalling data dirty buffers are tracked only in the
215 * journal. So although mm thinks everything is clean and
216 * ready for reaping the inode might still have some pages to
217 * write in the running transaction or waiting to be
218 * checkpointed. Thus calling jbd2_journal_invalidatepage()
219 * (via truncate_inode_pages()) to discard these buffers can
220 * cause data loss. Also even if we did not discard these
221 * buffers, we would have no way to find them after the inode
222 * is reaped and thus user could see stale data if he tries to
223 * read them before the transaction is checkpointed. So be
224 * careful and force everything to disk here... We use
225 * ei->i_datasync_tid to store the newest transaction
226 * containing inode's data.
228 * Note that directories do not have this problem because they
229 * don't use page cache.
231 if (inode->i_ino != EXT4_JOURNAL_INO &&
232 ext4_should_journal_data(inode) &&
233 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
234 inode->i_data.nrpages) {
235 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
236 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
238 jbd2_complete_transaction(journal, commit_tid);
239 filemap_write_and_wait(&inode->i_data);
241 truncate_inode_pages_final(&inode->i_data);
246 if (is_bad_inode(inode))
248 dquot_initialize(inode);
250 if (ext4_should_order_data(inode))
251 ext4_begin_ordered_truncate(inode, 0);
252 truncate_inode_pages_final(&inode->i_data);
255 * Protect us against freezing - iput() caller didn't have to have any
256 * protection against it. When we are in a running transaction though,
257 * we are already protected against freezing and we cannot grab further
258 * protection due to lock ordering constraints.
260 if (!ext4_journal_current_handle()) {
261 sb_start_intwrite(inode->i_sb);
262 freeze_protected = true;
265 if (!IS_NOQUOTA(inode))
266 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
269 * Block bitmap, group descriptor, and inode are accounted in both
270 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
272 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
273 ext4_blocks_for_truncate(inode) + extra_credits - 3);
274 if (IS_ERR(handle)) {
275 ext4_std_error(inode->i_sb, PTR_ERR(handle));
277 * If we're going to skip the normal cleanup, we still need to
278 * make sure that the in-core orphan linked list is properly
281 ext4_orphan_del(NULL, inode);
282 if (freeze_protected)
283 sb_end_intwrite(inode->i_sb);
288 ext4_handle_sync(handle);
291 * Set inode->i_size to 0 before calling ext4_truncate(). We need
292 * special handling of symlinks here because i_size is used to
293 * determine whether ext4_inode_info->i_data contains symlink data or
294 * block mappings. Setting i_size to 0 will remove its fast symlink
295 * status. Erase i_data so that it becomes a valid empty block map.
297 if (ext4_inode_is_fast_symlink(inode))
298 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
300 err = ext4_mark_inode_dirty(handle, inode);
302 ext4_warning(inode->i_sb,
303 "couldn't mark inode dirty (err %d)", err);
306 if (inode->i_blocks) {
307 err = ext4_truncate(inode);
309 ext4_error(inode->i_sb,
310 "couldn't truncate inode %lu (err %d)",
316 /* Remove xattr references. */
317 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
320 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
322 ext4_journal_stop(handle);
323 ext4_orphan_del(NULL, inode);
324 if (freeze_protected)
325 sb_end_intwrite(inode->i_sb);
326 ext4_xattr_inode_array_free(ea_inode_array);
331 * Kill off the orphan record which ext4_truncate created.
332 * AKPM: I think this can be inside the above `if'.
333 * Note that ext4_orphan_del() has to be able to cope with the
334 * deletion of a non-existent orphan - this is because we don't
335 * know if ext4_truncate() actually created an orphan record.
336 * (Well, we could do this if we need to, but heck - it works)
338 ext4_orphan_del(handle, inode);
339 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
342 * One subtle ordering requirement: if anything has gone wrong
343 * (transaction abort, IO errors, whatever), then we can still
344 * do these next steps (the fs will already have been marked as
345 * having errors), but we can't free the inode if the mark_dirty
348 if (ext4_mark_inode_dirty(handle, inode))
349 /* If that failed, just do the required in-core inode clear. */
350 ext4_clear_inode(inode);
352 ext4_free_inode(handle, inode);
353 ext4_journal_stop(handle);
354 if (freeze_protected)
355 sb_end_intwrite(inode->i_sb);
356 ext4_xattr_inode_array_free(ea_inode_array);
359 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
363 qsize_t *ext4_get_reserved_space(struct inode *inode)
365 return &EXT4_I(inode)->i_reserved_quota;
370 * Called with i_data_sem down, which is important since we can call
371 * ext4_discard_preallocations() from here.
373 void ext4_da_update_reserve_space(struct inode *inode,
374 int used, int quota_claim)
376 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
377 struct ext4_inode_info *ei = EXT4_I(inode);
379 spin_lock(&ei->i_block_reservation_lock);
380 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
381 if (unlikely(used > ei->i_reserved_data_blocks)) {
382 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
383 "with only %d reserved data blocks",
384 __func__, inode->i_ino, used,
385 ei->i_reserved_data_blocks);
387 used = ei->i_reserved_data_blocks;
390 /* Update per-inode reservations */
391 ei->i_reserved_data_blocks -= used;
392 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
394 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
396 /* Update quota subsystem for data blocks */
398 dquot_claim_block(inode, EXT4_C2B(sbi, used));
401 * We did fallocate with an offset that is already delayed
402 * allocated. So on delayed allocated writeback we should
403 * not re-claim the quota for fallocated blocks.
405 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
409 * If we have done all the pending block allocations and if
410 * there aren't any writers on the inode, we can discard the
411 * inode's preallocations.
413 if ((ei->i_reserved_data_blocks == 0) &&
414 !inode_is_open_for_write(inode))
415 ext4_discard_preallocations(inode);
418 static int __check_block_validity(struct inode *inode, const char *func,
420 struct ext4_map_blocks *map)
422 if (ext4_has_feature_journal(inode->i_sb) &&
424 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
426 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
428 ext4_error_inode(inode, func, line, map->m_pblk,
429 "lblock %lu mapped to illegal pblock %llu "
430 "(length %d)", (unsigned long) map->m_lblk,
431 map->m_pblk, map->m_len);
432 return -EFSCORRUPTED;
437 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
442 if (IS_ENCRYPTED(inode))
443 return fscrypt_zeroout_range(inode, lblk, pblk, len);
445 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
452 #define check_block_validity(inode, map) \
453 __check_block_validity((inode), __func__, __LINE__, (map))
455 #ifdef ES_AGGRESSIVE_TEST
456 static void ext4_map_blocks_es_recheck(handle_t *handle,
458 struct ext4_map_blocks *es_map,
459 struct ext4_map_blocks *map,
466 * There is a race window that the result is not the same.
467 * e.g. xfstests #223 when dioread_nolock enables. The reason
468 * is that we lookup a block mapping in extent status tree with
469 * out taking i_data_sem. So at the time the unwritten extent
470 * could be converted.
472 down_read(&EXT4_I(inode)->i_data_sem);
473 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
474 retval = ext4_ext_map_blocks(handle, inode, map, flags &
475 EXT4_GET_BLOCKS_KEEP_SIZE);
477 retval = ext4_ind_map_blocks(handle, inode, map, flags &
478 EXT4_GET_BLOCKS_KEEP_SIZE);
480 up_read((&EXT4_I(inode)->i_data_sem));
483 * We don't check m_len because extent will be collpased in status
484 * tree. So the m_len might not equal.
486 if (es_map->m_lblk != map->m_lblk ||
487 es_map->m_flags != map->m_flags ||
488 es_map->m_pblk != map->m_pblk) {
489 printk("ES cache assertion failed for inode: %lu "
490 "es_cached ex [%d/%d/%llu/%x] != "
491 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
492 inode->i_ino, es_map->m_lblk, es_map->m_len,
493 es_map->m_pblk, es_map->m_flags, map->m_lblk,
494 map->m_len, map->m_pblk, map->m_flags,
498 #endif /* ES_AGGRESSIVE_TEST */
501 * The ext4_map_blocks() function tries to look up the requested blocks,
502 * and returns if the blocks are already mapped.
504 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
505 * and store the allocated blocks in the result buffer head and mark it
508 * If file type is extents based, it will call ext4_ext_map_blocks(),
509 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
512 * On success, it returns the number of blocks being mapped or allocated. if
513 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
514 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
516 * It returns 0 if plain look up failed (blocks have not been allocated), in
517 * that case, @map is returned as unmapped but we still do fill map->m_len to
518 * indicate the length of a hole starting at map->m_lblk.
520 * It returns the error in case of allocation failure.
522 int ext4_map_blocks(handle_t *handle, struct inode *inode,
523 struct ext4_map_blocks *map, int flags)
525 struct extent_status es;
528 #ifdef ES_AGGRESSIVE_TEST
529 struct ext4_map_blocks orig_map;
531 memcpy(&orig_map, map, sizeof(*map));
535 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
536 "logical block %lu\n", inode->i_ino, flags, map->m_len,
537 (unsigned long) map->m_lblk);
540 * ext4_map_blocks returns an int, and m_len is an unsigned int
542 if (unlikely(map->m_len > INT_MAX))
543 map->m_len = INT_MAX;
545 /* We can handle the block number less than EXT_MAX_BLOCKS */
546 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
547 return -EFSCORRUPTED;
549 /* Lookup extent status tree firstly */
550 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
551 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
552 map->m_pblk = ext4_es_pblock(&es) +
553 map->m_lblk - es.es_lblk;
554 map->m_flags |= ext4_es_is_written(&es) ?
555 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
556 retval = es.es_len - (map->m_lblk - es.es_lblk);
557 if (retval > map->m_len)
560 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
562 retval = es.es_len - (map->m_lblk - es.es_lblk);
563 if (retval > map->m_len)
570 #ifdef ES_AGGRESSIVE_TEST
571 ext4_map_blocks_es_recheck(handle, inode, map,
578 * Try to see if we can get the block without requesting a new
581 down_read(&EXT4_I(inode)->i_data_sem);
582 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
583 retval = ext4_ext_map_blocks(handle, inode, map, flags &
584 EXT4_GET_BLOCKS_KEEP_SIZE);
586 retval = ext4_ind_map_blocks(handle, inode, map, flags &
587 EXT4_GET_BLOCKS_KEEP_SIZE);
592 if (unlikely(retval != map->m_len)) {
593 ext4_warning(inode->i_sb,
594 "ES len assertion failed for inode "
595 "%lu: retval %d != map->m_len %d",
596 inode->i_ino, retval, map->m_len);
600 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
601 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
602 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
603 !(status & EXTENT_STATUS_WRITTEN) &&
604 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
605 map->m_lblk + map->m_len - 1))
606 status |= EXTENT_STATUS_DELAYED;
607 ret = ext4_es_insert_extent(inode, map->m_lblk,
608 map->m_len, map->m_pblk, status);
612 up_read((&EXT4_I(inode)->i_data_sem));
615 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
616 ret = check_block_validity(inode, map);
621 /* If it is only a block(s) look up */
622 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
626 * Returns if the blocks have already allocated
628 * Note that if blocks have been preallocated
629 * ext4_ext_get_block() returns the create = 0
630 * with buffer head unmapped.
632 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
634 * If we need to convert extent to unwritten
635 * we continue and do the actual work in
636 * ext4_ext_map_blocks()
638 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
642 * Here we clear m_flags because after allocating an new extent,
643 * it will be set again.
645 map->m_flags &= ~EXT4_MAP_FLAGS;
648 * New blocks allocate and/or writing to unwritten extent
649 * will possibly result in updating i_data, so we take
650 * the write lock of i_data_sem, and call get_block()
651 * with create == 1 flag.
653 down_write(&EXT4_I(inode)->i_data_sem);
656 * We need to check for EXT4 here because migrate
657 * could have changed the inode type in between
659 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
660 retval = ext4_ext_map_blocks(handle, inode, map, flags);
662 retval = ext4_ind_map_blocks(handle, inode, map, flags);
664 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
666 * We allocated new blocks which will result in
667 * i_data's format changing. Force the migrate
668 * to fail by clearing migrate flags
670 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
677 if (unlikely(retval != map->m_len)) {
678 ext4_warning(inode->i_sb,
679 "ES len assertion failed for inode "
680 "%lu: retval %d != map->m_len %d",
681 inode->i_ino, retval, map->m_len);
686 * We have to zeroout blocks before inserting them into extent
687 * status tree. Otherwise someone could look them up there and
688 * use them before they are really zeroed. We also have to
689 * unmap metadata before zeroing as otherwise writeback can
690 * overwrite zeros with stale data from block device.
692 if (flags & EXT4_GET_BLOCKS_ZERO &&
693 map->m_flags & EXT4_MAP_MAPPED &&
694 map->m_flags & EXT4_MAP_NEW) {
695 ret = ext4_issue_zeroout(inode, map->m_lblk,
696 map->m_pblk, map->m_len);
704 * If the extent has been zeroed out, we don't need to update
705 * extent status tree.
707 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
708 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
709 if (ext4_es_is_written(&es))
712 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
713 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
714 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
715 !(status & EXTENT_STATUS_WRITTEN) &&
716 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
717 map->m_lblk + map->m_len - 1))
718 status |= EXTENT_STATUS_DELAYED;
719 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
720 map->m_pblk, status);
728 up_write((&EXT4_I(inode)->i_data_sem));
729 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
730 ret = check_block_validity(inode, map);
735 * Inodes with freshly allocated blocks where contents will be
736 * visible after transaction commit must be on transaction's
739 if (map->m_flags & EXT4_MAP_NEW &&
740 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
741 !(flags & EXT4_GET_BLOCKS_ZERO) &&
742 !ext4_is_quota_file(inode) &&
743 ext4_should_order_data(inode)) {
745 (loff_t)map->m_lblk << inode->i_blkbits;
746 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
748 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
749 ret = ext4_jbd2_inode_add_wait(handle, inode,
752 ret = ext4_jbd2_inode_add_write(handle, inode,
762 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
763 * we have to be careful as someone else may be manipulating b_state as well.
765 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
767 unsigned long old_state;
768 unsigned long new_state;
770 flags &= EXT4_MAP_FLAGS;
772 /* Dummy buffer_head? Set non-atomically. */
774 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
778 * Someone else may be modifying b_state. Be careful! This is ugly but
779 * once we get rid of using bh as a container for mapping information
780 * to pass to / from get_block functions, this can go away.
783 old_state = READ_ONCE(bh->b_state);
784 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
786 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
789 static int _ext4_get_block(struct inode *inode, sector_t iblock,
790 struct buffer_head *bh, int flags)
792 struct ext4_map_blocks map;
795 if (ext4_has_inline_data(inode))
799 map.m_len = bh->b_size >> inode->i_blkbits;
801 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
804 map_bh(bh, inode->i_sb, map.m_pblk);
805 ext4_update_bh_state(bh, map.m_flags);
806 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
808 } else if (ret == 0) {
809 /* hole case, need to fill in bh->b_size */
810 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
815 int ext4_get_block(struct inode *inode, sector_t iblock,
816 struct buffer_head *bh, int create)
818 return _ext4_get_block(inode, iblock, bh,
819 create ? EXT4_GET_BLOCKS_CREATE : 0);
823 * Get block function used when preparing for buffered write if we require
824 * creating an unwritten extent if blocks haven't been allocated. The extent
825 * will be converted to written after the IO is complete.
827 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
828 struct buffer_head *bh_result, int create)
830 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
831 inode->i_ino, create);
832 return _ext4_get_block(inode, iblock, bh_result,
833 EXT4_GET_BLOCKS_IO_CREATE_EXT);
836 /* Maximum number of blocks we map for direct IO at once. */
837 #define DIO_MAX_BLOCKS 4096
840 * Get blocks function for the cases that need to start a transaction -
841 * generally difference cases of direct IO and DAX IO. It also handles retries
844 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
845 struct buffer_head *bh_result, int flags)
852 /* Trim mapping request to maximum we can map at once for DIO */
853 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
854 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
855 dio_credits = ext4_chunk_trans_blocks(inode,
856 bh_result->b_size >> inode->i_blkbits);
858 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
860 return PTR_ERR(handle);
862 ret = _ext4_get_block(inode, iblock, bh_result, flags);
863 ext4_journal_stop(handle);
865 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
870 /* Get block function for DIO reads and writes to inodes without extents */
871 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
872 struct buffer_head *bh, int create)
874 /* We don't expect handle for direct IO */
875 WARN_ON_ONCE(ext4_journal_current_handle());
878 return _ext4_get_block(inode, iblock, bh, 0);
879 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
883 * Get block function for AIO DIO writes when we create unwritten extent if
884 * blocks are not allocated yet. The extent will be converted to written
885 * after IO is complete.
887 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
888 sector_t iblock, struct buffer_head *bh_result, int create)
892 /* We don't expect handle for direct IO */
893 WARN_ON_ONCE(ext4_journal_current_handle());
895 ret = ext4_get_block_trans(inode, iblock, bh_result,
896 EXT4_GET_BLOCKS_IO_CREATE_EXT);
899 * When doing DIO using unwritten extents, we need io_end to convert
900 * unwritten extents to written on IO completion. We allocate io_end
901 * once we spot unwritten extent and store it in b_private. Generic
902 * DIO code keeps b_private set and furthermore passes the value to
903 * our completion callback in 'private' argument.
905 if (!ret && buffer_unwritten(bh_result)) {
906 if (!bh_result->b_private) {
907 ext4_io_end_t *io_end;
909 io_end = ext4_init_io_end(inode, GFP_KERNEL);
912 bh_result->b_private = io_end;
913 ext4_set_io_unwritten_flag(inode, io_end);
915 set_buffer_defer_completion(bh_result);
922 * Get block function for non-AIO DIO writes when we create unwritten extent if
923 * blocks are not allocated yet. The extent will be converted to written
924 * after IO is complete by ext4_direct_IO_write().
926 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
927 sector_t iblock, struct buffer_head *bh_result, int create)
931 /* We don't expect handle for direct IO */
932 WARN_ON_ONCE(ext4_journal_current_handle());
934 ret = ext4_get_block_trans(inode, iblock, bh_result,
935 EXT4_GET_BLOCKS_IO_CREATE_EXT);
938 * Mark inode as having pending DIO writes to unwritten extents.
939 * ext4_direct_IO_write() checks this flag and converts extents to
942 if (!ret && buffer_unwritten(bh_result))
943 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
948 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
949 struct buffer_head *bh_result, int create)
953 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
954 inode->i_ino, create);
955 /* We don't expect handle for direct IO */
956 WARN_ON_ONCE(ext4_journal_current_handle());
958 ret = _ext4_get_block(inode, iblock, bh_result, 0);
960 * Blocks should have been preallocated! ext4_file_write_iter() checks
963 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
970 * `handle' can be NULL if create is zero
972 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
973 ext4_lblk_t block, int map_flags)
975 struct ext4_map_blocks map;
976 struct buffer_head *bh;
977 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
980 J_ASSERT(handle != NULL || create == 0);
984 err = ext4_map_blocks(handle, inode, &map, map_flags);
987 return create ? ERR_PTR(-ENOSPC) : NULL;
991 bh = sb_getblk(inode->i_sb, map.m_pblk);
993 return ERR_PTR(-ENOMEM);
994 if (map.m_flags & EXT4_MAP_NEW) {
995 J_ASSERT(create != 0);
996 J_ASSERT(handle != NULL);
999 * Now that we do not always journal data, we should
1000 * keep in mind whether this should always journal the
1001 * new buffer as metadata. For now, regular file
1002 * writes use ext4_get_block instead, so it's not a
1006 BUFFER_TRACE(bh, "call get_create_access");
1007 err = ext4_journal_get_create_access(handle, bh);
1008 if (unlikely(err)) {
1012 if (!buffer_uptodate(bh)) {
1013 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1014 set_buffer_uptodate(bh);
1017 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1018 err = ext4_handle_dirty_metadata(handle, inode, bh);
1022 BUFFER_TRACE(bh, "not a new buffer");
1026 return ERR_PTR(err);
1029 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1030 ext4_lblk_t block, int map_flags)
1032 struct buffer_head *bh;
1034 bh = ext4_getblk(handle, inode, block, map_flags);
1037 if (!bh || ext4_buffer_uptodate(bh))
1039 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1041 if (buffer_uptodate(bh))
1044 return ERR_PTR(-EIO);
1047 /* Read a contiguous batch of blocks. */
1048 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1049 bool wait, struct buffer_head **bhs)
1053 for (i = 0; i < bh_count; i++) {
1054 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1055 if (IS_ERR(bhs[i])) {
1056 err = PTR_ERR(bhs[i]);
1062 for (i = 0; i < bh_count; i++)
1063 /* Note that NULL bhs[i] is valid because of holes. */
1064 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
1065 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1071 for (i = 0; i < bh_count; i++)
1073 wait_on_buffer(bhs[i]);
1075 for (i = 0; i < bh_count; i++) {
1076 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1084 for (i = 0; i < bh_count; i++) {
1091 int ext4_walk_page_buffers(handle_t *handle,
1092 struct buffer_head *head,
1096 int (*fn)(handle_t *handle,
1097 struct buffer_head *bh))
1099 struct buffer_head *bh;
1100 unsigned block_start, block_end;
1101 unsigned blocksize = head->b_size;
1103 struct buffer_head *next;
1105 for (bh = head, block_start = 0;
1106 ret == 0 && (bh != head || !block_start);
1107 block_start = block_end, bh = next) {
1108 next = bh->b_this_page;
1109 block_end = block_start + blocksize;
1110 if (block_end <= from || block_start >= to) {
1111 if (partial && !buffer_uptodate(bh))
1115 err = (*fn)(handle, bh);
1123 * To preserve ordering, it is essential that the hole instantiation and
1124 * the data write be encapsulated in a single transaction. We cannot
1125 * close off a transaction and start a new one between the ext4_get_block()
1126 * and the commit_write(). So doing the jbd2_journal_start at the start of
1127 * prepare_write() is the right place.
1129 * Also, this function can nest inside ext4_writepage(). In that case, we
1130 * *know* that ext4_writepage() has generated enough buffer credits to do the
1131 * whole page. So we won't block on the journal in that case, which is good,
1132 * because the caller may be PF_MEMALLOC.
1134 * By accident, ext4 can be reentered when a transaction is open via
1135 * quota file writes. If we were to commit the transaction while thus
1136 * reentered, there can be a deadlock - we would be holding a quota
1137 * lock, and the commit would never complete if another thread had a
1138 * transaction open and was blocking on the quota lock - a ranking
1141 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1142 * will _not_ run commit under these circumstances because handle->h_ref
1143 * is elevated. We'll still have enough credits for the tiny quotafile
1146 int do_journal_get_write_access(handle_t *handle,
1147 struct buffer_head *bh)
1149 int dirty = buffer_dirty(bh);
1152 if (!buffer_mapped(bh) || buffer_freed(bh))
1155 * __block_write_begin() could have dirtied some buffers. Clean
1156 * the dirty bit as jbd2_journal_get_write_access() could complain
1157 * otherwise about fs integrity issues. Setting of the dirty bit
1158 * by __block_write_begin() isn't a real problem here as we clear
1159 * the bit before releasing a page lock and thus writeback cannot
1160 * ever write the buffer.
1163 clear_buffer_dirty(bh);
1164 BUFFER_TRACE(bh, "get write access");
1165 ret = ext4_journal_get_write_access(handle, bh);
1167 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1171 #ifdef CONFIG_FS_ENCRYPTION
1172 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1173 get_block_t *get_block)
1175 unsigned from = pos & (PAGE_SIZE - 1);
1176 unsigned to = from + len;
1177 struct inode *inode = page->mapping->host;
1178 unsigned block_start, block_end;
1181 unsigned blocksize = inode->i_sb->s_blocksize;
1183 struct buffer_head *bh, *head, *wait[2];
1187 BUG_ON(!PageLocked(page));
1188 BUG_ON(from > PAGE_SIZE);
1189 BUG_ON(to > PAGE_SIZE);
1192 if (!page_has_buffers(page))
1193 create_empty_buffers(page, blocksize, 0);
1194 head = page_buffers(page);
1195 bbits = ilog2(blocksize);
1196 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1198 for (bh = head, block_start = 0; bh != head || !block_start;
1199 block++, block_start = block_end, bh = bh->b_this_page) {
1200 block_end = block_start + blocksize;
1201 if (block_end <= from || block_start >= to) {
1202 if (PageUptodate(page)) {
1203 if (!buffer_uptodate(bh))
1204 set_buffer_uptodate(bh);
1209 clear_buffer_new(bh);
1210 if (!buffer_mapped(bh)) {
1211 WARN_ON(bh->b_size != blocksize);
1212 err = get_block(inode, block, bh, 1);
1215 if (buffer_new(bh)) {
1216 if (PageUptodate(page)) {
1217 clear_buffer_new(bh);
1218 set_buffer_uptodate(bh);
1219 mark_buffer_dirty(bh);
1222 if (block_end > to || block_start < from)
1223 zero_user_segments(page, to, block_end,
1228 if (PageUptodate(page)) {
1229 if (!buffer_uptodate(bh))
1230 set_buffer_uptodate(bh);
1233 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1234 !buffer_unwritten(bh) &&
1235 (block_start < from || block_end > to)) {
1236 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1237 wait[nr_wait++] = bh;
1241 * If we issued read requests, let them complete.
1243 for (i = 0; i < nr_wait; i++) {
1244 wait_on_buffer(wait[i]);
1245 if (!buffer_uptodate(wait[i]))
1248 if (unlikely(err)) {
1249 page_zero_new_buffers(page, from, to);
1250 } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
1251 for (i = 0; i < nr_wait; i++) {
1254 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1255 bh_offset(wait[i]));
1257 clear_buffer_uptodate(wait[i]);
1267 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1268 loff_t pos, unsigned len, unsigned flags,
1269 struct page **pagep, void **fsdata)
1271 struct inode *inode = mapping->host;
1272 int ret, needed_blocks;
1279 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1282 trace_ext4_write_begin(inode, pos, len, flags);
1284 * Reserve one block more for addition to orphan list in case
1285 * we allocate blocks but write fails for some reason
1287 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1288 index = pos >> PAGE_SHIFT;
1289 from = pos & (PAGE_SIZE - 1);
1292 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1293 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1302 * grab_cache_page_write_begin() can take a long time if the
1303 * system is thrashing due to memory pressure, or if the page
1304 * is being written back. So grab it first before we start
1305 * the transaction handle. This also allows us to allocate
1306 * the page (if needed) without using GFP_NOFS.
1309 page = grab_cache_page_write_begin(mapping, index, flags);
1313 * The same as page allocation, we prealloc buffer heads before
1314 * starting the handle.
1316 if (!page_has_buffers(page))
1317 create_empty_buffers(page, inode->i_sb->s_blocksize, 0);
1322 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1323 if (IS_ERR(handle)) {
1325 return PTR_ERR(handle);
1329 if (page->mapping != mapping) {
1330 /* The page got truncated from under us */
1333 ext4_journal_stop(handle);
1336 /* In case writeback began while the page was unlocked */
1337 wait_for_stable_page(page);
1339 #ifdef CONFIG_FS_ENCRYPTION
1340 if (ext4_should_dioread_nolock(inode))
1341 ret = ext4_block_write_begin(page, pos, len,
1342 ext4_get_block_unwritten);
1344 ret = ext4_block_write_begin(page, pos, len,
1347 if (ext4_should_dioread_nolock(inode))
1348 ret = __block_write_begin(page, pos, len,
1349 ext4_get_block_unwritten);
1351 ret = __block_write_begin(page, pos, len, ext4_get_block);
1353 if (!ret && ext4_should_journal_data(inode)) {
1354 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1356 do_journal_get_write_access);
1360 bool extended = (pos + len > inode->i_size) &&
1361 !ext4_verity_in_progress(inode);
1365 * __block_write_begin may have instantiated a few blocks
1366 * outside i_size. Trim these off again. Don't need
1367 * i_size_read because we hold i_mutex.
1369 * Add inode to orphan list in case we crash before
1372 if (extended && ext4_can_truncate(inode))
1373 ext4_orphan_add(handle, inode);
1375 ext4_journal_stop(handle);
1377 ext4_truncate_failed_write(inode);
1379 * If truncate failed early the inode might
1380 * still be on the orphan list; we need to
1381 * make sure the inode is removed from the
1382 * orphan list in that case.
1385 ext4_orphan_del(NULL, inode);
1388 if (ret == -ENOSPC &&
1389 ext4_should_retry_alloc(inode->i_sb, &retries))
1398 /* For write_end() in data=journal mode */
1399 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1402 if (!buffer_mapped(bh) || buffer_freed(bh))
1404 set_buffer_uptodate(bh);
1405 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1406 clear_buffer_meta(bh);
1407 clear_buffer_prio(bh);
1412 * We need to pick up the new inode size which generic_commit_write gave us
1413 * `file' can be NULL - eg, when called from page_symlink().
1415 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1416 * buffers are managed internally.
1418 static int ext4_write_end(struct file *file,
1419 struct address_space *mapping,
1420 loff_t pos, unsigned len, unsigned copied,
1421 struct page *page, void *fsdata)
1423 handle_t *handle = ext4_journal_current_handle();
1424 struct inode *inode = mapping->host;
1425 loff_t old_size = inode->i_size;
1427 int i_size_changed = 0;
1428 int inline_data = ext4_has_inline_data(inode);
1429 bool verity = ext4_verity_in_progress(inode);
1431 trace_ext4_write_end(inode, pos, len, copied);
1433 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1434 ret = ext4_write_inline_data_end(inode, pos, len,
1444 copied = block_write_end(file, mapping, pos,
1445 len, copied, page, fsdata);
1447 * it's important to update i_size while still holding page lock:
1448 * page writeout could otherwise come in and zero beyond i_size.
1450 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1451 * blocks are being written past EOF, so skip the i_size update.
1454 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1458 if (old_size < pos && !verity)
1459 pagecache_isize_extended(inode, old_size, pos);
1461 * Don't mark the inode dirty under page lock. First, it unnecessarily
1462 * makes the holding time of page lock longer. Second, it forces lock
1463 * ordering of page lock and transaction start for journaling
1466 if (i_size_changed || inline_data)
1467 ext4_mark_inode_dirty(handle, inode);
1470 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1471 /* if we have allocated more blocks and copied
1472 * less. We will have blocks allocated outside
1473 * inode->i_size. So truncate them
1475 ext4_orphan_add(handle, inode);
1477 ret2 = ext4_journal_stop(handle);
1481 if (pos + len > inode->i_size && !verity) {
1482 ext4_truncate_failed_write(inode);
1484 * If truncate failed early the inode might still be
1485 * on the orphan list; we need to make sure the inode
1486 * is removed from the orphan list in that case.
1489 ext4_orphan_del(NULL, inode);
1492 return ret ? ret : copied;
1496 * This is a private version of page_zero_new_buffers() which doesn't
1497 * set the buffer to be dirty, since in data=journalled mode we need
1498 * to call ext4_handle_dirty_metadata() instead.
1500 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1502 unsigned from, unsigned to)
1504 unsigned int block_start = 0, block_end;
1505 struct buffer_head *head, *bh;
1507 bh = head = page_buffers(page);
1509 block_end = block_start + bh->b_size;
1510 if (buffer_new(bh)) {
1511 if (block_end > from && block_start < to) {
1512 if (!PageUptodate(page)) {
1513 unsigned start, size;
1515 start = max(from, block_start);
1516 size = min(to, block_end) - start;
1518 zero_user(page, start, size);
1519 write_end_fn(handle, bh);
1521 clear_buffer_new(bh);
1524 block_start = block_end;
1525 bh = bh->b_this_page;
1526 } while (bh != head);
1529 static int ext4_journalled_write_end(struct file *file,
1530 struct address_space *mapping,
1531 loff_t pos, unsigned len, unsigned copied,
1532 struct page *page, void *fsdata)
1534 handle_t *handle = ext4_journal_current_handle();
1535 struct inode *inode = mapping->host;
1536 loff_t old_size = inode->i_size;
1540 int size_changed = 0;
1541 int inline_data = ext4_has_inline_data(inode);
1542 bool verity = ext4_verity_in_progress(inode);
1544 trace_ext4_journalled_write_end(inode, pos, len, copied);
1545 from = pos & (PAGE_SIZE - 1);
1548 BUG_ON(!ext4_handle_valid(handle));
1551 ret = ext4_write_inline_data_end(inode, pos, len,
1560 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1562 ext4_journalled_zero_new_buffers(handle, page, from, to);
1564 if (unlikely(copied < len))
1565 ext4_journalled_zero_new_buffers(handle, page,
1567 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1568 from + copied, &partial,
1571 SetPageUptodate(page);
1574 size_changed = ext4_update_inode_size(inode, pos + copied);
1575 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1576 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1580 if (old_size < pos && !verity)
1581 pagecache_isize_extended(inode, old_size, pos);
1583 if (size_changed || inline_data) {
1584 ret2 = ext4_mark_inode_dirty(handle, inode);
1590 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1591 /* if we have allocated more blocks and copied
1592 * less. We will have blocks allocated outside
1593 * inode->i_size. So truncate them
1595 ext4_orphan_add(handle, inode);
1597 ret2 = ext4_journal_stop(handle);
1600 if (pos + len > inode->i_size && !verity) {
1601 ext4_truncate_failed_write(inode);
1603 * If truncate failed early the inode might still be
1604 * on the orphan list; we need to make sure the inode
1605 * is removed from the orphan list in that case.
1608 ext4_orphan_del(NULL, inode);
1611 return ret ? ret : copied;
1615 * Reserve space for a single cluster
1617 static int ext4_da_reserve_space(struct inode *inode)
1619 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1620 struct ext4_inode_info *ei = EXT4_I(inode);
1624 * We will charge metadata quota at writeout time; this saves
1625 * us from metadata over-estimation, though we may go over by
1626 * a small amount in the end. Here we just reserve for data.
1628 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1632 spin_lock(&ei->i_block_reservation_lock);
1633 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1634 spin_unlock(&ei->i_block_reservation_lock);
1635 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1638 ei->i_reserved_data_blocks++;
1639 trace_ext4_da_reserve_space(inode);
1640 spin_unlock(&ei->i_block_reservation_lock);
1642 return 0; /* success */
1645 void ext4_da_release_space(struct inode *inode, int to_free)
1647 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1648 struct ext4_inode_info *ei = EXT4_I(inode);
1651 return; /* Nothing to release, exit */
1653 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1655 trace_ext4_da_release_space(inode, to_free);
1656 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1658 * if there aren't enough reserved blocks, then the
1659 * counter is messed up somewhere. Since this
1660 * function is called from invalidate page, it's
1661 * harmless to return without any action.
1663 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1664 "ino %lu, to_free %d with only %d reserved "
1665 "data blocks", inode->i_ino, to_free,
1666 ei->i_reserved_data_blocks);
1668 to_free = ei->i_reserved_data_blocks;
1670 ei->i_reserved_data_blocks -= to_free;
1672 /* update fs dirty data blocks counter */
1673 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1675 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1677 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1681 * Delayed allocation stuff
1684 struct mpage_da_data {
1685 struct inode *inode;
1686 struct writeback_control *wbc;
1688 pgoff_t first_page; /* The first page to write */
1689 pgoff_t next_page; /* Current page to examine */
1690 pgoff_t last_page; /* Last page to examine */
1692 * Extent to map - this can be after first_page because that can be
1693 * fully mapped. We somewhat abuse m_flags to store whether the extent
1694 * is delalloc or unwritten.
1696 struct ext4_map_blocks map;
1697 struct ext4_io_submit io_submit; /* IO submission data */
1698 unsigned int do_map:1;
1701 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1706 struct pagevec pvec;
1707 struct inode *inode = mpd->inode;
1708 struct address_space *mapping = inode->i_mapping;
1710 /* This is necessary when next_page == 0. */
1711 if (mpd->first_page >= mpd->next_page)
1714 index = mpd->first_page;
1715 end = mpd->next_page - 1;
1717 ext4_lblk_t start, last;
1718 start = index << (PAGE_SHIFT - inode->i_blkbits);
1719 last = end << (PAGE_SHIFT - inode->i_blkbits);
1722 * avoid racing with extent status tree scans made by
1723 * ext4_insert_delayed_block()
1725 down_write(&EXT4_I(inode)->i_data_sem);
1726 ext4_es_remove_extent(inode, start, last - start + 1);
1727 up_write(&EXT4_I(inode)->i_data_sem);
1730 pagevec_init(&pvec);
1731 while (index <= end) {
1732 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1735 for (i = 0; i < nr_pages; i++) {
1736 struct page *page = pvec.pages[i];
1738 BUG_ON(!PageLocked(page));
1739 BUG_ON(PageWriteback(page));
1741 if (page_mapped(page))
1742 clear_page_dirty_for_io(page);
1743 block_invalidatepage(page, 0, PAGE_SIZE);
1744 ClearPageUptodate(page);
1748 pagevec_release(&pvec);
1752 static void ext4_print_free_blocks(struct inode *inode)
1754 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1755 struct super_block *sb = inode->i_sb;
1756 struct ext4_inode_info *ei = EXT4_I(inode);
1758 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1759 EXT4_C2B(EXT4_SB(inode->i_sb),
1760 ext4_count_free_clusters(sb)));
1761 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1762 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1763 (long long) EXT4_C2B(EXT4_SB(sb),
1764 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1765 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1766 (long long) EXT4_C2B(EXT4_SB(sb),
1767 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1768 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1769 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1770 ei->i_reserved_data_blocks);
1774 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1776 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1780 * ext4_insert_delayed_block - adds a delayed block to the extents status
1781 * tree, incrementing the reserved cluster/block
1782 * count or making a pending reservation
1785 * @inode - file containing the newly added block
1786 * @lblk - logical block to be added
1788 * Returns 0 on success, negative error code on failure.
1790 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1792 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1794 bool allocated = false;
1795 bool reserved = false;
1798 * If the cluster containing lblk is shared with a delayed,
1799 * written, or unwritten extent in a bigalloc file system, it's
1800 * already been accounted for and does not need to be reserved.
1801 * A pending reservation must be made for the cluster if it's
1802 * shared with a written or unwritten extent and doesn't already
1803 * have one. Written and unwritten extents can be purged from the
1804 * extents status tree if the system is under memory pressure, so
1805 * it's necessary to examine the extent tree if a search of the
1806 * extents status tree doesn't get a match.
1808 if (sbi->s_cluster_ratio == 1) {
1809 ret = ext4_da_reserve_space(inode);
1810 if (ret != 0) /* ENOSPC */
1813 } else { /* bigalloc */
1814 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1815 if (!ext4_es_scan_clu(inode,
1816 &ext4_es_is_mapped, lblk)) {
1817 ret = ext4_clu_mapped(inode,
1818 EXT4_B2C(sbi, lblk));
1822 ret = ext4_da_reserve_space(inode);
1823 if (ret != 0) /* ENOSPC */
1835 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1836 if (ret && reserved)
1837 ext4_da_release_space(inode, 1);
1844 * This function is grabs code from the very beginning of
1845 * ext4_map_blocks, but assumes that the caller is from delayed write
1846 * time. This function looks up the requested blocks and sets the
1847 * buffer delay bit under the protection of i_data_sem.
1849 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1850 struct ext4_map_blocks *map,
1851 struct buffer_head *bh)
1853 struct extent_status es;
1855 sector_t invalid_block = ~((sector_t) 0xffff);
1856 #ifdef ES_AGGRESSIVE_TEST
1857 struct ext4_map_blocks orig_map;
1859 memcpy(&orig_map, map, sizeof(*map));
1862 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1866 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1867 "logical block %lu\n", inode->i_ino, map->m_len,
1868 (unsigned long) map->m_lblk);
1870 /* Lookup extent status tree firstly */
1871 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1872 if (ext4_es_is_hole(&es)) {
1874 down_read(&EXT4_I(inode)->i_data_sem);
1879 * Delayed extent could be allocated by fallocate.
1880 * So we need to check it.
1882 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1883 map_bh(bh, inode->i_sb, invalid_block);
1885 set_buffer_delay(bh);
1889 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1890 retval = es.es_len - (iblock - es.es_lblk);
1891 if (retval > map->m_len)
1892 retval = map->m_len;
1893 map->m_len = retval;
1894 if (ext4_es_is_written(&es))
1895 map->m_flags |= EXT4_MAP_MAPPED;
1896 else if (ext4_es_is_unwritten(&es))
1897 map->m_flags |= EXT4_MAP_UNWRITTEN;
1901 #ifdef ES_AGGRESSIVE_TEST
1902 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1908 * Try to see if we can get the block without requesting a new
1909 * file system block.
1911 down_read(&EXT4_I(inode)->i_data_sem);
1912 if (ext4_has_inline_data(inode))
1914 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1915 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1917 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1924 * XXX: __block_prepare_write() unmaps passed block,
1928 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1934 map_bh(bh, inode->i_sb, invalid_block);
1936 set_buffer_delay(bh);
1937 } else if (retval > 0) {
1939 unsigned int status;
1941 if (unlikely(retval != map->m_len)) {
1942 ext4_warning(inode->i_sb,
1943 "ES len assertion failed for inode "
1944 "%lu: retval %d != map->m_len %d",
1945 inode->i_ino, retval, map->m_len);
1949 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1950 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1951 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1952 map->m_pblk, status);
1958 up_read((&EXT4_I(inode)->i_data_sem));
1964 * This is a special get_block_t callback which is used by
1965 * ext4_da_write_begin(). It will either return mapped block or
1966 * reserve space for a single block.
1968 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1969 * We also have b_blocknr = -1 and b_bdev initialized properly
1971 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1972 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1973 * initialized properly.
1975 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1976 struct buffer_head *bh, int create)
1978 struct ext4_map_blocks map;
1981 BUG_ON(create == 0);
1982 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1984 map.m_lblk = iblock;
1988 * first, we need to know whether the block is allocated already
1989 * preallocated blocks are unmapped but should treated
1990 * the same as allocated blocks.
1992 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1996 map_bh(bh, inode->i_sb, map.m_pblk);
1997 ext4_update_bh_state(bh, map.m_flags);
1999 if (buffer_unwritten(bh)) {
2000 /* A delayed write to unwritten bh should be marked
2001 * new and mapped. Mapped ensures that we don't do
2002 * get_block multiple times when we write to the same
2003 * offset and new ensures that we do proper zero out
2004 * for partial write.
2007 set_buffer_mapped(bh);
2012 static int bget_one(handle_t *handle, struct buffer_head *bh)
2018 static int bput_one(handle_t *handle, struct buffer_head *bh)
2024 static int __ext4_journalled_writepage(struct page *page,
2027 struct address_space *mapping = page->mapping;
2028 struct inode *inode = mapping->host;
2029 struct buffer_head *page_bufs = NULL;
2030 handle_t *handle = NULL;
2031 int ret = 0, err = 0;
2032 int inline_data = ext4_has_inline_data(inode);
2033 struct buffer_head *inode_bh = NULL;
2035 ClearPageChecked(page);
2038 BUG_ON(page->index != 0);
2039 BUG_ON(len > ext4_get_max_inline_size(inode));
2040 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2041 if (inode_bh == NULL)
2044 page_bufs = page_buffers(page);
2049 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2053 * We need to release the page lock before we start the
2054 * journal, so grab a reference so the page won't disappear
2055 * out from under us.
2060 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2061 ext4_writepage_trans_blocks(inode));
2062 if (IS_ERR(handle)) {
2063 ret = PTR_ERR(handle);
2065 goto out_no_pagelock;
2067 BUG_ON(!ext4_handle_valid(handle));
2071 if (page->mapping != mapping) {
2072 /* The page got truncated from under us */
2073 ext4_journal_stop(handle);
2079 ret = ext4_mark_inode_dirty(handle, inode);
2081 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2082 do_journal_get_write_access);
2084 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2089 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2090 err = ext4_journal_stop(handle);
2094 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2098 if (!inline_data && page_bufs)
2099 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2106 * Note that we don't need to start a transaction unless we're journaling data
2107 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2108 * need to file the inode to the transaction's list in ordered mode because if
2109 * we are writing back data added by write(), the inode is already there and if
2110 * we are writing back data modified via mmap(), no one guarantees in which
2111 * transaction the data will hit the disk. In case we are journaling data, we
2112 * cannot start transaction directly because transaction start ranks above page
2113 * lock so we have to do some magic.
2115 * This function can get called via...
2116 * - ext4_writepages after taking page lock (have journal handle)
2117 * - journal_submit_inode_data_buffers (no journal handle)
2118 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2119 * - grab_page_cache when doing write_begin (have journal handle)
2121 * We don't do any block allocation in this function. If we have page with
2122 * multiple blocks we need to write those buffer_heads that are mapped. This
2123 * is important for mmaped based write. So if we do with blocksize 1K
2124 * truncate(f, 1024);
2125 * a = mmap(f, 0, 4096);
2127 * truncate(f, 4096);
2128 * we have in the page first buffer_head mapped via page_mkwrite call back
2129 * but other buffer_heads would be unmapped but dirty (dirty done via the
2130 * do_wp_page). So writepage should write the first block. If we modify
2131 * the mmap area beyond 1024 we will again get a page_fault and the
2132 * page_mkwrite callback will do the block allocation and mark the
2133 * buffer_heads mapped.
2135 * We redirty the page if we have any buffer_heads that is either delay or
2136 * unwritten in the page.
2138 * We can get recursively called as show below.
2140 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2143 * But since we don't do any block allocation we should not deadlock.
2144 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2146 static int ext4_writepage(struct page *page,
2147 struct writeback_control *wbc)
2152 struct buffer_head *page_bufs = NULL;
2153 struct inode *inode = page->mapping->host;
2154 struct ext4_io_submit io_submit;
2155 bool keep_towrite = false;
2157 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2158 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
2163 trace_ext4_writepage(page);
2164 size = i_size_read(inode);
2165 if (page->index == size >> PAGE_SHIFT &&
2166 !ext4_verity_in_progress(inode))
2167 len = size & ~PAGE_MASK;
2171 /* Should never happen but for bugs in other kernel subsystems */
2172 if (!page_has_buffers(page)) {
2173 ext4_warning_inode(inode,
2174 "page %lu does not have buffers attached", page->index);
2175 ClearPageDirty(page);
2180 page_bufs = page_buffers(page);
2182 * We cannot do block allocation or other extent handling in this
2183 * function. If there are buffers needing that, we have to redirty
2184 * the page. But we may reach here when we do a journal commit via
2185 * journal_submit_inode_data_buffers() and in that case we must write
2186 * allocated buffers to achieve data=ordered mode guarantees.
2188 * Also, if there is only one buffer per page (the fs block
2189 * size == the page size), if one buffer needs block
2190 * allocation or needs to modify the extent tree to clear the
2191 * unwritten flag, we know that the page can't be written at
2192 * all, so we might as well refuse the write immediately.
2193 * Unfortunately if the block size != page size, we can't as
2194 * easily detect this case using ext4_walk_page_buffers(), but
2195 * for the extremely common case, this is an optimization that
2196 * skips a useless round trip through ext4_bio_write_page().
2198 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2199 ext4_bh_delay_or_unwritten)) {
2200 redirty_page_for_writepage(wbc, page);
2201 if ((current->flags & PF_MEMALLOC) ||
2202 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2204 * For memory cleaning there's no point in writing only
2205 * some buffers. So just bail out. Warn if we came here
2206 * from direct reclaim.
2208 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2213 keep_towrite = true;
2216 if (PageChecked(page) && ext4_should_journal_data(inode))
2218 * It's mmapped pagecache. Add buffers and journal it. There
2219 * doesn't seem much point in redirtying the page here.
2221 return __ext4_journalled_writepage(page, len);
2223 ext4_io_submit_init(&io_submit, wbc);
2224 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2225 if (!io_submit.io_end) {
2226 redirty_page_for_writepage(wbc, page);
2230 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2231 ext4_io_submit(&io_submit);
2232 /* Drop io_end reference we got from init */
2233 ext4_put_io_end_defer(io_submit.io_end);
2237 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2243 BUG_ON(page->index != mpd->first_page);
2244 clear_page_dirty_for_io(page);
2246 * We have to be very careful here! Nothing protects writeback path
2247 * against i_size changes and the page can be writeably mapped into
2248 * page tables. So an application can be growing i_size and writing
2249 * data through mmap while writeback runs. clear_page_dirty_for_io()
2250 * write-protects our page in page tables and the page cannot get
2251 * written to again until we release page lock. So only after
2252 * clear_page_dirty_for_io() we are safe to sample i_size for
2253 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2254 * on the barrier provided by TestClearPageDirty in
2255 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2256 * after page tables are updated.
2258 size = i_size_read(mpd->inode);
2259 if (page->index == size >> PAGE_SHIFT &&
2260 !ext4_verity_in_progress(mpd->inode))
2261 len = size & ~PAGE_MASK;
2264 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2266 mpd->wbc->nr_to_write--;
2272 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2275 * mballoc gives us at most this number of blocks...
2276 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2277 * The rest of mballoc seems to handle chunks up to full group size.
2279 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2282 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2284 * @mpd - extent of blocks
2285 * @lblk - logical number of the block in the file
2286 * @bh - buffer head we want to add to the extent
2288 * The function is used to collect contig. blocks in the same state. If the
2289 * buffer doesn't require mapping for writeback and we haven't started the
2290 * extent of buffers to map yet, the function returns 'true' immediately - the
2291 * caller can write the buffer right away. Otherwise the function returns true
2292 * if the block has been added to the extent, false if the block couldn't be
2295 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2296 struct buffer_head *bh)
2298 struct ext4_map_blocks *map = &mpd->map;
2300 /* Buffer that doesn't need mapping for writeback? */
2301 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2302 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2303 /* So far no extent to map => we write the buffer right away */
2304 if (map->m_len == 0)
2309 /* First block in the extent? */
2310 if (map->m_len == 0) {
2311 /* We cannot map unless handle is started... */
2316 map->m_flags = bh->b_state & BH_FLAGS;
2320 /* Don't go larger than mballoc is willing to allocate */
2321 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2324 /* Can we merge the block to our big extent? */
2325 if (lblk == map->m_lblk + map->m_len &&
2326 (bh->b_state & BH_FLAGS) == map->m_flags) {
2334 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2336 * @mpd - extent of blocks for mapping
2337 * @head - the first buffer in the page
2338 * @bh - buffer we should start processing from
2339 * @lblk - logical number of the block in the file corresponding to @bh
2341 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2342 * the page for IO if all buffers in this page were mapped and there's no
2343 * accumulated extent of buffers to map or add buffers in the page to the
2344 * extent of buffers to map. The function returns 1 if the caller can continue
2345 * by processing the next page, 0 if it should stop adding buffers to the
2346 * extent to map because we cannot extend it anymore. It can also return value
2347 * < 0 in case of error during IO submission.
2349 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2350 struct buffer_head *head,
2351 struct buffer_head *bh,
2354 struct inode *inode = mpd->inode;
2356 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2357 >> inode->i_blkbits;
2359 if (ext4_verity_in_progress(inode))
2360 blocks = EXT_MAX_BLOCKS;
2363 BUG_ON(buffer_locked(bh));
2365 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2366 /* Found extent to map? */
2369 /* Buffer needs mapping and handle is not started? */
2372 /* Everything mapped so far and we hit EOF */
2375 } while (lblk++, (bh = bh->b_this_page) != head);
2376 /* So far everything mapped? Submit the page for IO. */
2377 if (mpd->map.m_len == 0) {
2378 err = mpage_submit_page(mpd, head->b_page);
2382 return lblk < blocks;
2386 * mpage_map_buffers - update buffers corresponding to changed extent and
2387 * submit fully mapped pages for IO
2389 * @mpd - description of extent to map, on return next extent to map
2391 * Scan buffers corresponding to changed extent (we expect corresponding pages
2392 * to be already locked) and update buffer state according to new extent state.
2393 * We map delalloc buffers to their physical location, clear unwritten bits,
2394 * and mark buffers as uninit when we perform writes to unwritten extents
2395 * and do extent conversion after IO is finished. If the last page is not fully
2396 * mapped, we update @map to the next extent in the last page that needs
2397 * mapping. Otherwise we submit the page for IO.
2399 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2401 struct pagevec pvec;
2403 struct inode *inode = mpd->inode;
2404 struct buffer_head *head, *bh;
2405 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2411 start = mpd->map.m_lblk >> bpp_bits;
2412 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2413 lblk = start << bpp_bits;
2414 pblock = mpd->map.m_pblk;
2416 pagevec_init(&pvec);
2417 while (start <= end) {
2418 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2422 for (i = 0; i < nr_pages; i++) {
2423 struct page *page = pvec.pages[i];
2425 bh = head = page_buffers(page);
2427 if (lblk < mpd->map.m_lblk)
2429 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2431 * Buffer after end of mapped extent.
2432 * Find next buffer in the page to map.
2435 mpd->map.m_flags = 0;
2437 * FIXME: If dioread_nolock supports
2438 * blocksize < pagesize, we need to make
2439 * sure we add size mapped so far to
2440 * io_end->size as the following call
2441 * can submit the page for IO.
2443 err = mpage_process_page_bufs(mpd, head,
2445 pagevec_release(&pvec);
2450 if (buffer_delay(bh)) {
2451 clear_buffer_delay(bh);
2452 bh->b_blocknr = pblock++;
2454 clear_buffer_unwritten(bh);
2455 } while (lblk++, (bh = bh->b_this_page) != head);
2458 * FIXME: This is going to break if dioread_nolock
2459 * supports blocksize < pagesize as we will try to
2460 * convert potentially unmapped parts of inode.
2462 mpd->io_submit.io_end->size += PAGE_SIZE;
2463 /* Page fully mapped - let IO run! */
2464 err = mpage_submit_page(mpd, page);
2466 pagevec_release(&pvec);
2470 pagevec_release(&pvec);
2472 /* Extent fully mapped and matches with page boundary. We are done. */
2474 mpd->map.m_flags = 0;
2478 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2480 struct inode *inode = mpd->inode;
2481 struct ext4_map_blocks *map = &mpd->map;
2482 int get_blocks_flags;
2483 int err, dioread_nolock;
2485 trace_ext4_da_write_pages_extent(inode, map);
2487 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2488 * to convert an unwritten extent to be initialized (in the case
2489 * where we have written into one or more preallocated blocks). It is
2490 * possible that we're going to need more metadata blocks than
2491 * previously reserved. However we must not fail because we're in
2492 * writeback and there is nothing we can do about it so it might result
2493 * in data loss. So use reserved blocks to allocate metadata if
2496 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2497 * the blocks in question are delalloc blocks. This indicates
2498 * that the blocks and quotas has already been checked when
2499 * the data was copied into the page cache.
2501 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2502 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2503 EXT4_GET_BLOCKS_IO_SUBMIT;
2504 dioread_nolock = ext4_should_dioread_nolock(inode);
2506 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2507 if (map->m_flags & (1 << BH_Delay))
2508 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2510 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2513 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2514 if (!mpd->io_submit.io_end->handle &&
2515 ext4_handle_valid(handle)) {
2516 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2517 handle->h_rsv_handle = NULL;
2519 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2522 BUG_ON(map->m_len == 0);
2527 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2528 * mpd->len and submit pages underlying it for IO
2530 * @handle - handle for journal operations
2531 * @mpd - extent to map
2532 * @give_up_on_write - we set this to true iff there is a fatal error and there
2533 * is no hope of writing the data. The caller should discard
2534 * dirty pages to avoid infinite loops.
2536 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2537 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2538 * them to initialized or split the described range from larger unwritten
2539 * extent. Note that we need not map all the described range since allocation
2540 * can return less blocks or the range is covered by more unwritten extents. We
2541 * cannot map more because we are limited by reserved transaction credits. On
2542 * the other hand we always make sure that the last touched page is fully
2543 * mapped so that it can be written out (and thus forward progress is
2544 * guaranteed). After mapping we submit all mapped pages for IO.
2546 static int mpage_map_and_submit_extent(handle_t *handle,
2547 struct mpage_da_data *mpd,
2548 bool *give_up_on_write)
2550 struct inode *inode = mpd->inode;
2551 struct ext4_map_blocks *map = &mpd->map;
2556 mpd->io_submit.io_end->offset =
2557 ((loff_t)map->m_lblk) << inode->i_blkbits;
2559 err = mpage_map_one_extent(handle, mpd);
2561 struct super_block *sb = inode->i_sb;
2563 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2564 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2565 goto invalidate_dirty_pages;
2567 * Let the uper layers retry transient errors.
2568 * In the case of ENOSPC, if ext4_count_free_blocks()
2569 * is non-zero, a commit should free up blocks.
2571 if ((err == -ENOMEM) ||
2572 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2574 goto update_disksize;
2577 ext4_msg(sb, KERN_CRIT,
2578 "Delayed block allocation failed for "
2579 "inode %lu at logical offset %llu with"
2580 " max blocks %u with error %d",
2582 (unsigned long long)map->m_lblk,
2583 (unsigned)map->m_len, -err);
2584 ext4_msg(sb, KERN_CRIT,
2585 "This should not happen!! Data will "
2588 ext4_print_free_blocks(inode);
2589 invalidate_dirty_pages:
2590 *give_up_on_write = true;
2595 * Update buffer state, submit mapped pages, and get us new
2598 err = mpage_map_and_submit_buffers(mpd);
2600 goto update_disksize;
2601 } while (map->m_len);
2605 * Update on-disk size after IO is submitted. Races with
2606 * truncate are avoided by checking i_size under i_data_sem.
2608 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2609 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2613 down_write(&EXT4_I(inode)->i_data_sem);
2614 i_size = i_size_read(inode);
2615 if (disksize > i_size)
2617 if (disksize > EXT4_I(inode)->i_disksize)
2618 EXT4_I(inode)->i_disksize = disksize;
2619 up_write(&EXT4_I(inode)->i_data_sem);
2620 err2 = ext4_mark_inode_dirty(handle, inode);
2622 ext4_error(inode->i_sb,
2623 "Failed to mark inode %lu dirty",
2632 * Calculate the total number of credits to reserve for one writepages
2633 * iteration. This is called from ext4_writepages(). We map an extent of
2634 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2635 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2636 * bpp - 1 blocks in bpp different extents.
2638 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2640 int bpp = ext4_journal_blocks_per_page(inode);
2642 return ext4_meta_trans_blocks(inode,
2643 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2647 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2648 * and underlying extent to map
2650 * @mpd - where to look for pages
2652 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2653 * IO immediately. When we find a page which isn't mapped we start accumulating
2654 * extent of buffers underlying these pages that needs mapping (formed by
2655 * either delayed or unwritten buffers). We also lock the pages containing
2656 * these buffers. The extent found is returned in @mpd structure (starting at
2657 * mpd->lblk with length mpd->len blocks).
2659 * Note that this function can attach bios to one io_end structure which are
2660 * neither logically nor physically contiguous. Although it may seem as an
2661 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2662 * case as we need to track IO to all buffers underlying a page in one io_end.
2664 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2666 struct address_space *mapping = mpd->inode->i_mapping;
2667 struct pagevec pvec;
2668 unsigned int nr_pages;
2669 long left = mpd->wbc->nr_to_write;
2670 pgoff_t index = mpd->first_page;
2671 pgoff_t end = mpd->last_page;
2674 int blkbits = mpd->inode->i_blkbits;
2676 struct buffer_head *head;
2678 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2679 tag = PAGECACHE_TAG_TOWRITE;
2681 tag = PAGECACHE_TAG_DIRTY;
2683 pagevec_init(&pvec);
2685 mpd->next_page = index;
2686 while (index <= end) {
2687 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2692 for (i = 0; i < nr_pages; i++) {
2693 struct page *page = pvec.pages[i];
2696 * Accumulated enough dirty pages? This doesn't apply
2697 * to WB_SYNC_ALL mode. For integrity sync we have to
2698 * keep going because someone may be concurrently
2699 * dirtying pages, and we might have synced a lot of
2700 * newly appeared dirty pages, but have not synced all
2701 * of the old dirty pages.
2703 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2706 /* If we can't merge this page, we are done. */
2707 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2712 * If the page is no longer dirty, or its mapping no
2713 * longer corresponds to inode we are writing (which
2714 * means it has been truncated or invalidated), or the
2715 * page is already under writeback and we are not doing
2716 * a data integrity writeback, skip the page
2718 if (!PageDirty(page) ||
2719 (PageWriteback(page) &&
2720 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2721 unlikely(page->mapping != mapping)) {
2726 wait_on_page_writeback(page);
2727 BUG_ON(PageWriteback(page));
2730 * Should never happen but for buggy code in
2731 * other subsystems that call
2732 * set_page_dirty() without properly warning
2733 * the file system first. See [1] for more
2736 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2738 if (!page_has_buffers(page)) {
2739 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2740 ClearPageDirty(page);
2745 if (mpd->map.m_len == 0)
2746 mpd->first_page = page->index;
2747 mpd->next_page = page->index + 1;
2748 /* Add all dirty buffers to mpd */
2749 lblk = ((ext4_lblk_t)page->index) <<
2750 (PAGE_SHIFT - blkbits);
2751 head = page_buffers(page);
2752 err = mpage_process_page_bufs(mpd, head, head, lblk);
2758 pagevec_release(&pvec);
2763 pagevec_release(&pvec);
2767 static int ext4_writepages(struct address_space *mapping,
2768 struct writeback_control *wbc)
2770 pgoff_t writeback_index = 0;
2771 long nr_to_write = wbc->nr_to_write;
2772 int range_whole = 0;
2774 handle_t *handle = NULL;
2775 struct mpage_da_data mpd;
2776 struct inode *inode = mapping->host;
2777 int needed_blocks, rsv_blocks = 0, ret = 0;
2778 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2780 struct blk_plug plug;
2781 bool give_up_on_write = false;
2783 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2786 percpu_down_read(&sbi->s_writepages_rwsem);
2787 trace_ext4_writepages(inode, wbc);
2790 * No pages to write? This is mainly a kludge to avoid starting
2791 * a transaction for special inodes like journal inode on last iput()
2792 * because that could violate lock ordering on umount
2794 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2795 goto out_writepages;
2797 if (ext4_should_journal_data(inode)) {
2798 ret = generic_writepages(mapping, wbc);
2799 goto out_writepages;
2803 * If the filesystem has aborted, it is read-only, so return
2804 * right away instead of dumping stack traces later on that
2805 * will obscure the real source of the problem. We test
2806 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2807 * the latter could be true if the filesystem is mounted
2808 * read-only, and in that case, ext4_writepages should
2809 * *never* be called, so if that ever happens, we would want
2812 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2813 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2815 goto out_writepages;
2819 * If we have inline data and arrive here, it means that
2820 * we will soon create the block for the 1st page, so
2821 * we'd better clear the inline data here.
2823 if (ext4_has_inline_data(inode)) {
2824 /* Just inode will be modified... */
2825 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2826 if (IS_ERR(handle)) {
2827 ret = PTR_ERR(handle);
2828 goto out_writepages;
2830 BUG_ON(ext4_test_inode_state(inode,
2831 EXT4_STATE_MAY_INLINE_DATA));
2832 ext4_destroy_inline_data(handle, inode);
2833 ext4_journal_stop(handle);
2836 if (ext4_should_dioread_nolock(inode)) {
2838 * We may need to convert up to one extent per block in
2839 * the page and we may dirty the inode.
2841 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2842 PAGE_SIZE >> inode->i_blkbits);
2845 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2848 if (wbc->range_cyclic) {
2849 writeback_index = mapping->writeback_index;
2850 if (writeback_index)
2852 mpd.first_page = writeback_index;
2855 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2856 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2861 ext4_io_submit_init(&mpd.io_submit, wbc);
2863 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2864 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2866 blk_start_plug(&plug);
2869 * First writeback pages that don't need mapping - we can avoid
2870 * starting a transaction unnecessarily and also avoid being blocked
2871 * in the block layer on device congestion while having transaction
2875 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2876 if (!mpd.io_submit.io_end) {
2880 ret = mpage_prepare_extent_to_map(&mpd);
2881 /* Unlock pages we didn't use */
2882 mpage_release_unused_pages(&mpd, false);
2883 /* Submit prepared bio */
2884 ext4_io_submit(&mpd.io_submit);
2885 ext4_put_io_end_defer(mpd.io_submit.io_end);
2886 mpd.io_submit.io_end = NULL;
2890 while (!done && mpd.first_page <= mpd.last_page) {
2891 /* For each extent of pages we use new io_end */
2892 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2893 if (!mpd.io_submit.io_end) {
2899 * We have two constraints: We find one extent to map and we
2900 * must always write out whole page (makes a difference when
2901 * blocksize < pagesize) so that we don't block on IO when we
2902 * try to write out the rest of the page. Journalled mode is
2903 * not supported by delalloc.
2905 BUG_ON(ext4_should_journal_data(inode));
2906 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2908 /* start a new transaction */
2909 handle = ext4_journal_start_with_reserve(inode,
2910 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2911 if (IS_ERR(handle)) {
2912 ret = PTR_ERR(handle);
2913 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2914 "%ld pages, ino %lu; err %d", __func__,
2915 wbc->nr_to_write, inode->i_ino, ret);
2916 /* Release allocated io_end */
2917 ext4_put_io_end(mpd.io_submit.io_end);
2918 mpd.io_submit.io_end = NULL;
2923 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2924 ret = mpage_prepare_extent_to_map(&mpd);
2927 ret = mpage_map_and_submit_extent(handle, &mpd,
2931 * We scanned the whole range (or exhausted
2932 * nr_to_write), submitted what was mapped and
2933 * didn't find anything needing mapping. We are
2940 * Caution: If the handle is synchronous,
2941 * ext4_journal_stop() can wait for transaction commit
2942 * to finish which may depend on writeback of pages to
2943 * complete or on page lock to be released. In that
2944 * case, we have to wait until after after we have
2945 * submitted all the IO, released page locks we hold,
2946 * and dropped io_end reference (for extent conversion
2947 * to be able to complete) before stopping the handle.
2949 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2950 ext4_journal_stop(handle);
2954 /* Unlock pages we didn't use */
2955 mpage_release_unused_pages(&mpd, give_up_on_write);
2956 /* Submit prepared bio */
2957 ext4_io_submit(&mpd.io_submit);
2960 * Drop our io_end reference we got from init. We have
2961 * to be careful and use deferred io_end finishing if
2962 * we are still holding the transaction as we can
2963 * release the last reference to io_end which may end
2964 * up doing unwritten extent conversion.
2967 ext4_put_io_end_defer(mpd.io_submit.io_end);
2968 ext4_journal_stop(handle);
2970 ext4_put_io_end(mpd.io_submit.io_end);
2971 mpd.io_submit.io_end = NULL;
2973 if (ret == -ENOSPC && sbi->s_journal) {
2975 * Commit the transaction which would
2976 * free blocks released in the transaction
2979 jbd2_journal_force_commit_nested(sbi->s_journal);
2983 /* Fatal error - ENOMEM, EIO... */
2988 blk_finish_plug(&plug);
2989 if (!ret && !cycled && wbc->nr_to_write > 0) {
2991 mpd.last_page = writeback_index - 1;
2997 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2999 * Set the writeback_index so that range_cyclic
3000 * mode will write it back later
3002 mapping->writeback_index = mpd.first_page;
3005 trace_ext4_writepages_result(inode, wbc, ret,
3006 nr_to_write - wbc->nr_to_write);
3007 percpu_up_read(&sbi->s_writepages_rwsem);
3011 static int ext4_dax_writepages(struct address_space *mapping,
3012 struct writeback_control *wbc)
3015 long nr_to_write = wbc->nr_to_write;
3016 struct inode *inode = mapping->host;
3017 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
3019 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3022 percpu_down_read(&sbi->s_writepages_rwsem);
3023 trace_ext4_writepages(inode, wbc);
3025 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
3026 trace_ext4_writepages_result(inode, wbc, ret,
3027 nr_to_write - wbc->nr_to_write);
3028 percpu_up_read(&sbi->s_writepages_rwsem);
3032 static int ext4_nonda_switch(struct super_block *sb)
3034 s64 free_clusters, dirty_clusters;
3035 struct ext4_sb_info *sbi = EXT4_SB(sb);
3038 * switch to non delalloc mode if we are running low
3039 * on free block. The free block accounting via percpu
3040 * counters can get slightly wrong with percpu_counter_batch getting
3041 * accumulated on each CPU without updating global counters
3042 * Delalloc need an accurate free block accounting. So switch
3043 * to non delalloc when we are near to error range.
3046 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3048 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3050 * Start pushing delalloc when 1/2 of free blocks are dirty.
3052 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3053 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3055 if (2 * free_clusters < 3 * dirty_clusters ||
3056 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3058 * free block count is less than 150% of dirty blocks
3059 * or free blocks is less than watermark
3066 /* We always reserve for an inode update; the superblock could be there too */
3067 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3069 if (likely(ext4_has_feature_large_file(inode->i_sb)))
3072 if (pos + len <= 0x7fffffffULL)
3075 /* We might need to update the superblock to set LARGE_FILE */
3079 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3080 loff_t pos, unsigned len, unsigned flags,
3081 struct page **pagep, void **fsdata)
3083 int ret, retries = 0;
3086 struct inode *inode = mapping->host;
3089 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3092 index = pos >> PAGE_SHIFT;
3094 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
3095 ext4_verity_in_progress(inode)) {
3096 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3097 return ext4_write_begin(file, mapping, pos,
3098 len, flags, pagep, fsdata);
3100 *fsdata = (void *)0;
3101 trace_ext4_da_write_begin(inode, pos, len, flags);
3103 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3104 ret = ext4_da_write_inline_data_begin(mapping, inode,
3114 * grab_cache_page_write_begin() can take a long time if the
3115 * system is thrashing due to memory pressure, or if the page
3116 * is being written back. So grab it first before we start
3117 * the transaction handle. This also allows us to allocate
3118 * the page (if needed) without using GFP_NOFS.
3121 page = grab_cache_page_write_begin(mapping, index, flags);
3127 * With delayed allocation, we don't log the i_disksize update
3128 * if there is delayed block allocation. But we still need
3129 * to journalling the i_disksize update if writes to the end
3130 * of file which has an already mapped buffer.
3133 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3134 ext4_da_write_credits(inode, pos, len));
3135 if (IS_ERR(handle)) {
3137 return PTR_ERR(handle);
3141 if (page->mapping != mapping) {
3142 /* The page got truncated from under us */
3145 ext4_journal_stop(handle);
3148 /* In case writeback began while the page was unlocked */
3149 wait_for_stable_page(page);
3151 #ifdef CONFIG_FS_ENCRYPTION
3152 ret = ext4_block_write_begin(page, pos, len,
3153 ext4_da_get_block_prep);
3155 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3159 ext4_journal_stop(handle);
3161 * block_write_begin may have instantiated a few blocks
3162 * outside i_size. Trim these off again. Don't need
3163 * i_size_read because we hold i_mutex.
3165 if (pos + len > inode->i_size)
3166 ext4_truncate_failed_write(inode);
3168 if (ret == -ENOSPC &&
3169 ext4_should_retry_alloc(inode->i_sb, &retries))
3181 * Check if we should update i_disksize
3182 * when write to the end of file but not require block allocation
3184 static int ext4_da_should_update_i_disksize(struct page *page,
3185 unsigned long offset)
3187 struct buffer_head *bh;
3188 struct inode *inode = page->mapping->host;
3192 bh = page_buffers(page);
3193 idx = offset >> inode->i_blkbits;
3195 for (i = 0; i < idx; i++)
3196 bh = bh->b_this_page;
3198 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3203 static int ext4_da_write_end(struct file *file,
3204 struct address_space *mapping,
3205 loff_t pos, unsigned len, unsigned copied,
3206 struct page *page, void *fsdata)
3208 struct inode *inode = mapping->host;
3210 handle_t *handle = ext4_journal_current_handle();
3212 unsigned long start, end;
3213 int write_mode = (int)(unsigned long)fsdata;
3215 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3216 return ext4_write_end(file, mapping, pos,
3217 len, copied, page, fsdata);
3219 trace_ext4_da_write_end(inode, pos, len, copied);
3220 start = pos & (PAGE_SIZE - 1);
3221 end = start + copied - 1;
3224 * generic_write_end() will run mark_inode_dirty() if i_size
3225 * changes. So let's piggyback the i_disksize mark_inode_dirty
3228 new_i_size = pos + copied;
3229 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3230 if (ext4_has_inline_data(inode) ||
3231 ext4_da_should_update_i_disksize(page, end)) {
3232 ext4_update_i_disksize(inode, new_i_size);
3233 /* We need to mark inode dirty even if
3234 * new_i_size is less that inode->i_size
3235 * bu greater than i_disksize.(hint delalloc)
3237 ext4_mark_inode_dirty(handle, inode);
3241 if (write_mode != CONVERT_INLINE_DATA &&
3242 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3243 ext4_has_inline_data(inode))
3244 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3247 ret2 = generic_write_end(file, mapping, pos, len, copied,
3253 ret2 = ext4_journal_stop(handle);
3257 return ret ? ret : copied;
3261 * Force all delayed allocation blocks to be allocated for a given inode.
3263 int ext4_alloc_da_blocks(struct inode *inode)
3265 trace_ext4_alloc_da_blocks(inode);
3267 if (!EXT4_I(inode)->i_reserved_data_blocks)
3271 * We do something simple for now. The filemap_flush() will
3272 * also start triggering a write of the data blocks, which is
3273 * not strictly speaking necessary (and for users of
3274 * laptop_mode, not even desirable). However, to do otherwise
3275 * would require replicating code paths in:
3277 * ext4_writepages() ->
3278 * write_cache_pages() ---> (via passed in callback function)
3279 * __mpage_da_writepage() -->
3280 * mpage_add_bh_to_extent()
3281 * mpage_da_map_blocks()
3283 * The problem is that write_cache_pages(), located in
3284 * mm/page-writeback.c, marks pages clean in preparation for
3285 * doing I/O, which is not desirable if we're not planning on
3288 * We could call write_cache_pages(), and then redirty all of
3289 * the pages by calling redirty_page_for_writepage() but that
3290 * would be ugly in the extreme. So instead we would need to
3291 * replicate parts of the code in the above functions,
3292 * simplifying them because we wouldn't actually intend to
3293 * write out the pages, but rather only collect contiguous
3294 * logical block extents, call the multi-block allocator, and
3295 * then update the buffer heads with the block allocations.
3297 * For now, though, we'll cheat by calling filemap_flush(),
3298 * which will map the blocks, and start the I/O, but not
3299 * actually wait for the I/O to complete.
3301 return filemap_flush(inode->i_mapping);
3305 * bmap() is special. It gets used by applications such as lilo and by
3306 * the swapper to find the on-disk block of a specific piece of data.
3308 * Naturally, this is dangerous if the block concerned is still in the
3309 * journal. If somebody makes a swapfile on an ext4 data-journaling
3310 * filesystem and enables swap, then they may get a nasty shock when the
3311 * data getting swapped to that swapfile suddenly gets overwritten by
3312 * the original zero's written out previously to the journal and
3313 * awaiting writeback in the kernel's buffer cache.
3315 * So, if we see any bmap calls here on a modified, data-journaled file,
3316 * take extra steps to flush any blocks which might be in the cache.
3318 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3320 struct inode *inode = mapping->host;
3325 * We can get here for an inline file via the FIBMAP ioctl
3327 if (ext4_has_inline_data(inode))
3330 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3331 test_opt(inode->i_sb, DELALLOC)) {
3333 * With delalloc we want to sync the file
3334 * so that we can make sure we allocate
3337 filemap_write_and_wait(mapping);
3340 if (EXT4_JOURNAL(inode) &&
3341 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3343 * This is a REALLY heavyweight approach, but the use of
3344 * bmap on dirty files is expected to be extremely rare:
3345 * only if we run lilo or swapon on a freshly made file
3346 * do we expect this to happen.
3348 * (bmap requires CAP_SYS_RAWIO so this does not
3349 * represent an unprivileged user DOS attack --- we'd be
3350 * in trouble if mortal users could trigger this path at
3353 * NB. EXT4_STATE_JDATA is not set on files other than
3354 * regular files. If somebody wants to bmap a directory
3355 * or symlink and gets confused because the buffer
3356 * hasn't yet been flushed to disk, they deserve
3357 * everything they get.
3360 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3361 journal = EXT4_JOURNAL(inode);
3362 jbd2_journal_lock_updates(journal);
3363 err = jbd2_journal_flush(journal);
3364 jbd2_journal_unlock_updates(journal);
3370 return generic_block_bmap(mapping, block, ext4_get_block);
3373 static int ext4_readpage(struct file *file, struct page *page)
3376 struct inode *inode = page->mapping->host;
3378 trace_ext4_readpage(page);
3380 if (ext4_has_inline_data(inode))
3381 ret = ext4_readpage_inline(inode, page);
3384 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3391 ext4_readpages(struct file *file, struct address_space *mapping,
3392 struct list_head *pages, unsigned nr_pages)
3394 struct inode *inode = mapping->host;
3396 /* If the file has inline data, no need to do readpages. */
3397 if (ext4_has_inline_data(inode))
3400 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3403 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3404 unsigned int length)
3406 trace_ext4_invalidatepage(page, offset, length);
3408 /* No journalling happens on data buffers when this function is used */
3409 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3411 block_invalidatepage(page, offset, length);
3414 static int __ext4_journalled_invalidatepage(struct page *page,
3415 unsigned int offset,
3416 unsigned int length)
3418 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3420 trace_ext4_journalled_invalidatepage(page, offset, length);
3423 * If it's a full truncate we just forget about the pending dirtying
3425 if (offset == 0 && length == PAGE_SIZE)
3426 ClearPageChecked(page);
3428 return jbd2_journal_invalidatepage(journal, page, offset, length);
3431 /* Wrapper for aops... */
3432 static void ext4_journalled_invalidatepage(struct page *page,
3433 unsigned int offset,
3434 unsigned int length)
3436 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3439 static int ext4_releasepage(struct page *page, gfp_t wait)
3441 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3443 trace_ext4_releasepage(page);
3445 /* Page has dirty journalled data -> cannot release */
3446 if (PageChecked(page))
3449 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3451 return try_to_free_buffers(page);
3454 static bool ext4_inode_datasync_dirty(struct inode *inode)
3456 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3459 return !jbd2_transaction_committed(journal,
3460 EXT4_I(inode)->i_datasync_tid);
3461 /* Any metadata buffers to write? */
3462 if (!list_empty(&inode->i_mapping->private_list))
3464 return inode->i_state & I_DIRTY_DATASYNC;
3467 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3468 unsigned flags, struct iomap *iomap)
3470 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3471 unsigned int blkbits = inode->i_blkbits;
3472 unsigned long first_block, last_block;
3473 struct ext4_map_blocks map;
3474 bool delalloc = false;
3477 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3479 first_block = offset >> blkbits;
3480 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3481 EXT4_MAX_LOGICAL_BLOCK);
3483 if (flags & IOMAP_REPORT) {
3484 if (ext4_has_inline_data(inode)) {
3485 ret = ext4_inline_data_iomap(inode, iomap);
3486 if (ret != -EAGAIN) {
3487 if (ret == 0 && offset >= iomap->length)
3493 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3497 map.m_lblk = first_block;
3498 map.m_len = last_block - first_block + 1;
3500 if (flags & IOMAP_REPORT) {
3501 ret = ext4_map_blocks(NULL, inode, &map, 0);
3506 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3507 struct extent_status es;
3509 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3510 map.m_lblk, end, &es);
3512 if (!es.es_len || es.es_lblk > end) {
3513 /* entire range is a hole */
3514 } else if (es.es_lblk > map.m_lblk) {
3515 /* range starts with a hole */
3516 map.m_len = es.es_lblk - map.m_lblk;
3518 ext4_lblk_t offs = 0;
3520 if (es.es_lblk < map.m_lblk)
3521 offs = map.m_lblk - es.es_lblk;
3522 map.m_lblk = es.es_lblk + offs;
3523 map.m_len = es.es_len - offs;
3527 } else if (flags & IOMAP_WRITE) {
3532 /* Trim mapping request to maximum we can map at once for DIO */
3533 if (map.m_len > DIO_MAX_BLOCKS)
3534 map.m_len = DIO_MAX_BLOCKS;
3535 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3538 * Either we allocate blocks and then we don't get unwritten
3539 * extent so we have reserved enough credits, or the blocks
3540 * are already allocated and unwritten and in that case
3541 * extent conversion fits in the credits as well.
3543 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3546 return PTR_ERR(handle);
3548 ret = ext4_map_blocks(handle, inode, &map,
3549 EXT4_GET_BLOCKS_CREATE_ZERO);
3551 ext4_journal_stop(handle);
3552 if (ret == -ENOSPC &&
3553 ext4_should_retry_alloc(inode->i_sb, &retries))
3559 * If we added blocks beyond i_size, we need to make sure they
3560 * will get truncated if we crash before updating i_size in
3561 * ext4_iomap_end(). For faults we don't need to do that (and
3562 * even cannot because for orphan list operations inode_lock is
3563 * required) - if we happen to instantiate block beyond i_size,
3564 * it is because we race with truncate which has already added
3565 * the inode to the orphan list.
3567 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3568 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3571 err = ext4_orphan_add(handle, inode);
3573 ext4_journal_stop(handle);
3577 ext4_journal_stop(handle);
3579 ret = ext4_map_blocks(NULL, inode, &map, 0);
3585 * Writes that span EOF might trigger an I/O size update on completion,
3586 * so consider them to be dirty for the purposes of O_DSYNC, even if
3587 * there is no other metadata changes being made or are pending here.
3590 if (ext4_inode_datasync_dirty(inode) ||
3591 offset + length > i_size_read(inode))
3592 iomap->flags |= IOMAP_F_DIRTY;
3593 iomap->bdev = inode->i_sb->s_bdev;
3594 iomap->dax_dev = sbi->s_daxdev;
3595 iomap->offset = (u64)first_block << blkbits;
3596 iomap->length = (u64)map.m_len << blkbits;
3599 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3600 iomap->addr = IOMAP_NULL_ADDR;
3602 if (map.m_flags & EXT4_MAP_MAPPED) {
3603 iomap->type = IOMAP_MAPPED;
3604 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3605 iomap->type = IOMAP_UNWRITTEN;
3610 iomap->addr = (u64)map.m_pblk << blkbits;
3613 if (map.m_flags & EXT4_MAP_NEW)
3614 iomap->flags |= IOMAP_F_NEW;
3619 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3620 ssize_t written, unsigned flags, struct iomap *iomap)
3624 int blkbits = inode->i_blkbits;
3625 bool truncate = false;
3627 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3630 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3631 if (IS_ERR(handle)) {
3632 ret = PTR_ERR(handle);
3635 if (ext4_update_inode_size(inode, offset + written))
3636 ext4_mark_inode_dirty(handle, inode);
3638 * We may need to truncate allocated but not written blocks beyond EOF.
3640 if (iomap->offset + iomap->length >
3641 ALIGN(inode->i_size, 1 << blkbits)) {
3642 ext4_lblk_t written_blk, end_blk;
3644 written_blk = (offset + written) >> blkbits;
3645 end_blk = (offset + length) >> blkbits;
3646 if (written_blk < end_blk && ext4_can_truncate(inode))
3650 * Remove inode from orphan list if we were extending a inode and
3651 * everything went fine.
3653 if (!truncate && inode->i_nlink &&
3654 !list_empty(&EXT4_I(inode)->i_orphan))
3655 ext4_orphan_del(handle, inode);
3656 ext4_journal_stop(handle);
3658 ext4_truncate_failed_write(inode);
3661 * If truncate failed early the inode might still be on the
3662 * orphan list; we need to make sure the inode is removed from
3663 * the orphan list in that case.
3666 ext4_orphan_del(NULL, inode);
3671 const struct iomap_ops ext4_iomap_ops = {
3672 .iomap_begin = ext4_iomap_begin,
3673 .iomap_end = ext4_iomap_end,
3676 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3677 ssize_t size, void *private)
3679 ext4_io_end_t *io_end = private;
3681 /* if not async direct IO just return */
3685 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3686 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3687 io_end, io_end->inode->i_ino, iocb, offset, size);
3690 * Error during AIO DIO. We cannot convert unwritten extents as the
3691 * data was not written. Just clear the unwritten flag and drop io_end.
3694 ext4_clear_io_unwritten_flag(io_end);
3697 io_end->offset = offset;
3698 io_end->size = size;
3699 ext4_put_io_end(io_end);
3705 * Handling of direct IO writes.
3707 * For ext4 extent files, ext4 will do direct-io write even to holes,
3708 * preallocated extents, and those write extend the file, no need to
3709 * fall back to buffered IO.
3711 * For holes, we fallocate those blocks, mark them as unwritten
3712 * If those blocks were preallocated, we mark sure they are split, but
3713 * still keep the range to write as unwritten.
3715 * The unwritten extents will be converted to written when DIO is completed.
3716 * For async direct IO, since the IO may still pending when return, we
3717 * set up an end_io call back function, which will do the conversion
3718 * when async direct IO completed.
3720 * If the O_DIRECT write will extend the file then add this inode to the
3721 * orphan list. So recovery will truncate it back to the original size
3722 * if the machine crashes during the write.
3725 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3727 struct file *file = iocb->ki_filp;
3728 struct inode *inode = file->f_mapping->host;
3729 struct ext4_inode_info *ei = EXT4_I(inode);
3731 loff_t offset = iocb->ki_pos;
3732 size_t count = iov_iter_count(iter);
3734 get_block_t *get_block_func = NULL;
3736 loff_t final_size = offset + count;
3740 if (final_size > inode->i_size || final_size > ei->i_disksize) {
3741 /* Credits for sb + inode write */
3742 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3743 if (IS_ERR(handle)) {
3744 ret = PTR_ERR(handle);
3747 ret = ext4_orphan_add(handle, inode);
3749 ext4_journal_stop(handle);
3753 ext4_update_i_disksize(inode, inode->i_size);
3754 ext4_journal_stop(handle);
3757 BUG_ON(iocb->private == NULL);
3760 * Make all waiters for direct IO properly wait also for extent
3761 * conversion. This also disallows race between truncate() and
3762 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3764 inode_dio_begin(inode);
3766 /* If we do a overwrite dio, i_mutex locking can be released */
3767 overwrite = *((int *)iocb->private);
3770 inode_unlock(inode);
3773 * For extent mapped files we could direct write to holes and fallocate.
3775 * Allocated blocks to fill the hole are marked as unwritten to prevent
3776 * parallel buffered read to expose the stale data before DIO complete
3779 * As to previously fallocated extents, ext4 get_block will just simply
3780 * mark the buffer mapped but still keep the extents unwritten.
3782 * For non AIO case, we will convert those unwritten extents to written
3783 * after return back from blockdev_direct_IO. That way we save us from
3784 * allocating io_end structure and also the overhead of offloading
3785 * the extent convertion to a workqueue.
3787 * For async DIO, the conversion needs to be deferred when the
3788 * IO is completed. The ext4 end_io callback function will be
3789 * called to take care of the conversion work. Here for async
3790 * case, we allocate an io_end structure to hook to the iocb.
3792 iocb->private = NULL;
3794 get_block_func = ext4_dio_get_block_overwrite;
3795 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3796 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3797 get_block_func = ext4_dio_get_block;
3798 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3799 } else if (is_sync_kiocb(iocb)) {
3800 get_block_func = ext4_dio_get_block_unwritten_sync;
3801 dio_flags = DIO_LOCKING;
3803 get_block_func = ext4_dio_get_block_unwritten_async;
3804 dio_flags = DIO_LOCKING;
3806 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3807 get_block_func, ext4_end_io_dio, NULL,
3810 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3811 EXT4_STATE_DIO_UNWRITTEN)) {
3814 * for non AIO case, since the IO is already
3815 * completed, we could do the conversion right here
3817 err = ext4_convert_unwritten_extents(NULL, inode,
3821 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3824 inode_dio_end(inode);
3825 /* take i_mutex locking again if we do a ovewrite dio */
3829 if (ret < 0 && final_size > inode->i_size)
3830 ext4_truncate_failed_write(inode);
3832 /* Handle extending of i_size after direct IO write */
3836 /* Credits for sb + inode write */
3837 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3838 if (IS_ERR(handle)) {
3840 * We wrote the data but cannot extend
3841 * i_size. Bail out. In async io case, we do
3842 * not return error here because we have
3843 * already submmitted the corresponding
3844 * bio. Returning error here makes the caller
3845 * think that this IO is done and failed
3846 * resulting in race with bio's completion
3850 ret = PTR_ERR(handle);
3852 ext4_orphan_del(NULL, inode);
3857 ext4_orphan_del(handle, inode);
3859 loff_t end = offset + ret;
3860 if (end > inode->i_size || end > ei->i_disksize) {
3861 ext4_update_i_disksize(inode, end);
3862 if (end > inode->i_size)
3863 i_size_write(inode, end);
3865 * We're going to return a positive `ret'
3866 * here due to non-zero-length I/O, so there's
3867 * no way of reporting error returns from
3868 * ext4_mark_inode_dirty() to userspace. So
3871 ext4_mark_inode_dirty(handle, inode);
3874 err = ext4_journal_stop(handle);
3882 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3884 struct address_space *mapping = iocb->ki_filp->f_mapping;
3885 struct inode *inode = mapping->host;
3886 size_t count = iov_iter_count(iter);
3888 loff_t offset = iocb->ki_pos;
3889 loff_t size = i_size_read(inode);
3895 * Shared inode_lock is enough for us - it protects against concurrent
3896 * writes & truncates and since we take care of writing back page cache,
3897 * we are protected against page writeback as well.
3899 if (iocb->ki_flags & IOCB_NOWAIT) {
3900 if (!inode_trylock_shared(inode))
3903 inode_lock_shared(inode);
3906 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3907 iocb->ki_pos + count - 1);
3910 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3911 iter, ext4_dio_get_block, NULL, NULL, 0);
3913 inode_unlock_shared(inode);
3917 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3919 struct file *file = iocb->ki_filp;
3920 struct inode *inode = file->f_mapping->host;
3921 size_t count = iov_iter_count(iter);
3922 loff_t offset = iocb->ki_pos;
3925 #ifdef CONFIG_FS_ENCRYPTION
3926 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
3929 if (fsverity_active(inode))
3933 * If we are doing data journalling we don't support O_DIRECT
3935 if (ext4_should_journal_data(inode))
3938 /* Let buffer I/O handle the inline data case. */
3939 if (ext4_has_inline_data(inode))
3942 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3943 if (iov_iter_rw(iter) == READ)
3944 ret = ext4_direct_IO_read(iocb, iter);
3946 ret = ext4_direct_IO_write(iocb, iter);
3947 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3952 * Pages can be marked dirty completely asynchronously from ext4's journalling
3953 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3954 * much here because ->set_page_dirty is called under VFS locks. The page is
3955 * not necessarily locked.
3957 * We cannot just dirty the page and leave attached buffers clean, because the
3958 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3959 * or jbddirty because all the journalling code will explode.
3961 * So what we do is to mark the page "pending dirty" and next time writepage
3962 * is called, propagate that into the buffers appropriately.
3964 static int ext4_journalled_set_page_dirty(struct page *page)
3966 SetPageChecked(page);
3967 return __set_page_dirty_nobuffers(page);
3970 static int ext4_set_page_dirty(struct page *page)
3972 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3973 WARN_ON_ONCE(!page_has_buffers(page));
3974 return __set_page_dirty_buffers(page);
3977 static const struct address_space_operations ext4_aops = {
3978 .readpage = ext4_readpage,
3979 .readpages = ext4_readpages,
3980 .writepage = ext4_writepage,
3981 .writepages = ext4_writepages,
3982 .write_begin = ext4_write_begin,
3983 .write_end = ext4_write_end,
3984 .set_page_dirty = ext4_set_page_dirty,
3986 .invalidatepage = ext4_invalidatepage,
3987 .releasepage = ext4_releasepage,
3988 .direct_IO = ext4_direct_IO,
3989 .migratepage = buffer_migrate_page,
3990 .is_partially_uptodate = block_is_partially_uptodate,
3991 .error_remove_page = generic_error_remove_page,
3994 static const struct address_space_operations ext4_journalled_aops = {
3995 .readpage = ext4_readpage,
3996 .readpages = ext4_readpages,
3997 .writepage = ext4_writepage,
3998 .writepages = ext4_writepages,
3999 .write_begin = ext4_write_begin,
4000 .write_end = ext4_journalled_write_end,
4001 .set_page_dirty = ext4_journalled_set_page_dirty,
4003 .invalidatepage = ext4_journalled_invalidatepage,
4004 .releasepage = ext4_releasepage,
4005 .direct_IO = ext4_direct_IO,
4006 .is_partially_uptodate = block_is_partially_uptodate,
4007 .error_remove_page = generic_error_remove_page,
4010 static const struct address_space_operations ext4_da_aops = {
4011 .readpage = ext4_readpage,
4012 .readpages = ext4_readpages,
4013 .writepage = ext4_writepage,
4014 .writepages = ext4_writepages,
4015 .write_begin = ext4_da_write_begin,
4016 .write_end = ext4_da_write_end,
4017 .set_page_dirty = ext4_set_page_dirty,
4019 .invalidatepage = ext4_invalidatepage,
4020 .releasepage = ext4_releasepage,
4021 .direct_IO = ext4_direct_IO,
4022 .migratepage = buffer_migrate_page,
4023 .is_partially_uptodate = block_is_partially_uptodate,
4024 .error_remove_page = generic_error_remove_page,
4027 static const struct address_space_operations ext4_dax_aops = {
4028 .writepages = ext4_dax_writepages,
4029 .direct_IO = noop_direct_IO,
4030 .set_page_dirty = noop_set_page_dirty,
4032 .invalidatepage = noop_invalidatepage,
4035 void ext4_set_aops(struct inode *inode)
4037 switch (ext4_inode_journal_mode(inode)) {
4038 case EXT4_INODE_ORDERED_DATA_MODE:
4039 case EXT4_INODE_WRITEBACK_DATA_MODE:
4041 case EXT4_INODE_JOURNAL_DATA_MODE:
4042 inode->i_mapping->a_ops = &ext4_journalled_aops;
4048 inode->i_mapping->a_ops = &ext4_dax_aops;
4049 else if (test_opt(inode->i_sb, DELALLOC))
4050 inode->i_mapping->a_ops = &ext4_da_aops;
4052 inode->i_mapping->a_ops = &ext4_aops;
4055 static int __ext4_block_zero_page_range(handle_t *handle,
4056 struct address_space *mapping, loff_t from, loff_t length)
4058 ext4_fsblk_t index = from >> PAGE_SHIFT;
4059 unsigned offset = from & (PAGE_SIZE-1);
4060 unsigned blocksize, pos;
4062 struct inode *inode = mapping->host;
4063 struct buffer_head *bh;
4067 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4068 mapping_gfp_constraint(mapping, ~__GFP_FS));
4072 blocksize = inode->i_sb->s_blocksize;
4074 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4076 if (!page_has_buffers(page))
4077 create_empty_buffers(page, blocksize, 0);
4079 /* Find the buffer that contains "offset" */
4080 bh = page_buffers(page);
4082 while (offset >= pos) {
4083 bh = bh->b_this_page;
4087 if (buffer_freed(bh)) {
4088 BUFFER_TRACE(bh, "freed: skip");
4091 if (!buffer_mapped(bh)) {
4092 BUFFER_TRACE(bh, "unmapped");
4093 ext4_get_block(inode, iblock, bh, 0);
4094 /* unmapped? It's a hole - nothing to do */
4095 if (!buffer_mapped(bh)) {
4096 BUFFER_TRACE(bh, "still unmapped");
4101 /* Ok, it's mapped. Make sure it's up-to-date */
4102 if (PageUptodate(page))
4103 set_buffer_uptodate(bh);
4105 if (!buffer_uptodate(bh)) {
4107 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4109 /* Uhhuh. Read error. Complain and punt. */
4110 if (!buffer_uptodate(bh))
4112 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
4113 /* We expect the key to be set. */
4114 BUG_ON(!fscrypt_has_encryption_key(inode));
4115 WARN_ON_ONCE(fscrypt_decrypt_pagecache_blocks(
4116 page, blocksize, bh_offset(bh)));
4119 if (ext4_should_journal_data(inode)) {
4120 BUFFER_TRACE(bh, "get write access");
4121 err = ext4_journal_get_write_access(handle, bh);
4125 zero_user(page, offset, length);
4126 BUFFER_TRACE(bh, "zeroed end of block");
4128 if (ext4_should_journal_data(inode)) {
4129 err = ext4_handle_dirty_metadata(handle, inode, bh);
4132 mark_buffer_dirty(bh);
4133 if (ext4_should_order_data(inode))
4134 err = ext4_jbd2_inode_add_write(handle, inode, from,
4145 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4146 * starting from file offset 'from'. The range to be zero'd must
4147 * be contained with in one block. If the specified range exceeds
4148 * the end of the block it will be shortened to end of the block
4149 * that cooresponds to 'from'
4151 static int ext4_block_zero_page_range(handle_t *handle,
4152 struct address_space *mapping, loff_t from, loff_t length)
4154 struct inode *inode = mapping->host;
4155 unsigned offset = from & (PAGE_SIZE-1);
4156 unsigned blocksize = inode->i_sb->s_blocksize;
4157 unsigned max = blocksize - (offset & (blocksize - 1));
4160 * correct length if it does not fall between
4161 * 'from' and the end of the block
4163 if (length > max || length < 0)
4166 if (IS_DAX(inode)) {
4167 return iomap_zero_range(inode, from, length, NULL,
4170 return __ext4_block_zero_page_range(handle, mapping, from, length);
4174 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4175 * up to the end of the block which corresponds to `from'.
4176 * This required during truncate. We need to physically zero the tail end
4177 * of that block so it doesn't yield old data if the file is later grown.
4179 static int ext4_block_truncate_page(handle_t *handle,
4180 struct address_space *mapping, loff_t from)
4182 unsigned offset = from & (PAGE_SIZE-1);
4185 struct inode *inode = mapping->host;
4187 /* If we are processing an encrypted inode during orphan list handling */
4188 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
4191 blocksize = inode->i_sb->s_blocksize;
4192 length = blocksize - (offset & (blocksize - 1));
4194 return ext4_block_zero_page_range(handle, mapping, from, length);
4197 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4198 loff_t lstart, loff_t length)
4200 struct super_block *sb = inode->i_sb;
4201 struct address_space *mapping = inode->i_mapping;
4202 unsigned partial_start, partial_end;
4203 ext4_fsblk_t start, end;
4204 loff_t byte_end = (lstart + length - 1);
4207 partial_start = lstart & (sb->s_blocksize - 1);
4208 partial_end = byte_end & (sb->s_blocksize - 1);
4210 start = lstart >> sb->s_blocksize_bits;
4211 end = byte_end >> sb->s_blocksize_bits;
4213 /* Handle partial zero within the single block */
4215 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4216 err = ext4_block_zero_page_range(handle, mapping,
4220 /* Handle partial zero out on the start of the range */
4221 if (partial_start) {
4222 err = ext4_block_zero_page_range(handle, mapping,
4223 lstart, sb->s_blocksize);
4227 /* Handle partial zero out on the end of the range */
4228 if (partial_end != sb->s_blocksize - 1)
4229 err = ext4_block_zero_page_range(handle, mapping,
4230 byte_end - partial_end,
4235 int ext4_can_truncate(struct inode *inode)
4237 if (S_ISREG(inode->i_mode))
4239 if (S_ISDIR(inode->i_mode))
4241 if (S_ISLNK(inode->i_mode))
4242 return !ext4_inode_is_fast_symlink(inode);
4247 * We have to make sure i_disksize gets properly updated before we truncate
4248 * page cache due to hole punching or zero range. Otherwise i_disksize update
4249 * can get lost as it may have been postponed to submission of writeback but
4250 * that will never happen after we truncate page cache.
4252 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4256 loff_t size = i_size_read(inode);
4258 WARN_ON(!inode_is_locked(inode));
4259 if (offset > size || offset + len < size)
4262 if (EXT4_I(inode)->i_disksize >= size)
4265 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4267 return PTR_ERR(handle);
4268 ext4_update_i_disksize(inode, size);
4269 ext4_mark_inode_dirty(handle, inode);
4270 ext4_journal_stop(handle);
4275 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4277 up_write(&ei->i_mmap_sem);
4279 down_write(&ei->i_mmap_sem);
4282 int ext4_break_layouts(struct inode *inode)
4284 struct ext4_inode_info *ei = EXT4_I(inode);
4288 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4292 page = dax_layout_busy_page(inode->i_mapping);
4296 error = ___wait_var_event(&page->_refcount,
4297 atomic_read(&page->_refcount) == 1,
4298 TASK_INTERRUPTIBLE, 0, 0,
4299 ext4_wait_dax_page(ei));
4300 } while (error == 0);
4306 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4307 * associated with the given offset and length
4309 * @inode: File inode
4310 * @offset: The offset where the hole will begin
4311 * @len: The length of the hole
4313 * Returns: 0 on success or negative on failure
4316 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4318 struct super_block *sb = inode->i_sb;
4319 ext4_lblk_t first_block, stop_block;
4320 struct address_space *mapping = inode->i_mapping;
4321 loff_t first_block_offset, last_block_offset, max_length;
4322 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4324 unsigned int credits;
4327 if (!S_ISREG(inode->i_mode))
4330 trace_ext4_punch_hole(inode, offset, length, 0);
4333 * Write out all dirty pages to avoid race conditions
4334 * Then release them.
4336 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4337 ret = filemap_write_and_wait_range(mapping, offset,
4338 offset + length - 1);
4345 /* No need to punch hole beyond i_size */
4346 if (offset >= inode->i_size)
4350 * If the hole extends beyond i_size, set the hole
4351 * to end after the page that contains i_size
4353 if (offset + length > inode->i_size) {
4354 length = inode->i_size +
4355 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4360 * For punch hole the length + offset needs to be within one block
4361 * before last range. Adjust the length if it goes beyond that limit.
4363 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4364 if (offset + length > max_length)
4365 length = max_length - offset;
4367 if (offset & (sb->s_blocksize - 1) ||
4368 (offset + length) & (sb->s_blocksize - 1)) {
4370 * Attach jinode to inode for jbd2 if we do any zeroing of
4373 ret = ext4_inode_attach_jinode(inode);
4379 /* Wait all existing dio workers, newcomers will block on i_mutex */
4380 inode_dio_wait(inode);
4383 * Prevent page faults from reinstantiating pages we have released from
4386 down_write(&EXT4_I(inode)->i_mmap_sem);
4388 ret = ext4_break_layouts(inode);
4392 first_block_offset = round_up(offset, sb->s_blocksize);
4393 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4395 /* Now release the pages and zero block aligned part of pages*/
4396 if (last_block_offset > first_block_offset) {
4397 ret = ext4_update_disksize_before_punch(inode, offset, length);
4400 truncate_pagecache_range(inode, first_block_offset,
4404 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4405 credits = ext4_writepage_trans_blocks(inode);
4407 credits = ext4_blocks_for_truncate(inode);
4408 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4409 if (IS_ERR(handle)) {
4410 ret = PTR_ERR(handle);
4411 ext4_std_error(sb, ret);
4415 ret = ext4_zero_partial_blocks(handle, inode, offset,
4420 first_block = (offset + sb->s_blocksize - 1) >>
4421 EXT4_BLOCK_SIZE_BITS(sb);
4422 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4424 /* If there are blocks to remove, do it */
4425 if (stop_block > first_block) {
4427 down_write(&EXT4_I(inode)->i_data_sem);
4428 ext4_discard_preallocations(inode);
4430 ret = ext4_es_remove_extent(inode, first_block,
4431 stop_block - first_block);
4433 up_write(&EXT4_I(inode)->i_data_sem);
4437 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4438 ret = ext4_ext_remove_space(inode, first_block,
4441 ret = ext4_ind_remove_space(handle, inode, first_block,
4444 up_write(&EXT4_I(inode)->i_data_sem);
4447 ext4_handle_sync(handle);
4449 inode->i_mtime = inode->i_ctime = current_time(inode);
4450 ext4_mark_inode_dirty(handle, inode);
4452 ext4_update_inode_fsync_trans(handle, inode, 1);
4454 ext4_journal_stop(handle);
4456 up_write(&EXT4_I(inode)->i_mmap_sem);
4458 inode_unlock(inode);
4462 int ext4_inode_attach_jinode(struct inode *inode)
4464 struct ext4_inode_info *ei = EXT4_I(inode);
4465 struct jbd2_inode *jinode;
4467 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4470 jinode = jbd2_alloc_inode(GFP_KERNEL);
4471 spin_lock(&inode->i_lock);
4474 spin_unlock(&inode->i_lock);
4477 ei->jinode = jinode;
4478 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4481 spin_unlock(&inode->i_lock);
4482 if (unlikely(jinode != NULL))
4483 jbd2_free_inode(jinode);
4490 * We block out ext4_get_block() block instantiations across the entire
4491 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4492 * simultaneously on behalf of the same inode.
4494 * As we work through the truncate and commit bits of it to the journal there
4495 * is one core, guiding principle: the file's tree must always be consistent on
4496 * disk. We must be able to restart the truncate after a crash.
4498 * The file's tree may be transiently inconsistent in memory (although it
4499 * probably isn't), but whenever we close off and commit a journal transaction,
4500 * the contents of (the filesystem + the journal) must be consistent and
4501 * restartable. It's pretty simple, really: bottom up, right to left (although
4502 * left-to-right works OK too).
4504 * Note that at recovery time, journal replay occurs *before* the restart of
4505 * truncate against the orphan inode list.
4507 * The committed inode has the new, desired i_size (which is the same as
4508 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4509 * that this inode's truncate did not complete and it will again call
4510 * ext4_truncate() to have another go. So there will be instantiated blocks
4511 * to the right of the truncation point in a crashed ext4 filesystem. But
4512 * that's fine - as long as they are linked from the inode, the post-crash
4513 * ext4_truncate() run will find them and release them.
4515 int ext4_truncate(struct inode *inode)
4517 struct ext4_inode_info *ei = EXT4_I(inode);
4518 unsigned int credits;
4521 struct address_space *mapping = inode->i_mapping;
4524 * There is a possibility that we're either freeing the inode
4525 * or it's a completely new inode. In those cases we might not
4526 * have i_mutex locked because it's not necessary.
4528 if (!(inode->i_state & (I_NEW|I_FREEING)))
4529 WARN_ON(!inode_is_locked(inode));
4530 trace_ext4_truncate_enter(inode);
4532 if (!ext4_can_truncate(inode))
4535 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4537 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4538 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4540 if (ext4_has_inline_data(inode)) {
4543 err = ext4_inline_data_truncate(inode, &has_inline);
4544 if (err || has_inline)
4548 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4549 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4550 err = ext4_inode_attach_jinode(inode);
4555 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4556 credits = ext4_writepage_trans_blocks(inode);
4558 credits = ext4_blocks_for_truncate(inode);
4560 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4561 if (IS_ERR(handle)) {
4562 err = PTR_ERR(handle);
4566 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4567 ext4_block_truncate_page(handle, mapping, inode->i_size);
4570 * We add the inode to the orphan list, so that if this
4571 * truncate spans multiple transactions, and we crash, we will
4572 * resume the truncate when the filesystem recovers. It also
4573 * marks the inode dirty, to catch the new size.
4575 * Implication: the file must always be in a sane, consistent
4576 * truncatable state while each transaction commits.
4578 err = ext4_orphan_add(handle, inode);
4582 down_write(&EXT4_I(inode)->i_data_sem);
4584 ext4_discard_preallocations(inode);
4586 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4587 err = ext4_ext_truncate(handle, inode);
4589 ext4_ind_truncate(handle, inode);
4591 up_write(&ei->i_data_sem);
4596 ext4_handle_sync(handle);
4600 * If this was a simple ftruncate() and the file will remain alive,
4601 * then we need to clear up the orphan record which we created above.
4602 * However, if this was a real unlink then we were called by
4603 * ext4_evict_inode(), and we allow that function to clean up the
4604 * orphan info for us.
4607 ext4_orphan_del(handle, inode);
4609 inode->i_mtime = inode->i_ctime = current_time(inode);
4610 ext4_mark_inode_dirty(handle, inode);
4611 ext4_journal_stop(handle);
4614 trace_ext4_truncate_exit(inode);
4619 * ext4_get_inode_loc returns with an extra refcount against the inode's
4620 * underlying buffer_head on success. If 'in_mem' is true, we have all
4621 * data in memory that is needed to recreate the on-disk version of this
4624 static int __ext4_get_inode_loc(struct inode *inode,
4625 struct ext4_iloc *iloc, int in_mem)
4627 struct ext4_group_desc *gdp;
4628 struct buffer_head *bh;
4629 struct super_block *sb = inode->i_sb;
4631 struct blk_plug plug;
4632 int inodes_per_block, inode_offset;
4635 if (inode->i_ino < EXT4_ROOT_INO ||
4636 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4637 return -EFSCORRUPTED;
4639 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4640 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4645 * Figure out the offset within the block group inode table
4647 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4648 inode_offset = ((inode->i_ino - 1) %
4649 EXT4_INODES_PER_GROUP(sb));
4650 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4652 block = ext4_inode_table(sb, gdp);
4653 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4654 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4655 ext4_error(sb, "Invalid inode table block %llu in "
4656 "block_group %u", block, iloc->block_group);
4657 return -EFSCORRUPTED;
4659 block += (inode_offset / inodes_per_block);
4661 bh = sb_getblk(sb, block);
4664 if (!buffer_uptodate(bh)) {
4668 * If the buffer has the write error flag, we have failed
4669 * to write out another inode in the same block. In this
4670 * case, we don't have to read the block because we may
4671 * read the old inode data successfully.
4673 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4674 set_buffer_uptodate(bh);
4676 if (buffer_uptodate(bh)) {
4677 /* someone brought it uptodate while we waited */
4683 * If we have all information of the inode in memory and this
4684 * is the only valid inode in the block, we need not read the
4688 struct buffer_head *bitmap_bh;
4691 start = inode_offset & ~(inodes_per_block - 1);
4693 /* Is the inode bitmap in cache? */
4694 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4695 if (unlikely(!bitmap_bh))
4699 * If the inode bitmap isn't in cache then the
4700 * optimisation may end up performing two reads instead
4701 * of one, so skip it.
4703 if (!buffer_uptodate(bitmap_bh)) {
4707 for (i = start; i < start + inodes_per_block; i++) {
4708 if (i == inode_offset)
4710 if (ext4_test_bit(i, bitmap_bh->b_data))
4714 if (i == start + inodes_per_block) {
4715 /* all other inodes are free, so skip I/O */
4716 memset(bh->b_data, 0, bh->b_size);
4717 set_buffer_uptodate(bh);
4725 * If we need to do any I/O, try to pre-readahead extra
4726 * blocks from the inode table.
4728 blk_start_plug(&plug);
4729 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4730 ext4_fsblk_t b, end, table;
4732 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4734 table = ext4_inode_table(sb, gdp);
4735 /* s_inode_readahead_blks is always a power of 2 */
4736 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4740 num = EXT4_INODES_PER_GROUP(sb);
4741 if (ext4_has_group_desc_csum(sb))
4742 num -= ext4_itable_unused_count(sb, gdp);
4743 table += num / inodes_per_block;
4747 sb_breadahead_unmovable(sb, b++);
4751 * There are other valid inodes in the buffer, this inode
4752 * has in-inode xattrs, or we don't have this inode in memory.
4753 * Read the block from disk.
4755 trace_ext4_load_inode(inode);
4757 bh->b_end_io = end_buffer_read_sync;
4758 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4759 blk_finish_plug(&plug);
4761 if (!buffer_uptodate(bh)) {
4762 EXT4_ERROR_INODE_BLOCK(inode, block,
4763 "unable to read itable block");
4773 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4775 /* We have all inode data except xattrs in memory here. */
4776 return __ext4_get_inode_loc(inode, iloc,
4777 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4780 static bool ext4_should_use_dax(struct inode *inode)
4782 if (!test_opt(inode->i_sb, DAX))
4784 if (!S_ISREG(inode->i_mode))
4786 if (ext4_should_journal_data(inode))
4788 if (ext4_has_inline_data(inode))
4790 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4792 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4797 void ext4_set_inode_flags(struct inode *inode)
4799 unsigned int flags = EXT4_I(inode)->i_flags;
4800 unsigned int new_fl = 0;
4802 if (flags & EXT4_SYNC_FL)
4804 if (flags & EXT4_APPEND_FL)
4806 if (flags & EXT4_IMMUTABLE_FL)
4807 new_fl |= S_IMMUTABLE;
4808 if (flags & EXT4_NOATIME_FL)
4809 new_fl |= S_NOATIME;
4810 if (flags & EXT4_DIRSYNC_FL)
4811 new_fl |= S_DIRSYNC;
4812 if (ext4_should_use_dax(inode))
4814 if (flags & EXT4_ENCRYPT_FL)
4815 new_fl |= S_ENCRYPTED;
4816 if (flags & EXT4_CASEFOLD_FL)
4817 new_fl |= S_CASEFOLD;
4818 if (flags & EXT4_VERITY_FL)
4820 inode_set_flags(inode, new_fl,
4821 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4822 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4825 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4826 struct ext4_inode_info *ei)
4829 struct inode *inode = &(ei->vfs_inode);
4830 struct super_block *sb = inode->i_sb;
4832 if (ext4_has_feature_huge_file(sb)) {
4833 /* we are using combined 48 bit field */
4834 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4835 le32_to_cpu(raw_inode->i_blocks_lo);
4836 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4837 /* i_blocks represent file system block size */
4838 return i_blocks << (inode->i_blkbits - 9);
4843 return le32_to_cpu(raw_inode->i_blocks_lo);
4847 static inline int ext4_iget_extra_inode(struct inode *inode,
4848 struct ext4_inode *raw_inode,
4849 struct ext4_inode_info *ei)
4851 __le32 *magic = (void *)raw_inode +
4852 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4854 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4855 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4858 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4859 err = ext4_find_inline_data_nolock(inode);
4860 if (!err && ext4_has_inline_data(inode))
4861 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4864 EXT4_I(inode)->i_inline_off = 0;
4868 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4870 if (!ext4_has_feature_project(inode->i_sb))
4872 *projid = EXT4_I(inode)->i_projid;
4877 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4878 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4881 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4883 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4884 inode_set_iversion_raw(inode, val);
4886 inode_set_iversion_queried(inode, val);
4888 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4890 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4891 return inode_peek_iversion_raw(inode);
4893 return inode_peek_iversion(inode);
4896 static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags)
4899 if (flags & EXT4_IGET_EA_INODE) {
4900 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4901 return "missing EA_INODE flag";
4902 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4903 EXT4_I(inode)->i_file_acl)
4904 return "ea_inode with extended attributes";
4906 if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4907 return "unexpected EA_INODE flag";
4909 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD))
4910 return "unexpected bad inode w/o EXT4_IGET_BAD";
4914 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4915 ext4_iget_flags flags, const char *function,
4918 struct ext4_iloc iloc;
4919 struct ext4_inode *raw_inode;
4920 struct ext4_inode_info *ei;
4921 struct inode *inode;
4922 const char *err_str;
4923 journal_t *journal = EXT4_SB(sb)->s_journal;
4931 if ((!(flags & EXT4_IGET_SPECIAL) &&
4932 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4933 (ino < EXT4_ROOT_INO) ||
4934 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4935 if (flags & EXT4_IGET_HANDLE)
4936 return ERR_PTR(-ESTALE);
4937 __ext4_error(sb, function, line,
4938 "inode #%lu: comm %s: iget: illegal inode #",
4939 ino, current->comm);
4940 return ERR_PTR(-EFSCORRUPTED);
4943 inode = iget_locked(sb, ino);
4945 return ERR_PTR(-ENOMEM);
4946 if (!(inode->i_state & I_NEW)) {
4947 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4948 ext4_error_inode(inode, function, line, 0, err_str);
4950 return ERR_PTR(-EFSCORRUPTED);
4958 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4961 raw_inode = ext4_raw_inode(&iloc);
4963 if ((flags & EXT4_IGET_HANDLE) &&
4964 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4969 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4970 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4971 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4972 EXT4_INODE_SIZE(inode->i_sb) ||
4973 (ei->i_extra_isize & 3)) {
4974 ext4_error_inode(inode, function, line, 0,
4975 "iget: bad extra_isize %u "
4978 EXT4_INODE_SIZE(inode->i_sb));
4979 ret = -EFSCORRUPTED;
4983 ei->i_extra_isize = 0;
4985 /* Precompute checksum seed for inode metadata */
4986 if (ext4_has_metadata_csum(sb)) {
4987 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4989 __le32 inum = cpu_to_le32(inode->i_ino);
4990 __le32 gen = raw_inode->i_generation;
4991 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4993 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4997 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4998 ext4_error_inode(inode, function, line, 0,
4999 "iget: checksum invalid");
5004 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5005 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5006 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5007 if (ext4_has_feature_project(sb) &&
5008 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5009 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5010 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
5012 i_projid = EXT4_DEF_PROJID;
5014 if (!(test_opt(inode->i_sb, NO_UID32))) {
5015 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5016 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5018 i_uid_write(inode, i_uid);
5019 i_gid_write(inode, i_gid);
5020 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
5021 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
5023 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
5024 ei->i_inline_off = 0;
5025 ei->i_dir_start_lookup = 0;
5026 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5027 /* We now have enough fields to check if the inode was active or not.
5028 * This is needed because nfsd might try to access dead inodes
5029 * the test is that same one that e2fsck uses
5030 * NeilBrown 1999oct15
5032 if (inode->i_nlink == 0) {
5033 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
5034 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
5035 ino != EXT4_BOOT_LOADER_INO) {
5036 /* this inode is deleted or unallocated */
5037 if (flags & EXT4_IGET_SPECIAL) {
5038 ext4_error_inode(inode, function, line, 0,
5039 "iget: special inode unallocated");
5040 ret = -EFSCORRUPTED;
5045 /* The only unlinked inodes we let through here have
5046 * valid i_mode and are being read by the orphan
5047 * recovery code: that's fine, we're about to complete
5048 * the process of deleting those.
5049 * OR it is the EXT4_BOOT_LOADER_INO which is
5050 * not initialized on a new filesystem. */
5052 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5053 ext4_set_inode_flags(inode);
5054 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5055 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5056 if (ext4_has_feature_64bit(sb))
5058 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5059 inode->i_size = ext4_isize(sb, raw_inode);
5060 if ((size = i_size_read(inode)) < 0) {
5061 ext4_error_inode(inode, function, line, 0,
5062 "iget: bad i_size value: %lld", size);
5063 ret = -EFSCORRUPTED;
5067 * If dir_index is not enabled but there's dir with INDEX flag set,
5068 * we'd normally treat htree data as empty space. But with metadata
5069 * checksumming that corrupts checksums so forbid that.
5071 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
5072 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
5073 ext4_error_inode(inode, function, line, 0,
5074 "iget: Dir with htree data on filesystem without dir_index feature.");
5075 ret = -EFSCORRUPTED;
5078 ei->i_disksize = inode->i_size;
5080 ei->i_reserved_quota = 0;
5082 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5083 ei->i_block_group = iloc.block_group;
5084 ei->i_last_alloc_group = ~0;
5086 * NOTE! The in-memory inode i_data array is in little-endian order
5087 * even on big-endian machines: we do NOT byteswap the block numbers!
5089 for (block = 0; block < EXT4_N_BLOCKS; block++)
5090 ei->i_data[block] = raw_inode->i_block[block];
5091 INIT_LIST_HEAD(&ei->i_orphan);
5094 * Set transaction id's of transactions that have to be committed
5095 * to finish f[data]sync. We set them to currently running transaction
5096 * as we cannot be sure that the inode or some of its metadata isn't
5097 * part of the transaction - the inode could have been reclaimed and
5098 * now it is reread from disk.
5101 transaction_t *transaction;
5104 read_lock(&journal->j_state_lock);
5105 if (journal->j_running_transaction)
5106 transaction = journal->j_running_transaction;
5108 transaction = journal->j_committing_transaction;
5110 tid = transaction->t_tid;
5112 tid = journal->j_commit_sequence;
5113 read_unlock(&journal->j_state_lock);
5114 ei->i_sync_tid = tid;
5115 ei->i_datasync_tid = tid;
5118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5119 if (ei->i_extra_isize == 0) {
5120 /* The extra space is currently unused. Use it. */
5121 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5122 ei->i_extra_isize = sizeof(struct ext4_inode) -
5123 EXT4_GOOD_OLD_INODE_SIZE;
5125 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5131 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5132 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5133 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5134 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5136 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5137 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5139 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5140 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5142 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5144 ext4_inode_set_iversion_queried(inode, ivers);
5148 if (ei->i_file_acl &&
5149 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5150 ext4_error_inode(inode, function, line, 0,
5151 "iget: bad extended attribute block %llu",
5153 ret = -EFSCORRUPTED;
5155 } else if (!ext4_has_inline_data(inode)) {
5156 /* validate the block references in the inode */
5157 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5158 (S_ISLNK(inode->i_mode) &&
5159 !ext4_inode_is_fast_symlink(inode))) {
5160 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5161 ret = ext4_ext_check_inode(inode);
5163 ret = ext4_ind_check_inode(inode);
5169 if (S_ISREG(inode->i_mode)) {
5170 inode->i_op = &ext4_file_inode_operations;
5171 inode->i_fop = &ext4_file_operations;
5172 ext4_set_aops(inode);
5173 } else if (S_ISDIR(inode->i_mode)) {
5174 inode->i_op = &ext4_dir_inode_operations;
5175 inode->i_fop = &ext4_dir_operations;
5176 } else if (S_ISLNK(inode->i_mode)) {
5177 /* VFS does not allow setting these so must be corruption */
5178 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5179 ext4_error_inode(inode, function, line, 0,
5180 "iget: immutable or append flags "
5181 "not allowed on symlinks");
5182 ret = -EFSCORRUPTED;
5185 if (IS_ENCRYPTED(inode)) {
5186 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5187 ext4_set_aops(inode);
5188 } else if (ext4_inode_is_fast_symlink(inode)) {
5189 inode->i_link = (char *)ei->i_data;
5190 inode->i_op = &ext4_fast_symlink_inode_operations;
5191 nd_terminate_link(ei->i_data, inode->i_size,
5192 sizeof(ei->i_data) - 1);
5194 inode->i_op = &ext4_symlink_inode_operations;
5195 ext4_set_aops(inode);
5197 inode_nohighmem(inode);
5198 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5199 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5200 inode->i_op = &ext4_special_inode_operations;
5201 if (raw_inode->i_block[0])
5202 init_special_inode(inode, inode->i_mode,
5203 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5205 init_special_inode(inode, inode->i_mode,
5206 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5207 } else if (ino == EXT4_BOOT_LOADER_INO) {
5208 make_bad_inode(inode);
5210 ret = -EFSCORRUPTED;
5211 ext4_error_inode(inode, function, line, 0,
5212 "iget: bogus i_mode (%o)", inode->i_mode);
5215 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
5216 ext4_error_inode(inode, function, line, 0,
5217 "casefold flag without casefold feature");
5218 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
5219 ext4_error_inode(inode, function, line, 0, err_str);
5220 ret = -EFSCORRUPTED;
5225 unlock_new_inode(inode);
5231 return ERR_PTR(ret);
5234 static int ext4_inode_blocks_set(handle_t *handle,
5235 struct ext4_inode *raw_inode,
5236 struct ext4_inode_info *ei)
5238 struct inode *inode = &(ei->vfs_inode);
5239 u64 i_blocks = READ_ONCE(inode->i_blocks);
5240 struct super_block *sb = inode->i_sb;
5242 if (i_blocks <= ~0U) {
5244 * i_blocks can be represented in a 32 bit variable
5245 * as multiple of 512 bytes
5247 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5248 raw_inode->i_blocks_high = 0;
5249 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5252 if (!ext4_has_feature_huge_file(sb))
5255 if (i_blocks <= 0xffffffffffffULL) {
5257 * i_blocks can be represented in a 48 bit variable
5258 * as multiple of 512 bytes
5260 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5261 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5262 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5264 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5265 /* i_block is stored in file system block size */
5266 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5267 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5268 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5273 struct other_inode {
5274 unsigned long orig_ino;
5275 struct ext4_inode *raw_inode;
5278 static int other_inode_match(struct inode * inode, unsigned long ino,
5281 struct other_inode *oi = (struct other_inode *) data;
5283 if ((inode->i_ino != ino) ||
5284 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5286 ((inode->i_state & I_DIRTY_TIME) == 0))
5288 spin_lock(&inode->i_lock);
5289 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5290 I_DIRTY_INODE)) == 0) &&
5291 (inode->i_state & I_DIRTY_TIME)) {
5292 struct ext4_inode_info *ei = EXT4_I(inode);
5294 inode->i_state &= ~I_DIRTY_TIME;
5295 spin_unlock(&inode->i_lock);
5297 spin_lock(&ei->i_raw_lock);
5298 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5299 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5300 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5301 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5302 spin_unlock(&ei->i_raw_lock);
5303 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5306 spin_unlock(&inode->i_lock);
5311 * Opportunistically update the other time fields for other inodes in
5312 * the same inode table block.
5314 static void ext4_update_other_inodes_time(struct super_block *sb,
5315 unsigned long orig_ino, char *buf)
5317 struct other_inode oi;
5319 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5320 int inode_size = EXT4_INODE_SIZE(sb);
5322 oi.orig_ino = orig_ino;
5324 * Calculate the first inode in the inode table block. Inode
5325 * numbers are one-based. That is, the first inode in a block
5326 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5328 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5329 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5330 if (ino == orig_ino)
5332 oi.raw_inode = (struct ext4_inode *) buf;
5333 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5338 * Post the struct inode info into an on-disk inode location in the
5339 * buffer-cache. This gobbles the caller's reference to the
5340 * buffer_head in the inode location struct.
5342 * The caller must have write access to iloc->bh.
5344 static int ext4_do_update_inode(handle_t *handle,
5345 struct inode *inode,
5346 struct ext4_iloc *iloc)
5348 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5349 struct ext4_inode_info *ei = EXT4_I(inode);
5350 struct buffer_head *bh = iloc->bh;
5351 struct super_block *sb = inode->i_sb;
5353 int need_datasync = 0, set_large_file = 0;
5358 spin_lock(&ei->i_raw_lock);
5360 /* For fields not tracked in the in-memory inode,
5361 * initialise them to zero for new inodes. */
5362 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5363 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5365 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5367 spin_unlock(&ei->i_raw_lock);
5371 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5372 i_uid = i_uid_read(inode);
5373 i_gid = i_gid_read(inode);
5374 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5375 if (!(test_opt(inode->i_sb, NO_UID32))) {
5376 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5377 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5379 * Fix up interoperability with old kernels. Otherwise, old inodes get
5380 * re-used with the upper 16 bits of the uid/gid intact
5382 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5383 raw_inode->i_uid_high = 0;
5384 raw_inode->i_gid_high = 0;
5386 raw_inode->i_uid_high =
5387 cpu_to_le16(high_16_bits(i_uid));
5388 raw_inode->i_gid_high =
5389 cpu_to_le16(high_16_bits(i_gid));
5392 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5393 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5394 raw_inode->i_uid_high = 0;
5395 raw_inode->i_gid_high = 0;
5397 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5399 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5400 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5401 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5402 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5404 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5405 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5406 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5407 raw_inode->i_file_acl_high =
5408 cpu_to_le16(ei->i_file_acl >> 32);
5409 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5410 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5411 ext4_isize_set(raw_inode, ei->i_disksize);
5414 if (ei->i_disksize > 0x7fffffffULL) {
5415 if (!ext4_has_feature_large_file(sb) ||
5416 EXT4_SB(sb)->s_es->s_rev_level ==
5417 cpu_to_le32(EXT4_GOOD_OLD_REV))
5420 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5421 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5422 if (old_valid_dev(inode->i_rdev)) {
5423 raw_inode->i_block[0] =
5424 cpu_to_le32(old_encode_dev(inode->i_rdev));
5425 raw_inode->i_block[1] = 0;
5427 raw_inode->i_block[0] = 0;
5428 raw_inode->i_block[1] =
5429 cpu_to_le32(new_encode_dev(inode->i_rdev));
5430 raw_inode->i_block[2] = 0;
5432 } else if (!ext4_has_inline_data(inode)) {
5433 for (block = 0; block < EXT4_N_BLOCKS; block++)
5434 raw_inode->i_block[block] = ei->i_data[block];
5437 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5438 u64 ivers = ext4_inode_peek_iversion(inode);
5440 raw_inode->i_disk_version = cpu_to_le32(ivers);
5441 if (ei->i_extra_isize) {
5442 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5443 raw_inode->i_version_hi =
5444 cpu_to_le32(ivers >> 32);
5445 raw_inode->i_extra_isize =
5446 cpu_to_le16(ei->i_extra_isize);
5450 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5451 i_projid != EXT4_DEF_PROJID);
5453 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5454 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5455 raw_inode->i_projid = cpu_to_le32(i_projid);
5457 ext4_inode_csum_set(inode, raw_inode, ei);
5458 spin_unlock(&ei->i_raw_lock);
5459 if (inode->i_sb->s_flags & SB_LAZYTIME)
5460 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5463 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5464 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5467 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5468 if (set_large_file) {
5469 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5470 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5473 ext4_set_feature_large_file(sb);
5474 ext4_handle_sync(handle);
5475 err = ext4_handle_dirty_super(handle, sb);
5477 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5480 ext4_std_error(inode->i_sb, err);
5485 * ext4_write_inode()
5487 * We are called from a few places:
5489 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5490 * Here, there will be no transaction running. We wait for any running
5491 * transaction to commit.
5493 * - Within flush work (sys_sync(), kupdate and such).
5494 * We wait on commit, if told to.
5496 * - Within iput_final() -> write_inode_now()
5497 * We wait on commit, if told to.
5499 * In all cases it is actually safe for us to return without doing anything,
5500 * because the inode has been copied into a raw inode buffer in
5501 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5504 * Note that we are absolutely dependent upon all inode dirtiers doing the
5505 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5506 * which we are interested.
5508 * It would be a bug for them to not do this. The code:
5510 * mark_inode_dirty(inode)
5512 * inode->i_size = expr;
5514 * is in error because write_inode() could occur while `stuff()' is running,
5515 * and the new i_size will be lost. Plus the inode will no longer be on the
5516 * superblock's dirty inode list.
5518 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5522 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5523 sb_rdonly(inode->i_sb))
5526 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5529 if (EXT4_SB(inode->i_sb)->s_journal) {
5530 if (ext4_journal_current_handle()) {
5531 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5537 * No need to force transaction in WB_SYNC_NONE mode. Also
5538 * ext4_sync_fs() will force the commit after everything is
5541 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5544 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5545 EXT4_I(inode)->i_sync_tid);
5547 struct ext4_iloc iloc;
5549 err = __ext4_get_inode_loc(inode, &iloc, 0);
5553 * sync(2) will flush the whole buffer cache. No need to do
5554 * it here separately for each inode.
5556 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5557 sync_dirty_buffer(iloc.bh);
5558 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5559 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5560 "IO error syncing inode");
5569 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5570 * buffers that are attached to a page stradding i_size and are undergoing
5571 * commit. In that case we have to wait for commit to finish and try again.
5573 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5577 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5578 tid_t commit_tid = 0;
5581 offset = inode->i_size & (PAGE_SIZE - 1);
5583 * If the page is fully truncated, we don't need to wait for any commit
5584 * (and we even should not as __ext4_journalled_invalidatepage() may
5585 * strip all buffers from the page but keep the page dirty which can then
5586 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5587 * buffers). Also we don't need to wait for any commit if all buffers in
5588 * the page remain valid. This is most beneficial for the common case of
5589 * blocksize == PAGESIZE.
5591 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5594 page = find_lock_page(inode->i_mapping,
5595 inode->i_size >> PAGE_SHIFT);
5598 ret = __ext4_journalled_invalidatepage(page, offset,
5599 PAGE_SIZE - offset);
5605 read_lock(&journal->j_state_lock);
5606 if (journal->j_committing_transaction)
5607 commit_tid = journal->j_committing_transaction->t_tid;
5608 read_unlock(&journal->j_state_lock);
5610 jbd2_log_wait_commit(journal, commit_tid);
5617 * Called from notify_change.
5619 * We want to trap VFS attempts to truncate the file as soon as
5620 * possible. In particular, we want to make sure that when the VFS
5621 * shrinks i_size, we put the inode on the orphan list and modify
5622 * i_disksize immediately, so that during the subsequent flushing of
5623 * dirty pages and freeing of disk blocks, we can guarantee that any
5624 * commit will leave the blocks being flushed in an unused state on
5625 * disk. (On recovery, the inode will get truncated and the blocks will
5626 * be freed, so we have a strong guarantee that no future commit will
5627 * leave these blocks visible to the user.)
5629 * Another thing we have to assure is that if we are in ordered mode
5630 * and inode is still attached to the committing transaction, we must
5631 * we start writeout of all the dirty pages which are being truncated.
5632 * This way we are sure that all the data written in the previous
5633 * transaction are already on disk (truncate waits for pages under
5636 * Called with inode->i_mutex down.
5638 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5640 struct inode *inode = d_inode(dentry);
5643 const unsigned int ia_valid = attr->ia_valid;
5645 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5648 if (unlikely(IS_IMMUTABLE(inode)))
5651 if (unlikely(IS_APPEND(inode) &&
5652 (ia_valid & (ATTR_MODE | ATTR_UID |
5653 ATTR_GID | ATTR_TIMES_SET))))
5656 error = setattr_prepare(dentry, attr);
5660 error = fscrypt_prepare_setattr(dentry, attr);
5664 error = fsverity_prepare_setattr(dentry, attr);
5668 if (is_quota_modification(inode, attr)) {
5669 error = dquot_initialize(inode);
5673 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5674 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5677 /* (user+group)*(old+new) structure, inode write (sb,
5678 * inode block, ? - but truncate inode update has it) */
5679 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5680 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5681 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5682 if (IS_ERR(handle)) {
5683 error = PTR_ERR(handle);
5687 /* dquot_transfer() calls back ext4_get_inode_usage() which
5688 * counts xattr inode references.
5690 down_read(&EXT4_I(inode)->xattr_sem);
5691 error = dquot_transfer(inode, attr);
5692 up_read(&EXT4_I(inode)->xattr_sem);
5695 ext4_journal_stop(handle);
5698 /* Update corresponding info in inode so that everything is in
5699 * one transaction */
5700 if (attr->ia_valid & ATTR_UID)
5701 inode->i_uid = attr->ia_uid;
5702 if (attr->ia_valid & ATTR_GID)
5703 inode->i_gid = attr->ia_gid;
5704 error = ext4_mark_inode_dirty(handle, inode);
5705 ext4_journal_stop(handle);
5708 if (attr->ia_valid & ATTR_SIZE) {
5710 loff_t oldsize = inode->i_size;
5711 loff_t old_disksize;
5712 int shrink = (attr->ia_size < inode->i_size);
5714 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5715 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5717 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5720 if (!S_ISREG(inode->i_mode))
5723 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5724 inode_inc_iversion(inode);
5727 if (ext4_should_order_data(inode)) {
5728 error = ext4_begin_ordered_truncate(inode,
5734 * Blocks are going to be removed from the inode. Wait
5735 * for dio in flight.
5737 inode_dio_wait(inode);
5740 down_write(&EXT4_I(inode)->i_mmap_sem);
5742 rc = ext4_break_layouts(inode);
5744 up_write(&EXT4_I(inode)->i_mmap_sem);
5748 if (attr->ia_size != inode->i_size) {
5749 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5750 if (IS_ERR(handle)) {
5751 error = PTR_ERR(handle);
5754 if (ext4_handle_valid(handle) && shrink) {
5755 error = ext4_orphan_add(handle, inode);
5759 * Update c/mtime on truncate up, ext4_truncate() will
5760 * update c/mtime in shrink case below
5763 inode->i_mtime = current_time(inode);
5764 inode->i_ctime = inode->i_mtime;
5766 down_write(&EXT4_I(inode)->i_data_sem);
5767 old_disksize = EXT4_I(inode)->i_disksize;
5768 EXT4_I(inode)->i_disksize = attr->ia_size;
5769 rc = ext4_mark_inode_dirty(handle, inode);
5773 * We have to update i_size under i_data_sem together
5774 * with i_disksize to avoid races with writeback code
5775 * running ext4_wb_update_i_disksize().
5778 i_size_write(inode, attr->ia_size);
5780 EXT4_I(inode)->i_disksize = old_disksize;
5781 up_write(&EXT4_I(inode)->i_data_sem);
5782 ext4_journal_stop(handle);
5786 pagecache_isize_extended(inode, oldsize,
5788 } else if (ext4_should_journal_data(inode)) {
5789 ext4_wait_for_tail_page_commit(inode);
5794 * Truncate pagecache after we've waited for commit
5795 * in data=journal mode to make pages freeable.
5797 truncate_pagecache(inode, inode->i_size);
5799 * Call ext4_truncate() even if i_size didn't change to
5800 * truncate possible preallocated blocks.
5802 if (attr->ia_size <= oldsize) {
5803 rc = ext4_truncate(inode);
5808 up_write(&EXT4_I(inode)->i_mmap_sem);
5812 setattr_copy(inode, attr);
5813 mark_inode_dirty(inode);
5817 * If the call to ext4_truncate failed to get a transaction handle at
5818 * all, we need to clean up the in-core orphan list manually.
5820 if (orphan && inode->i_nlink)
5821 ext4_orphan_del(NULL, inode);
5823 if (!error && (ia_valid & ATTR_MODE))
5824 rc = posix_acl_chmod(inode, inode->i_mode);
5827 ext4_std_error(inode->i_sb, error);
5833 int ext4_getattr(const struct path *path, struct kstat *stat,
5834 u32 request_mask, unsigned int query_flags)
5836 struct inode *inode = d_inode(path->dentry);
5837 struct ext4_inode *raw_inode;
5838 struct ext4_inode_info *ei = EXT4_I(inode);
5841 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5842 stat->result_mask |= STATX_BTIME;
5843 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5844 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5847 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5848 if (flags & EXT4_APPEND_FL)
5849 stat->attributes |= STATX_ATTR_APPEND;
5850 if (flags & EXT4_COMPR_FL)
5851 stat->attributes |= STATX_ATTR_COMPRESSED;
5852 if (flags & EXT4_ENCRYPT_FL)
5853 stat->attributes |= STATX_ATTR_ENCRYPTED;
5854 if (flags & EXT4_IMMUTABLE_FL)
5855 stat->attributes |= STATX_ATTR_IMMUTABLE;
5856 if (flags & EXT4_NODUMP_FL)
5857 stat->attributes |= STATX_ATTR_NODUMP;
5859 stat->attributes_mask |= (STATX_ATTR_APPEND |
5860 STATX_ATTR_COMPRESSED |
5861 STATX_ATTR_ENCRYPTED |
5862 STATX_ATTR_IMMUTABLE |
5865 generic_fillattr(inode, stat);
5869 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5870 u32 request_mask, unsigned int query_flags)
5872 struct inode *inode = d_inode(path->dentry);
5873 u64 delalloc_blocks;
5875 ext4_getattr(path, stat, request_mask, query_flags);
5878 * If there is inline data in the inode, the inode will normally not
5879 * have data blocks allocated (it may have an external xattr block).
5880 * Report at least one sector for such files, so tools like tar, rsync,
5881 * others don't incorrectly think the file is completely sparse.
5883 if (unlikely(ext4_has_inline_data(inode)))
5884 stat->blocks += (stat->size + 511) >> 9;
5887 * We can't update i_blocks if the block allocation is delayed
5888 * otherwise in the case of system crash before the real block
5889 * allocation is done, we will have i_blocks inconsistent with
5890 * on-disk file blocks.
5891 * We always keep i_blocks updated together with real
5892 * allocation. But to not confuse with user, stat
5893 * will return the blocks that include the delayed allocation
5894 * blocks for this file.
5896 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5897 EXT4_I(inode)->i_reserved_data_blocks);
5898 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5902 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5905 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5906 return ext4_ind_trans_blocks(inode, lblocks);
5907 return ext4_ext_index_trans_blocks(inode, pextents);
5911 * Account for index blocks, block groups bitmaps and block group
5912 * descriptor blocks if modify datablocks and index blocks
5913 * worse case, the indexs blocks spread over different block groups
5915 * If datablocks are discontiguous, they are possible to spread over
5916 * different block groups too. If they are contiguous, with flexbg,
5917 * they could still across block group boundary.
5919 * Also account for superblock, inode, quota and xattr blocks
5921 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5924 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5930 * How many index blocks need to touch to map @lblocks logical blocks
5931 * to @pextents physical extents?
5933 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5938 * Now let's see how many group bitmaps and group descriptors need
5941 groups = idxblocks + pextents;
5943 if (groups > ngroups)
5945 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5946 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5948 /* bitmaps and block group descriptor blocks */
5949 ret += groups + gdpblocks;
5951 /* Blocks for super block, inode, quota and xattr blocks */
5952 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5958 * Calculate the total number of credits to reserve to fit
5959 * the modification of a single pages into a single transaction,
5960 * which may include multiple chunks of block allocations.
5962 * This could be called via ext4_write_begin()
5964 * We need to consider the worse case, when
5965 * one new block per extent.
5967 int ext4_writepage_trans_blocks(struct inode *inode)
5969 int bpp = ext4_journal_blocks_per_page(inode);
5972 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5974 /* Account for data blocks for journalled mode */
5975 if (ext4_should_journal_data(inode))
5981 * Calculate the journal credits for a chunk of data modification.
5983 * This is called from DIO, fallocate or whoever calling
5984 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5986 * journal buffers for data blocks are not included here, as DIO
5987 * and fallocate do no need to journal data buffers.
5989 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5991 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5995 * The caller must have previously called ext4_reserve_inode_write().
5996 * Give this, we know that the caller already has write access to iloc->bh.
5998 int ext4_mark_iloc_dirty(handle_t *handle,
5999 struct inode *inode, struct ext4_iloc *iloc)
6003 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
6007 if (IS_I_VERSION(inode))
6008 inode_inc_iversion(inode);
6010 /* the do_update_inode consumes one bh->b_count */
6013 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
6014 err = ext4_do_update_inode(handle, inode, iloc);
6020 * On success, We end up with an outstanding reference count against
6021 * iloc->bh. This _must_ be cleaned up later.
6025 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
6026 struct ext4_iloc *iloc)
6030 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
6033 err = ext4_get_inode_loc(inode, iloc);
6035 BUFFER_TRACE(iloc->bh, "get_write_access");
6036 err = ext4_journal_get_write_access(handle, iloc->bh);
6042 ext4_std_error(inode->i_sb, err);
6046 static int __ext4_expand_extra_isize(struct inode *inode,
6047 unsigned int new_extra_isize,
6048 struct ext4_iloc *iloc,
6049 handle_t *handle, int *no_expand)
6051 struct ext4_inode *raw_inode;
6052 struct ext4_xattr_ibody_header *header;
6053 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
6054 struct ext4_inode_info *ei = EXT4_I(inode);
6057 /* this was checked at iget time, but double check for good measure */
6058 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
6059 (ei->i_extra_isize & 3)) {
6060 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
6062 EXT4_INODE_SIZE(inode->i_sb));
6063 return -EFSCORRUPTED;
6065 if ((new_extra_isize < ei->i_extra_isize) ||
6066 (new_extra_isize < 4) ||
6067 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
6068 return -EINVAL; /* Should never happen */
6070 raw_inode = ext4_raw_inode(iloc);
6072 header = IHDR(inode, raw_inode);
6074 /* No extended attributes present */
6075 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
6076 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
6077 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
6078 EXT4_I(inode)->i_extra_isize, 0,
6079 new_extra_isize - EXT4_I(inode)->i_extra_isize);
6080 EXT4_I(inode)->i_extra_isize = new_extra_isize;
6085 * We may need to allocate external xattr block so we need quotas
6086 * initialized. Here we can be called with various locks held so we
6087 * cannot affort to initialize quotas ourselves. So just bail.
6089 if (dquot_initialize_needed(inode))
6092 /* try to expand with EAs present */
6093 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
6097 * Inode size expansion failed; don't try again
6106 * Expand an inode by new_extra_isize bytes.
6107 * Returns 0 on success or negative error number on failure.
6109 static int ext4_try_to_expand_extra_isize(struct inode *inode,
6110 unsigned int new_extra_isize,
6111 struct ext4_iloc iloc,
6117 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
6121 * In nojournal mode, we can immediately attempt to expand
6122 * the inode. When journaled, we first need to obtain extra
6123 * buffer credits since we may write into the EA block
6124 * with this same handle. If journal_extend fails, then it will
6125 * only result in a minor loss of functionality for that inode.
6126 * If this is felt to be critical, then e2fsck should be run to
6127 * force a large enough s_min_extra_isize.
6129 if (ext4_handle_valid(handle) &&
6130 jbd2_journal_extend(handle,
6131 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
6134 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
6137 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
6138 handle, &no_expand);
6139 ext4_write_unlock_xattr(inode, &no_expand);
6144 int ext4_expand_extra_isize(struct inode *inode,
6145 unsigned int new_extra_isize,
6146 struct ext4_iloc *iloc)
6152 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
6157 handle = ext4_journal_start(inode, EXT4_HT_INODE,
6158 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6159 if (IS_ERR(handle)) {
6160 error = PTR_ERR(handle);
6165 ext4_write_lock_xattr(inode, &no_expand);
6167 BUFFER_TRACE(iloc->bh, "get_write_access");
6168 error = ext4_journal_get_write_access(handle, iloc->bh);
6174 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6175 handle, &no_expand);
6177 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6182 ext4_write_unlock_xattr(inode, &no_expand);
6183 ext4_journal_stop(handle);
6188 * What we do here is to mark the in-core inode as clean with respect to inode
6189 * dirtiness (it may still be data-dirty).
6190 * This means that the in-core inode may be reaped by prune_icache
6191 * without having to perform any I/O. This is a very good thing,
6192 * because *any* task may call prune_icache - even ones which
6193 * have a transaction open against a different journal.
6195 * Is this cheating? Not really. Sure, we haven't written the
6196 * inode out, but prune_icache isn't a user-visible syncing function.
6197 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6198 * we start and wait on commits.
6200 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6202 struct ext4_iloc iloc;
6203 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6207 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6208 err = ext4_reserve_inode_write(handle, inode, &iloc);
6212 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6213 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6216 return ext4_mark_iloc_dirty(handle, inode, &iloc);
6220 * ext4_dirty_inode() is called from __mark_inode_dirty()
6222 * We're really interested in the case where a file is being extended.
6223 * i_size has been changed by generic_commit_write() and we thus need
6224 * to include the updated inode in the current transaction.
6226 * Also, dquot_alloc_block() will always dirty the inode when blocks
6227 * are allocated to the file.
6229 * If the inode is marked synchronous, we don't honour that here - doing
6230 * so would cause a commit on atime updates, which we don't bother doing.
6231 * We handle synchronous inodes at the highest possible level.
6233 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6234 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6235 * to copy into the on-disk inode structure are the timestamp files.
6237 void ext4_dirty_inode(struct inode *inode, int flags)
6241 if (flags == I_DIRTY_TIME)
6243 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6247 ext4_mark_inode_dirty(handle, inode);
6249 ext4_journal_stop(handle);
6254 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6259 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6262 * We have to be very careful here: changing a data block's
6263 * journaling status dynamically is dangerous. If we write a
6264 * data block to the journal, change the status and then delete
6265 * that block, we risk forgetting to revoke the old log record
6266 * from the journal and so a subsequent replay can corrupt data.
6267 * So, first we make sure that the journal is empty and that
6268 * nobody is changing anything.
6271 journal = EXT4_JOURNAL(inode);
6274 if (is_journal_aborted(journal))
6277 /* Wait for all existing dio workers */
6278 inode_dio_wait(inode);
6281 * Before flushing the journal and switching inode's aops, we have
6282 * to flush all dirty data the inode has. There can be outstanding
6283 * delayed allocations, there can be unwritten extents created by
6284 * fallocate or buffered writes in dioread_nolock mode covered by
6285 * dirty data which can be converted only after flushing the dirty
6286 * data (and journalled aops don't know how to handle these cases).
6289 down_write(&EXT4_I(inode)->i_mmap_sem);
6290 err = filemap_write_and_wait(inode->i_mapping);
6292 up_write(&EXT4_I(inode)->i_mmap_sem);
6297 percpu_down_write(&sbi->s_writepages_rwsem);
6298 jbd2_journal_lock_updates(journal);
6301 * OK, there are no updates running now, and all cached data is
6302 * synced to disk. We are now in a completely consistent state
6303 * which doesn't have anything in the journal, and we know that
6304 * no filesystem updates are running, so it is safe to modify
6305 * the inode's in-core data-journaling state flag now.
6309 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6311 err = jbd2_journal_flush(journal);
6313 jbd2_journal_unlock_updates(journal);
6314 percpu_up_write(&sbi->s_writepages_rwsem);
6317 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6319 ext4_set_aops(inode);
6321 jbd2_journal_unlock_updates(journal);
6322 percpu_up_write(&sbi->s_writepages_rwsem);
6325 up_write(&EXT4_I(inode)->i_mmap_sem);
6327 /* Finally we can mark the inode as dirty. */
6329 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6331 return PTR_ERR(handle);
6333 err = ext4_mark_inode_dirty(handle, inode);
6334 ext4_handle_sync(handle);
6335 ext4_journal_stop(handle);
6336 ext4_std_error(inode->i_sb, err);
6341 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6343 return !buffer_mapped(bh);
6346 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6348 struct vm_area_struct *vma = vmf->vma;
6349 struct page *page = vmf->page;
6354 struct file *file = vma->vm_file;
6355 struct inode *inode = file_inode(file);
6356 struct address_space *mapping = inode->i_mapping;
6358 get_block_t *get_block;
6361 if (unlikely(IS_IMMUTABLE(inode)))
6362 return VM_FAULT_SIGBUS;
6364 sb_start_pagefault(inode->i_sb);
6365 file_update_time(vma->vm_file);
6367 down_read(&EXT4_I(inode)->i_mmap_sem);
6369 err = ext4_convert_inline_data(inode);
6373 /* Delalloc case is easy... */
6374 if (test_opt(inode->i_sb, DELALLOC) &&
6375 !ext4_should_journal_data(inode) &&
6376 !ext4_nonda_switch(inode->i_sb)) {
6378 err = block_page_mkwrite(vma, vmf,
6379 ext4_da_get_block_prep);
6380 } while (err == -ENOSPC &&
6381 ext4_should_retry_alloc(inode->i_sb, &retries));
6386 size = i_size_read(inode);
6387 /* Page got truncated from under us? */
6388 if (page->mapping != mapping || page_offset(page) > size) {
6390 ret = VM_FAULT_NOPAGE;
6394 if (page->index == size >> PAGE_SHIFT)
6395 len = size & ~PAGE_MASK;
6399 * Return if we have all the buffers mapped. This avoids the need to do
6400 * journal_start/journal_stop which can block and take a long time
6402 if (page_has_buffers(page)) {
6403 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6405 ext4_bh_unmapped)) {
6406 /* Wait so that we don't change page under IO */
6407 wait_for_stable_page(page);
6408 ret = VM_FAULT_LOCKED;
6413 /* OK, we need to fill the hole... */
6414 if (ext4_should_dioread_nolock(inode))
6415 get_block = ext4_get_block_unwritten;
6417 get_block = ext4_get_block;
6419 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6420 ext4_writepage_trans_blocks(inode));
6421 if (IS_ERR(handle)) {
6422 ret = VM_FAULT_SIGBUS;
6425 err = block_page_mkwrite(vma, vmf, get_block);
6426 if (!err && ext4_should_journal_data(inode)) {
6427 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6428 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6430 ret = VM_FAULT_SIGBUS;
6431 ext4_journal_stop(handle);
6434 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6436 ext4_journal_stop(handle);
6437 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6440 ret = block_page_mkwrite_return(err);
6442 up_read(&EXT4_I(inode)->i_mmap_sem);
6443 sb_end_pagefault(inode->i_sb);
6447 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6449 struct inode *inode = file_inode(vmf->vma->vm_file);
6452 down_read(&EXT4_I(inode)->i_mmap_sem);
6453 ret = filemap_fault(vmf);
6454 up_read(&EXT4_I(inode)->i_mmap_sem);