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)
199 int extra_credits = 3;
200 struct ext4_xattr_inode_array *ea_inode_array = NULL;
202 trace_ext4_evict_inode(inode);
204 if (inode->i_nlink) {
206 * When journalling data dirty buffers are tracked only in the
207 * journal. So although mm thinks everything is clean and
208 * ready for reaping the inode might still have some pages to
209 * write in the running transaction or waiting to be
210 * checkpointed. Thus calling jbd2_journal_invalidatepage()
211 * (via truncate_inode_pages()) to discard these buffers can
212 * cause data loss. Also even if we did not discard these
213 * buffers, we would have no way to find them after the inode
214 * is reaped and thus user could see stale data if he tries to
215 * read them before the transaction is checkpointed. So be
216 * careful and force everything to disk here... We use
217 * ei->i_datasync_tid to store the newest transaction
218 * containing inode's data.
220 * Note that directories do not have this problem because they
221 * don't use page cache.
223 if (inode->i_ino != EXT4_JOURNAL_INO &&
224 ext4_should_journal_data(inode) &&
225 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
226 inode->i_data.nrpages) {
227 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
228 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
230 jbd2_complete_transaction(journal, commit_tid);
231 filemap_write_and_wait(&inode->i_data);
233 truncate_inode_pages_final(&inode->i_data);
238 if (is_bad_inode(inode))
240 dquot_initialize(inode);
242 if (ext4_should_order_data(inode))
243 ext4_begin_ordered_truncate(inode, 0);
244 truncate_inode_pages_final(&inode->i_data);
247 * Protect us against freezing - iput() caller didn't have to have any
248 * protection against it
250 sb_start_intwrite(inode->i_sb);
252 if (!IS_NOQUOTA(inode))
253 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
255 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
256 ext4_blocks_for_truncate(inode)+extra_credits);
257 if (IS_ERR(handle)) {
258 ext4_std_error(inode->i_sb, PTR_ERR(handle));
260 * If we're going to skip the normal cleanup, we still need to
261 * make sure that the in-core orphan linked list is properly
264 ext4_orphan_del(NULL, inode);
265 sb_end_intwrite(inode->i_sb);
270 ext4_handle_sync(handle);
273 * Set inode->i_size to 0 before calling ext4_truncate(). We need
274 * special handling of symlinks here because i_size is used to
275 * determine whether ext4_inode_info->i_data contains symlink data or
276 * block mappings. Setting i_size to 0 will remove its fast symlink
277 * status. Erase i_data so that it becomes a valid empty block map.
279 if (ext4_inode_is_fast_symlink(inode))
280 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
282 err = ext4_mark_inode_dirty(handle, inode);
284 ext4_warning(inode->i_sb,
285 "couldn't mark inode dirty (err %d)", err);
288 if (inode->i_blocks) {
289 err = ext4_truncate(inode);
291 ext4_error(inode->i_sb,
292 "couldn't truncate inode %lu (err %d)",
298 /* Remove xattr references. */
299 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
302 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
304 ext4_journal_stop(handle);
305 ext4_orphan_del(NULL, inode);
306 sb_end_intwrite(inode->i_sb);
307 ext4_xattr_inode_array_free(ea_inode_array);
312 * Kill off the orphan record which ext4_truncate created.
313 * AKPM: I think this can be inside the above `if'.
314 * Note that ext4_orphan_del() has to be able to cope with the
315 * deletion of a non-existent orphan - this is because we don't
316 * know if ext4_truncate() actually created an orphan record.
317 * (Well, we could do this if we need to, but heck - it works)
319 ext4_orphan_del(handle, inode);
320 EXT4_I(inode)->i_dtime = get_seconds();
323 * One subtle ordering requirement: if anything has gone wrong
324 * (transaction abort, IO errors, whatever), then we can still
325 * do these next steps (the fs will already have been marked as
326 * having errors), but we can't free the inode if the mark_dirty
329 if (ext4_mark_inode_dirty(handle, inode))
330 /* If that failed, just do the required in-core inode clear. */
331 ext4_clear_inode(inode);
333 ext4_free_inode(handle, inode);
334 ext4_journal_stop(handle);
335 sb_end_intwrite(inode->i_sb);
336 ext4_xattr_inode_array_free(ea_inode_array);
339 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
343 qsize_t *ext4_get_reserved_space(struct inode *inode)
345 return &EXT4_I(inode)->i_reserved_quota;
350 * Called with i_data_sem down, which is important since we can call
351 * ext4_discard_preallocations() from here.
353 void ext4_da_update_reserve_space(struct inode *inode,
354 int used, int quota_claim)
356 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
357 struct ext4_inode_info *ei = EXT4_I(inode);
359 spin_lock(&ei->i_block_reservation_lock);
360 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
361 if (unlikely(used > ei->i_reserved_data_blocks)) {
362 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
363 "with only %d reserved data blocks",
364 __func__, inode->i_ino, used,
365 ei->i_reserved_data_blocks);
367 used = ei->i_reserved_data_blocks;
370 /* Update per-inode reservations */
371 ei->i_reserved_data_blocks -= used;
372 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
374 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
376 /* Update quota subsystem for data blocks */
378 dquot_claim_block(inode, EXT4_C2B(sbi, used));
381 * We did fallocate with an offset that is already delayed
382 * allocated. So on delayed allocated writeback we should
383 * not re-claim the quota for fallocated blocks.
385 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
389 * If we have done all the pending block allocations and if
390 * there aren't any writers on the inode, we can discard the
391 * inode's preallocations.
393 if ((ei->i_reserved_data_blocks == 0) &&
394 (atomic_read(&inode->i_writecount) == 0))
395 ext4_discard_preallocations(inode);
398 static int __check_block_validity(struct inode *inode, const char *func,
400 struct ext4_map_blocks *map)
402 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
404 ext4_error_inode(inode, func, line, map->m_pblk,
405 "lblock %lu mapped to illegal pblock "
406 "(length %d)", (unsigned long) map->m_lblk,
408 return -EFSCORRUPTED;
413 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
418 if (ext4_encrypted_inode(inode))
419 return fscrypt_zeroout_range(inode, lblk, pblk, len);
421 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
428 #define check_block_validity(inode, map) \
429 __check_block_validity((inode), __func__, __LINE__, (map))
431 #ifdef ES_AGGRESSIVE_TEST
432 static void ext4_map_blocks_es_recheck(handle_t *handle,
434 struct ext4_map_blocks *es_map,
435 struct ext4_map_blocks *map,
442 * There is a race window that the result is not the same.
443 * e.g. xfstests #223 when dioread_nolock enables. The reason
444 * is that we lookup a block mapping in extent status tree with
445 * out taking i_data_sem. So at the time the unwritten extent
446 * could be converted.
448 down_read(&EXT4_I(inode)->i_data_sem);
449 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
450 retval = ext4_ext_map_blocks(handle, inode, map, flags &
451 EXT4_GET_BLOCKS_KEEP_SIZE);
453 retval = ext4_ind_map_blocks(handle, inode, map, flags &
454 EXT4_GET_BLOCKS_KEEP_SIZE);
456 up_read((&EXT4_I(inode)->i_data_sem));
459 * We don't check m_len because extent will be collpased in status
460 * tree. So the m_len might not equal.
462 if (es_map->m_lblk != map->m_lblk ||
463 es_map->m_flags != map->m_flags ||
464 es_map->m_pblk != map->m_pblk) {
465 printk("ES cache assertion failed for inode: %lu "
466 "es_cached ex [%d/%d/%llu/%x] != "
467 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
468 inode->i_ino, es_map->m_lblk, es_map->m_len,
469 es_map->m_pblk, es_map->m_flags, map->m_lblk,
470 map->m_len, map->m_pblk, map->m_flags,
474 #endif /* ES_AGGRESSIVE_TEST */
477 * The ext4_map_blocks() function tries to look up the requested blocks,
478 * and returns if the blocks are already mapped.
480 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
481 * and store the allocated blocks in the result buffer head and mark it
484 * If file type is extents based, it will call ext4_ext_map_blocks(),
485 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
488 * On success, it returns the number of blocks being mapped or allocated. if
489 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
490 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
492 * It returns 0 if plain look up failed (blocks have not been allocated), in
493 * that case, @map is returned as unmapped but we still do fill map->m_len to
494 * indicate the length of a hole starting at map->m_lblk.
496 * It returns the error in case of allocation failure.
498 int ext4_map_blocks(handle_t *handle, struct inode *inode,
499 struct ext4_map_blocks *map, int flags)
501 struct extent_status es;
504 #ifdef ES_AGGRESSIVE_TEST
505 struct ext4_map_blocks orig_map;
507 memcpy(&orig_map, map, sizeof(*map));
511 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
512 "logical block %lu\n", inode->i_ino, flags, map->m_len,
513 (unsigned long) map->m_lblk);
516 * ext4_map_blocks returns an int, and m_len is an unsigned int
518 if (unlikely(map->m_len > INT_MAX))
519 map->m_len = INT_MAX;
521 /* We can handle the block number less than EXT_MAX_BLOCKS */
522 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
523 return -EFSCORRUPTED;
525 /* Lookup extent status tree firstly */
526 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
527 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
528 map->m_pblk = ext4_es_pblock(&es) +
529 map->m_lblk - es.es_lblk;
530 map->m_flags |= ext4_es_is_written(&es) ?
531 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
532 retval = es.es_len - (map->m_lblk - es.es_lblk);
533 if (retval > map->m_len)
536 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
538 retval = es.es_len - (map->m_lblk - es.es_lblk);
539 if (retval > map->m_len)
546 #ifdef ES_AGGRESSIVE_TEST
547 ext4_map_blocks_es_recheck(handle, inode, map,
554 * Try to see if we can get the block without requesting a new
557 down_read(&EXT4_I(inode)->i_data_sem);
558 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
559 retval = ext4_ext_map_blocks(handle, inode, map, flags &
560 EXT4_GET_BLOCKS_KEEP_SIZE);
562 retval = ext4_ind_map_blocks(handle, inode, map, flags &
563 EXT4_GET_BLOCKS_KEEP_SIZE);
568 if (unlikely(retval != map->m_len)) {
569 ext4_warning(inode->i_sb,
570 "ES len assertion failed for inode "
571 "%lu: retval %d != map->m_len %d",
572 inode->i_ino, retval, map->m_len);
576 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
577 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
578 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
579 !(status & EXTENT_STATUS_WRITTEN) &&
580 ext4_find_delalloc_range(inode, map->m_lblk,
581 map->m_lblk + map->m_len - 1))
582 status |= EXTENT_STATUS_DELAYED;
583 ret = ext4_es_insert_extent(inode, map->m_lblk,
584 map->m_len, map->m_pblk, status);
588 up_read((&EXT4_I(inode)->i_data_sem));
591 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
592 ret = check_block_validity(inode, map);
597 /* If it is only a block(s) look up */
598 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
602 * Returns if the blocks have already allocated
604 * Note that if blocks have been preallocated
605 * ext4_ext_get_block() returns the create = 0
606 * with buffer head unmapped.
608 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
610 * If we need to convert extent to unwritten
611 * we continue and do the actual work in
612 * ext4_ext_map_blocks()
614 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
618 * Here we clear m_flags because after allocating an new extent,
619 * it will be set again.
621 map->m_flags &= ~EXT4_MAP_FLAGS;
624 * New blocks allocate and/or writing to unwritten extent
625 * will possibly result in updating i_data, so we take
626 * the write lock of i_data_sem, and call get_block()
627 * with create == 1 flag.
629 down_write(&EXT4_I(inode)->i_data_sem);
632 * We need to check for EXT4 here because migrate
633 * could have changed the inode type in between
635 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
636 retval = ext4_ext_map_blocks(handle, inode, map, flags);
638 retval = ext4_ind_map_blocks(handle, inode, map, flags);
640 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
642 * We allocated new blocks which will result in
643 * i_data's format changing. Force the migrate
644 * to fail by clearing migrate flags
646 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
650 * Update reserved blocks/metadata blocks after successful
651 * block allocation which had been deferred till now. We don't
652 * support fallocate for non extent files. So we can update
653 * reserve space here.
656 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
657 ext4_da_update_reserve_space(inode, retval, 1);
663 if (unlikely(retval != map->m_len)) {
664 ext4_warning(inode->i_sb,
665 "ES len assertion failed for inode "
666 "%lu: retval %d != map->m_len %d",
667 inode->i_ino, retval, map->m_len);
672 * We have to zeroout blocks before inserting them into extent
673 * status tree. Otherwise someone could look them up there and
674 * use them before they are really zeroed. We also have to
675 * unmap metadata before zeroing as otherwise writeback can
676 * overwrite zeros with stale data from block device.
678 if (flags & EXT4_GET_BLOCKS_ZERO &&
679 map->m_flags & EXT4_MAP_MAPPED &&
680 map->m_flags & EXT4_MAP_NEW) {
681 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
683 ret = ext4_issue_zeroout(inode, map->m_lblk,
684 map->m_pblk, map->m_len);
692 * If the extent has been zeroed out, we don't need to update
693 * extent status tree.
695 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
696 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
697 if (ext4_es_is_written(&es))
700 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
701 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
702 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
703 !(status & EXTENT_STATUS_WRITTEN) &&
704 ext4_find_delalloc_range(inode, map->m_lblk,
705 map->m_lblk + map->m_len - 1))
706 status |= EXTENT_STATUS_DELAYED;
707 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
708 map->m_pblk, status);
716 up_write((&EXT4_I(inode)->i_data_sem));
717 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
718 ret = check_block_validity(inode, map);
723 * Inodes with freshly allocated blocks where contents will be
724 * visible after transaction commit must be on transaction's
727 if (map->m_flags & EXT4_MAP_NEW &&
728 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
729 !(flags & EXT4_GET_BLOCKS_ZERO) &&
730 !ext4_is_quota_file(inode) &&
731 ext4_should_order_data(inode)) {
732 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
733 ret = ext4_jbd2_inode_add_wait(handle, inode);
735 ret = ext4_jbd2_inode_add_write(handle, inode);
744 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
745 * we have to be careful as someone else may be manipulating b_state as well.
747 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
749 unsigned long old_state;
750 unsigned long new_state;
752 flags &= EXT4_MAP_FLAGS;
754 /* Dummy buffer_head? Set non-atomically. */
756 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
760 * Someone else may be modifying b_state. Be careful! This is ugly but
761 * once we get rid of using bh as a container for mapping information
762 * to pass to / from get_block functions, this can go away.
765 old_state = READ_ONCE(bh->b_state);
766 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
768 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
771 static int _ext4_get_block(struct inode *inode, sector_t iblock,
772 struct buffer_head *bh, int flags)
774 struct ext4_map_blocks map;
777 if (ext4_has_inline_data(inode))
781 map.m_len = bh->b_size >> inode->i_blkbits;
783 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
786 map_bh(bh, inode->i_sb, map.m_pblk);
787 ext4_update_bh_state(bh, map.m_flags);
788 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
790 } else if (ret == 0) {
791 /* hole case, need to fill in bh->b_size */
792 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
797 int ext4_get_block(struct inode *inode, sector_t iblock,
798 struct buffer_head *bh, int create)
800 return _ext4_get_block(inode, iblock, bh,
801 create ? EXT4_GET_BLOCKS_CREATE : 0);
805 * Get block function used when preparing for buffered write if we require
806 * creating an unwritten extent if blocks haven't been allocated. The extent
807 * will be converted to written after the IO is complete.
809 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
810 struct buffer_head *bh_result, int create)
812 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
813 inode->i_ino, create);
814 return _ext4_get_block(inode, iblock, bh_result,
815 EXT4_GET_BLOCKS_IO_CREATE_EXT);
818 /* Maximum number of blocks we map for direct IO at once. */
819 #define DIO_MAX_BLOCKS 4096
822 * Get blocks function for the cases that need to start a transaction -
823 * generally difference cases of direct IO and DAX IO. It also handles retries
826 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
827 struct buffer_head *bh_result, int flags)
834 /* Trim mapping request to maximum we can map at once for DIO */
835 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
836 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
837 dio_credits = ext4_chunk_trans_blocks(inode,
838 bh_result->b_size >> inode->i_blkbits);
840 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
842 return PTR_ERR(handle);
844 ret = _ext4_get_block(inode, iblock, bh_result, flags);
845 ext4_journal_stop(handle);
847 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
852 /* Get block function for DIO reads and writes to inodes without extents */
853 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
854 struct buffer_head *bh, int create)
856 /* We don't expect handle for direct IO */
857 WARN_ON_ONCE(ext4_journal_current_handle());
860 return _ext4_get_block(inode, iblock, bh, 0);
861 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
865 * Get block function for AIO DIO writes when we create unwritten extent if
866 * blocks are not allocated yet. The extent will be converted to written
867 * after IO is complete.
869 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
870 sector_t iblock, struct buffer_head *bh_result, int create)
874 /* We don't expect handle for direct IO */
875 WARN_ON_ONCE(ext4_journal_current_handle());
877 ret = ext4_get_block_trans(inode, iblock, bh_result,
878 EXT4_GET_BLOCKS_IO_CREATE_EXT);
881 * When doing DIO using unwritten extents, we need io_end to convert
882 * unwritten extents to written on IO completion. We allocate io_end
883 * once we spot unwritten extent and store it in b_private. Generic
884 * DIO code keeps b_private set and furthermore passes the value to
885 * our completion callback in 'private' argument.
887 if (!ret && buffer_unwritten(bh_result)) {
888 if (!bh_result->b_private) {
889 ext4_io_end_t *io_end;
891 io_end = ext4_init_io_end(inode, GFP_KERNEL);
894 bh_result->b_private = io_end;
895 ext4_set_io_unwritten_flag(inode, io_end);
897 set_buffer_defer_completion(bh_result);
904 * Get block function for non-AIO DIO writes when we create unwritten extent if
905 * blocks are not allocated yet. The extent will be converted to written
906 * after IO is complete by ext4_direct_IO_write().
908 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
909 sector_t iblock, struct buffer_head *bh_result, int create)
913 /* We don't expect handle for direct IO */
914 WARN_ON_ONCE(ext4_journal_current_handle());
916 ret = ext4_get_block_trans(inode, iblock, bh_result,
917 EXT4_GET_BLOCKS_IO_CREATE_EXT);
920 * Mark inode as having pending DIO writes to unwritten extents.
921 * ext4_direct_IO_write() checks this flag and converts extents to
924 if (!ret && buffer_unwritten(bh_result))
925 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
930 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
931 struct buffer_head *bh_result, int create)
935 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
936 inode->i_ino, create);
937 /* We don't expect handle for direct IO */
938 WARN_ON_ONCE(ext4_journal_current_handle());
940 ret = _ext4_get_block(inode, iblock, bh_result, 0);
942 * Blocks should have been preallocated! ext4_file_write_iter() checks
945 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
952 * `handle' can be NULL if create is zero
954 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
955 ext4_lblk_t block, int map_flags)
957 struct ext4_map_blocks map;
958 struct buffer_head *bh;
959 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
962 J_ASSERT(handle != NULL || create == 0);
966 err = ext4_map_blocks(handle, inode, &map, map_flags);
969 return create ? ERR_PTR(-ENOSPC) : NULL;
973 bh = sb_getblk(inode->i_sb, map.m_pblk);
975 return ERR_PTR(-ENOMEM);
976 if (map.m_flags & EXT4_MAP_NEW) {
977 J_ASSERT(create != 0);
978 J_ASSERT(handle != NULL);
981 * Now that we do not always journal data, we should
982 * keep in mind whether this should always journal the
983 * new buffer as metadata. For now, regular file
984 * writes use ext4_get_block instead, so it's not a
988 BUFFER_TRACE(bh, "call get_create_access");
989 err = ext4_journal_get_create_access(handle, bh);
994 if (!buffer_uptodate(bh)) {
995 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
996 set_buffer_uptodate(bh);
999 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1000 err = ext4_handle_dirty_metadata(handle, inode, bh);
1004 BUFFER_TRACE(bh, "not a new buffer");
1008 return ERR_PTR(err);
1011 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1012 ext4_lblk_t block, int map_flags)
1014 struct buffer_head *bh;
1016 bh = ext4_getblk(handle, inode, block, map_flags);
1019 if (!bh || buffer_uptodate(bh))
1021 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1023 if (buffer_uptodate(bh))
1026 return ERR_PTR(-EIO);
1029 /* Read a contiguous batch of blocks. */
1030 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1031 bool wait, struct buffer_head **bhs)
1035 for (i = 0; i < bh_count; i++) {
1036 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1037 if (IS_ERR(bhs[i])) {
1038 err = PTR_ERR(bhs[i]);
1044 for (i = 0; i < bh_count; i++)
1045 /* Note that NULL bhs[i] is valid because of holes. */
1046 if (bhs[i] && !buffer_uptodate(bhs[i]))
1047 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1053 for (i = 0; i < bh_count; i++)
1055 wait_on_buffer(bhs[i]);
1057 for (i = 0; i < bh_count; i++) {
1058 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1066 for (i = 0; i < bh_count; i++) {
1073 int ext4_walk_page_buffers(handle_t *handle,
1074 struct buffer_head *head,
1078 int (*fn)(handle_t *handle,
1079 struct buffer_head *bh))
1081 struct buffer_head *bh;
1082 unsigned block_start, block_end;
1083 unsigned blocksize = head->b_size;
1085 struct buffer_head *next;
1087 for (bh = head, block_start = 0;
1088 ret == 0 && (bh != head || !block_start);
1089 block_start = block_end, bh = next) {
1090 next = bh->b_this_page;
1091 block_end = block_start + blocksize;
1092 if (block_end <= from || block_start >= to) {
1093 if (partial && !buffer_uptodate(bh))
1097 err = (*fn)(handle, bh);
1105 * To preserve ordering, it is essential that the hole instantiation and
1106 * the data write be encapsulated in a single transaction. We cannot
1107 * close off a transaction and start a new one between the ext4_get_block()
1108 * and the commit_write(). So doing the jbd2_journal_start at the start of
1109 * prepare_write() is the right place.
1111 * Also, this function can nest inside ext4_writepage(). In that case, we
1112 * *know* that ext4_writepage() has generated enough buffer credits to do the
1113 * whole page. So we won't block on the journal in that case, which is good,
1114 * because the caller may be PF_MEMALLOC.
1116 * By accident, ext4 can be reentered when a transaction is open via
1117 * quota file writes. If we were to commit the transaction while thus
1118 * reentered, there can be a deadlock - we would be holding a quota
1119 * lock, and the commit would never complete if another thread had a
1120 * transaction open and was blocking on the quota lock - a ranking
1123 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1124 * will _not_ run commit under these circumstances because handle->h_ref
1125 * is elevated. We'll still have enough credits for the tiny quotafile
1128 int do_journal_get_write_access(handle_t *handle,
1129 struct buffer_head *bh)
1131 int dirty = buffer_dirty(bh);
1134 if (!buffer_mapped(bh) || buffer_freed(bh))
1137 * __block_write_begin() could have dirtied some buffers. Clean
1138 * the dirty bit as jbd2_journal_get_write_access() could complain
1139 * otherwise about fs integrity issues. Setting of the dirty bit
1140 * by __block_write_begin() isn't a real problem here as we clear
1141 * the bit before releasing a page lock and thus writeback cannot
1142 * ever write the buffer.
1145 clear_buffer_dirty(bh);
1146 BUFFER_TRACE(bh, "get write access");
1147 ret = ext4_journal_get_write_access(handle, bh);
1149 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1153 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1154 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1155 get_block_t *get_block)
1157 unsigned from = pos & (PAGE_SIZE - 1);
1158 unsigned to = from + len;
1159 struct inode *inode = page->mapping->host;
1160 unsigned block_start, block_end;
1163 unsigned blocksize = inode->i_sb->s_blocksize;
1165 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1166 bool decrypt = false;
1168 BUG_ON(!PageLocked(page));
1169 BUG_ON(from > PAGE_SIZE);
1170 BUG_ON(to > PAGE_SIZE);
1173 if (!page_has_buffers(page))
1174 create_empty_buffers(page, blocksize, 0);
1175 head = page_buffers(page);
1176 bbits = ilog2(blocksize);
1177 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1179 for (bh = head, block_start = 0; bh != head || !block_start;
1180 block++, block_start = block_end, bh = bh->b_this_page) {
1181 block_end = block_start + blocksize;
1182 if (block_end <= from || block_start >= to) {
1183 if (PageUptodate(page)) {
1184 if (!buffer_uptodate(bh))
1185 set_buffer_uptodate(bh);
1190 clear_buffer_new(bh);
1191 if (!buffer_mapped(bh)) {
1192 WARN_ON(bh->b_size != blocksize);
1193 err = get_block(inode, block, bh, 1);
1196 if (buffer_new(bh)) {
1197 clean_bdev_bh_alias(bh);
1198 if (PageUptodate(page)) {
1199 clear_buffer_new(bh);
1200 set_buffer_uptodate(bh);
1201 mark_buffer_dirty(bh);
1204 if (block_end > to || block_start < from)
1205 zero_user_segments(page, to, block_end,
1210 if (PageUptodate(page)) {
1211 if (!buffer_uptodate(bh))
1212 set_buffer_uptodate(bh);
1215 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1216 !buffer_unwritten(bh) &&
1217 (block_start < from || block_end > to)) {
1218 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1220 decrypt = ext4_encrypted_inode(inode) &&
1221 S_ISREG(inode->i_mode);
1225 * If we issued read requests, let them complete.
1227 while (wait_bh > wait) {
1228 wait_on_buffer(*--wait_bh);
1229 if (!buffer_uptodate(*wait_bh))
1233 page_zero_new_buffers(page, from, to);
1235 err = fscrypt_decrypt_page(page->mapping->host, page,
1236 PAGE_SIZE, 0, page->index);
1241 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1242 loff_t pos, unsigned len, unsigned flags,
1243 struct page **pagep, void **fsdata)
1245 struct inode *inode = mapping->host;
1246 int ret, needed_blocks;
1253 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1256 trace_ext4_write_begin(inode, pos, len, flags);
1258 * Reserve one block more for addition to orphan list in case
1259 * we allocate blocks but write fails for some reason
1261 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1262 index = pos >> PAGE_SHIFT;
1263 from = pos & (PAGE_SIZE - 1);
1266 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1267 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1276 * grab_cache_page_write_begin() can take a long time if the
1277 * system is thrashing due to memory pressure, or if the page
1278 * is being written back. So grab it first before we start
1279 * the transaction handle. This also allows us to allocate
1280 * the page (if needed) without using GFP_NOFS.
1283 page = grab_cache_page_write_begin(mapping, index, flags);
1289 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1290 if (IS_ERR(handle)) {
1292 return PTR_ERR(handle);
1296 if (page->mapping != mapping) {
1297 /* The page got truncated from under us */
1300 ext4_journal_stop(handle);
1303 /* In case writeback began while the page was unlocked */
1304 wait_for_stable_page(page);
1306 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1307 if (ext4_should_dioread_nolock(inode))
1308 ret = ext4_block_write_begin(page, pos, len,
1309 ext4_get_block_unwritten);
1311 ret = ext4_block_write_begin(page, pos, len,
1314 if (ext4_should_dioread_nolock(inode))
1315 ret = __block_write_begin(page, pos, len,
1316 ext4_get_block_unwritten);
1318 ret = __block_write_begin(page, pos, len, ext4_get_block);
1320 if (!ret && ext4_should_journal_data(inode)) {
1321 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1323 do_journal_get_write_access);
1329 * __block_write_begin may have instantiated a few blocks
1330 * outside i_size. Trim these off again. Don't need
1331 * i_size_read because we hold i_mutex.
1333 * Add inode to orphan list in case we crash before
1336 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1337 ext4_orphan_add(handle, inode);
1339 ext4_journal_stop(handle);
1340 if (pos + len > inode->i_size) {
1341 ext4_truncate_failed_write(inode);
1343 * If truncate failed early the inode might
1344 * still be on the orphan list; we need to
1345 * make sure the inode is removed from the
1346 * orphan list in that case.
1349 ext4_orphan_del(NULL, inode);
1352 if (ret == -ENOSPC &&
1353 ext4_should_retry_alloc(inode->i_sb, &retries))
1362 /* For write_end() in data=journal mode */
1363 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1366 if (!buffer_mapped(bh) || buffer_freed(bh))
1368 set_buffer_uptodate(bh);
1369 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1370 clear_buffer_meta(bh);
1371 clear_buffer_prio(bh);
1376 * We need to pick up the new inode size which generic_commit_write gave us
1377 * `file' can be NULL - eg, when called from page_symlink().
1379 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1380 * buffers are managed internally.
1382 static int ext4_write_end(struct file *file,
1383 struct address_space *mapping,
1384 loff_t pos, unsigned len, unsigned copied,
1385 struct page *page, void *fsdata)
1387 handle_t *handle = ext4_journal_current_handle();
1388 struct inode *inode = mapping->host;
1389 loff_t old_size = inode->i_size;
1391 int i_size_changed = 0;
1393 trace_ext4_write_end(inode, pos, len, copied);
1394 if (ext4_has_inline_data(inode)) {
1395 ret = ext4_write_inline_data_end(inode, pos, len,
1404 copied = block_write_end(file, mapping, pos,
1405 len, copied, page, fsdata);
1407 * it's important to update i_size while still holding page lock:
1408 * page writeout could otherwise come in and zero beyond i_size.
1410 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1415 pagecache_isize_extended(inode, old_size, pos);
1417 * Don't mark the inode dirty under page lock. First, it unnecessarily
1418 * makes the holding time of page lock longer. Second, it forces lock
1419 * ordering of page lock and transaction start for journaling
1423 ext4_mark_inode_dirty(handle, inode);
1425 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1426 /* if we have allocated more blocks and copied
1427 * less. We will have blocks allocated outside
1428 * inode->i_size. So truncate them
1430 ext4_orphan_add(handle, inode);
1432 ret2 = ext4_journal_stop(handle);
1436 if (pos + len > inode->i_size) {
1437 ext4_truncate_failed_write(inode);
1439 * If truncate failed early the inode might still be
1440 * on the orphan list; we need to make sure the inode
1441 * is removed from the orphan list in that case.
1444 ext4_orphan_del(NULL, inode);
1447 return ret ? ret : copied;
1451 * This is a private version of page_zero_new_buffers() which doesn't
1452 * set the buffer to be dirty, since in data=journalled mode we need
1453 * to call ext4_handle_dirty_metadata() instead.
1455 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1457 unsigned from, unsigned to)
1459 unsigned int block_start = 0, block_end;
1460 struct buffer_head *head, *bh;
1462 bh = head = page_buffers(page);
1464 block_end = block_start + bh->b_size;
1465 if (buffer_new(bh)) {
1466 if (block_end > from && block_start < to) {
1467 if (!PageUptodate(page)) {
1468 unsigned start, size;
1470 start = max(from, block_start);
1471 size = min(to, block_end) - start;
1473 zero_user(page, start, size);
1474 write_end_fn(handle, bh);
1476 clear_buffer_new(bh);
1479 block_start = block_end;
1480 bh = bh->b_this_page;
1481 } while (bh != head);
1484 static int ext4_journalled_write_end(struct file *file,
1485 struct address_space *mapping,
1486 loff_t pos, unsigned len, unsigned copied,
1487 struct page *page, void *fsdata)
1489 handle_t *handle = ext4_journal_current_handle();
1490 struct inode *inode = mapping->host;
1491 loff_t old_size = inode->i_size;
1495 int size_changed = 0;
1497 trace_ext4_journalled_write_end(inode, pos, len, copied);
1498 from = pos & (PAGE_SIZE - 1);
1501 BUG_ON(!ext4_handle_valid(handle));
1503 if (ext4_has_inline_data(inode)) {
1504 ret = ext4_write_inline_data_end(inode, pos, len,
1512 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1514 ext4_journalled_zero_new_buffers(handle, page, from, to);
1516 if (unlikely(copied < len))
1517 ext4_journalled_zero_new_buffers(handle, page,
1519 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1520 from + copied, &partial,
1523 SetPageUptodate(page);
1525 size_changed = ext4_update_inode_size(inode, pos + copied);
1526 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1527 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1532 pagecache_isize_extended(inode, old_size, pos);
1535 ret2 = ext4_mark_inode_dirty(handle, inode);
1540 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1541 /* if we have allocated more blocks and copied
1542 * less. We will have blocks allocated outside
1543 * inode->i_size. So truncate them
1545 ext4_orphan_add(handle, inode);
1548 ret2 = ext4_journal_stop(handle);
1551 if (pos + len > inode->i_size) {
1552 ext4_truncate_failed_write(inode);
1554 * If truncate failed early the inode might still be
1555 * on the orphan list; we need to make sure the inode
1556 * is removed from the orphan list in that case.
1559 ext4_orphan_del(NULL, inode);
1562 return ret ? ret : copied;
1566 * Reserve space for a single cluster
1568 static int ext4_da_reserve_space(struct inode *inode)
1570 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1571 struct ext4_inode_info *ei = EXT4_I(inode);
1575 * We will charge metadata quota at writeout time; this saves
1576 * us from metadata over-estimation, though we may go over by
1577 * a small amount in the end. Here we just reserve for data.
1579 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1583 spin_lock(&ei->i_block_reservation_lock);
1584 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1585 spin_unlock(&ei->i_block_reservation_lock);
1586 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1589 ei->i_reserved_data_blocks++;
1590 trace_ext4_da_reserve_space(inode);
1591 spin_unlock(&ei->i_block_reservation_lock);
1593 return 0; /* success */
1596 static void ext4_da_release_space(struct inode *inode, int to_free)
1598 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1599 struct ext4_inode_info *ei = EXT4_I(inode);
1602 return; /* Nothing to release, exit */
1604 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1606 trace_ext4_da_release_space(inode, to_free);
1607 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1609 * if there aren't enough reserved blocks, then the
1610 * counter is messed up somewhere. Since this
1611 * function is called from invalidate page, it's
1612 * harmless to return without any action.
1614 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1615 "ino %lu, to_free %d with only %d reserved "
1616 "data blocks", inode->i_ino, to_free,
1617 ei->i_reserved_data_blocks);
1619 to_free = ei->i_reserved_data_blocks;
1621 ei->i_reserved_data_blocks -= to_free;
1623 /* update fs dirty data blocks counter */
1624 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1626 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1628 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1631 static void ext4_da_page_release_reservation(struct page *page,
1632 unsigned int offset,
1633 unsigned int length)
1635 int to_release = 0, contiguous_blks = 0;
1636 struct buffer_head *head, *bh;
1637 unsigned int curr_off = 0;
1638 struct inode *inode = page->mapping->host;
1639 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1640 unsigned int stop = offset + length;
1644 BUG_ON(stop > PAGE_SIZE || stop < length);
1646 head = page_buffers(page);
1649 unsigned int next_off = curr_off + bh->b_size;
1651 if (next_off > stop)
1654 if ((offset <= curr_off) && (buffer_delay(bh))) {
1657 clear_buffer_delay(bh);
1658 } else if (contiguous_blks) {
1659 lblk = page->index <<
1660 (PAGE_SHIFT - inode->i_blkbits);
1661 lblk += (curr_off >> inode->i_blkbits) -
1663 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1664 contiguous_blks = 0;
1666 curr_off = next_off;
1667 } while ((bh = bh->b_this_page) != head);
1669 if (contiguous_blks) {
1670 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1671 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1672 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1675 /* If we have released all the blocks belonging to a cluster, then we
1676 * need to release the reserved space for that cluster. */
1677 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1678 while (num_clusters > 0) {
1679 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1680 ((num_clusters - 1) << sbi->s_cluster_bits);
1681 if (sbi->s_cluster_ratio == 1 ||
1682 !ext4_find_delalloc_cluster(inode, lblk))
1683 ext4_da_release_space(inode, 1);
1690 * Delayed allocation stuff
1693 struct mpage_da_data {
1694 struct inode *inode;
1695 struct writeback_control *wbc;
1697 pgoff_t first_page; /* The first page to write */
1698 pgoff_t next_page; /* Current page to examine */
1699 pgoff_t last_page; /* Last page to examine */
1701 * Extent to map - this can be after first_page because that can be
1702 * fully mapped. We somewhat abuse m_flags to store whether the extent
1703 * is delalloc or unwritten.
1705 struct ext4_map_blocks map;
1706 struct ext4_io_submit io_submit; /* IO submission data */
1707 unsigned int do_map:1;
1710 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1715 struct pagevec pvec;
1716 struct inode *inode = mpd->inode;
1717 struct address_space *mapping = inode->i_mapping;
1719 /* This is necessary when next_page == 0. */
1720 if (mpd->first_page >= mpd->next_page)
1723 index = mpd->first_page;
1724 end = mpd->next_page - 1;
1726 ext4_lblk_t start, last;
1727 start = index << (PAGE_SHIFT - inode->i_blkbits);
1728 last = end << (PAGE_SHIFT - inode->i_blkbits);
1729 ext4_es_remove_extent(inode, start, last - start + 1);
1732 pagevec_init(&pvec);
1733 while (index <= end) {
1734 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1737 for (i = 0; i < nr_pages; i++) {
1738 struct page *page = pvec.pages[i];
1740 BUG_ON(!PageLocked(page));
1741 BUG_ON(PageWriteback(page));
1743 if (page_mapped(page))
1744 clear_page_dirty_for_io(page);
1745 block_invalidatepage(page, 0, PAGE_SIZE);
1746 ClearPageUptodate(page);
1750 pagevec_release(&pvec);
1754 static void ext4_print_free_blocks(struct inode *inode)
1756 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1757 struct super_block *sb = inode->i_sb;
1758 struct ext4_inode_info *ei = EXT4_I(inode);
1760 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1761 EXT4_C2B(EXT4_SB(inode->i_sb),
1762 ext4_count_free_clusters(sb)));
1763 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1764 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1765 (long long) EXT4_C2B(EXT4_SB(sb),
1766 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1767 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1768 (long long) EXT4_C2B(EXT4_SB(sb),
1769 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1770 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1771 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1772 ei->i_reserved_data_blocks);
1776 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1778 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1782 * This function is grabs code from the very beginning of
1783 * ext4_map_blocks, but assumes that the caller is from delayed write
1784 * time. This function looks up the requested blocks and sets the
1785 * buffer delay bit under the protection of i_data_sem.
1787 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1788 struct ext4_map_blocks *map,
1789 struct buffer_head *bh)
1791 struct extent_status es;
1793 sector_t invalid_block = ~((sector_t) 0xffff);
1794 #ifdef ES_AGGRESSIVE_TEST
1795 struct ext4_map_blocks orig_map;
1797 memcpy(&orig_map, map, sizeof(*map));
1800 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1804 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1805 "logical block %lu\n", inode->i_ino, map->m_len,
1806 (unsigned long) map->m_lblk);
1808 /* Lookup extent status tree firstly */
1809 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1810 if (ext4_es_is_hole(&es)) {
1812 down_read(&EXT4_I(inode)->i_data_sem);
1817 * Delayed extent could be allocated by fallocate.
1818 * So we need to check it.
1820 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1821 map_bh(bh, inode->i_sb, invalid_block);
1823 set_buffer_delay(bh);
1827 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1828 retval = es.es_len - (iblock - es.es_lblk);
1829 if (retval > map->m_len)
1830 retval = map->m_len;
1831 map->m_len = retval;
1832 if (ext4_es_is_written(&es))
1833 map->m_flags |= EXT4_MAP_MAPPED;
1834 else if (ext4_es_is_unwritten(&es))
1835 map->m_flags |= EXT4_MAP_UNWRITTEN;
1839 #ifdef ES_AGGRESSIVE_TEST
1840 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1846 * Try to see if we can get the block without requesting a new
1847 * file system block.
1849 down_read(&EXT4_I(inode)->i_data_sem);
1850 if (ext4_has_inline_data(inode))
1852 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1853 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1855 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1861 * XXX: __block_prepare_write() unmaps passed block,
1865 * If the block was allocated from previously allocated cluster,
1866 * then we don't need to reserve it again. However we still need
1867 * to reserve metadata for every block we're going to write.
1869 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1870 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1871 ret = ext4_da_reserve_space(inode);
1873 /* not enough space to reserve */
1879 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1880 ~0, EXTENT_STATUS_DELAYED);
1886 map_bh(bh, inode->i_sb, invalid_block);
1888 set_buffer_delay(bh);
1889 } else if (retval > 0) {
1891 unsigned int status;
1893 if (unlikely(retval != map->m_len)) {
1894 ext4_warning(inode->i_sb,
1895 "ES len assertion failed for inode "
1896 "%lu: retval %d != map->m_len %d",
1897 inode->i_ino, retval, map->m_len);
1901 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1902 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1903 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1904 map->m_pblk, status);
1910 up_read((&EXT4_I(inode)->i_data_sem));
1916 * This is a special get_block_t callback which is used by
1917 * ext4_da_write_begin(). It will either return mapped block or
1918 * reserve space for a single block.
1920 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1921 * We also have b_blocknr = -1 and b_bdev initialized properly
1923 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1924 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1925 * initialized properly.
1927 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1928 struct buffer_head *bh, int create)
1930 struct ext4_map_blocks map;
1933 BUG_ON(create == 0);
1934 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1936 map.m_lblk = iblock;
1940 * first, we need to know whether the block is allocated already
1941 * preallocated blocks are unmapped but should treated
1942 * the same as allocated blocks.
1944 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1948 map_bh(bh, inode->i_sb, map.m_pblk);
1949 ext4_update_bh_state(bh, map.m_flags);
1951 if (buffer_unwritten(bh)) {
1952 /* A delayed write to unwritten bh should be marked
1953 * new and mapped. Mapped ensures that we don't do
1954 * get_block multiple times when we write to the same
1955 * offset and new ensures that we do proper zero out
1956 * for partial write.
1959 set_buffer_mapped(bh);
1964 static int bget_one(handle_t *handle, struct buffer_head *bh)
1970 static int bput_one(handle_t *handle, struct buffer_head *bh)
1976 static int __ext4_journalled_writepage(struct page *page,
1979 struct address_space *mapping = page->mapping;
1980 struct inode *inode = mapping->host;
1981 struct buffer_head *page_bufs = NULL;
1982 handle_t *handle = NULL;
1983 int ret = 0, err = 0;
1984 int inline_data = ext4_has_inline_data(inode);
1985 struct buffer_head *inode_bh = NULL;
1987 ClearPageChecked(page);
1990 BUG_ON(page->index != 0);
1991 BUG_ON(len > ext4_get_max_inline_size(inode));
1992 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1993 if (inode_bh == NULL)
1996 page_bufs = page_buffers(page);
2001 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2005 * We need to release the page lock before we start the
2006 * journal, so grab a reference so the page won't disappear
2007 * out from under us.
2012 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2013 ext4_writepage_trans_blocks(inode));
2014 if (IS_ERR(handle)) {
2015 ret = PTR_ERR(handle);
2017 goto out_no_pagelock;
2019 BUG_ON(!ext4_handle_valid(handle));
2023 if (page->mapping != mapping) {
2024 /* The page got truncated from under us */
2025 ext4_journal_stop(handle);
2031 BUFFER_TRACE(inode_bh, "get write access");
2032 ret = ext4_journal_get_write_access(handle, inode_bh);
2034 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2037 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2038 do_journal_get_write_access);
2040 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2045 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2046 err = ext4_journal_stop(handle);
2050 if (!ext4_has_inline_data(inode))
2051 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2053 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2062 * Note that we don't need to start a transaction unless we're journaling data
2063 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2064 * need to file the inode to the transaction's list in ordered mode because if
2065 * we are writing back data added by write(), the inode is already there and if
2066 * we are writing back data modified via mmap(), no one guarantees in which
2067 * transaction the data will hit the disk. In case we are journaling data, we
2068 * cannot start transaction directly because transaction start ranks above page
2069 * lock so we have to do some magic.
2071 * This function can get called via...
2072 * - ext4_writepages after taking page lock (have journal handle)
2073 * - journal_submit_inode_data_buffers (no journal handle)
2074 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2075 * - grab_page_cache when doing write_begin (have journal handle)
2077 * We don't do any block allocation in this function. If we have page with
2078 * multiple blocks we need to write those buffer_heads that are mapped. This
2079 * is important for mmaped based write. So if we do with blocksize 1K
2080 * truncate(f, 1024);
2081 * a = mmap(f, 0, 4096);
2083 * truncate(f, 4096);
2084 * we have in the page first buffer_head mapped via page_mkwrite call back
2085 * but other buffer_heads would be unmapped but dirty (dirty done via the
2086 * do_wp_page). So writepage should write the first block. If we modify
2087 * the mmap area beyond 1024 we will again get a page_fault and the
2088 * page_mkwrite callback will do the block allocation and mark the
2089 * buffer_heads mapped.
2091 * We redirty the page if we have any buffer_heads that is either delay or
2092 * unwritten in the page.
2094 * We can get recursively called as show below.
2096 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2099 * But since we don't do any block allocation we should not deadlock.
2100 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2102 static int ext4_writepage(struct page *page,
2103 struct writeback_control *wbc)
2108 struct buffer_head *page_bufs = NULL;
2109 struct inode *inode = page->mapping->host;
2110 struct ext4_io_submit io_submit;
2111 bool keep_towrite = false;
2113 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2114 ext4_invalidatepage(page, 0, PAGE_SIZE);
2119 trace_ext4_writepage(page);
2120 size = i_size_read(inode);
2121 if (page->index == size >> PAGE_SHIFT)
2122 len = size & ~PAGE_MASK;
2126 page_bufs = page_buffers(page);
2128 * We cannot do block allocation or other extent handling in this
2129 * function. If there are buffers needing that, we have to redirty
2130 * the page. But we may reach here when we do a journal commit via
2131 * journal_submit_inode_data_buffers() and in that case we must write
2132 * allocated buffers to achieve data=ordered mode guarantees.
2134 * Also, if there is only one buffer per page (the fs block
2135 * size == the page size), if one buffer needs block
2136 * allocation or needs to modify the extent tree to clear the
2137 * unwritten flag, we know that the page can't be written at
2138 * all, so we might as well refuse the write immediately.
2139 * Unfortunately if the block size != page size, we can't as
2140 * easily detect this case using ext4_walk_page_buffers(), but
2141 * for the extremely common case, this is an optimization that
2142 * skips a useless round trip through ext4_bio_write_page().
2144 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2145 ext4_bh_delay_or_unwritten)) {
2146 redirty_page_for_writepage(wbc, page);
2147 if ((current->flags & PF_MEMALLOC) ||
2148 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2150 * For memory cleaning there's no point in writing only
2151 * some buffers. So just bail out. Warn if we came here
2152 * from direct reclaim.
2154 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2159 keep_towrite = true;
2162 if (PageChecked(page) && ext4_should_journal_data(inode))
2164 * It's mmapped pagecache. Add buffers and journal it. There
2165 * doesn't seem much point in redirtying the page here.
2167 return __ext4_journalled_writepage(page, len);
2169 ext4_io_submit_init(&io_submit, wbc);
2170 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2171 if (!io_submit.io_end) {
2172 redirty_page_for_writepage(wbc, page);
2176 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2177 ext4_io_submit(&io_submit);
2178 /* Drop io_end reference we got from init */
2179 ext4_put_io_end_defer(io_submit.io_end);
2183 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2189 BUG_ON(page->index != mpd->first_page);
2190 clear_page_dirty_for_io(page);
2192 * We have to be very careful here! Nothing protects writeback path
2193 * against i_size changes and the page can be writeably mapped into
2194 * page tables. So an application can be growing i_size and writing
2195 * data through mmap while writeback runs. clear_page_dirty_for_io()
2196 * write-protects our page in page tables and the page cannot get
2197 * written to again until we release page lock. So only after
2198 * clear_page_dirty_for_io() we are safe to sample i_size for
2199 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2200 * on the barrier provided by TestClearPageDirty in
2201 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2202 * after page tables are updated.
2204 size = i_size_read(mpd->inode);
2205 if (page->index == size >> PAGE_SHIFT)
2206 len = size & ~PAGE_MASK;
2209 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2211 mpd->wbc->nr_to_write--;
2217 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2220 * mballoc gives us at most this number of blocks...
2221 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2222 * The rest of mballoc seems to handle chunks up to full group size.
2224 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2227 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2229 * @mpd - extent of blocks
2230 * @lblk - logical number of the block in the file
2231 * @bh - buffer head we want to add to the extent
2233 * The function is used to collect contig. blocks in the same state. If the
2234 * buffer doesn't require mapping for writeback and we haven't started the
2235 * extent of buffers to map yet, the function returns 'true' immediately - the
2236 * caller can write the buffer right away. Otherwise the function returns true
2237 * if the block has been added to the extent, false if the block couldn't be
2240 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2241 struct buffer_head *bh)
2243 struct ext4_map_blocks *map = &mpd->map;
2245 /* Buffer that doesn't need mapping for writeback? */
2246 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2247 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2248 /* So far no extent to map => we write the buffer right away */
2249 if (map->m_len == 0)
2254 /* First block in the extent? */
2255 if (map->m_len == 0) {
2256 /* We cannot map unless handle is started... */
2261 map->m_flags = bh->b_state & BH_FLAGS;
2265 /* Don't go larger than mballoc is willing to allocate */
2266 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2269 /* Can we merge the block to our big extent? */
2270 if (lblk == map->m_lblk + map->m_len &&
2271 (bh->b_state & BH_FLAGS) == map->m_flags) {
2279 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2281 * @mpd - extent of blocks for mapping
2282 * @head - the first buffer in the page
2283 * @bh - buffer we should start processing from
2284 * @lblk - logical number of the block in the file corresponding to @bh
2286 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2287 * the page for IO if all buffers in this page were mapped and there's no
2288 * accumulated extent of buffers to map or add buffers in the page to the
2289 * extent of buffers to map. The function returns 1 if the caller can continue
2290 * by processing the next page, 0 if it should stop adding buffers to the
2291 * extent to map because we cannot extend it anymore. It can also return value
2292 * < 0 in case of error during IO submission.
2294 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2295 struct buffer_head *head,
2296 struct buffer_head *bh,
2299 struct inode *inode = mpd->inode;
2301 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2302 >> inode->i_blkbits;
2305 BUG_ON(buffer_locked(bh));
2307 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2308 /* Found extent to map? */
2311 /* Buffer needs mapping and handle is not started? */
2314 /* Everything mapped so far and we hit EOF */
2317 } while (lblk++, (bh = bh->b_this_page) != head);
2318 /* So far everything mapped? Submit the page for IO. */
2319 if (mpd->map.m_len == 0) {
2320 err = mpage_submit_page(mpd, head->b_page);
2324 return lblk < blocks;
2328 * mpage_map_buffers - update buffers corresponding to changed extent and
2329 * submit fully mapped pages for IO
2331 * @mpd - description of extent to map, on return next extent to map
2333 * Scan buffers corresponding to changed extent (we expect corresponding pages
2334 * to be already locked) and update buffer state according to new extent state.
2335 * We map delalloc buffers to their physical location, clear unwritten bits,
2336 * and mark buffers as uninit when we perform writes to unwritten extents
2337 * and do extent conversion after IO is finished. If the last page is not fully
2338 * mapped, we update @map to the next extent in the last page that needs
2339 * mapping. Otherwise we submit the page for IO.
2341 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2343 struct pagevec pvec;
2345 struct inode *inode = mpd->inode;
2346 struct buffer_head *head, *bh;
2347 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2353 start = mpd->map.m_lblk >> bpp_bits;
2354 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2355 lblk = start << bpp_bits;
2356 pblock = mpd->map.m_pblk;
2358 pagevec_init(&pvec);
2359 while (start <= end) {
2360 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2364 for (i = 0; i < nr_pages; i++) {
2365 struct page *page = pvec.pages[i];
2367 bh = head = page_buffers(page);
2369 if (lblk < mpd->map.m_lblk)
2371 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2373 * Buffer after end of mapped extent.
2374 * Find next buffer in the page to map.
2377 mpd->map.m_flags = 0;
2379 * FIXME: If dioread_nolock supports
2380 * blocksize < pagesize, we need to make
2381 * sure we add size mapped so far to
2382 * io_end->size as the following call
2383 * can submit the page for IO.
2385 err = mpage_process_page_bufs(mpd, head,
2387 pagevec_release(&pvec);
2392 if (buffer_delay(bh)) {
2393 clear_buffer_delay(bh);
2394 bh->b_blocknr = pblock++;
2396 clear_buffer_unwritten(bh);
2397 } while (lblk++, (bh = bh->b_this_page) != head);
2400 * FIXME: This is going to break if dioread_nolock
2401 * supports blocksize < pagesize as we will try to
2402 * convert potentially unmapped parts of inode.
2404 mpd->io_submit.io_end->size += PAGE_SIZE;
2405 /* Page fully mapped - let IO run! */
2406 err = mpage_submit_page(mpd, page);
2408 pagevec_release(&pvec);
2412 pagevec_release(&pvec);
2414 /* Extent fully mapped and matches with page boundary. We are done. */
2416 mpd->map.m_flags = 0;
2420 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2422 struct inode *inode = mpd->inode;
2423 struct ext4_map_blocks *map = &mpd->map;
2424 int get_blocks_flags;
2425 int err, dioread_nolock;
2427 trace_ext4_da_write_pages_extent(inode, map);
2429 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2430 * to convert an unwritten extent to be initialized (in the case
2431 * where we have written into one or more preallocated blocks). It is
2432 * possible that we're going to need more metadata blocks than
2433 * previously reserved. However we must not fail because we're in
2434 * writeback and there is nothing we can do about it so it might result
2435 * in data loss. So use reserved blocks to allocate metadata if
2438 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2439 * the blocks in question are delalloc blocks. This indicates
2440 * that the blocks and quotas has already been checked when
2441 * the data was copied into the page cache.
2443 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2444 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2445 EXT4_GET_BLOCKS_IO_SUBMIT;
2446 dioread_nolock = ext4_should_dioread_nolock(inode);
2448 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2449 if (map->m_flags & (1 << BH_Delay))
2450 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2452 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2455 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2456 if (!mpd->io_submit.io_end->handle &&
2457 ext4_handle_valid(handle)) {
2458 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2459 handle->h_rsv_handle = NULL;
2461 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2464 BUG_ON(map->m_len == 0);
2465 if (map->m_flags & EXT4_MAP_NEW) {
2466 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2473 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2474 * mpd->len and submit pages underlying it for IO
2476 * @handle - handle for journal operations
2477 * @mpd - extent to map
2478 * @give_up_on_write - we set this to true iff there is a fatal error and there
2479 * is no hope of writing the data. The caller should discard
2480 * dirty pages to avoid infinite loops.
2482 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2483 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2484 * them to initialized or split the described range from larger unwritten
2485 * extent. Note that we need not map all the described range since allocation
2486 * can return less blocks or the range is covered by more unwritten extents. We
2487 * cannot map more because we are limited by reserved transaction credits. On
2488 * the other hand we always make sure that the last touched page is fully
2489 * mapped so that it can be written out (and thus forward progress is
2490 * guaranteed). After mapping we submit all mapped pages for IO.
2492 static int mpage_map_and_submit_extent(handle_t *handle,
2493 struct mpage_da_data *mpd,
2494 bool *give_up_on_write)
2496 struct inode *inode = mpd->inode;
2497 struct ext4_map_blocks *map = &mpd->map;
2502 mpd->io_submit.io_end->offset =
2503 ((loff_t)map->m_lblk) << inode->i_blkbits;
2505 err = mpage_map_one_extent(handle, mpd);
2507 struct super_block *sb = inode->i_sb;
2509 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2510 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2511 goto invalidate_dirty_pages;
2513 * Let the uper layers retry transient errors.
2514 * In the case of ENOSPC, if ext4_count_free_blocks()
2515 * is non-zero, a commit should free up blocks.
2517 if ((err == -ENOMEM) ||
2518 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2520 goto update_disksize;
2523 ext4_msg(sb, KERN_CRIT,
2524 "Delayed block allocation failed for "
2525 "inode %lu at logical offset %llu with"
2526 " max blocks %u with error %d",
2528 (unsigned long long)map->m_lblk,
2529 (unsigned)map->m_len, -err);
2530 ext4_msg(sb, KERN_CRIT,
2531 "This should not happen!! Data will "
2534 ext4_print_free_blocks(inode);
2535 invalidate_dirty_pages:
2536 *give_up_on_write = true;
2541 * Update buffer state, submit mapped pages, and get us new
2544 err = mpage_map_and_submit_buffers(mpd);
2546 goto update_disksize;
2547 } while (map->m_len);
2551 * Update on-disk size after IO is submitted. Races with
2552 * truncate are avoided by checking i_size under i_data_sem.
2554 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2555 if (disksize > EXT4_I(inode)->i_disksize) {
2559 down_write(&EXT4_I(inode)->i_data_sem);
2560 i_size = i_size_read(inode);
2561 if (disksize > i_size)
2563 if (disksize > EXT4_I(inode)->i_disksize)
2564 EXT4_I(inode)->i_disksize = disksize;
2565 up_write(&EXT4_I(inode)->i_data_sem);
2566 err2 = ext4_mark_inode_dirty(handle, inode);
2568 ext4_error(inode->i_sb,
2569 "Failed to mark inode %lu dirty",
2578 * Calculate the total number of credits to reserve for one writepages
2579 * iteration. This is called from ext4_writepages(). We map an extent of
2580 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2581 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2582 * bpp - 1 blocks in bpp different extents.
2584 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2586 int bpp = ext4_journal_blocks_per_page(inode);
2588 return ext4_meta_trans_blocks(inode,
2589 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2593 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2594 * and underlying extent to map
2596 * @mpd - where to look for pages
2598 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2599 * IO immediately. When we find a page which isn't mapped we start accumulating
2600 * extent of buffers underlying these pages that needs mapping (formed by
2601 * either delayed or unwritten buffers). We also lock the pages containing
2602 * these buffers. The extent found is returned in @mpd structure (starting at
2603 * mpd->lblk with length mpd->len blocks).
2605 * Note that this function can attach bios to one io_end structure which are
2606 * neither logically nor physically contiguous. Although it may seem as an
2607 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2608 * case as we need to track IO to all buffers underlying a page in one io_end.
2610 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2612 struct address_space *mapping = mpd->inode->i_mapping;
2613 struct pagevec pvec;
2614 unsigned int nr_pages;
2615 long left = mpd->wbc->nr_to_write;
2616 pgoff_t index = mpd->first_page;
2617 pgoff_t end = mpd->last_page;
2620 int blkbits = mpd->inode->i_blkbits;
2622 struct buffer_head *head;
2624 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2625 tag = PAGECACHE_TAG_TOWRITE;
2627 tag = PAGECACHE_TAG_DIRTY;
2629 pagevec_init(&pvec);
2631 mpd->next_page = index;
2632 while (index <= end) {
2633 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2638 for (i = 0; i < nr_pages; i++) {
2639 struct page *page = pvec.pages[i];
2642 * Accumulated enough dirty pages? This doesn't apply
2643 * to WB_SYNC_ALL mode. For integrity sync we have to
2644 * keep going because someone may be concurrently
2645 * dirtying pages, and we might have synced a lot of
2646 * newly appeared dirty pages, but have not synced all
2647 * of the old dirty pages.
2649 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2652 /* If we can't merge this page, we are done. */
2653 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2658 * If the page is no longer dirty, or its mapping no
2659 * longer corresponds to inode we are writing (which
2660 * means it has been truncated or invalidated), or the
2661 * page is already under writeback and we are not doing
2662 * a data integrity writeback, skip the page
2664 if (!PageDirty(page) ||
2665 (PageWriteback(page) &&
2666 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2667 unlikely(page->mapping != mapping)) {
2672 wait_on_page_writeback(page);
2673 BUG_ON(PageWriteback(page));
2675 if (mpd->map.m_len == 0)
2676 mpd->first_page = page->index;
2677 mpd->next_page = page->index + 1;
2678 /* Add all dirty buffers to mpd */
2679 lblk = ((ext4_lblk_t)page->index) <<
2680 (PAGE_SHIFT - blkbits);
2681 head = page_buffers(page);
2682 err = mpage_process_page_bufs(mpd, head, head, lblk);
2688 pagevec_release(&pvec);
2693 pagevec_release(&pvec);
2697 static int ext4_writepages(struct address_space *mapping,
2698 struct writeback_control *wbc)
2700 pgoff_t writeback_index = 0;
2701 long nr_to_write = wbc->nr_to_write;
2702 int range_whole = 0;
2704 handle_t *handle = NULL;
2705 struct mpage_da_data mpd;
2706 struct inode *inode = mapping->host;
2707 int needed_blocks, rsv_blocks = 0, ret = 0;
2708 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2710 struct blk_plug plug;
2711 bool give_up_on_write = false;
2713 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2716 percpu_down_read(&sbi->s_journal_flag_rwsem);
2717 trace_ext4_writepages(inode, wbc);
2719 if (dax_mapping(mapping)) {
2720 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2722 goto out_writepages;
2726 * No pages to write? This is mainly a kludge to avoid starting
2727 * a transaction for special inodes like journal inode on last iput()
2728 * because that could violate lock ordering on umount
2730 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2731 goto out_writepages;
2733 if (ext4_should_journal_data(inode)) {
2734 ret = generic_writepages(mapping, wbc);
2735 goto out_writepages;
2739 * If the filesystem has aborted, it is read-only, so return
2740 * right away instead of dumping stack traces later on that
2741 * will obscure the real source of the problem. We test
2742 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2743 * the latter could be true if the filesystem is mounted
2744 * read-only, and in that case, ext4_writepages should
2745 * *never* be called, so if that ever happens, we would want
2748 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2749 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2751 goto out_writepages;
2754 if (ext4_should_dioread_nolock(inode)) {
2756 * We may need to convert up to one extent per block in
2757 * the page and we may dirty the inode.
2759 rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2763 * If we have inline data and arrive here, it means that
2764 * we will soon create the block for the 1st page, so
2765 * we'd better clear the inline data here.
2767 if (ext4_has_inline_data(inode)) {
2768 /* Just inode will be modified... */
2769 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2770 if (IS_ERR(handle)) {
2771 ret = PTR_ERR(handle);
2772 goto out_writepages;
2774 BUG_ON(ext4_test_inode_state(inode,
2775 EXT4_STATE_MAY_INLINE_DATA));
2776 ext4_destroy_inline_data(handle, inode);
2777 ext4_journal_stop(handle);
2780 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2783 if (wbc->range_cyclic) {
2784 writeback_index = mapping->writeback_index;
2785 if (writeback_index)
2787 mpd.first_page = writeback_index;
2790 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2791 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2796 ext4_io_submit_init(&mpd.io_submit, wbc);
2798 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2799 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2801 blk_start_plug(&plug);
2804 * First writeback pages that don't need mapping - we can avoid
2805 * starting a transaction unnecessarily and also avoid being blocked
2806 * in the block layer on device congestion while having transaction
2810 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2811 if (!mpd.io_submit.io_end) {
2815 ret = mpage_prepare_extent_to_map(&mpd);
2816 /* Submit prepared bio */
2817 ext4_io_submit(&mpd.io_submit);
2818 ext4_put_io_end_defer(mpd.io_submit.io_end);
2819 mpd.io_submit.io_end = NULL;
2820 /* Unlock pages we didn't use */
2821 mpage_release_unused_pages(&mpd, false);
2825 while (!done && mpd.first_page <= mpd.last_page) {
2826 /* For each extent of pages we use new io_end */
2827 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2828 if (!mpd.io_submit.io_end) {
2834 * We have two constraints: We find one extent to map and we
2835 * must always write out whole page (makes a difference when
2836 * blocksize < pagesize) so that we don't block on IO when we
2837 * try to write out the rest of the page. Journalled mode is
2838 * not supported by delalloc.
2840 BUG_ON(ext4_should_journal_data(inode));
2841 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2843 /* start a new transaction */
2844 handle = ext4_journal_start_with_reserve(inode,
2845 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2846 if (IS_ERR(handle)) {
2847 ret = PTR_ERR(handle);
2848 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2849 "%ld pages, ino %lu; err %d", __func__,
2850 wbc->nr_to_write, inode->i_ino, ret);
2851 /* Release allocated io_end */
2852 ext4_put_io_end(mpd.io_submit.io_end);
2853 mpd.io_submit.io_end = NULL;
2858 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2859 ret = mpage_prepare_extent_to_map(&mpd);
2862 ret = mpage_map_and_submit_extent(handle, &mpd,
2866 * We scanned the whole range (or exhausted
2867 * nr_to_write), submitted what was mapped and
2868 * didn't find anything needing mapping. We are
2875 * Caution: If the handle is synchronous,
2876 * ext4_journal_stop() can wait for transaction commit
2877 * to finish which may depend on writeback of pages to
2878 * complete or on page lock to be released. In that
2879 * case, we have to wait until after after we have
2880 * submitted all the IO, released page locks we hold,
2881 * and dropped io_end reference (for extent conversion
2882 * to be able to complete) before stopping the handle.
2884 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2885 ext4_journal_stop(handle);
2889 /* Submit prepared bio */
2890 ext4_io_submit(&mpd.io_submit);
2891 /* Unlock pages we didn't use */
2892 mpage_release_unused_pages(&mpd, give_up_on_write);
2894 * Drop our io_end reference we got from init. We have
2895 * to be careful and use deferred io_end finishing if
2896 * we are still holding the transaction as we can
2897 * release the last reference to io_end which may end
2898 * up doing unwritten extent conversion.
2901 ext4_put_io_end_defer(mpd.io_submit.io_end);
2902 ext4_journal_stop(handle);
2904 ext4_put_io_end(mpd.io_submit.io_end);
2905 mpd.io_submit.io_end = NULL;
2907 if (ret == -ENOSPC && sbi->s_journal) {
2909 * Commit the transaction which would
2910 * free blocks released in the transaction
2913 jbd2_journal_force_commit_nested(sbi->s_journal);
2917 /* Fatal error - ENOMEM, EIO... */
2922 blk_finish_plug(&plug);
2923 if (!ret && !cycled && wbc->nr_to_write > 0) {
2925 mpd.last_page = writeback_index - 1;
2931 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2933 * Set the writeback_index so that range_cyclic
2934 * mode will write it back later
2936 mapping->writeback_index = mpd.first_page;
2939 trace_ext4_writepages_result(inode, wbc, ret,
2940 nr_to_write - wbc->nr_to_write);
2941 percpu_up_read(&sbi->s_journal_flag_rwsem);
2945 static int ext4_nonda_switch(struct super_block *sb)
2947 s64 free_clusters, dirty_clusters;
2948 struct ext4_sb_info *sbi = EXT4_SB(sb);
2951 * switch to non delalloc mode if we are running low
2952 * on free block. The free block accounting via percpu
2953 * counters can get slightly wrong with percpu_counter_batch getting
2954 * accumulated on each CPU without updating global counters
2955 * Delalloc need an accurate free block accounting. So switch
2956 * to non delalloc when we are near to error range.
2959 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2961 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2963 * Start pushing delalloc when 1/2 of free blocks are dirty.
2965 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2966 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2968 if (2 * free_clusters < 3 * dirty_clusters ||
2969 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2971 * free block count is less than 150% of dirty blocks
2972 * or free blocks is less than watermark
2979 /* We always reserve for an inode update; the superblock could be there too */
2980 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2982 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2985 if (pos + len <= 0x7fffffffULL)
2988 /* We might need to update the superblock to set LARGE_FILE */
2992 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2993 loff_t pos, unsigned len, unsigned flags,
2994 struct page **pagep, void **fsdata)
2996 int ret, retries = 0;
2999 struct inode *inode = mapping->host;
3002 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3005 index = pos >> PAGE_SHIFT;
3007 if (ext4_nonda_switch(inode->i_sb) ||
3008 S_ISLNK(inode->i_mode)) {
3009 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3010 return ext4_write_begin(file, mapping, pos,
3011 len, flags, pagep, fsdata);
3013 *fsdata = (void *)0;
3014 trace_ext4_da_write_begin(inode, pos, len, flags);
3016 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3017 ret = ext4_da_write_inline_data_begin(mapping, inode,
3027 * grab_cache_page_write_begin() can take a long time if the
3028 * system is thrashing due to memory pressure, or if the page
3029 * is being written back. So grab it first before we start
3030 * the transaction handle. This also allows us to allocate
3031 * the page (if needed) without using GFP_NOFS.
3034 page = grab_cache_page_write_begin(mapping, index, flags);
3040 * With delayed allocation, we don't log the i_disksize update
3041 * if there is delayed block allocation. But we still need
3042 * to journalling the i_disksize update if writes to the end
3043 * of file which has an already mapped buffer.
3046 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3047 ext4_da_write_credits(inode, pos, len));
3048 if (IS_ERR(handle)) {
3050 return PTR_ERR(handle);
3054 if (page->mapping != mapping) {
3055 /* The page got truncated from under us */
3058 ext4_journal_stop(handle);
3061 /* In case writeback began while the page was unlocked */
3062 wait_for_stable_page(page);
3064 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3065 ret = ext4_block_write_begin(page, pos, len,
3066 ext4_da_get_block_prep);
3068 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3072 ext4_journal_stop(handle);
3074 * block_write_begin may have instantiated a few blocks
3075 * outside i_size. Trim these off again. Don't need
3076 * i_size_read because we hold i_mutex.
3078 if (pos + len > inode->i_size)
3079 ext4_truncate_failed_write(inode);
3081 if (ret == -ENOSPC &&
3082 ext4_should_retry_alloc(inode->i_sb, &retries))
3094 * Check if we should update i_disksize
3095 * when write to the end of file but not require block allocation
3097 static int ext4_da_should_update_i_disksize(struct page *page,
3098 unsigned long offset)
3100 struct buffer_head *bh;
3101 struct inode *inode = page->mapping->host;
3105 bh = page_buffers(page);
3106 idx = offset >> inode->i_blkbits;
3108 for (i = 0; i < idx; i++)
3109 bh = bh->b_this_page;
3111 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3116 static int ext4_da_write_end(struct file *file,
3117 struct address_space *mapping,
3118 loff_t pos, unsigned len, unsigned copied,
3119 struct page *page, void *fsdata)
3121 struct inode *inode = mapping->host;
3123 handle_t *handle = ext4_journal_current_handle();
3125 unsigned long start, end;
3126 int write_mode = (int)(unsigned long)fsdata;
3128 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3129 return ext4_write_end(file, mapping, pos,
3130 len, copied, page, fsdata);
3132 trace_ext4_da_write_end(inode, pos, len, copied);
3133 start = pos & (PAGE_SIZE - 1);
3134 end = start + copied - 1;
3137 * generic_write_end() will run mark_inode_dirty() if i_size
3138 * changes. So let's piggyback the i_disksize mark_inode_dirty
3141 new_i_size = pos + copied;
3142 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3143 if (ext4_has_inline_data(inode) ||
3144 ext4_da_should_update_i_disksize(page, end)) {
3145 ext4_update_i_disksize(inode, new_i_size);
3146 /* We need to mark inode dirty even if
3147 * new_i_size is less that inode->i_size
3148 * bu greater than i_disksize.(hint delalloc)
3150 ext4_mark_inode_dirty(handle, inode);
3154 if (write_mode != CONVERT_INLINE_DATA &&
3155 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3156 ext4_has_inline_data(inode))
3157 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3160 ret2 = generic_write_end(file, mapping, pos, len, copied,
3166 ret2 = ext4_journal_stop(handle);
3170 return ret ? ret : copied;
3173 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3174 unsigned int length)
3177 * Drop reserved blocks
3179 BUG_ON(!PageLocked(page));
3180 if (!page_has_buffers(page))
3183 ext4_da_page_release_reservation(page, offset, length);
3186 ext4_invalidatepage(page, offset, length);
3192 * Force all delayed allocation blocks to be allocated for a given inode.
3194 int ext4_alloc_da_blocks(struct inode *inode)
3196 trace_ext4_alloc_da_blocks(inode);
3198 if (!EXT4_I(inode)->i_reserved_data_blocks)
3202 * We do something simple for now. The filemap_flush() will
3203 * also start triggering a write of the data blocks, which is
3204 * not strictly speaking necessary (and for users of
3205 * laptop_mode, not even desirable). However, to do otherwise
3206 * would require replicating code paths in:
3208 * ext4_writepages() ->
3209 * write_cache_pages() ---> (via passed in callback function)
3210 * __mpage_da_writepage() -->
3211 * mpage_add_bh_to_extent()
3212 * mpage_da_map_blocks()
3214 * The problem is that write_cache_pages(), located in
3215 * mm/page-writeback.c, marks pages clean in preparation for
3216 * doing I/O, which is not desirable if we're not planning on
3219 * We could call write_cache_pages(), and then redirty all of
3220 * the pages by calling redirty_page_for_writepage() but that
3221 * would be ugly in the extreme. So instead we would need to
3222 * replicate parts of the code in the above functions,
3223 * simplifying them because we wouldn't actually intend to
3224 * write out the pages, but rather only collect contiguous
3225 * logical block extents, call the multi-block allocator, and
3226 * then update the buffer heads with the block allocations.
3228 * For now, though, we'll cheat by calling filemap_flush(),
3229 * which will map the blocks, and start the I/O, but not
3230 * actually wait for the I/O to complete.
3232 return filemap_flush(inode->i_mapping);
3236 * bmap() is special. It gets used by applications such as lilo and by
3237 * the swapper to find the on-disk block of a specific piece of data.
3239 * Naturally, this is dangerous if the block concerned is still in the
3240 * journal. If somebody makes a swapfile on an ext4 data-journaling
3241 * filesystem and enables swap, then they may get a nasty shock when the
3242 * data getting swapped to that swapfile suddenly gets overwritten by
3243 * the original zero's written out previously to the journal and
3244 * awaiting writeback in the kernel's buffer cache.
3246 * So, if we see any bmap calls here on a modified, data-journaled file,
3247 * take extra steps to flush any blocks which might be in the cache.
3249 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3251 struct inode *inode = mapping->host;
3256 * We can get here for an inline file via the FIBMAP ioctl
3258 if (ext4_has_inline_data(inode))
3261 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3262 test_opt(inode->i_sb, DELALLOC)) {
3264 * With delalloc we want to sync the file
3265 * so that we can make sure we allocate
3268 filemap_write_and_wait(mapping);
3271 if (EXT4_JOURNAL(inode) &&
3272 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3274 * This is a REALLY heavyweight approach, but the use of
3275 * bmap on dirty files is expected to be extremely rare:
3276 * only if we run lilo or swapon on a freshly made file
3277 * do we expect this to happen.
3279 * (bmap requires CAP_SYS_RAWIO so this does not
3280 * represent an unprivileged user DOS attack --- we'd be
3281 * in trouble if mortal users could trigger this path at
3284 * NB. EXT4_STATE_JDATA is not set on files other than
3285 * regular files. If somebody wants to bmap a directory
3286 * or symlink and gets confused because the buffer
3287 * hasn't yet been flushed to disk, they deserve
3288 * everything they get.
3291 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3292 journal = EXT4_JOURNAL(inode);
3293 jbd2_journal_lock_updates(journal);
3294 err = jbd2_journal_flush(journal);
3295 jbd2_journal_unlock_updates(journal);
3301 return generic_block_bmap(mapping, block, ext4_get_block);
3304 static int ext4_readpage(struct file *file, struct page *page)
3307 struct inode *inode = page->mapping->host;
3309 trace_ext4_readpage(page);
3311 if (ext4_has_inline_data(inode))
3312 ret = ext4_readpage_inline(inode, page);
3315 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3321 ext4_readpages(struct file *file, struct address_space *mapping,
3322 struct list_head *pages, unsigned nr_pages)
3324 struct inode *inode = mapping->host;
3326 /* If the file has inline data, no need to do readpages. */
3327 if (ext4_has_inline_data(inode))
3330 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3333 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3334 unsigned int length)
3336 trace_ext4_invalidatepage(page, offset, length);
3338 /* No journalling happens on data buffers when this function is used */
3339 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3341 block_invalidatepage(page, offset, length);
3344 static int __ext4_journalled_invalidatepage(struct page *page,
3345 unsigned int offset,
3346 unsigned int length)
3348 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3350 trace_ext4_journalled_invalidatepage(page, offset, length);
3353 * If it's a full truncate we just forget about the pending dirtying
3355 if (offset == 0 && length == PAGE_SIZE)
3356 ClearPageChecked(page);
3358 return jbd2_journal_invalidatepage(journal, page, offset, length);
3361 /* Wrapper for aops... */
3362 static void ext4_journalled_invalidatepage(struct page *page,
3363 unsigned int offset,
3364 unsigned int length)
3366 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3369 static int ext4_releasepage(struct page *page, gfp_t wait)
3371 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3373 trace_ext4_releasepage(page);
3375 /* Page has dirty journalled data -> cannot release */
3376 if (PageChecked(page))
3379 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3381 return try_to_free_buffers(page);
3384 static bool ext4_inode_datasync_dirty(struct inode *inode)
3386 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3389 return !jbd2_transaction_committed(journal,
3390 EXT4_I(inode)->i_datasync_tid);
3391 /* Any metadata buffers to write? */
3392 if (!list_empty(&inode->i_mapping->private_list))
3394 return inode->i_state & I_DIRTY_DATASYNC;
3397 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3398 unsigned flags, struct iomap *iomap)
3400 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3401 unsigned int blkbits = inode->i_blkbits;
3402 unsigned long first_block = offset >> blkbits;
3403 unsigned long last_block = (offset + length - 1) >> blkbits;
3404 struct ext4_map_blocks map;
3405 bool delalloc = false;
3409 if (flags & IOMAP_REPORT) {
3410 if (ext4_has_inline_data(inode)) {
3411 ret = ext4_inline_data_iomap(inode, iomap);
3412 if (ret != -EAGAIN) {
3413 if (ret == 0 && offset >= iomap->length)
3419 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3423 map.m_lblk = first_block;
3424 map.m_len = last_block - first_block + 1;
3426 if (flags & IOMAP_REPORT) {
3427 ret = ext4_map_blocks(NULL, inode, &map, 0);
3432 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3433 struct extent_status es;
3435 ext4_es_find_delayed_extent_range(inode, map.m_lblk, end, &es);
3437 if (!es.es_len || es.es_lblk > end) {
3438 /* entire range is a hole */
3439 } else if (es.es_lblk > map.m_lblk) {
3440 /* range starts with a hole */
3441 map.m_len = es.es_lblk - map.m_lblk;
3443 ext4_lblk_t offs = 0;
3445 if (es.es_lblk < map.m_lblk)
3446 offs = map.m_lblk - es.es_lblk;
3447 map.m_lblk = es.es_lblk + offs;
3448 map.m_len = es.es_len - offs;
3452 } else if (flags & IOMAP_WRITE) {
3457 /* Trim mapping request to maximum we can map at once for DIO */
3458 if (map.m_len > DIO_MAX_BLOCKS)
3459 map.m_len = DIO_MAX_BLOCKS;
3460 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3463 * Either we allocate blocks and then we don't get unwritten
3464 * extent so we have reserved enough credits, or the blocks
3465 * are already allocated and unwritten and in that case
3466 * extent conversion fits in the credits as well.
3468 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3471 return PTR_ERR(handle);
3473 ret = ext4_map_blocks(handle, inode, &map,
3474 EXT4_GET_BLOCKS_CREATE_ZERO);
3476 ext4_journal_stop(handle);
3477 if (ret == -ENOSPC &&
3478 ext4_should_retry_alloc(inode->i_sb, &retries))
3484 * If we added blocks beyond i_size, we need to make sure they
3485 * will get truncated if we crash before updating i_size in
3486 * ext4_iomap_end(). For faults we don't need to do that (and
3487 * even cannot because for orphan list operations inode_lock is
3488 * required) - if we happen to instantiate block beyond i_size,
3489 * it is because we race with truncate which has already added
3490 * the inode to the orphan list.
3492 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3493 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3496 err = ext4_orphan_add(handle, inode);
3498 ext4_journal_stop(handle);
3502 ext4_journal_stop(handle);
3504 ret = ext4_map_blocks(NULL, inode, &map, 0);
3510 if (ext4_inode_datasync_dirty(inode))
3511 iomap->flags |= IOMAP_F_DIRTY;
3512 iomap->bdev = inode->i_sb->s_bdev;
3513 iomap->dax_dev = sbi->s_daxdev;
3514 iomap->offset = (u64)first_block << blkbits;
3515 iomap->length = (u64)map.m_len << blkbits;
3518 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3519 iomap->addr = IOMAP_NULL_ADDR;
3521 if (map.m_flags & EXT4_MAP_MAPPED) {
3522 iomap->type = IOMAP_MAPPED;
3523 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3524 iomap->type = IOMAP_UNWRITTEN;
3529 iomap->addr = (u64)map.m_pblk << blkbits;
3532 if (map.m_flags & EXT4_MAP_NEW)
3533 iomap->flags |= IOMAP_F_NEW;
3538 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3539 ssize_t written, unsigned flags, struct iomap *iomap)
3543 int blkbits = inode->i_blkbits;
3544 bool truncate = false;
3546 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3549 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3550 if (IS_ERR(handle)) {
3551 ret = PTR_ERR(handle);
3554 if (ext4_update_inode_size(inode, offset + written))
3555 ext4_mark_inode_dirty(handle, inode);
3557 * We may need to truncate allocated but not written blocks beyond EOF.
3559 if (iomap->offset + iomap->length >
3560 ALIGN(inode->i_size, 1 << blkbits)) {
3561 ext4_lblk_t written_blk, end_blk;
3563 written_blk = (offset + written) >> blkbits;
3564 end_blk = (offset + length) >> blkbits;
3565 if (written_blk < end_blk && ext4_can_truncate(inode))
3569 * Remove inode from orphan list if we were extending a inode and
3570 * everything went fine.
3572 if (!truncate && inode->i_nlink &&
3573 !list_empty(&EXT4_I(inode)->i_orphan))
3574 ext4_orphan_del(handle, inode);
3575 ext4_journal_stop(handle);
3577 ext4_truncate_failed_write(inode);
3580 * If truncate failed early the inode might still be on the
3581 * orphan list; we need to make sure the inode is removed from
3582 * the orphan list in that case.
3585 ext4_orphan_del(NULL, inode);
3590 const struct iomap_ops ext4_iomap_ops = {
3591 .iomap_begin = ext4_iomap_begin,
3592 .iomap_end = ext4_iomap_end,
3595 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3596 ssize_t size, void *private)
3598 ext4_io_end_t *io_end = private;
3600 /* if not async direct IO just return */
3604 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3605 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3606 io_end, io_end->inode->i_ino, iocb, offset, size);
3609 * Error during AIO DIO. We cannot convert unwritten extents as the
3610 * data was not written. Just clear the unwritten flag and drop io_end.
3613 ext4_clear_io_unwritten_flag(io_end);
3616 io_end->offset = offset;
3617 io_end->size = size;
3618 ext4_put_io_end(io_end);
3624 * Handling of direct IO writes.
3626 * For ext4 extent files, ext4 will do direct-io write even to holes,
3627 * preallocated extents, and those write extend the file, no need to
3628 * fall back to buffered IO.
3630 * For holes, we fallocate those blocks, mark them as unwritten
3631 * If those blocks were preallocated, we mark sure they are split, but
3632 * still keep the range to write as unwritten.
3634 * The unwritten extents will be converted to written when DIO is completed.
3635 * For async direct IO, since the IO may still pending when return, we
3636 * set up an end_io call back function, which will do the conversion
3637 * when async direct IO completed.
3639 * If the O_DIRECT write will extend the file then add this inode to the
3640 * orphan list. So recovery will truncate it back to the original size
3641 * if the machine crashes during the write.
3644 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3646 struct file *file = iocb->ki_filp;
3647 struct inode *inode = file->f_mapping->host;
3648 struct ext4_inode_info *ei = EXT4_I(inode);
3650 loff_t offset = iocb->ki_pos;
3651 size_t count = iov_iter_count(iter);
3653 get_block_t *get_block_func = NULL;
3655 loff_t final_size = offset + count;
3659 if (final_size > inode->i_size || final_size > ei->i_disksize) {
3660 /* Credits for sb + inode write */
3661 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3662 if (IS_ERR(handle)) {
3663 ret = PTR_ERR(handle);
3666 ret = ext4_orphan_add(handle, inode);
3668 ext4_journal_stop(handle);
3672 ext4_update_i_disksize(inode, inode->i_size);
3673 ext4_journal_stop(handle);
3676 BUG_ON(iocb->private == NULL);
3679 * Make all waiters for direct IO properly wait also for extent
3680 * conversion. This also disallows race between truncate() and
3681 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3683 inode_dio_begin(inode);
3685 /* If we do a overwrite dio, i_mutex locking can be released */
3686 overwrite = *((int *)iocb->private);
3689 inode_unlock(inode);
3692 * For extent mapped files we could direct write to holes and fallocate.
3694 * Allocated blocks to fill the hole are marked as unwritten to prevent
3695 * parallel buffered read to expose the stale data before DIO complete
3698 * As to previously fallocated extents, ext4 get_block will just simply
3699 * mark the buffer mapped but still keep the extents unwritten.
3701 * For non AIO case, we will convert those unwritten extents to written
3702 * after return back from blockdev_direct_IO. That way we save us from
3703 * allocating io_end structure and also the overhead of offloading
3704 * the extent convertion to a workqueue.
3706 * For async DIO, the conversion needs to be deferred when the
3707 * IO is completed. The ext4 end_io callback function will be
3708 * called to take care of the conversion work. Here for async
3709 * case, we allocate an io_end structure to hook to the iocb.
3711 iocb->private = NULL;
3713 get_block_func = ext4_dio_get_block_overwrite;
3714 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3715 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3716 get_block_func = ext4_dio_get_block;
3717 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3718 } else if (is_sync_kiocb(iocb)) {
3719 get_block_func = ext4_dio_get_block_unwritten_sync;
3720 dio_flags = DIO_LOCKING;
3722 get_block_func = ext4_dio_get_block_unwritten_async;
3723 dio_flags = DIO_LOCKING;
3725 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3726 get_block_func, ext4_end_io_dio, NULL,
3729 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3730 EXT4_STATE_DIO_UNWRITTEN)) {
3733 * for non AIO case, since the IO is already
3734 * completed, we could do the conversion right here
3736 err = ext4_convert_unwritten_extents(NULL, inode,
3740 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3743 inode_dio_end(inode);
3744 /* take i_mutex locking again if we do a ovewrite dio */
3748 if (ret < 0 && final_size > inode->i_size)
3749 ext4_truncate_failed_write(inode);
3751 /* Handle extending of i_size after direct IO write */
3755 /* Credits for sb + inode write */
3756 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3757 if (IS_ERR(handle)) {
3759 * We wrote the data but cannot extend
3760 * i_size. Bail out. In async io case, we do
3761 * not return error here because we have
3762 * already submmitted the corresponding
3763 * bio. Returning error here makes the caller
3764 * think that this IO is done and failed
3765 * resulting in race with bio's completion
3769 ret = PTR_ERR(handle);
3771 ext4_orphan_del(NULL, inode);
3776 ext4_orphan_del(handle, inode);
3778 loff_t end = offset + ret;
3779 if (end > inode->i_size || end > ei->i_disksize) {
3780 ext4_update_i_disksize(inode, end);
3781 if (end > inode->i_size)
3782 i_size_write(inode, end);
3784 * We're going to return a positive `ret'
3785 * here due to non-zero-length I/O, so there's
3786 * no way of reporting error returns from
3787 * ext4_mark_inode_dirty() to userspace. So
3790 ext4_mark_inode_dirty(handle, inode);
3793 err = ext4_journal_stop(handle);
3801 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3803 struct address_space *mapping = iocb->ki_filp->f_mapping;
3804 struct inode *inode = mapping->host;
3805 size_t count = iov_iter_count(iter);
3809 * Shared inode_lock is enough for us - it protects against concurrent
3810 * writes & truncates and since we take care of writing back page cache,
3811 * we are protected against page writeback as well.
3813 inode_lock_shared(inode);
3814 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3815 iocb->ki_pos + count - 1);
3818 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3819 iter, ext4_dio_get_block, NULL, NULL, 0);
3821 inode_unlock_shared(inode);
3825 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3827 struct file *file = iocb->ki_filp;
3828 struct inode *inode = file->f_mapping->host;
3829 size_t count = iov_iter_count(iter);
3830 loff_t offset = iocb->ki_pos;
3833 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3834 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3839 * If we are doing data journalling we don't support O_DIRECT
3841 if (ext4_should_journal_data(inode))
3844 /* Let buffer I/O handle the inline data case. */
3845 if (ext4_has_inline_data(inode))
3848 /* DAX uses iomap path now */
3849 if (WARN_ON_ONCE(IS_DAX(inode)))
3852 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3853 if (iov_iter_rw(iter) == READ)
3854 ret = ext4_direct_IO_read(iocb, iter);
3856 ret = ext4_direct_IO_write(iocb, iter);
3857 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3862 * Pages can be marked dirty completely asynchronously from ext4's journalling
3863 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3864 * much here because ->set_page_dirty is called under VFS locks. The page is
3865 * not necessarily locked.
3867 * We cannot just dirty the page and leave attached buffers clean, because the
3868 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3869 * or jbddirty because all the journalling code will explode.
3871 * So what we do is to mark the page "pending dirty" and next time writepage
3872 * is called, propagate that into the buffers appropriately.
3874 static int ext4_journalled_set_page_dirty(struct page *page)
3876 SetPageChecked(page);
3877 return __set_page_dirty_nobuffers(page);
3880 static int ext4_set_page_dirty(struct page *page)
3882 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3883 WARN_ON_ONCE(!page_has_buffers(page));
3884 return __set_page_dirty_buffers(page);
3887 static const struct address_space_operations ext4_aops = {
3888 .readpage = ext4_readpage,
3889 .readpages = ext4_readpages,
3890 .writepage = ext4_writepage,
3891 .writepages = ext4_writepages,
3892 .write_begin = ext4_write_begin,
3893 .write_end = ext4_write_end,
3894 .set_page_dirty = ext4_set_page_dirty,
3896 .invalidatepage = ext4_invalidatepage,
3897 .releasepage = ext4_releasepage,
3898 .direct_IO = ext4_direct_IO,
3899 .migratepage = buffer_migrate_page,
3900 .is_partially_uptodate = block_is_partially_uptodate,
3901 .error_remove_page = generic_error_remove_page,
3904 static const struct address_space_operations ext4_journalled_aops = {
3905 .readpage = ext4_readpage,
3906 .readpages = ext4_readpages,
3907 .writepage = ext4_writepage,
3908 .writepages = ext4_writepages,
3909 .write_begin = ext4_write_begin,
3910 .write_end = ext4_journalled_write_end,
3911 .set_page_dirty = ext4_journalled_set_page_dirty,
3913 .invalidatepage = ext4_journalled_invalidatepage,
3914 .releasepage = ext4_releasepage,
3915 .direct_IO = ext4_direct_IO,
3916 .is_partially_uptodate = block_is_partially_uptodate,
3917 .error_remove_page = generic_error_remove_page,
3920 static const struct address_space_operations ext4_da_aops = {
3921 .readpage = ext4_readpage,
3922 .readpages = ext4_readpages,
3923 .writepage = ext4_writepage,
3924 .writepages = ext4_writepages,
3925 .write_begin = ext4_da_write_begin,
3926 .write_end = ext4_da_write_end,
3927 .set_page_dirty = ext4_set_page_dirty,
3929 .invalidatepage = ext4_da_invalidatepage,
3930 .releasepage = ext4_releasepage,
3931 .direct_IO = ext4_direct_IO,
3932 .migratepage = buffer_migrate_page,
3933 .is_partially_uptodate = block_is_partially_uptodate,
3934 .error_remove_page = generic_error_remove_page,
3937 void ext4_set_aops(struct inode *inode)
3939 switch (ext4_inode_journal_mode(inode)) {
3940 case EXT4_INODE_ORDERED_DATA_MODE:
3941 case EXT4_INODE_WRITEBACK_DATA_MODE:
3943 case EXT4_INODE_JOURNAL_DATA_MODE:
3944 inode->i_mapping->a_ops = &ext4_journalled_aops;
3949 if (test_opt(inode->i_sb, DELALLOC))
3950 inode->i_mapping->a_ops = &ext4_da_aops;
3952 inode->i_mapping->a_ops = &ext4_aops;
3955 static int __ext4_block_zero_page_range(handle_t *handle,
3956 struct address_space *mapping, loff_t from, loff_t length)
3958 ext4_fsblk_t index = from >> PAGE_SHIFT;
3959 unsigned offset = from & (PAGE_SIZE-1);
3960 unsigned blocksize, pos;
3962 struct inode *inode = mapping->host;
3963 struct buffer_head *bh;
3967 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3968 mapping_gfp_constraint(mapping, ~__GFP_FS));
3972 blocksize = inode->i_sb->s_blocksize;
3974 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3976 if (!page_has_buffers(page))
3977 create_empty_buffers(page, blocksize, 0);
3979 /* Find the buffer that contains "offset" */
3980 bh = page_buffers(page);
3982 while (offset >= pos) {
3983 bh = bh->b_this_page;
3987 if (buffer_freed(bh)) {
3988 BUFFER_TRACE(bh, "freed: skip");
3991 if (!buffer_mapped(bh)) {
3992 BUFFER_TRACE(bh, "unmapped");
3993 ext4_get_block(inode, iblock, bh, 0);
3994 /* unmapped? It's a hole - nothing to do */
3995 if (!buffer_mapped(bh)) {
3996 BUFFER_TRACE(bh, "still unmapped");
4001 /* Ok, it's mapped. Make sure it's up-to-date */
4002 if (PageUptodate(page))
4003 set_buffer_uptodate(bh);
4005 if (!buffer_uptodate(bh)) {
4007 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4009 /* Uhhuh. Read error. Complain and punt. */
4010 if (!buffer_uptodate(bh))
4012 if (S_ISREG(inode->i_mode) &&
4013 ext4_encrypted_inode(inode)) {
4014 /* We expect the key to be set. */
4015 BUG_ON(!fscrypt_has_encryption_key(inode));
4016 BUG_ON(blocksize != PAGE_SIZE);
4017 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4018 page, PAGE_SIZE, 0, page->index));
4021 if (ext4_should_journal_data(inode)) {
4022 BUFFER_TRACE(bh, "get write access");
4023 err = ext4_journal_get_write_access(handle, bh);
4027 zero_user(page, offset, length);
4028 BUFFER_TRACE(bh, "zeroed end of block");
4030 if (ext4_should_journal_data(inode)) {
4031 err = ext4_handle_dirty_metadata(handle, inode, bh);
4034 mark_buffer_dirty(bh);
4035 if (ext4_should_order_data(inode))
4036 err = ext4_jbd2_inode_add_write(handle, inode);
4046 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4047 * starting from file offset 'from'. The range to be zero'd must
4048 * be contained with in one block. If the specified range exceeds
4049 * the end of the block it will be shortened to end of the block
4050 * that cooresponds to 'from'
4052 static int ext4_block_zero_page_range(handle_t *handle,
4053 struct address_space *mapping, loff_t from, loff_t length)
4055 struct inode *inode = mapping->host;
4056 unsigned offset = from & (PAGE_SIZE-1);
4057 unsigned blocksize = inode->i_sb->s_blocksize;
4058 unsigned max = blocksize - (offset & (blocksize - 1));
4061 * correct length if it does not fall between
4062 * 'from' and the end of the block
4064 if (length > max || length < 0)
4067 if (IS_DAX(inode)) {
4068 return iomap_zero_range(inode, from, length, NULL,
4071 return __ext4_block_zero_page_range(handle, mapping, from, length);
4075 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4076 * up to the end of the block which corresponds to `from'.
4077 * This required during truncate. We need to physically zero the tail end
4078 * of that block so it doesn't yield old data if the file is later grown.
4080 static int ext4_block_truncate_page(handle_t *handle,
4081 struct address_space *mapping, loff_t from)
4083 unsigned offset = from & (PAGE_SIZE-1);
4086 struct inode *inode = mapping->host;
4088 /* If we are processing an encrypted inode during orphan list handling */
4089 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4092 blocksize = inode->i_sb->s_blocksize;
4093 length = blocksize - (offset & (blocksize - 1));
4095 return ext4_block_zero_page_range(handle, mapping, from, length);
4098 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4099 loff_t lstart, loff_t length)
4101 struct super_block *sb = inode->i_sb;
4102 struct address_space *mapping = inode->i_mapping;
4103 unsigned partial_start, partial_end;
4104 ext4_fsblk_t start, end;
4105 loff_t byte_end = (lstart + length - 1);
4108 partial_start = lstart & (sb->s_blocksize - 1);
4109 partial_end = byte_end & (sb->s_blocksize - 1);
4111 start = lstart >> sb->s_blocksize_bits;
4112 end = byte_end >> sb->s_blocksize_bits;
4114 /* Handle partial zero within the single block */
4116 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4117 err = ext4_block_zero_page_range(handle, mapping,
4121 /* Handle partial zero out on the start of the range */
4122 if (partial_start) {
4123 err = ext4_block_zero_page_range(handle, mapping,
4124 lstart, sb->s_blocksize);
4128 /* Handle partial zero out on the end of the range */
4129 if (partial_end != sb->s_blocksize - 1)
4130 err = ext4_block_zero_page_range(handle, mapping,
4131 byte_end - partial_end,
4136 int ext4_can_truncate(struct inode *inode)
4138 if (S_ISREG(inode->i_mode))
4140 if (S_ISDIR(inode->i_mode))
4142 if (S_ISLNK(inode->i_mode))
4143 return !ext4_inode_is_fast_symlink(inode);
4148 * We have to make sure i_disksize gets properly updated before we truncate
4149 * page cache due to hole punching or zero range. Otherwise i_disksize update
4150 * can get lost as it may have been postponed to submission of writeback but
4151 * that will never happen after we truncate page cache.
4153 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4157 loff_t size = i_size_read(inode);
4159 WARN_ON(!inode_is_locked(inode));
4160 if (offset > size || offset + len < size)
4163 if (EXT4_I(inode)->i_disksize >= size)
4166 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4168 return PTR_ERR(handle);
4169 ext4_update_i_disksize(inode, size);
4170 ext4_mark_inode_dirty(handle, inode);
4171 ext4_journal_stop(handle);
4177 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4178 * associated with the given offset and length
4180 * @inode: File inode
4181 * @offset: The offset where the hole will begin
4182 * @len: The length of the hole
4184 * Returns: 0 on success or negative on failure
4187 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4189 struct super_block *sb = inode->i_sb;
4190 ext4_lblk_t first_block, stop_block;
4191 struct address_space *mapping = inode->i_mapping;
4192 loff_t first_block_offset, last_block_offset;
4194 unsigned int credits;
4197 if (!S_ISREG(inode->i_mode))
4200 trace_ext4_punch_hole(inode, offset, length, 0);
4203 * Write out all dirty pages to avoid race conditions
4204 * Then release them.
4206 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4207 ret = filemap_write_and_wait_range(mapping, offset,
4208 offset + length - 1);
4215 /* No need to punch hole beyond i_size */
4216 if (offset >= inode->i_size)
4220 * If the hole extends beyond i_size, set the hole
4221 * to end after the page that contains i_size
4223 if (offset + length > inode->i_size) {
4224 length = inode->i_size +
4225 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4229 if (offset & (sb->s_blocksize - 1) ||
4230 (offset + length) & (sb->s_blocksize - 1)) {
4232 * Attach jinode to inode for jbd2 if we do any zeroing of
4235 ret = ext4_inode_attach_jinode(inode);
4241 /* Wait all existing dio workers, newcomers will block on i_mutex */
4242 inode_dio_wait(inode);
4245 * Prevent page faults from reinstantiating pages we have released from
4248 down_write(&EXT4_I(inode)->i_mmap_sem);
4249 first_block_offset = round_up(offset, sb->s_blocksize);
4250 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4252 /* Now release the pages and zero block aligned part of pages*/
4253 if (last_block_offset > first_block_offset) {
4254 ret = ext4_update_disksize_before_punch(inode, offset, length);
4257 truncate_pagecache_range(inode, first_block_offset,
4261 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4262 credits = ext4_writepage_trans_blocks(inode);
4264 credits = ext4_blocks_for_truncate(inode);
4265 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4266 if (IS_ERR(handle)) {
4267 ret = PTR_ERR(handle);
4268 ext4_std_error(sb, ret);
4272 ret = ext4_zero_partial_blocks(handle, inode, offset,
4277 first_block = (offset + sb->s_blocksize - 1) >>
4278 EXT4_BLOCK_SIZE_BITS(sb);
4279 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4281 /* If there are no blocks to remove, return now */
4282 if (first_block >= stop_block)
4285 down_write(&EXT4_I(inode)->i_data_sem);
4286 ext4_discard_preallocations(inode);
4288 ret = ext4_es_remove_extent(inode, first_block,
4289 stop_block - first_block);
4291 up_write(&EXT4_I(inode)->i_data_sem);
4295 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4296 ret = ext4_ext_remove_space(inode, first_block,
4299 ret = ext4_ind_remove_space(handle, inode, first_block,
4302 up_write(&EXT4_I(inode)->i_data_sem);
4304 ext4_handle_sync(handle);
4306 inode->i_mtime = inode->i_ctime = current_time(inode);
4307 ext4_mark_inode_dirty(handle, inode);
4309 ext4_update_inode_fsync_trans(handle, inode, 1);
4311 ext4_journal_stop(handle);
4313 up_write(&EXT4_I(inode)->i_mmap_sem);
4315 inode_unlock(inode);
4319 int ext4_inode_attach_jinode(struct inode *inode)
4321 struct ext4_inode_info *ei = EXT4_I(inode);
4322 struct jbd2_inode *jinode;
4324 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4327 jinode = jbd2_alloc_inode(GFP_KERNEL);
4328 spin_lock(&inode->i_lock);
4331 spin_unlock(&inode->i_lock);
4334 ei->jinode = jinode;
4335 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4338 spin_unlock(&inode->i_lock);
4339 if (unlikely(jinode != NULL))
4340 jbd2_free_inode(jinode);
4347 * We block out ext4_get_block() block instantiations across the entire
4348 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4349 * simultaneously on behalf of the same inode.
4351 * As we work through the truncate and commit bits of it to the journal there
4352 * is one core, guiding principle: the file's tree must always be consistent on
4353 * disk. We must be able to restart the truncate after a crash.
4355 * The file's tree may be transiently inconsistent in memory (although it
4356 * probably isn't), but whenever we close off and commit a journal transaction,
4357 * the contents of (the filesystem + the journal) must be consistent and
4358 * restartable. It's pretty simple, really: bottom up, right to left (although
4359 * left-to-right works OK too).
4361 * Note that at recovery time, journal replay occurs *before* the restart of
4362 * truncate against the orphan inode list.
4364 * The committed inode has the new, desired i_size (which is the same as
4365 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4366 * that this inode's truncate did not complete and it will again call
4367 * ext4_truncate() to have another go. So there will be instantiated blocks
4368 * to the right of the truncation point in a crashed ext4 filesystem. But
4369 * that's fine - as long as they are linked from the inode, the post-crash
4370 * ext4_truncate() run will find them and release them.
4372 int ext4_truncate(struct inode *inode)
4374 struct ext4_inode_info *ei = EXT4_I(inode);
4375 unsigned int credits;
4378 struct address_space *mapping = inode->i_mapping;
4381 * There is a possibility that we're either freeing the inode
4382 * or it's a completely new inode. In those cases we might not
4383 * have i_mutex locked because it's not necessary.
4385 if (!(inode->i_state & (I_NEW|I_FREEING)))
4386 WARN_ON(!inode_is_locked(inode));
4387 trace_ext4_truncate_enter(inode);
4389 if (!ext4_can_truncate(inode))
4392 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4394 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4395 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4397 if (ext4_has_inline_data(inode)) {
4400 err = ext4_inline_data_truncate(inode, &has_inline);
4407 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4408 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4409 if (ext4_inode_attach_jinode(inode) < 0)
4413 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4414 credits = ext4_writepage_trans_blocks(inode);
4416 credits = ext4_blocks_for_truncate(inode);
4418 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4420 return PTR_ERR(handle);
4422 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4423 ext4_block_truncate_page(handle, mapping, inode->i_size);
4426 * We add the inode to the orphan list, so that if this
4427 * truncate spans multiple transactions, and we crash, we will
4428 * resume the truncate when the filesystem recovers. It also
4429 * marks the inode dirty, to catch the new size.
4431 * Implication: the file must always be in a sane, consistent
4432 * truncatable state while each transaction commits.
4434 err = ext4_orphan_add(handle, inode);
4438 down_write(&EXT4_I(inode)->i_data_sem);
4440 ext4_discard_preallocations(inode);
4442 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4443 err = ext4_ext_truncate(handle, inode);
4445 ext4_ind_truncate(handle, inode);
4447 up_write(&ei->i_data_sem);
4452 ext4_handle_sync(handle);
4456 * If this was a simple ftruncate() and the file will remain alive,
4457 * then we need to clear up the orphan record which we created above.
4458 * However, if this was a real unlink then we were called by
4459 * ext4_evict_inode(), and we allow that function to clean up the
4460 * orphan info for us.
4463 ext4_orphan_del(handle, inode);
4465 inode->i_mtime = inode->i_ctime = current_time(inode);
4466 ext4_mark_inode_dirty(handle, inode);
4467 ext4_journal_stop(handle);
4469 trace_ext4_truncate_exit(inode);
4474 * ext4_get_inode_loc returns with an extra refcount against the inode's
4475 * underlying buffer_head on success. If 'in_mem' is true, we have all
4476 * data in memory that is needed to recreate the on-disk version of this
4479 static int __ext4_get_inode_loc(struct inode *inode,
4480 struct ext4_iloc *iloc, int in_mem)
4482 struct ext4_group_desc *gdp;
4483 struct buffer_head *bh;
4484 struct super_block *sb = inode->i_sb;
4486 int inodes_per_block, inode_offset;
4489 if (!ext4_valid_inum(sb, inode->i_ino))
4490 return -EFSCORRUPTED;
4492 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4493 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4498 * Figure out the offset within the block group inode table
4500 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4501 inode_offset = ((inode->i_ino - 1) %
4502 EXT4_INODES_PER_GROUP(sb));
4503 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4504 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4506 bh = sb_getblk(sb, block);
4509 if (!buffer_uptodate(bh)) {
4513 * If the buffer has the write error flag, we have failed
4514 * to write out another inode in the same block. In this
4515 * case, we don't have to read the block because we may
4516 * read the old inode data successfully.
4518 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4519 set_buffer_uptodate(bh);
4521 if (buffer_uptodate(bh)) {
4522 /* someone brought it uptodate while we waited */
4528 * If we have all information of the inode in memory and this
4529 * is the only valid inode in the block, we need not read the
4533 struct buffer_head *bitmap_bh;
4536 start = inode_offset & ~(inodes_per_block - 1);
4538 /* Is the inode bitmap in cache? */
4539 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4540 if (unlikely(!bitmap_bh))
4544 * If the inode bitmap isn't in cache then the
4545 * optimisation may end up performing two reads instead
4546 * of one, so skip it.
4548 if (!buffer_uptodate(bitmap_bh)) {
4552 for (i = start; i < start + inodes_per_block; i++) {
4553 if (i == inode_offset)
4555 if (ext4_test_bit(i, bitmap_bh->b_data))
4559 if (i == start + inodes_per_block) {
4560 /* all other inodes are free, so skip I/O */
4561 memset(bh->b_data, 0, bh->b_size);
4562 set_buffer_uptodate(bh);
4570 * If we need to do any I/O, try to pre-readahead extra
4571 * blocks from the inode table.
4573 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4574 ext4_fsblk_t b, end, table;
4576 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4578 table = ext4_inode_table(sb, gdp);
4579 /* s_inode_readahead_blks is always a power of 2 */
4580 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4584 num = EXT4_INODES_PER_GROUP(sb);
4585 if (ext4_has_group_desc_csum(sb))
4586 num -= ext4_itable_unused_count(sb, gdp);
4587 table += num / inodes_per_block;
4591 sb_breadahead(sb, b++);
4595 * There are other valid inodes in the buffer, this inode
4596 * has in-inode xattrs, or we don't have this inode in memory.
4597 * Read the block from disk.
4599 trace_ext4_load_inode(inode);
4601 bh->b_end_io = end_buffer_read_sync;
4602 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4604 if (!buffer_uptodate(bh)) {
4605 EXT4_ERROR_INODE_BLOCK(inode, block,
4606 "unable to read itable block");
4616 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4618 /* We have all inode data except xattrs in memory here. */
4619 return __ext4_get_inode_loc(inode, iloc,
4620 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4623 static bool ext4_should_use_dax(struct inode *inode)
4625 if (!test_opt(inode->i_sb, DAX))
4627 if (!S_ISREG(inode->i_mode))
4629 if (ext4_should_journal_data(inode))
4631 if (ext4_has_inline_data(inode))
4633 if (ext4_encrypted_inode(inode))
4638 void ext4_set_inode_flags(struct inode *inode)
4640 unsigned int flags = EXT4_I(inode)->i_flags;
4641 unsigned int new_fl = 0;
4643 if (flags & EXT4_SYNC_FL)
4645 if (flags & EXT4_APPEND_FL)
4647 if (flags & EXT4_IMMUTABLE_FL)
4648 new_fl |= S_IMMUTABLE;
4649 if (flags & EXT4_NOATIME_FL)
4650 new_fl |= S_NOATIME;
4651 if (flags & EXT4_DIRSYNC_FL)
4652 new_fl |= S_DIRSYNC;
4653 if (ext4_should_use_dax(inode))
4655 if (flags & EXT4_ENCRYPT_FL)
4656 new_fl |= S_ENCRYPTED;
4657 inode_set_flags(inode, new_fl,
4658 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4662 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4663 struct ext4_inode_info *ei)
4666 struct inode *inode = &(ei->vfs_inode);
4667 struct super_block *sb = inode->i_sb;
4669 if (ext4_has_feature_huge_file(sb)) {
4670 /* we are using combined 48 bit field */
4671 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4672 le32_to_cpu(raw_inode->i_blocks_lo);
4673 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4674 /* i_blocks represent file system block size */
4675 return i_blocks << (inode->i_blkbits - 9);
4680 return le32_to_cpu(raw_inode->i_blocks_lo);
4684 static inline void ext4_iget_extra_inode(struct inode *inode,
4685 struct ext4_inode *raw_inode,
4686 struct ext4_inode_info *ei)
4688 __le32 *magic = (void *)raw_inode +
4689 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4690 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4691 EXT4_INODE_SIZE(inode->i_sb) &&
4692 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4693 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4694 ext4_find_inline_data_nolock(inode);
4696 EXT4_I(inode)->i_inline_off = 0;
4699 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4701 if (!ext4_has_feature_project(inode->i_sb))
4703 *projid = EXT4_I(inode)->i_projid;
4707 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4709 struct ext4_iloc iloc;
4710 struct ext4_inode *raw_inode;
4711 struct ext4_inode_info *ei;
4712 struct inode *inode;
4713 journal_t *journal = EXT4_SB(sb)->s_journal;
4721 inode = iget_locked(sb, ino);
4723 return ERR_PTR(-ENOMEM);
4724 if (!(inode->i_state & I_NEW))
4730 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4733 raw_inode = ext4_raw_inode(&iloc);
4735 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4736 EXT4_ERROR_INODE(inode, "root inode unallocated");
4737 ret = -EFSCORRUPTED;
4741 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4742 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4743 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4744 EXT4_INODE_SIZE(inode->i_sb) ||
4745 (ei->i_extra_isize & 3)) {
4746 EXT4_ERROR_INODE(inode,
4747 "bad extra_isize %u (inode size %u)",
4749 EXT4_INODE_SIZE(inode->i_sb));
4750 ret = -EFSCORRUPTED;
4754 ei->i_extra_isize = 0;
4756 /* Precompute checksum seed for inode metadata */
4757 if (ext4_has_metadata_csum(sb)) {
4758 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4760 __le32 inum = cpu_to_le32(inode->i_ino);
4761 __le32 gen = raw_inode->i_generation;
4762 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4764 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4768 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4769 EXT4_ERROR_INODE(inode, "checksum invalid");
4774 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4775 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4776 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4777 if (ext4_has_feature_project(sb) &&
4778 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4779 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4780 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4782 i_projid = EXT4_DEF_PROJID;
4784 if (!(test_opt(inode->i_sb, NO_UID32))) {
4785 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4786 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4788 i_uid_write(inode, i_uid);
4789 i_gid_write(inode, i_gid);
4790 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4791 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4793 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4794 ei->i_inline_off = 0;
4795 ei->i_dir_start_lookup = 0;
4796 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4797 /* We now have enough fields to check if the inode was active or not.
4798 * This is needed because nfsd might try to access dead inodes
4799 * the test is that same one that e2fsck uses
4800 * NeilBrown 1999oct15
4802 if (inode->i_nlink == 0) {
4803 if ((inode->i_mode == 0 ||
4804 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4805 ino != EXT4_BOOT_LOADER_INO) {
4806 /* this inode is deleted */
4810 /* The only unlinked inodes we let through here have
4811 * valid i_mode and are being read by the orphan
4812 * recovery code: that's fine, we're about to complete
4813 * the process of deleting those.
4814 * OR it is the EXT4_BOOT_LOADER_INO which is
4815 * not initialized on a new filesystem. */
4817 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4818 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4819 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4820 if (ext4_has_feature_64bit(sb))
4822 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4823 inode->i_size = ext4_isize(sb, raw_inode);
4824 if ((size = i_size_read(inode)) < 0) {
4825 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4826 ret = -EFSCORRUPTED;
4829 ei->i_disksize = inode->i_size;
4831 ei->i_reserved_quota = 0;
4833 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4834 ei->i_block_group = iloc.block_group;
4835 ei->i_last_alloc_group = ~0;
4837 * NOTE! The in-memory inode i_data array is in little-endian order
4838 * even on big-endian machines: we do NOT byteswap the block numbers!
4840 for (block = 0; block < EXT4_N_BLOCKS; block++)
4841 ei->i_data[block] = raw_inode->i_block[block];
4842 INIT_LIST_HEAD(&ei->i_orphan);
4845 * Set transaction id's of transactions that have to be committed
4846 * to finish f[data]sync. We set them to currently running transaction
4847 * as we cannot be sure that the inode or some of its metadata isn't
4848 * part of the transaction - the inode could have been reclaimed and
4849 * now it is reread from disk.
4852 transaction_t *transaction;
4855 read_lock(&journal->j_state_lock);
4856 if (journal->j_running_transaction)
4857 transaction = journal->j_running_transaction;
4859 transaction = journal->j_committing_transaction;
4861 tid = transaction->t_tid;
4863 tid = journal->j_commit_sequence;
4864 read_unlock(&journal->j_state_lock);
4865 ei->i_sync_tid = tid;
4866 ei->i_datasync_tid = tid;
4869 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4870 if (ei->i_extra_isize == 0) {
4871 /* The extra space is currently unused. Use it. */
4872 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4873 ei->i_extra_isize = sizeof(struct ext4_inode) -
4874 EXT4_GOOD_OLD_INODE_SIZE;
4876 ext4_iget_extra_inode(inode, raw_inode, ei);
4880 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4881 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4882 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4883 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4885 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4886 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4888 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4889 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4891 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4893 inode_set_iversion_queried(inode, ivers);
4897 if (ei->i_file_acl &&
4898 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4899 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4901 ret = -EFSCORRUPTED;
4903 } else if (!ext4_has_inline_data(inode)) {
4904 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4905 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4906 (S_ISLNK(inode->i_mode) &&
4907 !ext4_inode_is_fast_symlink(inode))))
4908 /* Validate extent which is part of inode */
4909 ret = ext4_ext_check_inode(inode);
4910 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4911 (S_ISLNK(inode->i_mode) &&
4912 !ext4_inode_is_fast_symlink(inode))) {
4913 /* Validate block references which are part of inode */
4914 ret = ext4_ind_check_inode(inode);
4920 if (S_ISREG(inode->i_mode)) {
4921 inode->i_op = &ext4_file_inode_operations;
4922 inode->i_fop = &ext4_file_operations;
4923 ext4_set_aops(inode);
4924 } else if (S_ISDIR(inode->i_mode)) {
4925 inode->i_op = &ext4_dir_inode_operations;
4926 inode->i_fop = &ext4_dir_operations;
4927 } else if (S_ISLNK(inode->i_mode)) {
4928 if (ext4_encrypted_inode(inode)) {
4929 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4930 ext4_set_aops(inode);
4931 } else if (ext4_inode_is_fast_symlink(inode)) {
4932 inode->i_link = (char *)ei->i_data;
4933 inode->i_op = &ext4_fast_symlink_inode_operations;
4934 nd_terminate_link(ei->i_data, inode->i_size,
4935 sizeof(ei->i_data) - 1);
4937 inode->i_op = &ext4_symlink_inode_operations;
4938 ext4_set_aops(inode);
4940 inode_nohighmem(inode);
4941 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4942 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4943 inode->i_op = &ext4_special_inode_operations;
4944 if (raw_inode->i_block[0])
4945 init_special_inode(inode, inode->i_mode,
4946 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4948 init_special_inode(inode, inode->i_mode,
4949 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4950 } else if (ino == EXT4_BOOT_LOADER_INO) {
4951 make_bad_inode(inode);
4953 ret = -EFSCORRUPTED;
4954 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4958 ext4_set_inode_flags(inode);
4960 unlock_new_inode(inode);
4966 return ERR_PTR(ret);
4969 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4971 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4972 return ERR_PTR(-EFSCORRUPTED);
4973 return ext4_iget(sb, ino);
4976 static int ext4_inode_blocks_set(handle_t *handle,
4977 struct ext4_inode *raw_inode,
4978 struct ext4_inode_info *ei)
4980 struct inode *inode = &(ei->vfs_inode);
4981 u64 i_blocks = inode->i_blocks;
4982 struct super_block *sb = inode->i_sb;
4984 if (i_blocks <= ~0U) {
4986 * i_blocks can be represented in a 32 bit variable
4987 * as multiple of 512 bytes
4989 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4990 raw_inode->i_blocks_high = 0;
4991 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4994 if (!ext4_has_feature_huge_file(sb))
4997 if (i_blocks <= 0xffffffffffffULL) {
4999 * i_blocks can be represented in a 48 bit variable
5000 * as multiple of 512 bytes
5002 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5003 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5004 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5006 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5007 /* i_block is stored in file system block size */
5008 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5009 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5010 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5015 struct other_inode {
5016 unsigned long orig_ino;
5017 struct ext4_inode *raw_inode;
5020 static int other_inode_match(struct inode * inode, unsigned long ino,
5023 struct other_inode *oi = (struct other_inode *) data;
5025 if ((inode->i_ino != ino) ||
5026 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5027 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
5028 ((inode->i_state & I_DIRTY_TIME) == 0))
5030 spin_lock(&inode->i_lock);
5031 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5032 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
5033 (inode->i_state & I_DIRTY_TIME)) {
5034 struct ext4_inode_info *ei = EXT4_I(inode);
5036 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5037 spin_unlock(&inode->i_lock);
5039 spin_lock(&ei->i_raw_lock);
5040 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5041 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5042 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5043 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5044 spin_unlock(&ei->i_raw_lock);
5045 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5048 spin_unlock(&inode->i_lock);
5053 * Opportunistically update the other time fields for other inodes in
5054 * the same inode table block.
5056 static void ext4_update_other_inodes_time(struct super_block *sb,
5057 unsigned long orig_ino, char *buf)
5059 struct other_inode oi;
5061 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5062 int inode_size = EXT4_INODE_SIZE(sb);
5064 oi.orig_ino = orig_ino;
5066 * Calculate the first inode in the inode table block. Inode
5067 * numbers are one-based. That is, the first inode in a block
5068 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5070 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5071 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5072 if (ino == orig_ino)
5074 oi.raw_inode = (struct ext4_inode *) buf;
5075 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5080 * Post the struct inode info into an on-disk inode location in the
5081 * buffer-cache. This gobbles the caller's reference to the
5082 * buffer_head in the inode location struct.
5084 * The caller must have write access to iloc->bh.
5086 static int ext4_do_update_inode(handle_t *handle,
5087 struct inode *inode,
5088 struct ext4_iloc *iloc)
5090 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5091 struct ext4_inode_info *ei = EXT4_I(inode);
5092 struct buffer_head *bh = iloc->bh;
5093 struct super_block *sb = inode->i_sb;
5094 int err = 0, rc, block;
5095 int need_datasync = 0, set_large_file = 0;
5100 spin_lock(&ei->i_raw_lock);
5102 /* For fields not tracked in the in-memory inode,
5103 * initialise them to zero for new inodes. */
5104 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5105 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5107 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5108 i_uid = i_uid_read(inode);
5109 i_gid = i_gid_read(inode);
5110 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5111 if (!(test_opt(inode->i_sb, NO_UID32))) {
5112 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5113 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5115 * Fix up interoperability with old kernels. Otherwise, old inodes get
5116 * re-used with the upper 16 bits of the uid/gid intact
5118 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5119 raw_inode->i_uid_high = 0;
5120 raw_inode->i_gid_high = 0;
5122 raw_inode->i_uid_high =
5123 cpu_to_le16(high_16_bits(i_uid));
5124 raw_inode->i_gid_high =
5125 cpu_to_le16(high_16_bits(i_gid));
5128 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5129 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5130 raw_inode->i_uid_high = 0;
5131 raw_inode->i_gid_high = 0;
5133 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5135 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5136 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5137 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5138 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5140 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5142 spin_unlock(&ei->i_raw_lock);
5145 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5146 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5147 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5148 raw_inode->i_file_acl_high =
5149 cpu_to_le16(ei->i_file_acl >> 32);
5150 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5151 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5152 ext4_isize_set(raw_inode, ei->i_disksize);
5155 if (ei->i_disksize > 0x7fffffffULL) {
5156 if (!ext4_has_feature_large_file(sb) ||
5157 EXT4_SB(sb)->s_es->s_rev_level ==
5158 cpu_to_le32(EXT4_GOOD_OLD_REV))
5161 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5162 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5163 if (old_valid_dev(inode->i_rdev)) {
5164 raw_inode->i_block[0] =
5165 cpu_to_le32(old_encode_dev(inode->i_rdev));
5166 raw_inode->i_block[1] = 0;
5168 raw_inode->i_block[0] = 0;
5169 raw_inode->i_block[1] =
5170 cpu_to_le32(new_encode_dev(inode->i_rdev));
5171 raw_inode->i_block[2] = 0;
5173 } else if (!ext4_has_inline_data(inode)) {
5174 for (block = 0; block < EXT4_N_BLOCKS; block++)
5175 raw_inode->i_block[block] = ei->i_data[block];
5178 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5179 u64 ivers = inode_peek_iversion(inode);
5181 raw_inode->i_disk_version = cpu_to_le32(ivers);
5182 if (ei->i_extra_isize) {
5183 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5184 raw_inode->i_version_hi =
5185 cpu_to_le32(ivers >> 32);
5186 raw_inode->i_extra_isize =
5187 cpu_to_le16(ei->i_extra_isize);
5191 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5192 i_projid != EXT4_DEF_PROJID);
5194 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5195 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5196 raw_inode->i_projid = cpu_to_le32(i_projid);
5198 ext4_inode_csum_set(inode, raw_inode, ei);
5199 spin_unlock(&ei->i_raw_lock);
5200 if (inode->i_sb->s_flags & SB_LAZYTIME)
5201 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5204 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5205 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5208 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5209 if (set_large_file) {
5210 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5211 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5214 ext4_update_dynamic_rev(sb);
5215 ext4_set_feature_large_file(sb);
5216 ext4_handle_sync(handle);
5217 err = ext4_handle_dirty_super(handle, sb);
5219 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5222 ext4_std_error(inode->i_sb, err);
5227 * ext4_write_inode()
5229 * We are called from a few places:
5231 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5232 * Here, there will be no transaction running. We wait for any running
5233 * transaction to commit.
5235 * - Within flush work (sys_sync(), kupdate and such).
5236 * We wait on commit, if told to.
5238 * - Within iput_final() -> write_inode_now()
5239 * We wait on commit, if told to.
5241 * In all cases it is actually safe for us to return without doing anything,
5242 * because the inode has been copied into a raw inode buffer in
5243 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5246 * Note that we are absolutely dependent upon all inode dirtiers doing the
5247 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5248 * which we are interested.
5250 * It would be a bug for them to not do this. The code:
5252 * mark_inode_dirty(inode)
5254 * inode->i_size = expr;
5256 * is in error because write_inode() could occur while `stuff()' is running,
5257 * and the new i_size will be lost. Plus the inode will no longer be on the
5258 * superblock's dirty inode list.
5260 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5264 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5267 if (EXT4_SB(inode->i_sb)->s_journal) {
5268 if (ext4_journal_current_handle()) {
5269 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5275 * No need to force transaction in WB_SYNC_NONE mode. Also
5276 * ext4_sync_fs() will force the commit after everything is
5279 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5282 err = ext4_force_commit(inode->i_sb);
5284 struct ext4_iloc iloc;
5286 err = __ext4_get_inode_loc(inode, &iloc, 0);
5290 * sync(2) will flush the whole buffer cache. No need to do
5291 * it here separately for each inode.
5293 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5294 sync_dirty_buffer(iloc.bh);
5295 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5296 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5297 "IO error syncing inode");
5306 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5307 * buffers that are attached to a page stradding i_size and are undergoing
5308 * commit. In that case we have to wait for commit to finish and try again.
5310 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5314 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5315 tid_t commit_tid = 0;
5318 offset = inode->i_size & (PAGE_SIZE - 1);
5320 * All buffers in the last page remain valid? Then there's nothing to
5321 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5324 if (offset > PAGE_SIZE - i_blocksize(inode))
5327 page = find_lock_page(inode->i_mapping,
5328 inode->i_size >> PAGE_SHIFT);
5331 ret = __ext4_journalled_invalidatepage(page, offset,
5332 PAGE_SIZE - offset);
5338 read_lock(&journal->j_state_lock);
5339 if (journal->j_committing_transaction)
5340 commit_tid = journal->j_committing_transaction->t_tid;
5341 read_unlock(&journal->j_state_lock);
5343 jbd2_log_wait_commit(journal, commit_tid);
5350 * Called from notify_change.
5352 * We want to trap VFS attempts to truncate the file as soon as
5353 * possible. In particular, we want to make sure that when the VFS
5354 * shrinks i_size, we put the inode on the orphan list and modify
5355 * i_disksize immediately, so that during the subsequent flushing of
5356 * dirty pages and freeing of disk blocks, we can guarantee that any
5357 * commit will leave the blocks being flushed in an unused state on
5358 * disk. (On recovery, the inode will get truncated and the blocks will
5359 * be freed, so we have a strong guarantee that no future commit will
5360 * leave these blocks visible to the user.)
5362 * Another thing we have to assure is that if we are in ordered mode
5363 * and inode is still attached to the committing transaction, we must
5364 * we start writeout of all the dirty pages which are being truncated.
5365 * This way we are sure that all the data written in the previous
5366 * transaction are already on disk (truncate waits for pages under
5369 * Called with inode->i_mutex down.
5371 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5373 struct inode *inode = d_inode(dentry);
5376 const unsigned int ia_valid = attr->ia_valid;
5378 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5381 error = setattr_prepare(dentry, attr);
5385 error = fscrypt_prepare_setattr(dentry, attr);
5389 if (is_quota_modification(inode, attr)) {
5390 error = dquot_initialize(inode);
5394 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5395 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5398 /* (user+group)*(old+new) structure, inode write (sb,
5399 * inode block, ? - but truncate inode update has it) */
5400 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5401 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5402 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5403 if (IS_ERR(handle)) {
5404 error = PTR_ERR(handle);
5408 /* dquot_transfer() calls back ext4_get_inode_usage() which
5409 * counts xattr inode references.
5411 down_read(&EXT4_I(inode)->xattr_sem);
5412 error = dquot_transfer(inode, attr);
5413 up_read(&EXT4_I(inode)->xattr_sem);
5416 ext4_journal_stop(handle);
5419 /* Update corresponding info in inode so that everything is in
5420 * one transaction */
5421 if (attr->ia_valid & ATTR_UID)
5422 inode->i_uid = attr->ia_uid;
5423 if (attr->ia_valid & ATTR_GID)
5424 inode->i_gid = attr->ia_gid;
5425 error = ext4_mark_inode_dirty(handle, inode);
5426 ext4_journal_stop(handle);
5429 if (attr->ia_valid & ATTR_SIZE) {
5431 loff_t oldsize = inode->i_size;
5432 int shrink = (attr->ia_size <= inode->i_size);
5434 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5435 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5437 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5440 if (!S_ISREG(inode->i_mode))
5443 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5444 inode_inc_iversion(inode);
5446 if (ext4_should_order_data(inode) &&
5447 (attr->ia_size < inode->i_size)) {
5448 error = ext4_begin_ordered_truncate(inode,
5453 if (attr->ia_size != inode->i_size) {
5454 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5455 if (IS_ERR(handle)) {
5456 error = PTR_ERR(handle);
5459 if (ext4_handle_valid(handle) && shrink) {
5460 error = ext4_orphan_add(handle, inode);
5464 * Update c/mtime on truncate up, ext4_truncate() will
5465 * update c/mtime in shrink case below
5468 inode->i_mtime = current_time(inode);
5469 inode->i_ctime = inode->i_mtime;
5471 down_write(&EXT4_I(inode)->i_data_sem);
5472 EXT4_I(inode)->i_disksize = attr->ia_size;
5473 rc = ext4_mark_inode_dirty(handle, inode);
5477 * We have to update i_size under i_data_sem together
5478 * with i_disksize to avoid races with writeback code
5479 * running ext4_wb_update_i_disksize().
5482 i_size_write(inode, attr->ia_size);
5483 up_write(&EXT4_I(inode)->i_data_sem);
5484 ext4_journal_stop(handle);
5487 ext4_orphan_del(NULL, inode);
5492 pagecache_isize_extended(inode, oldsize, inode->i_size);
5495 * Blocks are going to be removed from the inode. Wait
5496 * for dio in flight. Temporarily disable
5497 * dioread_nolock to prevent livelock.
5500 if (!ext4_should_journal_data(inode)) {
5501 inode_dio_wait(inode);
5503 ext4_wait_for_tail_page_commit(inode);
5505 down_write(&EXT4_I(inode)->i_mmap_sem);
5507 * Truncate pagecache after we've waited for commit
5508 * in data=journal mode to make pages freeable.
5510 truncate_pagecache(inode, inode->i_size);
5512 rc = ext4_truncate(inode);
5516 up_write(&EXT4_I(inode)->i_mmap_sem);
5520 setattr_copy(inode, attr);
5521 mark_inode_dirty(inode);
5525 * If the call to ext4_truncate failed to get a transaction handle at
5526 * all, we need to clean up the in-core orphan list manually.
5528 if (orphan && inode->i_nlink)
5529 ext4_orphan_del(NULL, inode);
5531 if (!error && (ia_valid & ATTR_MODE))
5532 rc = posix_acl_chmod(inode, inode->i_mode);
5535 ext4_std_error(inode->i_sb, error);
5541 int ext4_getattr(const struct path *path, struct kstat *stat,
5542 u32 request_mask, unsigned int query_flags)
5544 struct inode *inode = d_inode(path->dentry);
5545 struct ext4_inode *raw_inode;
5546 struct ext4_inode_info *ei = EXT4_I(inode);
5549 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5550 stat->result_mask |= STATX_BTIME;
5551 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5552 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5555 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5556 if (flags & EXT4_APPEND_FL)
5557 stat->attributes |= STATX_ATTR_APPEND;
5558 if (flags & EXT4_COMPR_FL)
5559 stat->attributes |= STATX_ATTR_COMPRESSED;
5560 if (flags & EXT4_ENCRYPT_FL)
5561 stat->attributes |= STATX_ATTR_ENCRYPTED;
5562 if (flags & EXT4_IMMUTABLE_FL)
5563 stat->attributes |= STATX_ATTR_IMMUTABLE;
5564 if (flags & EXT4_NODUMP_FL)
5565 stat->attributes |= STATX_ATTR_NODUMP;
5567 stat->attributes_mask |= (STATX_ATTR_APPEND |
5568 STATX_ATTR_COMPRESSED |
5569 STATX_ATTR_ENCRYPTED |
5570 STATX_ATTR_IMMUTABLE |
5573 generic_fillattr(inode, stat);
5577 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5578 u32 request_mask, unsigned int query_flags)
5580 struct inode *inode = d_inode(path->dentry);
5581 u64 delalloc_blocks;
5583 ext4_getattr(path, stat, request_mask, query_flags);
5586 * If there is inline data in the inode, the inode will normally not
5587 * have data blocks allocated (it may have an external xattr block).
5588 * Report at least one sector for such files, so tools like tar, rsync,
5589 * others don't incorrectly think the file is completely sparse.
5591 if (unlikely(ext4_has_inline_data(inode)))
5592 stat->blocks += (stat->size + 511) >> 9;
5595 * We can't update i_blocks if the block allocation is delayed
5596 * otherwise in the case of system crash before the real block
5597 * allocation is done, we will have i_blocks inconsistent with
5598 * on-disk file blocks.
5599 * We always keep i_blocks updated together with real
5600 * allocation. But to not confuse with user, stat
5601 * will return the blocks that include the delayed allocation
5602 * blocks for this file.
5604 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5605 EXT4_I(inode)->i_reserved_data_blocks);
5606 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5610 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5613 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5614 return ext4_ind_trans_blocks(inode, lblocks);
5615 return ext4_ext_index_trans_blocks(inode, pextents);
5619 * Account for index blocks, block groups bitmaps and block group
5620 * descriptor blocks if modify datablocks and index blocks
5621 * worse case, the indexs blocks spread over different block groups
5623 * If datablocks are discontiguous, they are possible to spread over
5624 * different block groups too. If they are contiguous, with flexbg,
5625 * they could still across block group boundary.
5627 * Also account for superblock, inode, quota and xattr blocks
5629 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5632 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5638 * How many index blocks need to touch to map @lblocks logical blocks
5639 * to @pextents physical extents?
5641 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5646 * Now let's see how many group bitmaps and group descriptors need
5649 groups = idxblocks + pextents;
5651 if (groups > ngroups)
5653 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5654 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5656 /* bitmaps and block group descriptor blocks */
5657 ret += groups + gdpblocks;
5659 /* Blocks for super block, inode, quota and xattr blocks */
5660 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5666 * Calculate the total number of credits to reserve to fit
5667 * the modification of a single pages into a single transaction,
5668 * which may include multiple chunks of block allocations.
5670 * This could be called via ext4_write_begin()
5672 * We need to consider the worse case, when
5673 * one new block per extent.
5675 int ext4_writepage_trans_blocks(struct inode *inode)
5677 int bpp = ext4_journal_blocks_per_page(inode);
5680 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5682 /* Account for data blocks for journalled mode */
5683 if (ext4_should_journal_data(inode))
5689 * Calculate the journal credits for a chunk of data modification.
5691 * This is called from DIO, fallocate or whoever calling
5692 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5694 * journal buffers for data blocks are not included here, as DIO
5695 * and fallocate do no need to journal data buffers.
5697 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5699 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5703 * The caller must have previously called ext4_reserve_inode_write().
5704 * Give this, we know that the caller already has write access to iloc->bh.
5706 int ext4_mark_iloc_dirty(handle_t *handle,
5707 struct inode *inode, struct ext4_iloc *iloc)
5711 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5714 if (IS_I_VERSION(inode))
5715 inode_inc_iversion(inode);
5717 /* the do_update_inode consumes one bh->b_count */
5720 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5721 err = ext4_do_update_inode(handle, inode, iloc);
5727 * On success, We end up with an outstanding reference count against
5728 * iloc->bh. This _must_ be cleaned up later.
5732 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5733 struct ext4_iloc *iloc)
5737 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5740 err = ext4_get_inode_loc(inode, iloc);
5742 BUFFER_TRACE(iloc->bh, "get_write_access");
5743 err = ext4_journal_get_write_access(handle, iloc->bh);
5749 ext4_std_error(inode->i_sb, err);
5753 static int __ext4_expand_extra_isize(struct inode *inode,
5754 unsigned int new_extra_isize,
5755 struct ext4_iloc *iloc,
5756 handle_t *handle, int *no_expand)
5758 struct ext4_inode *raw_inode;
5759 struct ext4_xattr_ibody_header *header;
5762 raw_inode = ext4_raw_inode(iloc);
5764 header = IHDR(inode, raw_inode);
5766 /* No extended attributes present */
5767 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5768 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5769 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5770 EXT4_I(inode)->i_extra_isize, 0,
5771 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5772 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5776 /* try to expand with EAs present */
5777 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5781 * Inode size expansion failed; don't try again
5790 * Expand an inode by new_extra_isize bytes.
5791 * Returns 0 on success or negative error number on failure.
5793 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5794 unsigned int new_extra_isize,
5795 struct ext4_iloc iloc,
5801 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5805 * In nojournal mode, we can immediately attempt to expand
5806 * the inode. When journaled, we first need to obtain extra
5807 * buffer credits since we may write into the EA block
5808 * with this same handle. If journal_extend fails, then it will
5809 * only result in a minor loss of functionality for that inode.
5810 * If this is felt to be critical, then e2fsck should be run to
5811 * force a large enough s_min_extra_isize.
5813 if (ext4_handle_valid(handle) &&
5814 jbd2_journal_extend(handle,
5815 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5818 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5821 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5822 handle, &no_expand);
5823 ext4_write_unlock_xattr(inode, &no_expand);
5828 int ext4_expand_extra_isize(struct inode *inode,
5829 unsigned int new_extra_isize,
5830 struct ext4_iloc *iloc)
5836 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5841 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5842 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5843 if (IS_ERR(handle)) {
5844 error = PTR_ERR(handle);
5849 ext4_write_lock_xattr(inode, &no_expand);
5851 BUFFER_TRACE(iloc.bh, "get_write_access");
5852 error = ext4_journal_get_write_access(handle, iloc->bh);
5858 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5859 handle, &no_expand);
5861 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5865 ext4_write_unlock_xattr(inode, &no_expand);
5867 ext4_journal_stop(handle);
5872 * What we do here is to mark the in-core inode as clean with respect to inode
5873 * dirtiness (it may still be data-dirty).
5874 * This means that the in-core inode may be reaped by prune_icache
5875 * without having to perform any I/O. This is a very good thing,
5876 * because *any* task may call prune_icache - even ones which
5877 * have a transaction open against a different journal.
5879 * Is this cheating? Not really. Sure, we haven't written the
5880 * inode out, but prune_icache isn't a user-visible syncing function.
5881 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5882 * we start and wait on commits.
5884 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5886 struct ext4_iloc iloc;
5887 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5891 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5892 err = ext4_reserve_inode_write(handle, inode, &iloc);
5896 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5897 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5900 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5904 * ext4_dirty_inode() is called from __mark_inode_dirty()
5906 * We're really interested in the case where a file is being extended.
5907 * i_size has been changed by generic_commit_write() and we thus need
5908 * to include the updated inode in the current transaction.
5910 * Also, dquot_alloc_block() will always dirty the inode when blocks
5911 * are allocated to the file.
5913 * If the inode is marked synchronous, we don't honour that here - doing
5914 * so would cause a commit on atime updates, which we don't bother doing.
5915 * We handle synchronous inodes at the highest possible level.
5917 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5918 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5919 * to copy into the on-disk inode structure are the timestamp files.
5921 void ext4_dirty_inode(struct inode *inode, int flags)
5925 if (flags == I_DIRTY_TIME)
5927 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5931 ext4_mark_inode_dirty(handle, inode);
5933 ext4_journal_stop(handle);
5940 * Bind an inode's backing buffer_head into this transaction, to prevent
5941 * it from being flushed to disk early. Unlike
5942 * ext4_reserve_inode_write, this leaves behind no bh reference and
5943 * returns no iloc structure, so the caller needs to repeat the iloc
5944 * lookup to mark the inode dirty later.
5946 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5948 struct ext4_iloc iloc;
5952 err = ext4_get_inode_loc(inode, &iloc);
5954 BUFFER_TRACE(iloc.bh, "get_write_access");
5955 err = jbd2_journal_get_write_access(handle, iloc.bh);
5957 err = ext4_handle_dirty_metadata(handle,
5963 ext4_std_error(inode->i_sb, err);
5968 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5973 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5976 * We have to be very careful here: changing a data block's
5977 * journaling status dynamically is dangerous. If we write a
5978 * data block to the journal, change the status and then delete
5979 * that block, we risk forgetting to revoke the old log record
5980 * from the journal and so a subsequent replay can corrupt data.
5981 * So, first we make sure that the journal is empty and that
5982 * nobody is changing anything.
5985 journal = EXT4_JOURNAL(inode);
5988 if (is_journal_aborted(journal))
5991 /* Wait for all existing dio workers */
5992 inode_dio_wait(inode);
5995 * Before flushing the journal and switching inode's aops, we have
5996 * to flush all dirty data the inode has. There can be outstanding
5997 * delayed allocations, there can be unwritten extents created by
5998 * fallocate or buffered writes in dioread_nolock mode covered by
5999 * dirty data which can be converted only after flushing the dirty
6000 * data (and journalled aops don't know how to handle these cases).
6003 down_write(&EXT4_I(inode)->i_mmap_sem);
6004 err = filemap_write_and_wait(inode->i_mapping);
6006 up_write(&EXT4_I(inode)->i_mmap_sem);
6011 percpu_down_write(&sbi->s_journal_flag_rwsem);
6012 jbd2_journal_lock_updates(journal);
6015 * OK, there are no updates running now, and all cached data is
6016 * synced to disk. We are now in a completely consistent state
6017 * which doesn't have anything in the journal, and we know that
6018 * no filesystem updates are running, so it is safe to modify
6019 * the inode's in-core data-journaling state flag now.
6023 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6025 err = jbd2_journal_flush(journal);
6027 jbd2_journal_unlock_updates(journal);
6028 percpu_up_write(&sbi->s_journal_flag_rwsem);
6031 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6033 ext4_set_aops(inode);
6035 jbd2_journal_unlock_updates(journal);
6036 percpu_up_write(&sbi->s_journal_flag_rwsem);
6039 up_write(&EXT4_I(inode)->i_mmap_sem);
6041 /* Finally we can mark the inode as dirty. */
6043 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6045 return PTR_ERR(handle);
6047 err = ext4_mark_inode_dirty(handle, inode);
6048 ext4_handle_sync(handle);
6049 ext4_journal_stop(handle);
6050 ext4_std_error(inode->i_sb, err);
6055 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6057 return !buffer_mapped(bh);
6060 int ext4_page_mkwrite(struct vm_fault *vmf)
6062 struct vm_area_struct *vma = vmf->vma;
6063 struct page *page = vmf->page;
6067 struct file *file = vma->vm_file;
6068 struct inode *inode = file_inode(file);
6069 struct address_space *mapping = inode->i_mapping;
6071 get_block_t *get_block;
6074 sb_start_pagefault(inode->i_sb);
6075 file_update_time(vma->vm_file);
6077 down_read(&EXT4_I(inode)->i_mmap_sem);
6079 ret = ext4_convert_inline_data(inode);
6083 /* Delalloc case is easy... */
6084 if (test_opt(inode->i_sb, DELALLOC) &&
6085 !ext4_should_journal_data(inode) &&
6086 !ext4_nonda_switch(inode->i_sb)) {
6088 ret = block_page_mkwrite(vma, vmf,
6089 ext4_da_get_block_prep);
6090 } while (ret == -ENOSPC &&
6091 ext4_should_retry_alloc(inode->i_sb, &retries));
6096 size = i_size_read(inode);
6097 /* Page got truncated from under us? */
6098 if (page->mapping != mapping || page_offset(page) > size) {
6100 ret = VM_FAULT_NOPAGE;
6104 if (page->index == size >> PAGE_SHIFT)
6105 len = size & ~PAGE_MASK;
6109 * Return if we have all the buffers mapped. This avoids the need to do
6110 * journal_start/journal_stop which can block and take a long time
6112 if (page_has_buffers(page)) {
6113 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6115 ext4_bh_unmapped)) {
6116 /* Wait so that we don't change page under IO */
6117 wait_for_stable_page(page);
6118 ret = VM_FAULT_LOCKED;
6123 /* OK, we need to fill the hole... */
6124 if (ext4_should_dioread_nolock(inode))
6125 get_block = ext4_get_block_unwritten;
6127 get_block = ext4_get_block;
6129 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6130 ext4_writepage_trans_blocks(inode));
6131 if (IS_ERR(handle)) {
6132 ret = VM_FAULT_SIGBUS;
6135 ret = block_page_mkwrite(vma, vmf, get_block);
6136 if (!ret && ext4_should_journal_data(inode)) {
6137 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6138 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6140 ret = VM_FAULT_SIGBUS;
6141 ext4_journal_stop(handle);
6144 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6146 ext4_journal_stop(handle);
6147 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6150 ret = block_page_mkwrite_return(ret);
6152 up_read(&EXT4_I(inode)->i_mmap_sem);
6153 sb_end_pagefault(inode->i_sb);
6157 int ext4_filemap_fault(struct vm_fault *vmf)
6159 struct inode *inode = file_inode(vmf->vma->vm_file);
6162 down_read(&EXT4_I(inode)->i_mmap_sem);
6163 err = filemap_fault(vmf);
6164 up_read(&EXT4_I(inode)->i_mmap_sem);