2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/bitops.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = le16_to_cpu(raw->i_checksum_lo);
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = le16_to_cpu(raw->i_checksum_hi);
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = cpu_to_le16(csum_lo);
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = cpu_to_le16(csum_hi);
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !ext4_has_metadata_csum(inode->i_sb))
87 provided = le16_to_cpu(raw->i_checksum_lo);
88 calculated = ext4_inode_csum(inode, raw, ei);
89 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
90 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
91 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
95 return provided == calculated;
98 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
99 struct ext4_inode_info *ei)
103 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
104 cpu_to_le32(EXT4_OS_LINUX) ||
105 !ext4_has_metadata_csum(inode->i_sb))
108 csum = ext4_inode_csum(inode, raw, ei);
109 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
110 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
111 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
112 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
115 static inline int ext4_begin_ordered_truncate(struct inode *inode,
118 trace_ext4_begin_ordered_truncate(inode, new_size);
120 * If jinode is zero, then we never opened the file for
121 * writing, so there's no need to call
122 * jbd2_journal_begin_ordered_truncate() since there's no
123 * outstanding writes we need to flush.
125 if (!EXT4_I(inode)->jinode)
127 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
128 EXT4_I(inode)->jinode,
132 static void ext4_invalidatepage(struct page *page, unsigned int offset,
133 unsigned int length);
134 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
135 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
136 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
140 * Test whether an inode is a fast symlink.
142 int ext4_inode_is_fast_symlink(struct inode *inode)
144 int ea_blocks = EXT4_I(inode)->i_file_acl ?
145 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
147 if (ext4_has_inline_data(inode))
150 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode) == NULL);
170 jbd_debug(2, "restarting handle %p\n", handle);
171 up_write(&EXT4_I(inode)->i_data_sem);
172 ret = ext4_journal_restart(handle, nblocks);
173 down_write(&EXT4_I(inode)->i_data_sem);
174 ext4_discard_preallocations(inode);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode *inode)
187 trace_ext4_evict_inode(inode);
189 if (inode->i_nlink) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode) &&
209 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
210 inode->i_ino != EXT4_JOURNAL_INO) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_complete_transaction(journal, commit_tid);
215 filemap_write_and_wait(&inode->i_data);
217 truncate_inode_pages_final(&inode->i_data);
219 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
223 if (is_bad_inode(inode))
225 dquot_initialize(inode);
227 if (ext4_should_order_data(inode))
228 ext4_begin_ordered_truncate(inode, 0);
229 truncate_inode_pages_final(&inode->i_data);
231 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode->i_sb);
238 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
239 ext4_blocks_for_truncate(inode)+3);
240 if (IS_ERR(handle)) {
241 ext4_std_error(inode->i_sb, PTR_ERR(handle));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL, inode);
248 sb_end_intwrite(inode->i_sb);
253 ext4_handle_sync(handle);
255 err = ext4_mark_inode_dirty(handle, inode);
257 ext4_warning(inode->i_sb,
258 "couldn't mark inode dirty (err %d)", err);
262 ext4_truncate(inode);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle, 3)) {
271 err = ext4_journal_extend(handle, 3);
273 err = ext4_journal_restart(handle, 3);
275 ext4_warning(inode->i_sb,
276 "couldn't extend journal (err %d)", err);
278 ext4_journal_stop(handle);
279 ext4_orphan_del(NULL, inode);
280 sb_end_intwrite(inode->i_sb);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle, inode);
294 EXT4_I(inode)->i_dtime = get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle, inode))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode);
307 ext4_free_inode(handle, inode);
308 ext4_journal_stop(handle);
309 sb_end_intwrite(inode->i_sb);
312 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
316 qsize_t *ext4_get_reserved_space(struct inode *inode)
318 return &EXT4_I(inode)->i_reserved_quota;
323 * Called with i_data_sem down, which is important since we can call
324 * ext4_discard_preallocations() from here.
326 void ext4_da_update_reserve_space(struct inode *inode,
327 int used, int quota_claim)
329 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
330 struct ext4_inode_info *ei = EXT4_I(inode);
332 spin_lock(&ei->i_block_reservation_lock);
333 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
334 if (unlikely(used > ei->i_reserved_data_blocks)) {
335 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
336 "with only %d reserved data blocks",
337 __func__, inode->i_ino, used,
338 ei->i_reserved_data_blocks);
340 used = ei->i_reserved_data_blocks;
343 /* Update per-inode reservations */
344 ei->i_reserved_data_blocks -= used;
345 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
347 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
349 /* Update quota subsystem for data blocks */
351 dquot_claim_block(inode, EXT4_C2B(sbi, used));
354 * We did fallocate with an offset that is already delayed
355 * allocated. So on delayed allocated writeback we should
356 * not re-claim the quota for fallocated blocks.
358 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
362 * If we have done all the pending block allocations and if
363 * there aren't any writers on the inode, we can discard the
364 * inode's preallocations.
366 if ((ei->i_reserved_data_blocks == 0) &&
367 (atomic_read(&inode->i_writecount) == 0))
368 ext4_discard_preallocations(inode);
371 static int __check_block_validity(struct inode *inode, const char *func,
373 struct ext4_map_blocks *map)
375 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
377 ext4_error_inode(inode, func, line, map->m_pblk,
378 "lblock %lu mapped to illegal pblock "
379 "(length %d)", (unsigned long) map->m_lblk,
381 return -EFSCORRUPTED;
386 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
391 if (ext4_encrypted_inode(inode))
392 return ext4_encrypted_zeroout(inode, lblk, pblk, len);
394 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
401 #define check_block_validity(inode, map) \
402 __check_block_validity((inode), __func__, __LINE__, (map))
404 #ifdef ES_AGGRESSIVE_TEST
405 static void ext4_map_blocks_es_recheck(handle_t *handle,
407 struct ext4_map_blocks *es_map,
408 struct ext4_map_blocks *map,
415 * There is a race window that the result is not the same.
416 * e.g. xfstests #223 when dioread_nolock enables. The reason
417 * is that we lookup a block mapping in extent status tree with
418 * out taking i_data_sem. So at the time the unwritten extent
419 * could be converted.
421 down_read(&EXT4_I(inode)->i_data_sem);
422 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
423 retval = ext4_ext_map_blocks(handle, inode, map, flags &
424 EXT4_GET_BLOCKS_KEEP_SIZE);
426 retval = ext4_ind_map_blocks(handle, inode, map, flags &
427 EXT4_GET_BLOCKS_KEEP_SIZE);
429 up_read((&EXT4_I(inode)->i_data_sem));
432 * We don't check m_len because extent will be collpased in status
433 * tree. So the m_len might not equal.
435 if (es_map->m_lblk != map->m_lblk ||
436 es_map->m_flags != map->m_flags ||
437 es_map->m_pblk != map->m_pblk) {
438 printk("ES cache assertion failed for inode: %lu "
439 "es_cached ex [%d/%d/%llu/%x] != "
440 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
441 inode->i_ino, es_map->m_lblk, es_map->m_len,
442 es_map->m_pblk, es_map->m_flags, map->m_lblk,
443 map->m_len, map->m_pblk, map->m_flags,
447 #endif /* ES_AGGRESSIVE_TEST */
450 * The ext4_map_blocks() function tries to look up the requested blocks,
451 * and returns if the blocks are already mapped.
453 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
454 * and store the allocated blocks in the result buffer head and mark it
457 * If file type is extents based, it will call ext4_ext_map_blocks(),
458 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
461 * On success, it returns the number of blocks being mapped or allocated.
462 * if create==0 and the blocks are pre-allocated and unwritten block,
463 * the result buffer head is unmapped. If the create ==1, it will make sure
464 * the buffer head is mapped.
466 * It returns 0 if plain look up failed (blocks have not been allocated), in
467 * that case, buffer head is unmapped
469 * It returns the error in case of allocation failure.
471 int ext4_map_blocks(handle_t *handle, struct inode *inode,
472 struct ext4_map_blocks *map, int flags)
474 struct extent_status es;
477 #ifdef ES_AGGRESSIVE_TEST
478 struct ext4_map_blocks orig_map;
480 memcpy(&orig_map, map, sizeof(*map));
484 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
485 "logical block %lu\n", inode->i_ino, flags, map->m_len,
486 (unsigned long) map->m_lblk);
489 * ext4_map_blocks returns an int, and m_len is an unsigned int
491 if (unlikely(map->m_len > INT_MAX))
492 map->m_len = INT_MAX;
494 /* We can handle the block number less than EXT_MAX_BLOCKS */
495 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
496 return -EFSCORRUPTED;
498 /* Lookup extent status tree firstly */
499 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
500 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
501 map->m_pblk = ext4_es_pblock(&es) +
502 map->m_lblk - es.es_lblk;
503 map->m_flags |= ext4_es_is_written(&es) ?
504 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
505 retval = es.es_len - (map->m_lblk - es.es_lblk);
506 if (retval > map->m_len)
509 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
514 #ifdef ES_AGGRESSIVE_TEST
515 ext4_map_blocks_es_recheck(handle, inode, map,
522 * Try to see if we can get the block without requesting a new
525 down_read(&EXT4_I(inode)->i_data_sem);
526 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
527 retval = ext4_ext_map_blocks(handle, inode, map, flags &
528 EXT4_GET_BLOCKS_KEEP_SIZE);
530 retval = ext4_ind_map_blocks(handle, inode, map, flags &
531 EXT4_GET_BLOCKS_KEEP_SIZE);
536 if (unlikely(retval != map->m_len)) {
537 ext4_warning(inode->i_sb,
538 "ES len assertion failed for inode "
539 "%lu: retval %d != map->m_len %d",
540 inode->i_ino, retval, map->m_len);
544 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
545 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
546 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
547 !(status & EXTENT_STATUS_WRITTEN) &&
548 ext4_find_delalloc_range(inode, map->m_lblk,
549 map->m_lblk + map->m_len - 1))
550 status |= EXTENT_STATUS_DELAYED;
551 ret = ext4_es_insert_extent(inode, map->m_lblk,
552 map->m_len, map->m_pblk, status);
556 up_read((&EXT4_I(inode)->i_data_sem));
559 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
560 ret = check_block_validity(inode, map);
565 /* If it is only a block(s) look up */
566 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
570 * Returns if the blocks have already allocated
572 * Note that if blocks have been preallocated
573 * ext4_ext_get_block() returns the create = 0
574 * with buffer head unmapped.
576 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
578 * If we need to convert extent to unwritten
579 * we continue and do the actual work in
580 * ext4_ext_map_blocks()
582 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
586 * Here we clear m_flags because after allocating an new extent,
587 * it will be set again.
589 map->m_flags &= ~EXT4_MAP_FLAGS;
592 * New blocks allocate and/or writing to unwritten extent
593 * will possibly result in updating i_data, so we take
594 * the write lock of i_data_sem, and call get_block()
595 * with create == 1 flag.
597 down_write(&EXT4_I(inode)->i_data_sem);
600 * We need to check for EXT4 here because migrate
601 * could have changed the inode type in between
603 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
604 retval = ext4_ext_map_blocks(handle, inode, map, flags);
606 retval = ext4_ind_map_blocks(handle, inode, map, flags);
608 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
610 * We allocated new blocks which will result in
611 * i_data's format changing. Force the migrate
612 * to fail by clearing migrate flags
614 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
618 * Update reserved blocks/metadata blocks after successful
619 * block allocation which had been deferred till now. We don't
620 * support fallocate for non extent files. So we can update
621 * reserve space here.
624 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
625 ext4_da_update_reserve_space(inode, retval, 1);
631 if (unlikely(retval != map->m_len)) {
632 ext4_warning(inode->i_sb,
633 "ES len assertion failed for inode "
634 "%lu: retval %d != map->m_len %d",
635 inode->i_ino, retval, map->m_len);
640 * We have to zeroout blocks before inserting them into extent
641 * status tree. Otherwise someone could look them up there and
642 * use them before they are really zeroed.
644 if (flags & EXT4_GET_BLOCKS_ZERO &&
645 map->m_flags & EXT4_MAP_MAPPED &&
646 map->m_flags & EXT4_MAP_NEW) {
647 ret = ext4_issue_zeroout(inode, map->m_lblk,
648 map->m_pblk, map->m_len);
656 * If the extent has been zeroed out, we don't need to update
657 * extent status tree.
659 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
660 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
661 if (ext4_es_is_written(&es))
664 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
665 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
666 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
667 !(status & EXTENT_STATUS_WRITTEN) &&
668 ext4_find_delalloc_range(inode, map->m_lblk,
669 map->m_lblk + map->m_len - 1))
670 status |= EXTENT_STATUS_DELAYED;
671 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
672 map->m_pblk, status);
680 up_write((&EXT4_I(inode)->i_data_sem));
681 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
682 ret = check_block_validity(inode, map);
689 /* Maximum number of blocks we map for direct IO at once. */
690 #define DIO_MAX_BLOCKS 4096
692 static int _ext4_get_block(struct inode *inode, sector_t iblock,
693 struct buffer_head *bh, int flags)
695 handle_t *handle = ext4_journal_current_handle();
696 struct ext4_map_blocks map;
697 int ret = 0, started = 0;
700 if (ext4_has_inline_data(inode))
704 map.m_len = bh->b_size >> inode->i_blkbits;
706 if (flags && !handle) {
707 /* Direct IO write... */
708 if (map.m_len > DIO_MAX_BLOCKS)
709 map.m_len = DIO_MAX_BLOCKS;
710 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
711 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
713 if (IS_ERR(handle)) {
714 ret = PTR_ERR(handle);
720 ret = ext4_map_blocks(handle, inode, &map, flags);
722 ext4_io_end_t *io_end = ext4_inode_aio(inode);
724 map_bh(bh, inode->i_sb, map.m_pblk);
725 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
726 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
727 set_buffer_defer_completion(bh);
728 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
732 ext4_journal_stop(handle);
736 int ext4_get_block(struct inode *inode, sector_t iblock,
737 struct buffer_head *bh, int create)
739 return _ext4_get_block(inode, iblock, bh,
740 create ? EXT4_GET_BLOCKS_CREATE : 0);
744 * `handle' can be NULL if create is zero
746 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
747 ext4_lblk_t block, int map_flags)
749 struct ext4_map_blocks map;
750 struct buffer_head *bh;
751 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
754 J_ASSERT(handle != NULL || create == 0);
758 err = ext4_map_blocks(handle, inode, &map, map_flags);
761 return create ? ERR_PTR(-ENOSPC) : NULL;
765 bh = sb_getblk(inode->i_sb, map.m_pblk);
767 return ERR_PTR(-ENOMEM);
768 if (map.m_flags & EXT4_MAP_NEW) {
769 J_ASSERT(create != 0);
770 J_ASSERT(handle != NULL);
773 * Now that we do not always journal data, we should
774 * keep in mind whether this should always journal the
775 * new buffer as metadata. For now, regular file
776 * writes use ext4_get_block instead, so it's not a
780 BUFFER_TRACE(bh, "call get_create_access");
781 err = ext4_journal_get_create_access(handle, bh);
786 if (!buffer_uptodate(bh)) {
787 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
788 set_buffer_uptodate(bh);
791 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
792 err = ext4_handle_dirty_metadata(handle, inode, bh);
796 BUFFER_TRACE(bh, "not a new buffer");
803 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
804 ext4_lblk_t block, int map_flags)
806 struct buffer_head *bh;
808 bh = ext4_getblk(handle, inode, block, map_flags);
811 if (!bh || buffer_uptodate(bh))
813 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
815 if (buffer_uptodate(bh))
818 return ERR_PTR(-EIO);
821 int ext4_walk_page_buffers(handle_t *handle,
822 struct buffer_head *head,
826 int (*fn)(handle_t *handle,
827 struct buffer_head *bh))
829 struct buffer_head *bh;
830 unsigned block_start, block_end;
831 unsigned blocksize = head->b_size;
833 struct buffer_head *next;
835 for (bh = head, block_start = 0;
836 ret == 0 && (bh != head || !block_start);
837 block_start = block_end, bh = next) {
838 next = bh->b_this_page;
839 block_end = block_start + blocksize;
840 if (block_end <= from || block_start >= to) {
841 if (partial && !buffer_uptodate(bh))
845 err = (*fn)(handle, bh);
853 * To preserve ordering, it is essential that the hole instantiation and
854 * the data write be encapsulated in a single transaction. We cannot
855 * close off a transaction and start a new one between the ext4_get_block()
856 * and the commit_write(). So doing the jbd2_journal_start at the start of
857 * prepare_write() is the right place.
859 * Also, this function can nest inside ext4_writepage(). In that case, we
860 * *know* that ext4_writepage() has generated enough buffer credits to do the
861 * whole page. So we won't block on the journal in that case, which is good,
862 * because the caller may be PF_MEMALLOC.
864 * By accident, ext4 can be reentered when a transaction is open via
865 * quota file writes. If we were to commit the transaction while thus
866 * reentered, there can be a deadlock - we would be holding a quota
867 * lock, and the commit would never complete if another thread had a
868 * transaction open and was blocking on the quota lock - a ranking
871 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
872 * will _not_ run commit under these circumstances because handle->h_ref
873 * is elevated. We'll still have enough credits for the tiny quotafile
876 int do_journal_get_write_access(handle_t *handle,
877 struct buffer_head *bh)
879 int dirty = buffer_dirty(bh);
882 if (!buffer_mapped(bh) || buffer_freed(bh))
885 * __block_write_begin() could have dirtied some buffers. Clean
886 * the dirty bit as jbd2_journal_get_write_access() could complain
887 * otherwise about fs integrity issues. Setting of the dirty bit
888 * by __block_write_begin() isn't a real problem here as we clear
889 * the bit before releasing a page lock and thus writeback cannot
890 * ever write the buffer.
893 clear_buffer_dirty(bh);
894 BUFFER_TRACE(bh, "get write access");
895 ret = ext4_journal_get_write_access(handle, bh);
897 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
901 #ifdef CONFIG_EXT4_FS_ENCRYPTION
902 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
903 get_block_t *get_block)
905 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
906 unsigned to = from + len;
907 struct inode *inode = page->mapping->host;
908 unsigned block_start, block_end;
911 unsigned blocksize = inode->i_sb->s_blocksize;
913 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
914 bool decrypt = false;
916 BUG_ON(!PageLocked(page));
917 BUG_ON(from > PAGE_CACHE_SIZE);
918 BUG_ON(to > PAGE_CACHE_SIZE);
921 if (!page_has_buffers(page))
922 create_empty_buffers(page, blocksize, 0);
923 head = page_buffers(page);
924 bbits = ilog2(blocksize);
925 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
927 for (bh = head, block_start = 0; bh != head || !block_start;
928 block++, block_start = block_end, bh = bh->b_this_page) {
929 block_end = block_start + blocksize;
930 if (block_end <= from || block_start >= to) {
931 if (PageUptodate(page)) {
932 if (!buffer_uptodate(bh))
933 set_buffer_uptodate(bh);
938 clear_buffer_new(bh);
939 if (!buffer_mapped(bh)) {
940 WARN_ON(bh->b_size != blocksize);
941 err = get_block(inode, block, bh, 1);
944 if (buffer_new(bh)) {
945 unmap_underlying_metadata(bh->b_bdev,
947 if (PageUptodate(page)) {
948 clear_buffer_new(bh);
949 set_buffer_uptodate(bh);
950 mark_buffer_dirty(bh);
953 if (block_end > to || block_start < from)
954 zero_user_segments(page, to, block_end,
959 if (PageUptodate(page)) {
960 if (!buffer_uptodate(bh))
961 set_buffer_uptodate(bh);
964 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
965 !buffer_unwritten(bh) &&
966 (block_start < from || block_end > to)) {
967 ll_rw_block(READ, 1, &bh);
969 decrypt = ext4_encrypted_inode(inode) &&
970 S_ISREG(inode->i_mode);
974 * If we issued read requests, let them complete.
976 while (wait_bh > wait) {
977 wait_on_buffer(*--wait_bh);
978 if (!buffer_uptodate(*wait_bh))
982 page_zero_new_buffers(page, from, to);
984 err = ext4_decrypt(page);
989 static int ext4_write_begin(struct file *file, struct address_space *mapping,
990 loff_t pos, unsigned len, unsigned flags,
991 struct page **pagep, void **fsdata)
993 struct inode *inode = mapping->host;
994 int ret, needed_blocks;
1001 trace_ext4_write_begin(inode, pos, len, flags);
1003 * Reserve one block more for addition to orphan list in case
1004 * we allocate blocks but write fails for some reason
1006 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1007 index = pos >> PAGE_CACHE_SHIFT;
1008 from = pos & (PAGE_CACHE_SIZE - 1);
1011 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1012 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1021 * grab_cache_page_write_begin() can take a long time if the
1022 * system is thrashing due to memory pressure, or if the page
1023 * is being written back. So grab it first before we start
1024 * the transaction handle. This also allows us to allocate
1025 * the page (if needed) without using GFP_NOFS.
1028 page = grab_cache_page_write_begin(mapping, index, flags);
1034 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1035 if (IS_ERR(handle)) {
1036 page_cache_release(page);
1037 return PTR_ERR(handle);
1041 if (page->mapping != mapping) {
1042 /* The page got truncated from under us */
1044 page_cache_release(page);
1045 ext4_journal_stop(handle);
1048 /* In case writeback began while the page was unlocked */
1049 wait_for_stable_page(page);
1051 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1052 if (ext4_should_dioread_nolock(inode))
1053 ret = ext4_block_write_begin(page, pos, len,
1054 ext4_get_block_write);
1056 ret = ext4_block_write_begin(page, pos, len,
1059 if (ext4_should_dioread_nolock(inode))
1060 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1062 ret = __block_write_begin(page, pos, len, ext4_get_block);
1064 if (!ret && ext4_should_journal_data(inode)) {
1065 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1067 do_journal_get_write_access);
1073 * __block_write_begin may have instantiated a few blocks
1074 * outside i_size. Trim these off again. Don't need
1075 * i_size_read because we hold i_mutex.
1077 * Add inode to orphan list in case we crash before
1080 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1081 ext4_orphan_add(handle, inode);
1083 ext4_journal_stop(handle);
1084 if (pos + len > inode->i_size) {
1085 ext4_truncate_failed_write(inode);
1087 * If truncate failed early the inode might
1088 * still be on the orphan list; we need to
1089 * make sure the inode is removed from the
1090 * orphan list in that case.
1093 ext4_orphan_del(NULL, inode);
1096 if (ret == -ENOSPC &&
1097 ext4_should_retry_alloc(inode->i_sb, &retries))
1099 page_cache_release(page);
1106 /* For write_end() in data=journal mode */
1107 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1110 if (!buffer_mapped(bh) || buffer_freed(bh))
1112 set_buffer_uptodate(bh);
1113 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1114 clear_buffer_meta(bh);
1115 clear_buffer_prio(bh);
1120 * We need to pick up the new inode size which generic_commit_write gave us
1121 * `file' can be NULL - eg, when called from page_symlink().
1123 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1124 * buffers are managed internally.
1126 static int ext4_write_end(struct file *file,
1127 struct address_space *mapping,
1128 loff_t pos, unsigned len, unsigned copied,
1129 struct page *page, void *fsdata)
1131 handle_t *handle = ext4_journal_current_handle();
1132 struct inode *inode = mapping->host;
1133 loff_t old_size = inode->i_size;
1135 int i_size_changed = 0;
1137 trace_ext4_write_end(inode, pos, len, copied);
1138 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1139 ret = ext4_jbd2_file_inode(handle, inode);
1142 page_cache_release(page);
1147 if (ext4_has_inline_data(inode)) {
1148 ret = ext4_write_inline_data_end(inode, pos, len,
1154 copied = block_write_end(file, mapping, pos,
1155 len, copied, page, fsdata);
1157 * it's important to update i_size while still holding page lock:
1158 * page writeout could otherwise come in and zero beyond i_size.
1160 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1162 page_cache_release(page);
1165 pagecache_isize_extended(inode, old_size, pos);
1167 * Don't mark the inode dirty under page lock. First, it unnecessarily
1168 * makes the holding time of page lock longer. Second, it forces lock
1169 * ordering of page lock and transaction start for journaling
1173 ext4_mark_inode_dirty(handle, inode);
1175 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1176 /* if we have allocated more blocks and copied
1177 * less. We will have blocks allocated outside
1178 * inode->i_size. So truncate them
1180 ext4_orphan_add(handle, inode);
1182 ret2 = ext4_journal_stop(handle);
1186 if (pos + len > inode->i_size) {
1187 ext4_truncate_failed_write(inode);
1189 * If truncate failed early the inode might still be
1190 * on the orphan list; we need to make sure the inode
1191 * is removed from the orphan list in that case.
1194 ext4_orphan_del(NULL, inode);
1197 return ret ? ret : copied;
1201 * This is a private version of page_zero_new_buffers() which doesn't
1202 * set the buffer to be dirty, since in data=journalled mode we need
1203 * to call ext4_handle_dirty_metadata() instead.
1205 static void zero_new_buffers(struct page *page, unsigned from, unsigned to)
1207 unsigned int block_start = 0, block_end;
1208 struct buffer_head *head, *bh;
1210 bh = head = page_buffers(page);
1212 block_end = block_start + bh->b_size;
1213 if (buffer_new(bh)) {
1214 if (block_end > from && block_start < to) {
1215 if (!PageUptodate(page)) {
1216 unsigned start, size;
1218 start = max(from, block_start);
1219 size = min(to, block_end) - start;
1221 zero_user(page, start, size);
1222 set_buffer_uptodate(bh);
1224 clear_buffer_new(bh);
1227 block_start = block_end;
1228 bh = bh->b_this_page;
1229 } while (bh != head);
1232 static int ext4_journalled_write_end(struct file *file,
1233 struct address_space *mapping,
1234 loff_t pos, unsigned len, unsigned copied,
1235 struct page *page, void *fsdata)
1237 handle_t *handle = ext4_journal_current_handle();
1238 struct inode *inode = mapping->host;
1239 loff_t old_size = inode->i_size;
1243 int size_changed = 0;
1245 trace_ext4_journalled_write_end(inode, pos, len, copied);
1246 from = pos & (PAGE_CACHE_SIZE - 1);
1249 BUG_ON(!ext4_handle_valid(handle));
1251 if (ext4_has_inline_data(inode))
1252 copied = ext4_write_inline_data_end(inode, pos, len,
1256 if (!PageUptodate(page))
1258 zero_new_buffers(page, from+copied, to);
1261 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1262 to, &partial, write_end_fn);
1264 SetPageUptodate(page);
1266 size_changed = ext4_update_inode_size(inode, pos + copied);
1267 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1268 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1270 page_cache_release(page);
1273 pagecache_isize_extended(inode, old_size, pos);
1276 ret2 = ext4_mark_inode_dirty(handle, inode);
1281 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1282 /* if we have allocated more blocks and copied
1283 * less. We will have blocks allocated outside
1284 * inode->i_size. So truncate them
1286 ext4_orphan_add(handle, inode);
1288 ret2 = ext4_journal_stop(handle);
1291 if (pos + len > inode->i_size) {
1292 ext4_truncate_failed_write(inode);
1294 * If truncate failed early the inode might still be
1295 * on the orphan list; we need to make sure the inode
1296 * is removed from the orphan list in that case.
1299 ext4_orphan_del(NULL, inode);
1302 return ret ? ret : copied;
1306 * Reserve space for a single cluster
1308 static int ext4_da_reserve_space(struct inode *inode)
1310 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1311 struct ext4_inode_info *ei = EXT4_I(inode);
1315 * We will charge metadata quota at writeout time; this saves
1316 * us from metadata over-estimation, though we may go over by
1317 * a small amount in the end. Here we just reserve for data.
1319 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1323 spin_lock(&ei->i_block_reservation_lock);
1324 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1325 spin_unlock(&ei->i_block_reservation_lock);
1326 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1329 ei->i_reserved_data_blocks++;
1330 trace_ext4_da_reserve_space(inode);
1331 spin_unlock(&ei->i_block_reservation_lock);
1333 return 0; /* success */
1336 static void ext4_da_release_space(struct inode *inode, int to_free)
1338 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1339 struct ext4_inode_info *ei = EXT4_I(inode);
1342 return; /* Nothing to release, exit */
1344 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1346 trace_ext4_da_release_space(inode, to_free);
1347 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1349 * if there aren't enough reserved blocks, then the
1350 * counter is messed up somewhere. Since this
1351 * function is called from invalidate page, it's
1352 * harmless to return without any action.
1354 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1355 "ino %lu, to_free %d with only %d reserved "
1356 "data blocks", inode->i_ino, to_free,
1357 ei->i_reserved_data_blocks);
1359 to_free = ei->i_reserved_data_blocks;
1361 ei->i_reserved_data_blocks -= to_free;
1363 /* update fs dirty data blocks counter */
1364 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1366 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1368 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1371 static void ext4_da_page_release_reservation(struct page *page,
1372 unsigned int offset,
1373 unsigned int length)
1375 int to_release = 0, contiguous_blks = 0;
1376 struct buffer_head *head, *bh;
1377 unsigned int curr_off = 0;
1378 struct inode *inode = page->mapping->host;
1379 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1380 unsigned int stop = offset + length;
1384 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1386 head = page_buffers(page);
1389 unsigned int next_off = curr_off + bh->b_size;
1391 if (next_off > stop)
1394 if ((offset <= curr_off) && (buffer_delay(bh))) {
1397 clear_buffer_delay(bh);
1398 } else if (contiguous_blks) {
1399 lblk = page->index <<
1400 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1401 lblk += (curr_off >> inode->i_blkbits) -
1403 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1404 contiguous_blks = 0;
1406 curr_off = next_off;
1407 } while ((bh = bh->b_this_page) != head);
1409 if (contiguous_blks) {
1410 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1411 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1412 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1415 /* If we have released all the blocks belonging to a cluster, then we
1416 * need to release the reserved space for that cluster. */
1417 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1418 while (num_clusters > 0) {
1419 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1420 ((num_clusters - 1) << sbi->s_cluster_bits);
1421 if (sbi->s_cluster_ratio == 1 ||
1422 !ext4_find_delalloc_cluster(inode, lblk))
1423 ext4_da_release_space(inode, 1);
1430 * Delayed allocation stuff
1433 struct mpage_da_data {
1434 struct inode *inode;
1435 struct writeback_control *wbc;
1437 pgoff_t first_page; /* The first page to write */
1438 pgoff_t next_page; /* Current page to examine */
1439 pgoff_t last_page; /* Last page to examine */
1441 * Extent to map - this can be after first_page because that can be
1442 * fully mapped. We somewhat abuse m_flags to store whether the extent
1443 * is delalloc or unwritten.
1445 struct ext4_map_blocks map;
1446 struct ext4_io_submit io_submit; /* IO submission data */
1449 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1454 struct pagevec pvec;
1455 struct inode *inode = mpd->inode;
1456 struct address_space *mapping = inode->i_mapping;
1458 /* This is necessary when next_page == 0. */
1459 if (mpd->first_page >= mpd->next_page)
1462 index = mpd->first_page;
1463 end = mpd->next_page - 1;
1465 ext4_lblk_t start, last;
1466 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1467 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1468 ext4_es_remove_extent(inode, start, last - start + 1);
1471 pagevec_init(&pvec, 0);
1472 while (index <= end) {
1473 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1476 for (i = 0; i < nr_pages; i++) {
1477 struct page *page = pvec.pages[i];
1478 if (page->index > end)
1480 BUG_ON(!PageLocked(page));
1481 BUG_ON(PageWriteback(page));
1483 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1484 ClearPageUptodate(page);
1488 index = pvec.pages[nr_pages - 1]->index + 1;
1489 pagevec_release(&pvec);
1493 static void ext4_print_free_blocks(struct inode *inode)
1495 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1496 struct super_block *sb = inode->i_sb;
1497 struct ext4_inode_info *ei = EXT4_I(inode);
1499 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1500 EXT4_C2B(EXT4_SB(inode->i_sb),
1501 ext4_count_free_clusters(sb)));
1502 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1503 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1504 (long long) EXT4_C2B(EXT4_SB(sb),
1505 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1506 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1507 (long long) EXT4_C2B(EXT4_SB(sb),
1508 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1509 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1510 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1511 ei->i_reserved_data_blocks);
1515 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1517 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1521 * This function is grabs code from the very beginning of
1522 * ext4_map_blocks, but assumes that the caller is from delayed write
1523 * time. This function looks up the requested blocks and sets the
1524 * buffer delay bit under the protection of i_data_sem.
1526 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1527 struct ext4_map_blocks *map,
1528 struct buffer_head *bh)
1530 struct extent_status es;
1532 sector_t invalid_block = ~((sector_t) 0xffff);
1533 #ifdef ES_AGGRESSIVE_TEST
1534 struct ext4_map_blocks orig_map;
1536 memcpy(&orig_map, map, sizeof(*map));
1539 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1543 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1544 "logical block %lu\n", inode->i_ino, map->m_len,
1545 (unsigned long) map->m_lblk);
1547 /* Lookup extent status tree firstly */
1548 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1549 if (ext4_es_is_hole(&es)) {
1551 down_read(&EXT4_I(inode)->i_data_sem);
1556 * Delayed extent could be allocated by fallocate.
1557 * So we need to check it.
1559 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1560 map_bh(bh, inode->i_sb, invalid_block);
1562 set_buffer_delay(bh);
1566 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1567 retval = es.es_len - (iblock - es.es_lblk);
1568 if (retval > map->m_len)
1569 retval = map->m_len;
1570 map->m_len = retval;
1571 if (ext4_es_is_written(&es))
1572 map->m_flags |= EXT4_MAP_MAPPED;
1573 else if (ext4_es_is_unwritten(&es))
1574 map->m_flags |= EXT4_MAP_UNWRITTEN;
1578 #ifdef ES_AGGRESSIVE_TEST
1579 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1585 * Try to see if we can get the block without requesting a new
1586 * file system block.
1588 down_read(&EXT4_I(inode)->i_data_sem);
1589 if (ext4_has_inline_data(inode))
1591 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1592 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1594 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1600 * XXX: __block_prepare_write() unmaps passed block,
1604 * If the block was allocated from previously allocated cluster,
1605 * then we don't need to reserve it again. However we still need
1606 * to reserve metadata for every block we're going to write.
1608 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1609 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1610 ret = ext4_da_reserve_space(inode);
1612 /* not enough space to reserve */
1618 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1619 ~0, EXTENT_STATUS_DELAYED);
1625 map_bh(bh, inode->i_sb, invalid_block);
1627 set_buffer_delay(bh);
1628 } else if (retval > 0) {
1630 unsigned int status;
1632 if (unlikely(retval != map->m_len)) {
1633 ext4_warning(inode->i_sb,
1634 "ES len assertion failed for inode "
1635 "%lu: retval %d != map->m_len %d",
1636 inode->i_ino, retval, map->m_len);
1640 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1641 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1642 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1643 map->m_pblk, status);
1649 up_read((&EXT4_I(inode)->i_data_sem));
1655 * This is a special get_block_t callback which is used by
1656 * ext4_da_write_begin(). It will either return mapped block or
1657 * reserve space for a single block.
1659 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1660 * We also have b_blocknr = -1 and b_bdev initialized properly
1662 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1663 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1664 * initialized properly.
1666 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1667 struct buffer_head *bh, int create)
1669 struct ext4_map_blocks map;
1672 BUG_ON(create == 0);
1673 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1675 map.m_lblk = iblock;
1679 * first, we need to know whether the block is allocated already
1680 * preallocated blocks are unmapped but should treated
1681 * the same as allocated blocks.
1683 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1687 map_bh(bh, inode->i_sb, map.m_pblk);
1688 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1690 if (buffer_unwritten(bh)) {
1691 /* A delayed write to unwritten bh should be marked
1692 * new and mapped. Mapped ensures that we don't do
1693 * get_block multiple times when we write to the same
1694 * offset and new ensures that we do proper zero out
1695 * for partial write.
1698 set_buffer_mapped(bh);
1703 static int bget_one(handle_t *handle, struct buffer_head *bh)
1709 static int bput_one(handle_t *handle, struct buffer_head *bh)
1715 static int __ext4_journalled_writepage(struct page *page,
1718 struct address_space *mapping = page->mapping;
1719 struct inode *inode = mapping->host;
1720 struct buffer_head *page_bufs = NULL;
1721 handle_t *handle = NULL;
1722 int ret = 0, err = 0;
1723 int inline_data = ext4_has_inline_data(inode);
1724 struct buffer_head *inode_bh = NULL;
1726 ClearPageChecked(page);
1729 BUG_ON(page->index != 0);
1730 BUG_ON(len > ext4_get_max_inline_size(inode));
1731 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1732 if (inode_bh == NULL)
1735 page_bufs = page_buffers(page);
1740 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1744 * We need to release the page lock before we start the
1745 * journal, so grab a reference so the page won't disappear
1746 * out from under us.
1751 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1752 ext4_writepage_trans_blocks(inode));
1753 if (IS_ERR(handle)) {
1754 ret = PTR_ERR(handle);
1756 goto out_no_pagelock;
1758 BUG_ON(!ext4_handle_valid(handle));
1762 if (page->mapping != mapping) {
1763 /* The page got truncated from under us */
1764 ext4_journal_stop(handle);
1770 BUFFER_TRACE(inode_bh, "get write access");
1771 ret = ext4_journal_get_write_access(handle, inode_bh);
1773 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1776 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1777 do_journal_get_write_access);
1779 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1784 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1785 err = ext4_journal_stop(handle);
1789 if (!ext4_has_inline_data(inode))
1790 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1792 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1801 * Note that we don't need to start a transaction unless we're journaling data
1802 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1803 * need to file the inode to the transaction's list in ordered mode because if
1804 * we are writing back data added by write(), the inode is already there and if
1805 * we are writing back data modified via mmap(), no one guarantees in which
1806 * transaction the data will hit the disk. In case we are journaling data, we
1807 * cannot start transaction directly because transaction start ranks above page
1808 * lock so we have to do some magic.
1810 * This function can get called via...
1811 * - ext4_writepages after taking page lock (have journal handle)
1812 * - journal_submit_inode_data_buffers (no journal handle)
1813 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1814 * - grab_page_cache when doing write_begin (have journal handle)
1816 * We don't do any block allocation in this function. If we have page with
1817 * multiple blocks we need to write those buffer_heads that are mapped. This
1818 * is important for mmaped based write. So if we do with blocksize 1K
1819 * truncate(f, 1024);
1820 * a = mmap(f, 0, 4096);
1822 * truncate(f, 4096);
1823 * we have in the page first buffer_head mapped via page_mkwrite call back
1824 * but other buffer_heads would be unmapped but dirty (dirty done via the
1825 * do_wp_page). So writepage should write the first block. If we modify
1826 * the mmap area beyond 1024 we will again get a page_fault and the
1827 * page_mkwrite callback will do the block allocation and mark the
1828 * buffer_heads mapped.
1830 * We redirty the page if we have any buffer_heads that is either delay or
1831 * unwritten in the page.
1833 * We can get recursively called as show below.
1835 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1838 * But since we don't do any block allocation we should not deadlock.
1839 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1841 static int ext4_writepage(struct page *page,
1842 struct writeback_control *wbc)
1847 struct buffer_head *page_bufs = NULL;
1848 struct inode *inode = page->mapping->host;
1849 struct ext4_io_submit io_submit;
1850 bool keep_towrite = false;
1852 trace_ext4_writepage(page);
1853 size = i_size_read(inode);
1854 if (page->index == size >> PAGE_CACHE_SHIFT)
1855 len = size & ~PAGE_CACHE_MASK;
1857 len = PAGE_CACHE_SIZE;
1859 page_bufs = page_buffers(page);
1861 * We cannot do block allocation or other extent handling in this
1862 * function. If there are buffers needing that, we have to redirty
1863 * the page. But we may reach here when we do a journal commit via
1864 * journal_submit_inode_data_buffers() and in that case we must write
1865 * allocated buffers to achieve data=ordered mode guarantees.
1867 * Also, if there is only one buffer per page (the fs block
1868 * size == the page size), if one buffer needs block
1869 * allocation or needs to modify the extent tree to clear the
1870 * unwritten flag, we know that the page can't be written at
1871 * all, so we might as well refuse the write immediately.
1872 * Unfortunately if the block size != page size, we can't as
1873 * easily detect this case using ext4_walk_page_buffers(), but
1874 * for the extremely common case, this is an optimization that
1875 * skips a useless round trip through ext4_bio_write_page().
1877 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1878 ext4_bh_delay_or_unwritten)) {
1879 redirty_page_for_writepage(wbc, page);
1880 if ((current->flags & PF_MEMALLOC) ||
1881 (inode->i_sb->s_blocksize == PAGE_CACHE_SIZE)) {
1883 * For memory cleaning there's no point in writing only
1884 * some buffers. So just bail out. Warn if we came here
1885 * from direct reclaim.
1887 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1892 keep_towrite = true;
1895 if (PageChecked(page) && ext4_should_journal_data(inode))
1897 * It's mmapped pagecache. Add buffers and journal it. There
1898 * doesn't seem much point in redirtying the page here.
1900 return __ext4_journalled_writepage(page, len);
1902 ext4_io_submit_init(&io_submit, wbc);
1903 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1904 if (!io_submit.io_end) {
1905 redirty_page_for_writepage(wbc, page);
1909 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1910 ext4_io_submit(&io_submit);
1911 /* Drop io_end reference we got from init */
1912 ext4_put_io_end_defer(io_submit.io_end);
1916 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1919 loff_t size = i_size_read(mpd->inode);
1922 BUG_ON(page->index != mpd->first_page);
1923 if (page->index == size >> PAGE_CACHE_SHIFT)
1924 len = size & ~PAGE_CACHE_MASK;
1926 len = PAGE_CACHE_SIZE;
1927 clear_page_dirty_for_io(page);
1928 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1930 mpd->wbc->nr_to_write--;
1936 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1939 * mballoc gives us at most this number of blocks...
1940 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1941 * The rest of mballoc seems to handle chunks up to full group size.
1943 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1946 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1948 * @mpd - extent of blocks
1949 * @lblk - logical number of the block in the file
1950 * @bh - buffer head we want to add to the extent
1952 * The function is used to collect contig. blocks in the same state. If the
1953 * buffer doesn't require mapping for writeback and we haven't started the
1954 * extent of buffers to map yet, the function returns 'true' immediately - the
1955 * caller can write the buffer right away. Otherwise the function returns true
1956 * if the block has been added to the extent, false if the block couldn't be
1959 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1960 struct buffer_head *bh)
1962 struct ext4_map_blocks *map = &mpd->map;
1964 /* Buffer that doesn't need mapping for writeback? */
1965 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1966 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1967 /* So far no extent to map => we write the buffer right away */
1968 if (map->m_len == 0)
1973 /* First block in the extent? */
1974 if (map->m_len == 0) {
1977 map->m_flags = bh->b_state & BH_FLAGS;
1981 /* Don't go larger than mballoc is willing to allocate */
1982 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1985 /* Can we merge the block to our big extent? */
1986 if (lblk == map->m_lblk + map->m_len &&
1987 (bh->b_state & BH_FLAGS) == map->m_flags) {
1995 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1997 * @mpd - extent of blocks for mapping
1998 * @head - the first buffer in the page
1999 * @bh - buffer we should start processing from
2000 * @lblk - logical number of the block in the file corresponding to @bh
2002 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2003 * the page for IO if all buffers in this page were mapped and there's no
2004 * accumulated extent of buffers to map or add buffers in the page to the
2005 * extent of buffers to map. The function returns 1 if the caller can continue
2006 * by processing the next page, 0 if it should stop adding buffers to the
2007 * extent to map because we cannot extend it anymore. It can also return value
2008 * < 0 in case of error during IO submission.
2010 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2011 struct buffer_head *head,
2012 struct buffer_head *bh,
2015 struct inode *inode = mpd->inode;
2017 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2018 >> inode->i_blkbits;
2021 BUG_ON(buffer_locked(bh));
2023 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2024 /* Found extent to map? */
2027 /* Everything mapped so far and we hit EOF */
2030 } while (lblk++, (bh = bh->b_this_page) != head);
2031 /* So far everything mapped? Submit the page for IO. */
2032 if (mpd->map.m_len == 0) {
2033 err = mpage_submit_page(mpd, head->b_page);
2037 return lblk < blocks;
2041 * mpage_map_buffers - update buffers corresponding to changed extent and
2042 * submit fully mapped pages for IO
2044 * @mpd - description of extent to map, on return next extent to map
2046 * Scan buffers corresponding to changed extent (we expect corresponding pages
2047 * to be already locked) and update buffer state according to new extent state.
2048 * We map delalloc buffers to their physical location, clear unwritten bits,
2049 * and mark buffers as uninit when we perform writes to unwritten extents
2050 * and do extent conversion after IO is finished. If the last page is not fully
2051 * mapped, we update @map to the next extent in the last page that needs
2052 * mapping. Otherwise we submit the page for IO.
2054 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2056 struct pagevec pvec;
2058 struct inode *inode = mpd->inode;
2059 struct buffer_head *head, *bh;
2060 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2066 start = mpd->map.m_lblk >> bpp_bits;
2067 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2068 lblk = start << bpp_bits;
2069 pblock = mpd->map.m_pblk;
2071 pagevec_init(&pvec, 0);
2072 while (start <= end) {
2073 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2077 for (i = 0; i < nr_pages; i++) {
2078 struct page *page = pvec.pages[i];
2080 if (page->index > end)
2082 /* Up to 'end' pages must be contiguous */
2083 BUG_ON(page->index != start);
2084 bh = head = page_buffers(page);
2086 if (lblk < mpd->map.m_lblk)
2088 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2090 * Buffer after end of mapped extent.
2091 * Find next buffer in the page to map.
2094 mpd->map.m_flags = 0;
2096 * FIXME: If dioread_nolock supports
2097 * blocksize < pagesize, we need to make
2098 * sure we add size mapped so far to
2099 * io_end->size as the following call
2100 * can submit the page for IO.
2102 err = mpage_process_page_bufs(mpd, head,
2104 pagevec_release(&pvec);
2109 if (buffer_delay(bh)) {
2110 clear_buffer_delay(bh);
2111 bh->b_blocknr = pblock++;
2113 clear_buffer_unwritten(bh);
2114 } while (lblk++, (bh = bh->b_this_page) != head);
2117 * FIXME: This is going to break if dioread_nolock
2118 * supports blocksize < pagesize as we will try to
2119 * convert potentially unmapped parts of inode.
2121 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2122 /* Page fully mapped - let IO run! */
2123 err = mpage_submit_page(mpd, page);
2125 pagevec_release(&pvec);
2130 pagevec_release(&pvec);
2132 /* Extent fully mapped and matches with page boundary. We are done. */
2134 mpd->map.m_flags = 0;
2138 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2140 struct inode *inode = mpd->inode;
2141 struct ext4_map_blocks *map = &mpd->map;
2142 int get_blocks_flags;
2143 int err, dioread_nolock;
2145 trace_ext4_da_write_pages_extent(inode, map);
2147 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2148 * to convert an unwritten extent to be initialized (in the case
2149 * where we have written into one or more preallocated blocks). It is
2150 * possible that we're going to need more metadata blocks than
2151 * previously reserved. However we must not fail because we're in
2152 * writeback and there is nothing we can do about it so it might result
2153 * in data loss. So use reserved blocks to allocate metadata if
2156 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2157 * the blocks in question are delalloc blocks. This indicates
2158 * that the blocks and quotas has already been checked when
2159 * the data was copied into the page cache.
2161 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2162 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2163 dioread_nolock = ext4_should_dioread_nolock(inode);
2165 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2166 if (map->m_flags & (1 << BH_Delay))
2167 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2169 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2172 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2173 if (!mpd->io_submit.io_end->handle &&
2174 ext4_handle_valid(handle)) {
2175 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2176 handle->h_rsv_handle = NULL;
2178 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2181 BUG_ON(map->m_len == 0);
2182 if (map->m_flags & EXT4_MAP_NEW) {
2183 struct block_device *bdev = inode->i_sb->s_bdev;
2186 for (i = 0; i < map->m_len; i++)
2187 unmap_underlying_metadata(bdev, map->m_pblk + i);
2193 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2194 * mpd->len and submit pages underlying it for IO
2196 * @handle - handle for journal operations
2197 * @mpd - extent to map
2198 * @give_up_on_write - we set this to true iff there is a fatal error and there
2199 * is no hope of writing the data. The caller should discard
2200 * dirty pages to avoid infinite loops.
2202 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2203 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2204 * them to initialized or split the described range from larger unwritten
2205 * extent. Note that we need not map all the described range since allocation
2206 * can return less blocks or the range is covered by more unwritten extents. We
2207 * cannot map more because we are limited by reserved transaction credits. On
2208 * the other hand we always make sure that the last touched page is fully
2209 * mapped so that it can be written out (and thus forward progress is
2210 * guaranteed). After mapping we submit all mapped pages for IO.
2212 static int mpage_map_and_submit_extent(handle_t *handle,
2213 struct mpage_da_data *mpd,
2214 bool *give_up_on_write)
2216 struct inode *inode = mpd->inode;
2217 struct ext4_map_blocks *map = &mpd->map;
2222 mpd->io_submit.io_end->offset =
2223 ((loff_t)map->m_lblk) << inode->i_blkbits;
2225 err = mpage_map_one_extent(handle, mpd);
2227 struct super_block *sb = inode->i_sb;
2229 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2230 goto invalidate_dirty_pages;
2232 * Let the uper layers retry transient errors.
2233 * In the case of ENOSPC, if ext4_count_free_blocks()
2234 * is non-zero, a commit should free up blocks.
2236 if ((err == -ENOMEM) ||
2237 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2239 goto update_disksize;
2242 ext4_msg(sb, KERN_CRIT,
2243 "Delayed block allocation failed for "
2244 "inode %lu at logical offset %llu with"
2245 " max blocks %u with error %d",
2247 (unsigned long long)map->m_lblk,
2248 (unsigned)map->m_len, -err);
2249 ext4_msg(sb, KERN_CRIT,
2250 "This should not happen!! Data will "
2253 ext4_print_free_blocks(inode);
2254 invalidate_dirty_pages:
2255 *give_up_on_write = true;
2260 * Update buffer state, submit mapped pages, and get us new
2263 err = mpage_map_and_submit_buffers(mpd);
2265 goto update_disksize;
2266 } while (map->m_len);
2270 * Update on-disk size after IO is submitted. Races with
2271 * truncate are avoided by checking i_size under i_data_sem.
2273 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2274 if (disksize > EXT4_I(inode)->i_disksize) {
2278 down_write(&EXT4_I(inode)->i_data_sem);
2279 i_size = i_size_read(inode);
2280 if (disksize > i_size)
2282 if (disksize > EXT4_I(inode)->i_disksize)
2283 EXT4_I(inode)->i_disksize = disksize;
2284 err2 = ext4_mark_inode_dirty(handle, inode);
2285 up_write(&EXT4_I(inode)->i_data_sem);
2287 ext4_error(inode->i_sb,
2288 "Failed to mark inode %lu dirty",
2297 * Calculate the total number of credits to reserve for one writepages
2298 * iteration. This is called from ext4_writepages(). We map an extent of
2299 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2300 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2301 * bpp - 1 blocks in bpp different extents.
2303 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2305 int bpp = ext4_journal_blocks_per_page(inode);
2307 return ext4_meta_trans_blocks(inode,
2308 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2312 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2313 * and underlying extent to map
2315 * @mpd - where to look for pages
2317 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2318 * IO immediately. When we find a page which isn't mapped we start accumulating
2319 * extent of buffers underlying these pages that needs mapping (formed by
2320 * either delayed or unwritten buffers). We also lock the pages containing
2321 * these buffers. The extent found is returned in @mpd structure (starting at
2322 * mpd->lblk with length mpd->len blocks).
2324 * Note that this function can attach bios to one io_end structure which are
2325 * neither logically nor physically contiguous. Although it may seem as an
2326 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2327 * case as we need to track IO to all buffers underlying a page in one io_end.
2329 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2331 struct address_space *mapping = mpd->inode->i_mapping;
2332 struct pagevec pvec;
2333 unsigned int nr_pages;
2334 long left = mpd->wbc->nr_to_write;
2335 pgoff_t index = mpd->first_page;
2336 pgoff_t end = mpd->last_page;
2339 int blkbits = mpd->inode->i_blkbits;
2341 struct buffer_head *head;
2343 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2344 tag = PAGECACHE_TAG_TOWRITE;
2346 tag = PAGECACHE_TAG_DIRTY;
2348 pagevec_init(&pvec, 0);
2350 mpd->next_page = index;
2351 while (index <= end) {
2352 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2353 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2357 for (i = 0; i < nr_pages; i++) {
2358 struct page *page = pvec.pages[i];
2361 * At this point, the page may be truncated or
2362 * invalidated (changing page->mapping to NULL), or
2363 * even swizzled back from swapper_space to tmpfs file
2364 * mapping. However, page->index will not change
2365 * because we have a reference on the page.
2367 if (page->index > end)
2371 * Accumulated enough dirty pages? This doesn't apply
2372 * to WB_SYNC_ALL mode. For integrity sync we have to
2373 * keep going because someone may be concurrently
2374 * dirtying pages, and we might have synced a lot of
2375 * newly appeared dirty pages, but have not synced all
2376 * of the old dirty pages.
2378 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2381 /* If we can't merge this page, we are done. */
2382 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2387 * If the page is no longer dirty, or its mapping no
2388 * longer corresponds to inode we are writing (which
2389 * means it has been truncated or invalidated), or the
2390 * page is already under writeback and we are not doing
2391 * a data integrity writeback, skip the page
2393 if (!PageDirty(page) ||
2394 (PageWriteback(page) &&
2395 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2396 unlikely(page->mapping != mapping)) {
2401 wait_on_page_writeback(page);
2402 BUG_ON(PageWriteback(page));
2404 if (mpd->map.m_len == 0)
2405 mpd->first_page = page->index;
2406 mpd->next_page = page->index + 1;
2407 /* Add all dirty buffers to mpd */
2408 lblk = ((ext4_lblk_t)page->index) <<
2409 (PAGE_CACHE_SHIFT - blkbits);
2410 head = page_buffers(page);
2411 err = mpage_process_page_bufs(mpd, head, head, lblk);
2417 pagevec_release(&pvec);
2422 pagevec_release(&pvec);
2426 static int __writepage(struct page *page, struct writeback_control *wbc,
2429 struct address_space *mapping = data;
2430 int ret = ext4_writepage(page, wbc);
2431 mapping_set_error(mapping, ret);
2435 static int ext4_writepages(struct address_space *mapping,
2436 struct writeback_control *wbc)
2438 pgoff_t writeback_index = 0;
2439 long nr_to_write = wbc->nr_to_write;
2440 int range_whole = 0;
2442 handle_t *handle = NULL;
2443 struct mpage_da_data mpd;
2444 struct inode *inode = mapping->host;
2445 int needed_blocks, rsv_blocks = 0, ret = 0;
2446 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2448 struct blk_plug plug;
2449 bool give_up_on_write = false;
2451 trace_ext4_writepages(inode, wbc);
2454 * No pages to write? This is mainly a kludge to avoid starting
2455 * a transaction for special inodes like journal inode on last iput()
2456 * because that could violate lock ordering on umount
2458 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2459 goto out_writepages;
2461 if (ext4_should_journal_data(inode)) {
2462 struct blk_plug plug;
2464 blk_start_plug(&plug);
2465 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2466 blk_finish_plug(&plug);
2467 goto out_writepages;
2471 * If the filesystem has aborted, it is read-only, so return
2472 * right away instead of dumping stack traces later on that
2473 * will obscure the real source of the problem. We test
2474 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2475 * the latter could be true if the filesystem is mounted
2476 * read-only, and in that case, ext4_writepages should
2477 * *never* be called, so if that ever happens, we would want
2480 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2482 goto out_writepages;
2485 if (ext4_should_dioread_nolock(inode)) {
2487 * We may need to convert up to one extent per block in
2488 * the page and we may dirty the inode.
2490 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2494 * If we have inline data and arrive here, it means that
2495 * we will soon create the block for the 1st page, so
2496 * we'd better clear the inline data here.
2498 if (ext4_has_inline_data(inode)) {
2499 /* Just inode will be modified... */
2500 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2501 if (IS_ERR(handle)) {
2502 ret = PTR_ERR(handle);
2503 goto out_writepages;
2505 BUG_ON(ext4_test_inode_state(inode,
2506 EXT4_STATE_MAY_INLINE_DATA));
2507 ext4_destroy_inline_data(handle, inode);
2508 ext4_journal_stop(handle);
2511 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2514 if (wbc->range_cyclic) {
2515 writeback_index = mapping->writeback_index;
2516 if (writeback_index)
2518 mpd.first_page = writeback_index;
2521 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2522 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2527 ext4_io_submit_init(&mpd.io_submit, wbc);
2529 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2530 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2532 blk_start_plug(&plug);
2533 while (!done && mpd.first_page <= mpd.last_page) {
2534 /* For each extent of pages we use new io_end */
2535 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2536 if (!mpd.io_submit.io_end) {
2542 * We have two constraints: We find one extent to map and we
2543 * must always write out whole page (makes a difference when
2544 * blocksize < pagesize) so that we don't block on IO when we
2545 * try to write out the rest of the page. Journalled mode is
2546 * not supported by delalloc.
2548 BUG_ON(ext4_should_journal_data(inode));
2549 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2551 /* start a new transaction */
2552 handle = ext4_journal_start_with_reserve(inode,
2553 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2554 if (IS_ERR(handle)) {
2555 ret = PTR_ERR(handle);
2556 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2557 "%ld pages, ino %lu; err %d", __func__,
2558 wbc->nr_to_write, inode->i_ino, ret);
2559 /* Release allocated io_end */
2560 ext4_put_io_end(mpd.io_submit.io_end);
2564 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2565 ret = mpage_prepare_extent_to_map(&mpd);
2568 ret = mpage_map_and_submit_extent(handle, &mpd,
2572 * We scanned the whole range (or exhausted
2573 * nr_to_write), submitted what was mapped and
2574 * didn't find anything needing mapping. We are
2580 ext4_journal_stop(handle);
2581 /* Submit prepared bio */
2582 ext4_io_submit(&mpd.io_submit);
2583 /* Unlock pages we didn't use */
2584 mpage_release_unused_pages(&mpd, give_up_on_write);
2585 /* Drop our io_end reference we got from init */
2586 ext4_put_io_end(mpd.io_submit.io_end);
2588 if (ret == -ENOSPC && sbi->s_journal) {
2590 * Commit the transaction which would
2591 * free blocks released in the transaction
2594 jbd2_journal_force_commit_nested(sbi->s_journal);
2598 /* Fatal error - ENOMEM, EIO... */
2602 blk_finish_plug(&plug);
2603 if (!ret && !cycled && wbc->nr_to_write > 0) {
2605 mpd.last_page = writeback_index - 1;
2611 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2613 * Set the writeback_index so that range_cyclic
2614 * mode will write it back later
2616 mapping->writeback_index = mpd.first_page;
2619 trace_ext4_writepages_result(inode, wbc, ret,
2620 nr_to_write - wbc->nr_to_write);
2624 static int ext4_nonda_switch(struct super_block *sb)
2626 s64 free_clusters, dirty_clusters;
2627 struct ext4_sb_info *sbi = EXT4_SB(sb);
2630 * switch to non delalloc mode if we are running low
2631 * on free block. The free block accounting via percpu
2632 * counters can get slightly wrong with percpu_counter_batch getting
2633 * accumulated on each CPU without updating global counters
2634 * Delalloc need an accurate free block accounting. So switch
2635 * to non delalloc when we are near to error range.
2638 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2640 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2642 * Start pushing delalloc when 1/2 of free blocks are dirty.
2644 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2645 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2647 if (2 * free_clusters < 3 * dirty_clusters ||
2648 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2650 * free block count is less than 150% of dirty blocks
2651 * or free blocks is less than watermark
2658 /* We always reserve for an inode update; the superblock could be there too */
2659 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2661 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2664 if (pos + len <= 0x7fffffffULL)
2667 /* We might need to update the superblock to set LARGE_FILE */
2671 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2672 loff_t pos, unsigned len, unsigned flags,
2673 struct page **pagep, void **fsdata)
2675 int ret, retries = 0;
2678 struct inode *inode = mapping->host;
2681 index = pos >> PAGE_CACHE_SHIFT;
2683 if (ext4_nonda_switch(inode->i_sb)) {
2684 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2685 return ext4_write_begin(file, mapping, pos,
2686 len, flags, pagep, fsdata);
2688 *fsdata = (void *)0;
2689 trace_ext4_da_write_begin(inode, pos, len, flags);
2691 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2692 ret = ext4_da_write_inline_data_begin(mapping, inode,
2702 * grab_cache_page_write_begin() can take a long time if the
2703 * system is thrashing due to memory pressure, or if the page
2704 * is being written back. So grab it first before we start
2705 * the transaction handle. This also allows us to allocate
2706 * the page (if needed) without using GFP_NOFS.
2709 page = grab_cache_page_write_begin(mapping, index, flags);
2715 * With delayed allocation, we don't log the i_disksize update
2716 * if there is delayed block allocation. But we still need
2717 * to journalling the i_disksize update if writes to the end
2718 * of file which has an already mapped buffer.
2721 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2722 ext4_da_write_credits(inode, pos, len));
2723 if (IS_ERR(handle)) {
2724 page_cache_release(page);
2725 return PTR_ERR(handle);
2729 if (page->mapping != mapping) {
2730 /* The page got truncated from under us */
2732 page_cache_release(page);
2733 ext4_journal_stop(handle);
2736 /* In case writeback began while the page was unlocked */
2737 wait_for_stable_page(page);
2739 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2740 ret = ext4_block_write_begin(page, pos, len,
2741 ext4_da_get_block_prep);
2743 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2747 ext4_journal_stop(handle);
2749 * block_write_begin may have instantiated a few blocks
2750 * outside i_size. Trim these off again. Don't need
2751 * i_size_read because we hold i_mutex.
2753 if (pos + len > inode->i_size)
2754 ext4_truncate_failed_write(inode);
2756 if (ret == -ENOSPC &&
2757 ext4_should_retry_alloc(inode->i_sb, &retries))
2760 page_cache_release(page);
2769 * Check if we should update i_disksize
2770 * when write to the end of file but not require block allocation
2772 static int ext4_da_should_update_i_disksize(struct page *page,
2773 unsigned long offset)
2775 struct buffer_head *bh;
2776 struct inode *inode = page->mapping->host;
2780 bh = page_buffers(page);
2781 idx = offset >> inode->i_blkbits;
2783 for (i = 0; i < idx; i++)
2784 bh = bh->b_this_page;
2786 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2791 static int ext4_da_write_end(struct file *file,
2792 struct address_space *mapping,
2793 loff_t pos, unsigned len, unsigned copied,
2794 struct page *page, void *fsdata)
2796 struct inode *inode = mapping->host;
2798 handle_t *handle = ext4_journal_current_handle();
2800 unsigned long start, end;
2801 int write_mode = (int)(unsigned long)fsdata;
2803 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2804 return ext4_write_end(file, mapping, pos,
2805 len, copied, page, fsdata);
2807 trace_ext4_da_write_end(inode, pos, len, copied);
2808 start = pos & (PAGE_CACHE_SIZE - 1);
2809 end = start + copied - 1;
2812 * generic_write_end() will run mark_inode_dirty() if i_size
2813 * changes. So let's piggyback the i_disksize mark_inode_dirty
2816 new_i_size = pos + copied;
2817 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2818 if (ext4_has_inline_data(inode) ||
2819 ext4_da_should_update_i_disksize(page, end)) {
2820 ext4_update_i_disksize(inode, new_i_size);
2821 /* We need to mark inode dirty even if
2822 * new_i_size is less that inode->i_size
2823 * bu greater than i_disksize.(hint delalloc)
2825 ext4_mark_inode_dirty(handle, inode);
2829 if (write_mode != CONVERT_INLINE_DATA &&
2830 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2831 ext4_has_inline_data(inode))
2832 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2835 ret2 = generic_write_end(file, mapping, pos, len, copied,
2841 ret2 = ext4_journal_stop(handle);
2845 return ret ? ret : copied;
2848 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2849 unsigned int length)
2852 * Drop reserved blocks
2854 BUG_ON(!PageLocked(page));
2855 if (!page_has_buffers(page))
2858 ext4_da_page_release_reservation(page, offset, length);
2861 ext4_invalidatepage(page, offset, length);
2867 * Force all delayed allocation blocks to be allocated for a given inode.
2869 int ext4_alloc_da_blocks(struct inode *inode)
2871 trace_ext4_alloc_da_blocks(inode);
2873 if (!EXT4_I(inode)->i_reserved_data_blocks)
2877 * We do something simple for now. The filemap_flush() will
2878 * also start triggering a write of the data blocks, which is
2879 * not strictly speaking necessary (and for users of
2880 * laptop_mode, not even desirable). However, to do otherwise
2881 * would require replicating code paths in:
2883 * ext4_writepages() ->
2884 * write_cache_pages() ---> (via passed in callback function)
2885 * __mpage_da_writepage() -->
2886 * mpage_add_bh_to_extent()
2887 * mpage_da_map_blocks()
2889 * The problem is that write_cache_pages(), located in
2890 * mm/page-writeback.c, marks pages clean in preparation for
2891 * doing I/O, which is not desirable if we're not planning on
2894 * We could call write_cache_pages(), and then redirty all of
2895 * the pages by calling redirty_page_for_writepage() but that
2896 * would be ugly in the extreme. So instead we would need to
2897 * replicate parts of the code in the above functions,
2898 * simplifying them because we wouldn't actually intend to
2899 * write out the pages, but rather only collect contiguous
2900 * logical block extents, call the multi-block allocator, and
2901 * then update the buffer heads with the block allocations.
2903 * For now, though, we'll cheat by calling filemap_flush(),
2904 * which will map the blocks, and start the I/O, but not
2905 * actually wait for the I/O to complete.
2907 return filemap_flush(inode->i_mapping);
2911 * bmap() is special. It gets used by applications such as lilo and by
2912 * the swapper to find the on-disk block of a specific piece of data.
2914 * Naturally, this is dangerous if the block concerned is still in the
2915 * journal. If somebody makes a swapfile on an ext4 data-journaling
2916 * filesystem and enables swap, then they may get a nasty shock when the
2917 * data getting swapped to that swapfile suddenly gets overwritten by
2918 * the original zero's written out previously to the journal and
2919 * awaiting writeback in the kernel's buffer cache.
2921 * So, if we see any bmap calls here on a modified, data-journaled file,
2922 * take extra steps to flush any blocks which might be in the cache.
2924 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2926 struct inode *inode = mapping->host;
2931 * We can get here for an inline file via the FIBMAP ioctl
2933 if (ext4_has_inline_data(inode))
2936 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2937 test_opt(inode->i_sb, DELALLOC)) {
2939 * With delalloc we want to sync the file
2940 * so that we can make sure we allocate
2943 filemap_write_and_wait(mapping);
2946 if (EXT4_JOURNAL(inode) &&
2947 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2949 * This is a REALLY heavyweight approach, but the use of
2950 * bmap on dirty files is expected to be extremely rare:
2951 * only if we run lilo or swapon on a freshly made file
2952 * do we expect this to happen.
2954 * (bmap requires CAP_SYS_RAWIO so this does not
2955 * represent an unprivileged user DOS attack --- we'd be
2956 * in trouble if mortal users could trigger this path at
2959 * NB. EXT4_STATE_JDATA is not set on files other than
2960 * regular files. If somebody wants to bmap a directory
2961 * or symlink and gets confused because the buffer
2962 * hasn't yet been flushed to disk, they deserve
2963 * everything they get.
2966 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2967 journal = EXT4_JOURNAL(inode);
2968 jbd2_journal_lock_updates(journal);
2969 err = jbd2_journal_flush(journal);
2970 jbd2_journal_unlock_updates(journal);
2976 return generic_block_bmap(mapping, block, ext4_get_block);
2979 static int ext4_readpage(struct file *file, struct page *page)
2982 struct inode *inode = page->mapping->host;
2984 trace_ext4_readpage(page);
2986 if (ext4_has_inline_data(inode))
2987 ret = ext4_readpage_inline(inode, page);
2990 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
2996 ext4_readpages(struct file *file, struct address_space *mapping,
2997 struct list_head *pages, unsigned nr_pages)
2999 struct inode *inode = mapping->host;
3001 /* If the file has inline data, no need to do readpages. */
3002 if (ext4_has_inline_data(inode))
3005 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3008 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3009 unsigned int length)
3011 trace_ext4_invalidatepage(page, offset, length);
3013 /* No journalling happens on data buffers when this function is used */
3014 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3016 block_invalidatepage(page, offset, length);
3019 static int __ext4_journalled_invalidatepage(struct page *page,
3020 unsigned int offset,
3021 unsigned int length)
3023 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3025 trace_ext4_journalled_invalidatepage(page, offset, length);
3028 * If it's a full truncate we just forget about the pending dirtying
3030 if (offset == 0 && length == PAGE_CACHE_SIZE)
3031 ClearPageChecked(page);
3033 return jbd2_journal_invalidatepage(journal, page, offset, length);
3036 /* Wrapper for aops... */
3037 static void ext4_journalled_invalidatepage(struct page *page,
3038 unsigned int offset,
3039 unsigned int length)
3041 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3044 static int ext4_releasepage(struct page *page, gfp_t wait)
3046 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3048 trace_ext4_releasepage(page);
3050 /* Page has dirty journalled data -> cannot release */
3051 if (PageChecked(page))
3054 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3056 return try_to_free_buffers(page);
3060 * ext4_get_block used when preparing for a DIO write or buffer write.
3061 * We allocate an uinitialized extent if blocks haven't been allocated.
3062 * The extent will be converted to initialized after the IO is complete.
3064 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3065 struct buffer_head *bh_result, int create)
3067 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3068 inode->i_ino, create);
3069 return _ext4_get_block(inode, iblock, bh_result,
3070 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3073 static int ext4_get_block_overwrite(struct inode *inode, sector_t iblock,
3074 struct buffer_head *bh_result, int create)
3078 ext4_debug("ext4_get_block_overwrite: inode %lu, create flag %d\n",
3079 inode->i_ino, create);
3080 ret = _ext4_get_block(inode, iblock, bh_result, 0);
3082 * Blocks should have been preallocated! ext4_file_write_iter() checks
3085 WARN_ON_ONCE(!buffer_mapped(bh_result));
3090 #ifdef CONFIG_FS_DAX
3091 int ext4_dax_mmap_get_block(struct inode *inode, sector_t iblock,
3092 struct buffer_head *bh_result, int create)
3096 struct ext4_map_blocks map;
3097 handle_t *handle = NULL;
3100 ext4_debug("ext4_dax_mmap_get_block: inode %lu, create flag %d\n",
3101 inode->i_ino, create);
3102 map.m_lblk = iblock;
3103 map.m_len = bh_result->b_size >> inode->i_blkbits;
3104 credits = ext4_chunk_trans_blocks(inode, map.m_len);
3106 flags |= EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_CREATE_ZERO;
3107 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
3108 if (IS_ERR(handle)) {
3109 ret = PTR_ERR(handle);
3114 ret = ext4_map_blocks(handle, inode, &map, flags);
3116 err = ext4_journal_stop(handle);
3117 if (ret >= 0 && err < 0)
3122 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3126 * We are protected by i_mmap_sem so we know block cannot go
3127 * away from under us even though we dropped i_data_sem.
3128 * Convert extent to written and write zeros there.
3130 * Note: We may get here even when create == 0.
3132 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
3133 if (IS_ERR(handle)) {
3134 ret = PTR_ERR(handle);
3138 err = ext4_map_blocks(handle, inode, &map,
3139 EXT4_GET_BLOCKS_CONVERT | EXT4_GET_BLOCKS_CREATE_ZERO);
3142 err2 = ext4_journal_stop(handle);
3143 if (err2 < 0 && ret > 0)
3147 WARN_ON_ONCE(ret == 0 && create);
3149 map_bh(bh_result, inode->i_sb, map.m_pblk);
3150 bh_result->b_state = (bh_result->b_state & ~EXT4_MAP_FLAGS) |
3153 * At least for now we have to clear BH_New so that DAX code
3154 * doesn't attempt to zero blocks again in a racy way.
3156 bh_result->b_state &= ~(1 << BH_New);
3157 bh_result->b_size = map.m_len << inode->i_blkbits;
3164 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3165 ssize_t size, void *private)
3167 ext4_io_end_t *io_end = iocb->private;
3172 /* if not async direct IO just return */
3176 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3177 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3178 iocb->private, io_end->inode->i_ino, iocb, offset,
3181 iocb->private = NULL;
3182 io_end->offset = offset;
3183 io_end->size = size;
3184 ext4_put_io_end(io_end);
3190 * For ext4 extent files, ext4 will do direct-io write to holes,
3191 * preallocated extents, and those write extend the file, no need to
3192 * fall back to buffered IO.
3194 * For holes, we fallocate those blocks, mark them as unwritten
3195 * If those blocks were preallocated, we mark sure they are split, but
3196 * still keep the range to write as unwritten.
3198 * The unwritten extents will be converted to written when DIO is completed.
3199 * For async direct IO, since the IO may still pending when return, we
3200 * set up an end_io call back function, which will do the conversion
3201 * when async direct IO completed.
3203 * If the O_DIRECT write will extend the file then add this inode to the
3204 * orphan list. So recovery will truncate it back to the original size
3205 * if the machine crashes during the write.
3208 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3211 struct file *file = iocb->ki_filp;
3212 struct inode *inode = file->f_mapping->host;
3214 size_t count = iov_iter_count(iter);
3216 get_block_t *get_block_func = NULL;
3218 loff_t final_size = offset + count;
3219 ext4_io_end_t *io_end = NULL;
3221 /* Use the old path for reads and writes beyond i_size. */
3222 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3223 return ext4_ind_direct_IO(iocb, iter, offset);
3225 BUG_ON(iocb->private == NULL);
3228 * Make all waiters for direct IO properly wait also for extent
3229 * conversion. This also disallows race between truncate() and
3230 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3232 if (iov_iter_rw(iter) == WRITE)
3233 inode_dio_begin(inode);
3235 /* If we do a overwrite dio, i_mutex locking can be released */
3236 overwrite = *((int *)iocb->private);
3239 inode_unlock(inode);
3242 * We could direct write to holes and fallocate.
3244 * Allocated blocks to fill the hole are marked as
3245 * unwritten to prevent parallel buffered read to expose
3246 * the stale data before DIO complete the data IO.
3248 * As to previously fallocated extents, ext4 get_block will
3249 * just simply mark the buffer mapped but still keep the
3250 * extents unwritten.
3252 * For non AIO case, we will convert those unwritten extents
3253 * to written after return back from blockdev_direct_IO.
3255 * For async DIO, the conversion needs to be deferred when the
3256 * IO is completed. The ext4 end_io callback function will be
3257 * called to take care of the conversion work. Here for async
3258 * case, we allocate an io_end structure to hook to the iocb.
3260 iocb->private = NULL;
3261 ext4_inode_aio_set(inode, NULL);
3262 if (!is_sync_kiocb(iocb)) {
3263 io_end = ext4_init_io_end(inode, GFP_NOFS);
3269 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3271 iocb->private = ext4_get_io_end(io_end);
3273 * we save the io structure for current async direct
3274 * IO, so that later ext4_map_blocks() could flag the
3275 * io structure whether there is a unwritten extents
3276 * needs to be converted when IO is completed.
3278 ext4_inode_aio_set(inode, io_end);
3282 get_block_func = ext4_get_block_overwrite;
3284 get_block_func = ext4_get_block_write;
3285 dio_flags = DIO_LOCKING;
3287 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3288 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3291 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3292 ext4_end_io_dio, dio_flags);
3294 ret = __blockdev_direct_IO(iocb, inode,
3295 inode->i_sb->s_bdev, iter, offset,
3297 ext4_end_io_dio, NULL, dio_flags);
3300 * Put our reference to io_end. This can free the io_end structure e.g.
3301 * in sync IO case or in case of error. It can even perform extent
3302 * conversion if all bios we submitted finished before we got here.
3303 * Note that in that case iocb->private can be already set to NULL
3307 ext4_inode_aio_set(inode, NULL);
3308 ext4_put_io_end(io_end);
3310 * When no IO was submitted ext4_end_io_dio() was not
3311 * called so we have to put iocb's reference.
3313 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3314 WARN_ON(iocb->private != io_end);
3315 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3316 ext4_put_io_end(io_end);
3317 iocb->private = NULL;
3320 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3321 EXT4_STATE_DIO_UNWRITTEN)) {
3324 * for non AIO case, since the IO is already
3325 * completed, we could do the conversion right here
3327 err = ext4_convert_unwritten_extents(NULL, inode,
3331 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3335 if (iov_iter_rw(iter) == WRITE)
3336 inode_dio_end(inode);
3337 /* take i_mutex locking again if we do a ovewrite dio */
3344 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3347 struct file *file = iocb->ki_filp;
3348 struct inode *inode = file->f_mapping->host;
3349 size_t count = iov_iter_count(iter);
3352 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3353 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3358 * If we are doing data journalling we don't support O_DIRECT
3360 if (ext4_should_journal_data(inode))
3363 /* Let buffer I/O handle the inline data case. */
3364 if (ext4_has_inline_data(inode))
3367 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3368 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3369 ret = ext4_ext_direct_IO(iocb, iter, offset);
3371 ret = ext4_ind_direct_IO(iocb, iter, offset);
3372 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3377 * Pages can be marked dirty completely asynchronously from ext4's journalling
3378 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3379 * much here because ->set_page_dirty is called under VFS locks. The page is
3380 * not necessarily locked.
3382 * We cannot just dirty the page and leave attached buffers clean, because the
3383 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3384 * or jbddirty because all the journalling code will explode.
3386 * So what we do is to mark the page "pending dirty" and next time writepage
3387 * is called, propagate that into the buffers appropriately.
3389 static int ext4_journalled_set_page_dirty(struct page *page)
3391 SetPageChecked(page);
3392 return __set_page_dirty_nobuffers(page);
3395 static const struct address_space_operations ext4_aops = {
3396 .readpage = ext4_readpage,
3397 .readpages = ext4_readpages,
3398 .writepage = ext4_writepage,
3399 .writepages = ext4_writepages,
3400 .write_begin = ext4_write_begin,
3401 .write_end = ext4_write_end,
3403 .invalidatepage = ext4_invalidatepage,
3404 .releasepage = ext4_releasepage,
3405 .direct_IO = ext4_direct_IO,
3406 .migratepage = buffer_migrate_page,
3407 .is_partially_uptodate = block_is_partially_uptodate,
3408 .error_remove_page = generic_error_remove_page,
3411 static const struct address_space_operations ext4_journalled_aops = {
3412 .readpage = ext4_readpage,
3413 .readpages = ext4_readpages,
3414 .writepage = ext4_writepage,
3415 .writepages = ext4_writepages,
3416 .write_begin = ext4_write_begin,
3417 .write_end = ext4_journalled_write_end,
3418 .set_page_dirty = ext4_journalled_set_page_dirty,
3420 .invalidatepage = ext4_journalled_invalidatepage,
3421 .releasepage = ext4_releasepage,
3422 .direct_IO = ext4_direct_IO,
3423 .is_partially_uptodate = block_is_partially_uptodate,
3424 .error_remove_page = generic_error_remove_page,
3427 static const struct address_space_operations ext4_da_aops = {
3428 .readpage = ext4_readpage,
3429 .readpages = ext4_readpages,
3430 .writepage = ext4_writepage,
3431 .writepages = ext4_writepages,
3432 .write_begin = ext4_da_write_begin,
3433 .write_end = ext4_da_write_end,
3435 .invalidatepage = ext4_da_invalidatepage,
3436 .releasepage = ext4_releasepage,
3437 .direct_IO = ext4_direct_IO,
3438 .migratepage = buffer_migrate_page,
3439 .is_partially_uptodate = block_is_partially_uptodate,
3440 .error_remove_page = generic_error_remove_page,
3443 void ext4_set_aops(struct inode *inode)
3445 switch (ext4_inode_journal_mode(inode)) {
3446 case EXT4_INODE_ORDERED_DATA_MODE:
3447 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3449 case EXT4_INODE_WRITEBACK_DATA_MODE:
3450 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3452 case EXT4_INODE_JOURNAL_DATA_MODE:
3453 inode->i_mapping->a_ops = &ext4_journalled_aops;
3458 if (test_opt(inode->i_sb, DELALLOC))
3459 inode->i_mapping->a_ops = &ext4_da_aops;
3461 inode->i_mapping->a_ops = &ext4_aops;
3464 static int __ext4_block_zero_page_range(handle_t *handle,
3465 struct address_space *mapping, loff_t from, loff_t length)
3467 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3468 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3469 unsigned blocksize, pos;
3471 struct inode *inode = mapping->host;
3472 struct buffer_head *bh;
3476 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3477 mapping_gfp_constraint(mapping, ~__GFP_FS));
3481 blocksize = inode->i_sb->s_blocksize;
3483 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3485 if (!page_has_buffers(page))
3486 create_empty_buffers(page, blocksize, 0);
3488 /* Find the buffer that contains "offset" */
3489 bh = page_buffers(page);
3491 while (offset >= pos) {
3492 bh = bh->b_this_page;
3496 if (buffer_freed(bh)) {
3497 BUFFER_TRACE(bh, "freed: skip");
3500 if (!buffer_mapped(bh)) {
3501 BUFFER_TRACE(bh, "unmapped");
3502 ext4_get_block(inode, iblock, bh, 0);
3503 /* unmapped? It's a hole - nothing to do */
3504 if (!buffer_mapped(bh)) {
3505 BUFFER_TRACE(bh, "still unmapped");
3510 /* Ok, it's mapped. Make sure it's up-to-date */
3511 if (PageUptodate(page))
3512 set_buffer_uptodate(bh);
3514 if (!buffer_uptodate(bh)) {
3516 ll_rw_block(READ, 1, &bh);
3518 /* Uhhuh. Read error. Complain and punt. */
3519 if (!buffer_uptodate(bh))
3521 if (S_ISREG(inode->i_mode) &&
3522 ext4_encrypted_inode(inode)) {
3523 /* We expect the key to be set. */
3524 BUG_ON(!ext4_has_encryption_key(inode));
3525 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3526 WARN_ON_ONCE(ext4_decrypt(page));
3529 if (ext4_should_journal_data(inode)) {
3530 BUFFER_TRACE(bh, "get write access");
3531 err = ext4_journal_get_write_access(handle, bh);
3535 zero_user(page, offset, length);
3536 BUFFER_TRACE(bh, "zeroed end of block");
3538 if (ext4_should_journal_data(inode)) {
3539 err = ext4_handle_dirty_metadata(handle, inode, bh);
3542 mark_buffer_dirty(bh);
3543 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3544 err = ext4_jbd2_file_inode(handle, inode);
3549 page_cache_release(page);
3554 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3555 * starting from file offset 'from'. The range to be zero'd must
3556 * be contained with in one block. If the specified range exceeds
3557 * the end of the block it will be shortened to end of the block
3558 * that cooresponds to 'from'
3560 static int ext4_block_zero_page_range(handle_t *handle,
3561 struct address_space *mapping, loff_t from, loff_t length)
3563 struct inode *inode = mapping->host;
3564 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3565 unsigned blocksize = inode->i_sb->s_blocksize;
3566 unsigned max = blocksize - (offset & (blocksize - 1));
3569 * correct length if it does not fall between
3570 * 'from' and the end of the block
3572 if (length > max || length < 0)
3576 return dax_zero_page_range(inode, from, length, ext4_get_block);
3577 return __ext4_block_zero_page_range(handle, mapping, from, length);
3581 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3582 * up to the end of the block which corresponds to `from'.
3583 * This required during truncate. We need to physically zero the tail end
3584 * of that block so it doesn't yield old data if the file is later grown.
3586 static int ext4_block_truncate_page(handle_t *handle,
3587 struct address_space *mapping, loff_t from)
3589 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3592 struct inode *inode = mapping->host;
3594 blocksize = inode->i_sb->s_blocksize;
3595 length = blocksize - (offset & (blocksize - 1));
3597 return ext4_block_zero_page_range(handle, mapping, from, length);
3600 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3601 loff_t lstart, loff_t length)
3603 struct super_block *sb = inode->i_sb;
3604 struct address_space *mapping = inode->i_mapping;
3605 unsigned partial_start, partial_end;
3606 ext4_fsblk_t start, end;
3607 loff_t byte_end = (lstart + length - 1);
3610 partial_start = lstart & (sb->s_blocksize - 1);
3611 partial_end = byte_end & (sb->s_blocksize - 1);
3613 start = lstart >> sb->s_blocksize_bits;
3614 end = byte_end >> sb->s_blocksize_bits;
3616 /* Handle partial zero within the single block */
3618 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3619 err = ext4_block_zero_page_range(handle, mapping,
3623 /* Handle partial zero out on the start of the range */
3624 if (partial_start) {
3625 err = ext4_block_zero_page_range(handle, mapping,
3626 lstart, sb->s_blocksize);
3630 /* Handle partial zero out on the end of the range */
3631 if (partial_end != sb->s_blocksize - 1)
3632 err = ext4_block_zero_page_range(handle, mapping,
3633 byte_end - partial_end,
3638 int ext4_can_truncate(struct inode *inode)
3640 if (S_ISREG(inode->i_mode))
3642 if (S_ISDIR(inode->i_mode))
3644 if (S_ISLNK(inode->i_mode))
3645 return !ext4_inode_is_fast_symlink(inode);
3650 * We have to make sure i_disksize gets properly updated before we truncate
3651 * page cache due to hole punching or zero range. Otherwise i_disksize update
3652 * can get lost as it may have been postponed to submission of writeback but
3653 * that will never happen after we truncate page cache.
3655 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3659 loff_t size = i_size_read(inode);
3661 WARN_ON(!inode_is_locked(inode));
3662 if (offset > size || offset + len < size)
3665 if (EXT4_I(inode)->i_disksize >= size)
3668 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3670 return PTR_ERR(handle);
3671 ext4_update_i_disksize(inode, size);
3672 ext4_mark_inode_dirty(handle, inode);
3673 ext4_journal_stop(handle);
3679 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3680 * associated with the given offset and length
3682 * @inode: File inode
3683 * @offset: The offset where the hole will begin
3684 * @len: The length of the hole
3686 * Returns: 0 on success or negative on failure
3689 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3691 struct super_block *sb = inode->i_sb;
3692 ext4_lblk_t first_block, stop_block;
3693 struct address_space *mapping = inode->i_mapping;
3694 loff_t first_block_offset, last_block_offset;
3696 unsigned int credits;
3699 if (!S_ISREG(inode->i_mode))
3702 trace_ext4_punch_hole(inode, offset, length, 0);
3705 * Write out all dirty pages to avoid race conditions
3706 * Then release them.
3708 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3709 ret = filemap_write_and_wait_range(mapping, offset,
3710 offset + length - 1);
3717 /* No need to punch hole beyond i_size */
3718 if (offset >= inode->i_size)
3722 * If the hole extends beyond i_size, set the hole
3723 * to end after the page that contains i_size
3725 if (offset + length > inode->i_size) {
3726 length = inode->i_size +
3727 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3731 if (offset & (sb->s_blocksize - 1) ||
3732 (offset + length) & (sb->s_blocksize - 1)) {
3734 * Attach jinode to inode for jbd2 if we do any zeroing of
3737 ret = ext4_inode_attach_jinode(inode);
3743 /* Wait all existing dio workers, newcomers will block on i_mutex */
3744 ext4_inode_block_unlocked_dio(inode);
3745 inode_dio_wait(inode);
3748 * Prevent page faults from reinstantiating pages we have released from
3751 down_write(&EXT4_I(inode)->i_mmap_sem);
3752 first_block_offset = round_up(offset, sb->s_blocksize);
3753 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3755 /* Now release the pages and zero block aligned part of pages*/
3756 if (last_block_offset > first_block_offset) {
3757 ret = ext4_update_disksize_before_punch(inode, offset, length);
3760 truncate_pagecache_range(inode, first_block_offset,
3764 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3765 credits = ext4_writepage_trans_blocks(inode);
3767 credits = ext4_blocks_for_truncate(inode);
3768 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3769 if (IS_ERR(handle)) {
3770 ret = PTR_ERR(handle);
3771 ext4_std_error(sb, ret);
3775 ret = ext4_zero_partial_blocks(handle, inode, offset,
3780 first_block = (offset + sb->s_blocksize - 1) >>
3781 EXT4_BLOCK_SIZE_BITS(sb);
3782 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3784 /* If there are no blocks to remove, return now */
3785 if (first_block >= stop_block)
3788 down_write(&EXT4_I(inode)->i_data_sem);
3789 ext4_discard_preallocations(inode);
3791 ret = ext4_es_remove_extent(inode, first_block,
3792 stop_block - first_block);
3794 up_write(&EXT4_I(inode)->i_data_sem);
3798 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3799 ret = ext4_ext_remove_space(inode, first_block,
3802 ret = ext4_ind_remove_space(handle, inode, first_block,
3805 up_write(&EXT4_I(inode)->i_data_sem);
3807 ext4_handle_sync(handle);
3809 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3810 ext4_mark_inode_dirty(handle, inode);
3812 ext4_journal_stop(handle);
3814 up_write(&EXT4_I(inode)->i_mmap_sem);
3815 ext4_inode_resume_unlocked_dio(inode);
3817 inode_unlock(inode);
3821 int ext4_inode_attach_jinode(struct inode *inode)
3823 struct ext4_inode_info *ei = EXT4_I(inode);
3824 struct jbd2_inode *jinode;
3826 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3829 jinode = jbd2_alloc_inode(GFP_KERNEL);
3830 spin_lock(&inode->i_lock);
3833 spin_unlock(&inode->i_lock);
3836 ei->jinode = jinode;
3837 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3840 spin_unlock(&inode->i_lock);
3841 if (unlikely(jinode != NULL))
3842 jbd2_free_inode(jinode);
3849 * We block out ext4_get_block() block instantiations across the entire
3850 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3851 * simultaneously on behalf of the same inode.
3853 * As we work through the truncate and commit bits of it to the journal there
3854 * is one core, guiding principle: the file's tree must always be consistent on
3855 * disk. We must be able to restart the truncate after a crash.
3857 * The file's tree may be transiently inconsistent in memory (although it
3858 * probably isn't), but whenever we close off and commit a journal transaction,
3859 * the contents of (the filesystem + the journal) must be consistent and
3860 * restartable. It's pretty simple, really: bottom up, right to left (although
3861 * left-to-right works OK too).
3863 * Note that at recovery time, journal replay occurs *before* the restart of
3864 * truncate against the orphan inode list.
3866 * The committed inode has the new, desired i_size (which is the same as
3867 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3868 * that this inode's truncate did not complete and it will again call
3869 * ext4_truncate() to have another go. So there will be instantiated blocks
3870 * to the right of the truncation point in a crashed ext4 filesystem. But
3871 * that's fine - as long as they are linked from the inode, the post-crash
3872 * ext4_truncate() run will find them and release them.
3874 void ext4_truncate(struct inode *inode)
3876 struct ext4_inode_info *ei = EXT4_I(inode);
3877 unsigned int credits;
3879 struct address_space *mapping = inode->i_mapping;
3882 * There is a possibility that we're either freeing the inode
3883 * or it's a completely new inode. In those cases we might not
3884 * have i_mutex locked because it's not necessary.
3886 if (!(inode->i_state & (I_NEW|I_FREEING)))
3887 WARN_ON(!inode_is_locked(inode));
3888 trace_ext4_truncate_enter(inode);
3890 if (!ext4_can_truncate(inode))
3893 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3895 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3896 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3898 if (ext4_has_inline_data(inode)) {
3901 ext4_inline_data_truncate(inode, &has_inline);
3906 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3907 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3908 if (ext4_inode_attach_jinode(inode) < 0)
3912 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3913 credits = ext4_writepage_trans_blocks(inode);
3915 credits = ext4_blocks_for_truncate(inode);
3917 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3918 if (IS_ERR(handle)) {
3919 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3923 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3924 ext4_block_truncate_page(handle, mapping, inode->i_size);
3927 * We add the inode to the orphan list, so that if this
3928 * truncate spans multiple transactions, and we crash, we will
3929 * resume the truncate when the filesystem recovers. It also
3930 * marks the inode dirty, to catch the new size.
3932 * Implication: the file must always be in a sane, consistent
3933 * truncatable state while each transaction commits.
3935 if (ext4_orphan_add(handle, inode))
3938 down_write(&EXT4_I(inode)->i_data_sem);
3940 ext4_discard_preallocations(inode);
3942 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3943 ext4_ext_truncate(handle, inode);
3945 ext4_ind_truncate(handle, inode);
3947 up_write(&ei->i_data_sem);
3950 ext4_handle_sync(handle);
3954 * If this was a simple ftruncate() and the file will remain alive,
3955 * then we need to clear up the orphan record which we created above.
3956 * However, if this was a real unlink then we were called by
3957 * ext4_evict_inode(), and we allow that function to clean up the
3958 * orphan info for us.
3961 ext4_orphan_del(handle, inode);
3963 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3964 ext4_mark_inode_dirty(handle, inode);
3965 ext4_journal_stop(handle);
3967 trace_ext4_truncate_exit(inode);
3971 * ext4_get_inode_loc returns with an extra refcount against the inode's
3972 * underlying buffer_head on success. If 'in_mem' is true, we have all
3973 * data in memory that is needed to recreate the on-disk version of this
3976 static int __ext4_get_inode_loc(struct inode *inode,
3977 struct ext4_iloc *iloc, int in_mem)
3979 struct ext4_group_desc *gdp;
3980 struct buffer_head *bh;
3981 struct super_block *sb = inode->i_sb;
3983 int inodes_per_block, inode_offset;
3986 if (!ext4_valid_inum(sb, inode->i_ino))
3987 return -EFSCORRUPTED;
3989 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3990 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3995 * Figure out the offset within the block group inode table
3997 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3998 inode_offset = ((inode->i_ino - 1) %
3999 EXT4_INODES_PER_GROUP(sb));
4000 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4001 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4003 bh = sb_getblk(sb, block);
4006 if (!buffer_uptodate(bh)) {
4010 * If the buffer has the write error flag, we have failed
4011 * to write out another inode in the same block. In this
4012 * case, we don't have to read the block because we may
4013 * read the old inode data successfully.
4015 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4016 set_buffer_uptodate(bh);
4018 if (buffer_uptodate(bh)) {
4019 /* someone brought it uptodate while we waited */
4025 * If we have all information of the inode in memory and this
4026 * is the only valid inode in the block, we need not read the
4030 struct buffer_head *bitmap_bh;
4033 start = inode_offset & ~(inodes_per_block - 1);
4035 /* Is the inode bitmap in cache? */
4036 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4037 if (unlikely(!bitmap_bh))
4041 * If the inode bitmap isn't in cache then the
4042 * optimisation may end up performing two reads instead
4043 * of one, so skip it.
4045 if (!buffer_uptodate(bitmap_bh)) {
4049 for (i = start; i < start + inodes_per_block; i++) {
4050 if (i == inode_offset)
4052 if (ext4_test_bit(i, bitmap_bh->b_data))
4056 if (i == start + inodes_per_block) {
4057 /* all other inodes are free, so skip I/O */
4058 memset(bh->b_data, 0, bh->b_size);
4059 set_buffer_uptodate(bh);
4067 * If we need to do any I/O, try to pre-readahead extra
4068 * blocks from the inode table.
4070 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4071 ext4_fsblk_t b, end, table;
4073 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4075 table = ext4_inode_table(sb, gdp);
4076 /* s_inode_readahead_blks is always a power of 2 */
4077 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4081 num = EXT4_INODES_PER_GROUP(sb);
4082 if (ext4_has_group_desc_csum(sb))
4083 num -= ext4_itable_unused_count(sb, gdp);
4084 table += num / inodes_per_block;
4088 sb_breadahead(sb, b++);
4092 * There are other valid inodes in the buffer, this inode
4093 * has in-inode xattrs, or we don't have this inode in memory.
4094 * Read the block from disk.
4096 trace_ext4_load_inode(inode);
4098 bh->b_end_io = end_buffer_read_sync;
4099 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4101 if (!buffer_uptodate(bh)) {
4102 EXT4_ERROR_INODE_BLOCK(inode, block,
4103 "unable to read itable block");
4113 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4115 /* We have all inode data except xattrs in memory here. */
4116 return __ext4_get_inode_loc(inode, iloc,
4117 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4120 void ext4_set_inode_flags(struct inode *inode)
4122 unsigned int flags = EXT4_I(inode)->i_flags;
4123 unsigned int new_fl = 0;
4125 if (flags & EXT4_SYNC_FL)
4127 if (flags & EXT4_APPEND_FL)
4129 if (flags & EXT4_IMMUTABLE_FL)
4130 new_fl |= S_IMMUTABLE;
4131 if (flags & EXT4_NOATIME_FL)
4132 new_fl |= S_NOATIME;
4133 if (flags & EXT4_DIRSYNC_FL)
4134 new_fl |= S_DIRSYNC;
4135 if (test_opt(inode->i_sb, DAX))
4137 inode_set_flags(inode, new_fl,
4138 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4141 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4142 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4144 unsigned int vfs_fl;
4145 unsigned long old_fl, new_fl;
4148 vfs_fl = ei->vfs_inode.i_flags;
4149 old_fl = ei->i_flags;
4150 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4151 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4153 if (vfs_fl & S_SYNC)
4154 new_fl |= EXT4_SYNC_FL;
4155 if (vfs_fl & S_APPEND)
4156 new_fl |= EXT4_APPEND_FL;
4157 if (vfs_fl & S_IMMUTABLE)
4158 new_fl |= EXT4_IMMUTABLE_FL;
4159 if (vfs_fl & S_NOATIME)
4160 new_fl |= EXT4_NOATIME_FL;
4161 if (vfs_fl & S_DIRSYNC)
4162 new_fl |= EXT4_DIRSYNC_FL;
4163 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4166 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4167 struct ext4_inode_info *ei)
4170 struct inode *inode = &(ei->vfs_inode);
4171 struct super_block *sb = inode->i_sb;
4173 if (ext4_has_feature_huge_file(sb)) {
4174 /* we are using combined 48 bit field */
4175 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4176 le32_to_cpu(raw_inode->i_blocks_lo);
4177 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4178 /* i_blocks represent file system block size */
4179 return i_blocks << (inode->i_blkbits - 9);
4184 return le32_to_cpu(raw_inode->i_blocks_lo);
4188 static inline void ext4_iget_extra_inode(struct inode *inode,
4189 struct ext4_inode *raw_inode,
4190 struct ext4_inode_info *ei)
4192 __le32 *magic = (void *)raw_inode +
4193 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4194 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4195 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4196 ext4_find_inline_data_nolock(inode);
4198 EXT4_I(inode)->i_inline_off = 0;
4201 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4203 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, EXT4_FEATURE_RO_COMPAT_PROJECT))
4205 *projid = EXT4_I(inode)->i_projid;
4209 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4211 struct ext4_iloc iloc;
4212 struct ext4_inode *raw_inode;
4213 struct ext4_inode_info *ei;
4214 struct inode *inode;
4215 journal_t *journal = EXT4_SB(sb)->s_journal;
4222 inode = iget_locked(sb, ino);
4224 return ERR_PTR(-ENOMEM);
4225 if (!(inode->i_state & I_NEW))
4231 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4234 raw_inode = ext4_raw_inode(&iloc);
4236 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4237 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4238 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4239 EXT4_INODE_SIZE(inode->i_sb)) {
4240 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4241 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4242 EXT4_INODE_SIZE(inode->i_sb));
4243 ret = -EFSCORRUPTED;
4247 ei->i_extra_isize = 0;
4249 /* Precompute checksum seed for inode metadata */
4250 if (ext4_has_metadata_csum(sb)) {
4251 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4253 __le32 inum = cpu_to_le32(inode->i_ino);
4254 __le32 gen = raw_inode->i_generation;
4255 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4257 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4261 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4262 EXT4_ERROR_INODE(inode, "checksum invalid");
4267 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4268 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4269 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4270 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) &&
4271 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4272 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4273 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4275 i_projid = EXT4_DEF_PROJID;
4277 if (!(test_opt(inode->i_sb, NO_UID32))) {
4278 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4279 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4281 i_uid_write(inode, i_uid);
4282 i_gid_write(inode, i_gid);
4283 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4284 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4286 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4287 ei->i_inline_off = 0;
4288 ei->i_dir_start_lookup = 0;
4289 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4290 /* We now have enough fields to check if the inode was active or not.
4291 * This is needed because nfsd might try to access dead inodes
4292 * the test is that same one that e2fsck uses
4293 * NeilBrown 1999oct15
4295 if (inode->i_nlink == 0) {
4296 if ((inode->i_mode == 0 ||
4297 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4298 ino != EXT4_BOOT_LOADER_INO) {
4299 /* this inode is deleted */
4303 /* The only unlinked inodes we let through here have
4304 * valid i_mode and are being read by the orphan
4305 * recovery code: that's fine, we're about to complete
4306 * the process of deleting those.
4307 * OR it is the EXT4_BOOT_LOADER_INO which is
4308 * not initialized on a new filesystem. */
4310 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4311 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4312 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4313 if (ext4_has_feature_64bit(sb))
4315 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4316 inode->i_size = ext4_isize(raw_inode);
4317 ei->i_disksize = inode->i_size;
4319 ei->i_reserved_quota = 0;
4321 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4322 ei->i_block_group = iloc.block_group;
4323 ei->i_last_alloc_group = ~0;
4325 * NOTE! The in-memory inode i_data array is in little-endian order
4326 * even on big-endian machines: we do NOT byteswap the block numbers!
4328 for (block = 0; block < EXT4_N_BLOCKS; block++)
4329 ei->i_data[block] = raw_inode->i_block[block];
4330 INIT_LIST_HEAD(&ei->i_orphan);
4333 * Set transaction id's of transactions that have to be committed
4334 * to finish f[data]sync. We set them to currently running transaction
4335 * as we cannot be sure that the inode or some of its metadata isn't
4336 * part of the transaction - the inode could have been reclaimed and
4337 * now it is reread from disk.
4340 transaction_t *transaction;
4343 read_lock(&journal->j_state_lock);
4344 if (journal->j_running_transaction)
4345 transaction = journal->j_running_transaction;
4347 transaction = journal->j_committing_transaction;
4349 tid = transaction->t_tid;
4351 tid = journal->j_commit_sequence;
4352 read_unlock(&journal->j_state_lock);
4353 ei->i_sync_tid = tid;
4354 ei->i_datasync_tid = tid;
4357 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4358 if (ei->i_extra_isize == 0) {
4359 /* The extra space is currently unused. Use it. */
4360 ei->i_extra_isize = sizeof(struct ext4_inode) -
4361 EXT4_GOOD_OLD_INODE_SIZE;
4363 ext4_iget_extra_inode(inode, raw_inode, ei);
4367 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4368 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4369 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4370 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4372 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4373 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4374 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4375 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4377 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4382 if (ei->i_file_acl &&
4383 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4384 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4386 ret = -EFSCORRUPTED;
4388 } else if (!ext4_has_inline_data(inode)) {
4389 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4390 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4391 (S_ISLNK(inode->i_mode) &&
4392 !ext4_inode_is_fast_symlink(inode))))
4393 /* Validate extent which is part of inode */
4394 ret = ext4_ext_check_inode(inode);
4395 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4396 (S_ISLNK(inode->i_mode) &&
4397 !ext4_inode_is_fast_symlink(inode))) {
4398 /* Validate block references which are part of inode */
4399 ret = ext4_ind_check_inode(inode);
4405 if (S_ISREG(inode->i_mode)) {
4406 inode->i_op = &ext4_file_inode_operations;
4407 inode->i_fop = &ext4_file_operations;
4408 ext4_set_aops(inode);
4409 } else if (S_ISDIR(inode->i_mode)) {
4410 inode->i_op = &ext4_dir_inode_operations;
4411 inode->i_fop = &ext4_dir_operations;
4412 } else if (S_ISLNK(inode->i_mode)) {
4413 if (ext4_encrypted_inode(inode)) {
4414 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4415 ext4_set_aops(inode);
4416 } else if (ext4_inode_is_fast_symlink(inode)) {
4417 inode->i_link = (char *)ei->i_data;
4418 inode->i_op = &ext4_fast_symlink_inode_operations;
4419 nd_terminate_link(ei->i_data, inode->i_size,
4420 sizeof(ei->i_data) - 1);
4422 inode->i_op = &ext4_symlink_inode_operations;
4423 ext4_set_aops(inode);
4425 inode_nohighmem(inode);
4426 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4427 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4428 inode->i_op = &ext4_special_inode_operations;
4429 if (raw_inode->i_block[0])
4430 init_special_inode(inode, inode->i_mode,
4431 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4433 init_special_inode(inode, inode->i_mode,
4434 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4435 } else if (ino == EXT4_BOOT_LOADER_INO) {
4436 make_bad_inode(inode);
4438 ret = -EFSCORRUPTED;
4439 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4443 ext4_set_inode_flags(inode);
4444 unlock_new_inode(inode);
4450 return ERR_PTR(ret);
4453 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4455 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4456 return ERR_PTR(-EFSCORRUPTED);
4457 return ext4_iget(sb, ino);
4460 static int ext4_inode_blocks_set(handle_t *handle,
4461 struct ext4_inode *raw_inode,
4462 struct ext4_inode_info *ei)
4464 struct inode *inode = &(ei->vfs_inode);
4465 u64 i_blocks = inode->i_blocks;
4466 struct super_block *sb = inode->i_sb;
4468 if (i_blocks <= ~0U) {
4470 * i_blocks can be represented in a 32 bit variable
4471 * as multiple of 512 bytes
4473 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4474 raw_inode->i_blocks_high = 0;
4475 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4478 if (!ext4_has_feature_huge_file(sb))
4481 if (i_blocks <= 0xffffffffffffULL) {
4483 * i_blocks can be represented in a 48 bit variable
4484 * as multiple of 512 bytes
4486 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4487 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4488 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4490 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4491 /* i_block is stored in file system block size */
4492 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4493 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4494 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4499 struct other_inode {
4500 unsigned long orig_ino;
4501 struct ext4_inode *raw_inode;
4504 static int other_inode_match(struct inode * inode, unsigned long ino,
4507 struct other_inode *oi = (struct other_inode *) data;
4509 if ((inode->i_ino != ino) ||
4510 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4511 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4512 ((inode->i_state & I_DIRTY_TIME) == 0))
4514 spin_lock(&inode->i_lock);
4515 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4516 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4517 (inode->i_state & I_DIRTY_TIME)) {
4518 struct ext4_inode_info *ei = EXT4_I(inode);
4520 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4521 spin_unlock(&inode->i_lock);
4523 spin_lock(&ei->i_raw_lock);
4524 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4525 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4526 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4527 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4528 spin_unlock(&ei->i_raw_lock);
4529 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4532 spin_unlock(&inode->i_lock);
4537 * Opportunistically update the other time fields for other inodes in
4538 * the same inode table block.
4540 static void ext4_update_other_inodes_time(struct super_block *sb,
4541 unsigned long orig_ino, char *buf)
4543 struct other_inode oi;
4545 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4546 int inode_size = EXT4_INODE_SIZE(sb);
4548 oi.orig_ino = orig_ino;
4550 * Calculate the first inode in the inode table block. Inode
4551 * numbers are one-based. That is, the first inode in a block
4552 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4554 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4555 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4556 if (ino == orig_ino)
4558 oi.raw_inode = (struct ext4_inode *) buf;
4559 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4564 * Post the struct inode info into an on-disk inode location in the
4565 * buffer-cache. This gobbles the caller's reference to the
4566 * buffer_head in the inode location struct.
4568 * The caller must have write access to iloc->bh.
4570 static int ext4_do_update_inode(handle_t *handle,
4571 struct inode *inode,
4572 struct ext4_iloc *iloc)
4574 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4575 struct ext4_inode_info *ei = EXT4_I(inode);
4576 struct buffer_head *bh = iloc->bh;
4577 struct super_block *sb = inode->i_sb;
4578 int err = 0, rc, block;
4579 int need_datasync = 0, set_large_file = 0;
4584 spin_lock(&ei->i_raw_lock);
4586 /* For fields not tracked in the in-memory inode,
4587 * initialise them to zero for new inodes. */
4588 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4589 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4591 ext4_get_inode_flags(ei);
4592 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4593 i_uid = i_uid_read(inode);
4594 i_gid = i_gid_read(inode);
4595 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4596 if (!(test_opt(inode->i_sb, NO_UID32))) {
4597 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4598 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4600 * Fix up interoperability with old kernels. Otherwise, old inodes get
4601 * re-used with the upper 16 bits of the uid/gid intact
4604 raw_inode->i_uid_high =
4605 cpu_to_le16(high_16_bits(i_uid));
4606 raw_inode->i_gid_high =
4607 cpu_to_le16(high_16_bits(i_gid));
4609 raw_inode->i_uid_high = 0;
4610 raw_inode->i_gid_high = 0;
4613 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4614 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4615 raw_inode->i_uid_high = 0;
4616 raw_inode->i_gid_high = 0;
4618 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4620 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4621 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4622 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4623 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4625 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4627 spin_unlock(&ei->i_raw_lock);
4630 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4631 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4632 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4633 raw_inode->i_file_acl_high =
4634 cpu_to_le16(ei->i_file_acl >> 32);
4635 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4636 if (ei->i_disksize != ext4_isize(raw_inode)) {
4637 ext4_isize_set(raw_inode, ei->i_disksize);
4640 if (ei->i_disksize > 0x7fffffffULL) {
4641 if (!ext4_has_feature_large_file(sb) ||
4642 EXT4_SB(sb)->s_es->s_rev_level ==
4643 cpu_to_le32(EXT4_GOOD_OLD_REV))
4646 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4647 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4648 if (old_valid_dev(inode->i_rdev)) {
4649 raw_inode->i_block[0] =
4650 cpu_to_le32(old_encode_dev(inode->i_rdev));
4651 raw_inode->i_block[1] = 0;
4653 raw_inode->i_block[0] = 0;
4654 raw_inode->i_block[1] =
4655 cpu_to_le32(new_encode_dev(inode->i_rdev));
4656 raw_inode->i_block[2] = 0;
4658 } else if (!ext4_has_inline_data(inode)) {
4659 for (block = 0; block < EXT4_N_BLOCKS; block++)
4660 raw_inode->i_block[block] = ei->i_data[block];
4663 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4664 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4665 if (ei->i_extra_isize) {
4666 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4667 raw_inode->i_version_hi =
4668 cpu_to_le32(inode->i_version >> 32);
4669 raw_inode->i_extra_isize =
4670 cpu_to_le16(ei->i_extra_isize);
4674 BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
4675 EXT4_FEATURE_RO_COMPAT_PROJECT) &&
4676 i_projid != EXT4_DEF_PROJID);
4678 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4679 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4680 raw_inode->i_projid = cpu_to_le32(i_projid);
4682 ext4_inode_csum_set(inode, raw_inode, ei);
4683 spin_unlock(&ei->i_raw_lock);
4684 if (inode->i_sb->s_flags & MS_LAZYTIME)
4685 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4688 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4689 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4692 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4693 if (set_large_file) {
4694 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4695 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4698 ext4_update_dynamic_rev(sb);
4699 ext4_set_feature_large_file(sb);
4700 ext4_handle_sync(handle);
4701 err = ext4_handle_dirty_super(handle, sb);
4703 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4706 ext4_std_error(inode->i_sb, err);
4711 * ext4_write_inode()
4713 * We are called from a few places:
4715 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4716 * Here, there will be no transaction running. We wait for any running
4717 * transaction to commit.
4719 * - Within flush work (sys_sync(), kupdate and such).
4720 * We wait on commit, if told to.
4722 * - Within iput_final() -> write_inode_now()
4723 * We wait on commit, if told to.
4725 * In all cases it is actually safe for us to return without doing anything,
4726 * because the inode has been copied into a raw inode buffer in
4727 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4730 * Note that we are absolutely dependent upon all inode dirtiers doing the
4731 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4732 * which we are interested.
4734 * It would be a bug for them to not do this. The code:
4736 * mark_inode_dirty(inode)
4738 * inode->i_size = expr;
4740 * is in error because write_inode() could occur while `stuff()' is running,
4741 * and the new i_size will be lost. Plus the inode will no longer be on the
4742 * superblock's dirty inode list.
4744 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4748 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4751 if (EXT4_SB(inode->i_sb)->s_journal) {
4752 if (ext4_journal_current_handle()) {
4753 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4759 * No need to force transaction in WB_SYNC_NONE mode. Also
4760 * ext4_sync_fs() will force the commit after everything is
4763 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4766 err = ext4_force_commit(inode->i_sb);
4768 struct ext4_iloc iloc;
4770 err = __ext4_get_inode_loc(inode, &iloc, 0);
4774 * sync(2) will flush the whole buffer cache. No need to do
4775 * it here separately for each inode.
4777 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4778 sync_dirty_buffer(iloc.bh);
4779 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4780 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4781 "IO error syncing inode");
4790 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4791 * buffers that are attached to a page stradding i_size and are undergoing
4792 * commit. In that case we have to wait for commit to finish and try again.
4794 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4798 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4799 tid_t commit_tid = 0;
4802 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4804 * All buffers in the last page remain valid? Then there's nothing to
4805 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4808 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4811 page = find_lock_page(inode->i_mapping,
4812 inode->i_size >> PAGE_CACHE_SHIFT);
4815 ret = __ext4_journalled_invalidatepage(page, offset,
4816 PAGE_CACHE_SIZE - offset);
4818 page_cache_release(page);
4822 read_lock(&journal->j_state_lock);
4823 if (journal->j_committing_transaction)
4824 commit_tid = journal->j_committing_transaction->t_tid;
4825 read_unlock(&journal->j_state_lock);
4827 jbd2_log_wait_commit(journal, commit_tid);
4834 * Called from notify_change.
4836 * We want to trap VFS attempts to truncate the file as soon as
4837 * possible. In particular, we want to make sure that when the VFS
4838 * shrinks i_size, we put the inode on the orphan list and modify
4839 * i_disksize immediately, so that during the subsequent flushing of
4840 * dirty pages and freeing of disk blocks, we can guarantee that any
4841 * commit will leave the blocks being flushed in an unused state on
4842 * disk. (On recovery, the inode will get truncated and the blocks will
4843 * be freed, so we have a strong guarantee that no future commit will
4844 * leave these blocks visible to the user.)
4846 * Another thing we have to assure is that if we are in ordered mode
4847 * and inode is still attached to the committing transaction, we must
4848 * we start writeout of all the dirty pages which are being truncated.
4849 * This way we are sure that all the data written in the previous
4850 * transaction are already on disk (truncate waits for pages under
4853 * Called with inode->i_mutex down.
4855 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4857 struct inode *inode = d_inode(dentry);
4860 const unsigned int ia_valid = attr->ia_valid;
4862 error = inode_change_ok(inode, attr);
4866 if (is_quota_modification(inode, attr)) {
4867 error = dquot_initialize(inode);
4871 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4872 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4875 /* (user+group)*(old+new) structure, inode write (sb,
4876 * inode block, ? - but truncate inode update has it) */
4877 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4878 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4879 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4880 if (IS_ERR(handle)) {
4881 error = PTR_ERR(handle);
4884 error = dquot_transfer(inode, attr);
4886 ext4_journal_stop(handle);
4889 /* Update corresponding info in inode so that everything is in
4890 * one transaction */
4891 if (attr->ia_valid & ATTR_UID)
4892 inode->i_uid = attr->ia_uid;
4893 if (attr->ia_valid & ATTR_GID)
4894 inode->i_gid = attr->ia_gid;
4895 error = ext4_mark_inode_dirty(handle, inode);
4896 ext4_journal_stop(handle);
4899 if (attr->ia_valid & ATTR_SIZE) {
4901 loff_t oldsize = inode->i_size;
4902 int shrink = (attr->ia_size <= inode->i_size);
4904 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4905 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4907 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4910 if (!S_ISREG(inode->i_mode))
4913 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4914 inode_inc_iversion(inode);
4916 if (ext4_should_order_data(inode) &&
4917 (attr->ia_size < inode->i_size)) {
4918 error = ext4_begin_ordered_truncate(inode,
4923 if (attr->ia_size != inode->i_size) {
4924 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4925 if (IS_ERR(handle)) {
4926 error = PTR_ERR(handle);
4929 if (ext4_handle_valid(handle) && shrink) {
4930 error = ext4_orphan_add(handle, inode);
4934 * Update c/mtime on truncate up, ext4_truncate() will
4935 * update c/mtime in shrink case below
4938 inode->i_mtime = ext4_current_time(inode);
4939 inode->i_ctime = inode->i_mtime;
4941 down_write(&EXT4_I(inode)->i_data_sem);
4942 EXT4_I(inode)->i_disksize = attr->ia_size;
4943 rc = ext4_mark_inode_dirty(handle, inode);
4947 * We have to update i_size under i_data_sem together
4948 * with i_disksize to avoid races with writeback code
4949 * running ext4_wb_update_i_disksize().
4952 i_size_write(inode, attr->ia_size);
4953 up_write(&EXT4_I(inode)->i_data_sem);
4954 ext4_journal_stop(handle);
4957 ext4_orphan_del(NULL, inode);
4962 pagecache_isize_extended(inode, oldsize, inode->i_size);
4965 * Blocks are going to be removed from the inode. Wait
4966 * for dio in flight. Temporarily disable
4967 * dioread_nolock to prevent livelock.
4970 if (!ext4_should_journal_data(inode)) {
4971 ext4_inode_block_unlocked_dio(inode);
4972 inode_dio_wait(inode);
4973 ext4_inode_resume_unlocked_dio(inode);
4975 ext4_wait_for_tail_page_commit(inode);
4977 down_write(&EXT4_I(inode)->i_mmap_sem);
4979 * Truncate pagecache after we've waited for commit
4980 * in data=journal mode to make pages freeable.
4982 truncate_pagecache(inode, inode->i_size);
4984 ext4_truncate(inode);
4985 up_write(&EXT4_I(inode)->i_mmap_sem);
4989 setattr_copy(inode, attr);
4990 mark_inode_dirty(inode);
4994 * If the call to ext4_truncate failed to get a transaction handle at
4995 * all, we need to clean up the in-core orphan list manually.
4997 if (orphan && inode->i_nlink)
4998 ext4_orphan_del(NULL, inode);
5000 if (!rc && (ia_valid & ATTR_MODE))
5001 rc = posix_acl_chmod(inode, inode->i_mode);
5004 ext4_std_error(inode->i_sb, error);
5010 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5013 struct inode *inode;
5014 unsigned long long delalloc_blocks;
5016 inode = d_inode(dentry);
5017 generic_fillattr(inode, stat);
5020 * If there is inline data in the inode, the inode will normally not
5021 * have data blocks allocated (it may have an external xattr block).
5022 * Report at least one sector for such files, so tools like tar, rsync,
5023 * others doen't incorrectly think the file is completely sparse.
5025 if (unlikely(ext4_has_inline_data(inode)))
5026 stat->blocks += (stat->size + 511) >> 9;
5029 * We can't update i_blocks if the block allocation is delayed
5030 * otherwise in the case of system crash before the real block
5031 * allocation is done, we will have i_blocks inconsistent with
5032 * on-disk file blocks.
5033 * We always keep i_blocks updated together with real
5034 * allocation. But to not confuse with user, stat
5035 * will return the blocks that include the delayed allocation
5036 * blocks for this file.
5038 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5039 EXT4_I(inode)->i_reserved_data_blocks);
5040 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5044 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5047 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5048 return ext4_ind_trans_blocks(inode, lblocks);
5049 return ext4_ext_index_trans_blocks(inode, pextents);
5053 * Account for index blocks, block groups bitmaps and block group
5054 * descriptor blocks if modify datablocks and index blocks
5055 * worse case, the indexs blocks spread over different block groups
5057 * If datablocks are discontiguous, they are possible to spread over
5058 * different block groups too. If they are contiguous, with flexbg,
5059 * they could still across block group boundary.
5061 * Also account for superblock, inode, quota and xattr blocks
5063 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5066 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5072 * How many index blocks need to touch to map @lblocks logical blocks
5073 * to @pextents physical extents?
5075 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5080 * Now let's see how many group bitmaps and group descriptors need
5083 groups = idxblocks + pextents;
5085 if (groups > ngroups)
5087 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5088 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5090 /* bitmaps and block group descriptor blocks */
5091 ret += groups + gdpblocks;
5093 /* Blocks for super block, inode, quota and xattr blocks */
5094 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5100 * Calculate the total number of credits to reserve to fit
5101 * the modification of a single pages into a single transaction,
5102 * which may include multiple chunks of block allocations.
5104 * This could be called via ext4_write_begin()
5106 * We need to consider the worse case, when
5107 * one new block per extent.
5109 int ext4_writepage_trans_blocks(struct inode *inode)
5111 int bpp = ext4_journal_blocks_per_page(inode);
5114 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5116 /* Account for data blocks for journalled mode */
5117 if (ext4_should_journal_data(inode))
5123 * Calculate the journal credits for a chunk of data modification.
5125 * This is called from DIO, fallocate or whoever calling
5126 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5128 * journal buffers for data blocks are not included here, as DIO
5129 * and fallocate do no need to journal data buffers.
5131 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5133 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5137 * The caller must have previously called ext4_reserve_inode_write().
5138 * Give this, we know that the caller already has write access to iloc->bh.
5140 int ext4_mark_iloc_dirty(handle_t *handle,
5141 struct inode *inode, struct ext4_iloc *iloc)
5145 if (IS_I_VERSION(inode))
5146 inode_inc_iversion(inode);
5148 /* the do_update_inode consumes one bh->b_count */
5151 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5152 err = ext4_do_update_inode(handle, inode, iloc);
5158 * On success, We end up with an outstanding reference count against
5159 * iloc->bh. This _must_ be cleaned up later.
5163 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5164 struct ext4_iloc *iloc)
5168 err = ext4_get_inode_loc(inode, iloc);
5170 BUFFER_TRACE(iloc->bh, "get_write_access");
5171 err = ext4_journal_get_write_access(handle, iloc->bh);
5177 ext4_std_error(inode->i_sb, err);
5182 * Expand an inode by new_extra_isize bytes.
5183 * Returns 0 on success or negative error number on failure.
5185 static int ext4_expand_extra_isize(struct inode *inode,
5186 unsigned int new_extra_isize,
5187 struct ext4_iloc iloc,
5190 struct ext4_inode *raw_inode;
5191 struct ext4_xattr_ibody_header *header;
5193 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5196 raw_inode = ext4_raw_inode(&iloc);
5198 header = IHDR(inode, raw_inode);
5200 /* No extended attributes present */
5201 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5202 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5203 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5205 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5209 /* try to expand with EAs present */
5210 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5215 * What we do here is to mark the in-core inode as clean with respect to inode
5216 * dirtiness (it may still be data-dirty).
5217 * This means that the in-core inode may be reaped by prune_icache
5218 * without having to perform any I/O. This is a very good thing,
5219 * because *any* task may call prune_icache - even ones which
5220 * have a transaction open against a different journal.
5222 * Is this cheating? Not really. Sure, we haven't written the
5223 * inode out, but prune_icache isn't a user-visible syncing function.
5224 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5225 * we start and wait on commits.
5227 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5229 struct ext4_iloc iloc;
5230 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5231 static unsigned int mnt_count;
5235 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5236 err = ext4_reserve_inode_write(handle, inode, &iloc);
5237 if (ext4_handle_valid(handle) &&
5238 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5239 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5241 * We need extra buffer credits since we may write into EA block
5242 * with this same handle. If journal_extend fails, then it will
5243 * only result in a minor loss of functionality for that inode.
5244 * If this is felt to be critical, then e2fsck should be run to
5245 * force a large enough s_min_extra_isize.
5247 if ((jbd2_journal_extend(handle,
5248 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5249 ret = ext4_expand_extra_isize(inode,
5250 sbi->s_want_extra_isize,
5253 ext4_set_inode_state(inode,
5254 EXT4_STATE_NO_EXPAND);
5256 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5257 ext4_warning(inode->i_sb,
5258 "Unable to expand inode %lu. Delete"
5259 " some EAs or run e2fsck.",
5262 le16_to_cpu(sbi->s_es->s_mnt_count);
5268 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5273 * ext4_dirty_inode() is called from __mark_inode_dirty()
5275 * We're really interested in the case where a file is being extended.
5276 * i_size has been changed by generic_commit_write() and we thus need
5277 * to include the updated inode in the current transaction.
5279 * Also, dquot_alloc_block() will always dirty the inode when blocks
5280 * are allocated to the file.
5282 * If the inode is marked synchronous, we don't honour that here - doing
5283 * so would cause a commit on atime updates, which we don't bother doing.
5284 * We handle synchronous inodes at the highest possible level.
5286 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5287 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5288 * to copy into the on-disk inode structure are the timestamp files.
5290 void ext4_dirty_inode(struct inode *inode, int flags)
5294 if (flags == I_DIRTY_TIME)
5296 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5300 ext4_mark_inode_dirty(handle, inode);
5302 ext4_journal_stop(handle);
5309 * Bind an inode's backing buffer_head into this transaction, to prevent
5310 * it from being flushed to disk early. Unlike
5311 * ext4_reserve_inode_write, this leaves behind no bh reference and
5312 * returns no iloc structure, so the caller needs to repeat the iloc
5313 * lookup to mark the inode dirty later.
5315 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5317 struct ext4_iloc iloc;
5321 err = ext4_get_inode_loc(inode, &iloc);
5323 BUFFER_TRACE(iloc.bh, "get_write_access");
5324 err = jbd2_journal_get_write_access(handle, iloc.bh);
5326 err = ext4_handle_dirty_metadata(handle,
5332 ext4_std_error(inode->i_sb, err);
5337 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5344 * We have to be very careful here: changing a data block's
5345 * journaling status dynamically is dangerous. If we write a
5346 * data block to the journal, change the status and then delete
5347 * that block, we risk forgetting to revoke the old log record
5348 * from the journal and so a subsequent replay can corrupt data.
5349 * So, first we make sure that the journal is empty and that
5350 * nobody is changing anything.
5353 journal = EXT4_JOURNAL(inode);
5356 if (is_journal_aborted(journal))
5358 /* We have to allocate physical blocks for delalloc blocks
5359 * before flushing journal. otherwise delalloc blocks can not
5360 * be allocated any more. even more truncate on delalloc blocks
5361 * could trigger BUG by flushing delalloc blocks in journal.
5362 * There is no delalloc block in non-journal data mode.
5364 if (val && test_opt(inode->i_sb, DELALLOC)) {
5365 err = ext4_alloc_da_blocks(inode);
5370 /* Wait for all existing dio workers */
5371 ext4_inode_block_unlocked_dio(inode);
5372 inode_dio_wait(inode);
5374 jbd2_journal_lock_updates(journal);
5377 * OK, there are no updates running now, and all cached data is
5378 * synced to disk. We are now in a completely consistent state
5379 * which doesn't have anything in the journal, and we know that
5380 * no filesystem updates are running, so it is safe to modify
5381 * the inode's in-core data-journaling state flag now.
5385 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5387 err = jbd2_journal_flush(journal);
5389 jbd2_journal_unlock_updates(journal);
5390 ext4_inode_resume_unlocked_dio(inode);
5393 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5395 ext4_set_aops(inode);
5397 jbd2_journal_unlock_updates(journal);
5398 ext4_inode_resume_unlocked_dio(inode);
5400 /* Finally we can mark the inode as dirty. */
5402 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5404 return PTR_ERR(handle);
5406 err = ext4_mark_inode_dirty(handle, inode);
5407 ext4_handle_sync(handle);
5408 ext4_journal_stop(handle);
5409 ext4_std_error(inode->i_sb, err);
5414 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5416 return !buffer_mapped(bh);
5419 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5421 struct page *page = vmf->page;
5425 struct file *file = vma->vm_file;
5426 struct inode *inode = file_inode(file);
5427 struct address_space *mapping = inode->i_mapping;
5429 get_block_t *get_block;
5432 sb_start_pagefault(inode->i_sb);
5433 file_update_time(vma->vm_file);
5435 down_read(&EXT4_I(inode)->i_mmap_sem);
5436 /* Delalloc case is easy... */
5437 if (test_opt(inode->i_sb, DELALLOC) &&
5438 !ext4_should_journal_data(inode) &&
5439 !ext4_nonda_switch(inode->i_sb)) {
5441 ret = block_page_mkwrite(vma, vmf,
5442 ext4_da_get_block_prep);
5443 } while (ret == -ENOSPC &&
5444 ext4_should_retry_alloc(inode->i_sb, &retries));
5449 size = i_size_read(inode);
5450 /* Page got truncated from under us? */
5451 if (page->mapping != mapping || page_offset(page) > size) {
5453 ret = VM_FAULT_NOPAGE;
5457 if (page->index == size >> PAGE_CACHE_SHIFT)
5458 len = size & ~PAGE_CACHE_MASK;
5460 len = PAGE_CACHE_SIZE;
5462 * Return if we have all the buffers mapped. This avoids the need to do
5463 * journal_start/journal_stop which can block and take a long time
5465 if (page_has_buffers(page)) {
5466 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5468 ext4_bh_unmapped)) {
5469 /* Wait so that we don't change page under IO */
5470 wait_for_stable_page(page);
5471 ret = VM_FAULT_LOCKED;
5476 /* OK, we need to fill the hole... */
5477 if (ext4_should_dioread_nolock(inode))
5478 get_block = ext4_get_block_write;
5480 get_block = ext4_get_block;
5482 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5483 ext4_writepage_trans_blocks(inode));
5484 if (IS_ERR(handle)) {
5485 ret = VM_FAULT_SIGBUS;
5488 ret = block_page_mkwrite(vma, vmf, get_block);
5489 if (!ret && ext4_should_journal_data(inode)) {
5490 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5491 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5493 ret = VM_FAULT_SIGBUS;
5494 ext4_journal_stop(handle);
5497 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5499 ext4_journal_stop(handle);
5500 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5503 ret = block_page_mkwrite_return(ret);
5505 up_read(&EXT4_I(inode)->i_mmap_sem);
5506 sb_end_pagefault(inode->i_sb);
5510 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5512 struct inode *inode = file_inode(vma->vm_file);
5515 down_read(&EXT4_I(inode)->i_mmap_sem);
5516 err = filemap_fault(vma, vmf);
5517 up_read(&EXT4_I(inode)->i_mmap_sem);