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);
56 int offset = offsetof(struct ext4_inode, i_checksum_lo);
57 unsigned int csum_size = sizeof(dummy_csum);
59 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
60 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
62 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
63 EXT4_GOOD_OLD_INODE_SIZE - offset);
65 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
66 offset = offsetof(struct ext4_inode, i_checksum_hi);
67 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
68 EXT4_GOOD_OLD_INODE_SIZE,
69 offset - EXT4_GOOD_OLD_INODE_SIZE);
70 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
71 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
74 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
75 EXT4_INODE_SIZE(inode->i_sb) -
83 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
84 struct ext4_inode_info *ei)
86 __u32 provided, calculated;
88 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
89 cpu_to_le32(EXT4_OS_LINUX) ||
90 !ext4_has_metadata_csum(inode->i_sb))
93 provided = le16_to_cpu(raw->i_checksum_lo);
94 calculated = ext4_inode_csum(inode, raw, ei);
95 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
96 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
97 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
101 return provided == calculated;
104 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
105 struct ext4_inode_info *ei)
109 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
110 cpu_to_le32(EXT4_OS_LINUX) ||
111 !ext4_has_metadata_csum(inode->i_sb))
114 csum = ext4_inode_csum(inode, raw, ei);
115 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
116 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
117 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
118 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode *inode,
124 trace_ext4_begin_ordered_truncate(inode, new_size);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode)->jinode)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
134 EXT4_I(inode)->jinode,
138 static void ext4_invalidatepage(struct page *page, unsigned int offset,
139 unsigned int length);
140 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
141 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
142 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
146 * Test whether an inode is a fast symlink.
148 int ext4_inode_is_fast_symlink(struct inode *inode)
150 int ea_blocks = EXT4_I(inode)->i_file_acl ?
151 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
153 if (ext4_has_inline_data(inode))
156 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
160 * Restart the transaction associated with *handle. This does a commit,
161 * so before we call here everything must be consistently dirtied against
164 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
170 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
171 * moment, get_block can be called only for blocks inside i_size since
172 * page cache has been already dropped and writes are blocked by
173 * i_mutex. So we can safely drop the i_data_sem here.
175 BUG_ON(EXT4_JOURNAL(inode) == NULL);
176 jbd_debug(2, "restarting handle %p\n", handle);
177 up_write(&EXT4_I(inode)->i_data_sem);
178 ret = ext4_journal_restart(handle, nblocks);
179 down_write(&EXT4_I(inode)->i_data_sem);
180 ext4_discard_preallocations(inode);
186 * Called at the last iput() if i_nlink is zero.
188 void ext4_evict_inode(struct inode *inode)
193 trace_ext4_evict_inode(inode);
195 if (inode->i_nlink) {
197 * When journalling data dirty buffers are tracked only in the
198 * journal. So although mm thinks everything is clean and
199 * ready for reaping the inode might still have some pages to
200 * write in the running transaction or waiting to be
201 * checkpointed. Thus calling jbd2_journal_invalidatepage()
202 * (via truncate_inode_pages()) to discard these buffers can
203 * cause data loss. Also even if we did not discard these
204 * buffers, we would have no way to find them after the inode
205 * is reaped and thus user could see stale data if he tries to
206 * read them before the transaction is checkpointed. So be
207 * careful and force everything to disk here... We use
208 * ei->i_datasync_tid to store the newest transaction
209 * containing inode's data.
211 * Note that directories do not have this problem because they
212 * don't use page cache.
214 if (inode->i_ino != EXT4_JOURNAL_INO &&
215 ext4_should_journal_data(inode) &&
216 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
217 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
218 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
220 jbd2_complete_transaction(journal, commit_tid);
221 filemap_write_and_wait(&inode->i_data);
223 truncate_inode_pages_final(&inode->i_data);
225 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
229 if (is_bad_inode(inode))
231 dquot_initialize(inode);
233 if (ext4_should_order_data(inode))
234 ext4_begin_ordered_truncate(inode, 0);
235 truncate_inode_pages_final(&inode->i_data);
237 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
240 * Protect us against freezing - iput() caller didn't have to have any
241 * protection against it
243 sb_start_intwrite(inode->i_sb);
244 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
245 ext4_blocks_for_truncate(inode)+3);
246 if (IS_ERR(handle)) {
247 ext4_std_error(inode->i_sb, PTR_ERR(handle));
249 * If we're going to skip the normal cleanup, we still need to
250 * make sure that the in-core orphan linked list is properly
253 ext4_orphan_del(NULL, inode);
254 sb_end_intwrite(inode->i_sb);
259 ext4_handle_sync(handle);
261 err = ext4_mark_inode_dirty(handle, inode);
263 ext4_warning(inode->i_sb,
264 "couldn't mark inode dirty (err %d)", err);
268 ext4_truncate(inode);
271 * ext4_ext_truncate() doesn't reserve any slop when it
272 * restarts journal transactions; therefore there may not be
273 * enough credits left in the handle to remove the inode from
274 * the orphan list and set the dtime field.
276 if (!ext4_handle_has_enough_credits(handle, 3)) {
277 err = ext4_journal_extend(handle, 3);
279 err = ext4_journal_restart(handle, 3);
281 ext4_warning(inode->i_sb,
282 "couldn't extend journal (err %d)", err);
284 ext4_journal_stop(handle);
285 ext4_orphan_del(NULL, inode);
286 sb_end_intwrite(inode->i_sb);
292 * Kill off the orphan record which ext4_truncate created.
293 * AKPM: I think this can be inside the above `if'.
294 * Note that ext4_orphan_del() has to be able to cope with the
295 * deletion of a non-existent orphan - this is because we don't
296 * know if ext4_truncate() actually created an orphan record.
297 * (Well, we could do this if we need to, but heck - it works)
299 ext4_orphan_del(handle, inode);
300 EXT4_I(inode)->i_dtime = get_seconds();
303 * One subtle ordering requirement: if anything has gone wrong
304 * (transaction abort, IO errors, whatever), then we can still
305 * do these next steps (the fs will already have been marked as
306 * having errors), but we can't free the inode if the mark_dirty
309 if (ext4_mark_inode_dirty(handle, inode))
310 /* If that failed, just do the required in-core inode clear. */
311 ext4_clear_inode(inode);
313 ext4_free_inode(handle, inode);
314 ext4_journal_stop(handle);
315 sb_end_intwrite(inode->i_sb);
318 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
322 qsize_t *ext4_get_reserved_space(struct inode *inode)
324 return &EXT4_I(inode)->i_reserved_quota;
329 * Called with i_data_sem down, which is important since we can call
330 * ext4_discard_preallocations() from here.
332 void ext4_da_update_reserve_space(struct inode *inode,
333 int used, int quota_claim)
335 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
336 struct ext4_inode_info *ei = EXT4_I(inode);
338 spin_lock(&ei->i_block_reservation_lock);
339 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
340 if (unlikely(used > ei->i_reserved_data_blocks)) {
341 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
342 "with only %d reserved data blocks",
343 __func__, inode->i_ino, used,
344 ei->i_reserved_data_blocks);
346 used = ei->i_reserved_data_blocks;
349 /* Update per-inode reservations */
350 ei->i_reserved_data_blocks -= used;
351 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
353 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
355 /* Update quota subsystem for data blocks */
357 dquot_claim_block(inode, EXT4_C2B(sbi, used));
360 * We did fallocate with an offset that is already delayed
361 * allocated. So on delayed allocated writeback we should
362 * not re-claim the quota for fallocated blocks.
364 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
368 * If we have done all the pending block allocations and if
369 * there aren't any writers on the inode, we can discard the
370 * inode's preallocations.
372 if ((ei->i_reserved_data_blocks == 0) &&
373 (atomic_read(&inode->i_writecount) == 0))
374 ext4_discard_preallocations(inode);
377 static int __check_block_validity(struct inode *inode, const char *func,
379 struct ext4_map_blocks *map)
381 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
383 ext4_error_inode(inode, func, line, map->m_pblk,
384 "lblock %lu mapped to illegal pblock "
385 "(length %d)", (unsigned long) map->m_lblk,
387 return -EFSCORRUPTED;
392 #define check_block_validity(inode, map) \
393 __check_block_validity((inode), __func__, __LINE__, (map))
395 #ifdef ES_AGGRESSIVE_TEST
396 static void ext4_map_blocks_es_recheck(handle_t *handle,
398 struct ext4_map_blocks *es_map,
399 struct ext4_map_blocks *map,
406 * There is a race window that the result is not the same.
407 * e.g. xfstests #223 when dioread_nolock enables. The reason
408 * is that we lookup a block mapping in extent status tree with
409 * out taking i_data_sem. So at the time the unwritten extent
410 * could be converted.
412 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
413 down_read(&EXT4_I(inode)->i_data_sem);
414 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
415 retval = ext4_ext_map_blocks(handle, inode, map, flags &
416 EXT4_GET_BLOCKS_KEEP_SIZE);
418 retval = ext4_ind_map_blocks(handle, inode, map, flags &
419 EXT4_GET_BLOCKS_KEEP_SIZE);
421 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
422 up_read((&EXT4_I(inode)->i_data_sem));
425 * We don't check m_len because extent will be collpased in status
426 * tree. So the m_len might not equal.
428 if (es_map->m_lblk != map->m_lblk ||
429 es_map->m_flags != map->m_flags ||
430 es_map->m_pblk != map->m_pblk) {
431 printk("ES cache assertion failed for inode: %lu "
432 "es_cached ex [%d/%d/%llu/%x] != "
433 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
434 inode->i_ino, es_map->m_lblk, es_map->m_len,
435 es_map->m_pblk, es_map->m_flags, map->m_lblk,
436 map->m_len, map->m_pblk, map->m_flags,
440 #endif /* ES_AGGRESSIVE_TEST */
443 * The ext4_map_blocks() function tries to look up the requested blocks,
444 * and returns if the blocks are already mapped.
446 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
447 * and store the allocated blocks in the result buffer head and mark it
450 * If file type is extents based, it will call ext4_ext_map_blocks(),
451 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
454 * On success, it returns the number of blocks being mapped or allocated.
455 * if create==0 and the blocks are pre-allocated and unwritten block,
456 * the result buffer head is unmapped. If the create ==1, it will make sure
457 * the buffer head is mapped.
459 * It returns 0 if plain look up failed (blocks have not been allocated), in
460 * that case, buffer head is unmapped
462 * It returns the error in case of allocation failure.
464 int ext4_map_blocks(handle_t *handle, struct inode *inode,
465 struct ext4_map_blocks *map, int flags)
467 struct extent_status es;
470 #ifdef ES_AGGRESSIVE_TEST
471 struct ext4_map_blocks orig_map;
473 memcpy(&orig_map, map, sizeof(*map));
477 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
478 "logical block %lu\n", inode->i_ino, flags, map->m_len,
479 (unsigned long) map->m_lblk);
482 * ext4_map_blocks returns an int, and m_len is an unsigned int
484 if (unlikely(map->m_len > INT_MAX))
485 map->m_len = INT_MAX;
487 /* We can handle the block number less than EXT_MAX_BLOCKS */
488 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
489 return -EFSCORRUPTED;
491 /* Lookup extent status tree firstly */
492 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
493 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
494 map->m_pblk = ext4_es_pblock(&es) +
495 map->m_lblk - es.es_lblk;
496 map->m_flags |= ext4_es_is_written(&es) ?
497 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
498 retval = es.es_len - (map->m_lblk - es.es_lblk);
499 if (retval > map->m_len)
502 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
507 #ifdef ES_AGGRESSIVE_TEST
508 ext4_map_blocks_es_recheck(handle, inode, map,
515 * Try to see if we can get the block without requesting a new
518 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
519 down_read(&EXT4_I(inode)->i_data_sem);
520 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
521 retval = ext4_ext_map_blocks(handle, inode, map, flags &
522 EXT4_GET_BLOCKS_KEEP_SIZE);
524 retval = ext4_ind_map_blocks(handle, inode, map, flags &
525 EXT4_GET_BLOCKS_KEEP_SIZE);
530 if (unlikely(retval != map->m_len)) {
531 ext4_warning(inode->i_sb,
532 "ES len assertion failed for inode "
533 "%lu: retval %d != map->m_len %d",
534 inode->i_ino, retval, map->m_len);
538 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
539 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
540 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
541 !(status & EXTENT_STATUS_WRITTEN) &&
542 ext4_find_delalloc_range(inode, map->m_lblk,
543 map->m_lblk + map->m_len - 1))
544 status |= EXTENT_STATUS_DELAYED;
545 ret = ext4_es_insert_extent(inode, map->m_lblk,
546 map->m_len, map->m_pblk, status);
550 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
551 up_read((&EXT4_I(inode)->i_data_sem));
554 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
555 ret = check_block_validity(inode, map);
560 /* If it is only a block(s) look up */
561 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
565 * Returns if the blocks have already allocated
567 * Note that if blocks have been preallocated
568 * ext4_ext_get_block() returns the create = 0
569 * with buffer head unmapped.
571 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
573 * If we need to convert extent to unwritten
574 * we continue and do the actual work in
575 * ext4_ext_map_blocks()
577 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
581 * Here we clear m_flags because after allocating an new extent,
582 * it will be set again.
584 map->m_flags &= ~EXT4_MAP_FLAGS;
587 * New blocks allocate and/or writing to unwritten extent
588 * will possibly result in updating i_data, so we take
589 * the write lock of i_data_sem, and call get_block()
590 * with create == 1 flag.
592 down_write(&EXT4_I(inode)->i_data_sem);
595 * We need to check for EXT4 here because migrate
596 * could have changed the inode type in between
598 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
599 retval = ext4_ext_map_blocks(handle, inode, map, flags);
601 retval = ext4_ind_map_blocks(handle, inode, map, flags);
603 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
605 * We allocated new blocks which will result in
606 * i_data's format changing. Force the migrate
607 * to fail by clearing migrate flags
609 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
613 * Update reserved blocks/metadata blocks after successful
614 * block allocation which had been deferred till now. We don't
615 * support fallocate for non extent files. So we can update
616 * reserve space here.
619 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
620 ext4_da_update_reserve_space(inode, retval, 1);
626 if (unlikely(retval != map->m_len)) {
627 ext4_warning(inode->i_sb,
628 "ES len assertion failed for inode "
629 "%lu: retval %d != map->m_len %d",
630 inode->i_ino, retval, map->m_len);
635 * If the extent has been zeroed out, we don't need to update
636 * extent status tree.
638 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
639 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
640 if (ext4_es_is_written(&es))
643 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
644 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
645 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
646 !(status & EXTENT_STATUS_WRITTEN) &&
647 ext4_find_delalloc_range(inode, map->m_lblk,
648 map->m_lblk + map->m_len - 1))
649 status |= EXTENT_STATUS_DELAYED;
650 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
651 map->m_pblk, status);
657 up_write((&EXT4_I(inode)->i_data_sem));
658 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
659 ret = check_block_validity(inode, map);
667 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
668 * we have to be careful as someone else may be manipulating b_state as well.
670 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
672 unsigned long old_state;
673 unsigned long new_state;
675 flags &= EXT4_MAP_FLAGS;
677 /* Dummy buffer_head? Set non-atomically. */
679 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
683 * Someone else may be modifying b_state. Be careful! This is ugly but
684 * once we get rid of using bh as a container for mapping information
685 * to pass to / from get_block functions, this can go away.
688 old_state = READ_ONCE(bh->b_state);
689 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
691 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
694 /* Maximum number of blocks we map for direct IO at once. */
695 #define DIO_MAX_BLOCKS 4096
697 static int _ext4_get_block(struct inode *inode, sector_t iblock,
698 struct buffer_head *bh, int flags)
700 handle_t *handle = ext4_journal_current_handle();
701 struct ext4_map_blocks map;
702 int ret = 0, started = 0;
705 if (ext4_has_inline_data(inode))
709 map.m_len = bh->b_size >> inode->i_blkbits;
711 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
712 /* Direct IO write... */
713 if (map.m_len > DIO_MAX_BLOCKS)
714 map.m_len = DIO_MAX_BLOCKS;
715 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
716 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
718 if (IS_ERR(handle)) {
719 ret = PTR_ERR(handle);
725 ret = ext4_map_blocks(handle, inode, &map, flags);
727 ext4_io_end_t *io_end = ext4_inode_aio(inode);
729 map_bh(bh, inode->i_sb, map.m_pblk);
730 ext4_update_bh_state(bh, map.m_flags);
731 if (IS_DAX(inode) && buffer_unwritten(bh)) {
733 * dgc: I suspect unwritten conversion on ext4+DAX is
734 * fundamentally broken here when there are concurrent
735 * read/write in progress on this inode.
737 WARN_ON_ONCE(io_end);
738 bh->b_assoc_map = inode->i_mapping;
739 bh->b_private = (void *)(unsigned long)iblock;
741 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
742 set_buffer_defer_completion(bh);
743 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
747 ext4_journal_stop(handle);
751 int ext4_get_block(struct inode *inode, sector_t iblock,
752 struct buffer_head *bh, int create)
754 return _ext4_get_block(inode, iblock, bh,
755 create ? EXT4_GET_BLOCKS_CREATE : 0);
759 * `handle' can be NULL if create is zero
761 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
762 ext4_lblk_t block, int map_flags)
764 struct ext4_map_blocks map;
765 struct buffer_head *bh;
766 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
769 J_ASSERT(handle != NULL || create == 0);
773 err = ext4_map_blocks(handle, inode, &map, map_flags);
776 return create ? ERR_PTR(-ENOSPC) : NULL;
780 bh = sb_getblk(inode->i_sb, map.m_pblk);
782 return ERR_PTR(-ENOMEM);
783 if (map.m_flags & EXT4_MAP_NEW) {
784 J_ASSERT(create != 0);
785 J_ASSERT(handle != NULL);
788 * Now that we do not always journal data, we should
789 * keep in mind whether this should always journal the
790 * new buffer as metadata. For now, regular file
791 * writes use ext4_get_block instead, so it's not a
795 BUFFER_TRACE(bh, "call get_create_access");
796 err = ext4_journal_get_create_access(handle, bh);
801 if (!buffer_uptodate(bh)) {
802 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
803 set_buffer_uptodate(bh);
806 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
807 err = ext4_handle_dirty_metadata(handle, inode, bh);
811 BUFFER_TRACE(bh, "not a new buffer");
818 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
819 ext4_lblk_t block, int map_flags)
821 struct buffer_head *bh;
823 bh = ext4_getblk(handle, inode, block, map_flags);
826 if (!bh || buffer_uptodate(bh))
828 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
830 if (buffer_uptodate(bh))
833 return ERR_PTR(-EIO);
836 int ext4_walk_page_buffers(handle_t *handle,
837 struct buffer_head *head,
841 int (*fn)(handle_t *handle,
842 struct buffer_head *bh))
844 struct buffer_head *bh;
845 unsigned block_start, block_end;
846 unsigned blocksize = head->b_size;
848 struct buffer_head *next;
850 for (bh = head, block_start = 0;
851 ret == 0 && (bh != head || !block_start);
852 block_start = block_end, bh = next) {
853 next = bh->b_this_page;
854 block_end = block_start + blocksize;
855 if (block_end <= from || block_start >= to) {
856 if (partial && !buffer_uptodate(bh))
860 err = (*fn)(handle, bh);
868 * To preserve ordering, it is essential that the hole instantiation and
869 * the data write be encapsulated in a single transaction. We cannot
870 * close off a transaction and start a new one between the ext4_get_block()
871 * and the commit_write(). So doing the jbd2_journal_start at the start of
872 * prepare_write() is the right place.
874 * Also, this function can nest inside ext4_writepage(). In that case, we
875 * *know* that ext4_writepage() has generated enough buffer credits to do the
876 * whole page. So we won't block on the journal in that case, which is good,
877 * because the caller may be PF_MEMALLOC.
879 * By accident, ext4 can be reentered when a transaction is open via
880 * quota file writes. If we were to commit the transaction while thus
881 * reentered, there can be a deadlock - we would be holding a quota
882 * lock, and the commit would never complete if another thread had a
883 * transaction open and was blocking on the quota lock - a ranking
886 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
887 * will _not_ run commit under these circumstances because handle->h_ref
888 * is elevated. We'll still have enough credits for the tiny quotafile
891 int do_journal_get_write_access(handle_t *handle,
892 struct buffer_head *bh)
894 int dirty = buffer_dirty(bh);
897 if (!buffer_mapped(bh) || buffer_freed(bh))
900 * __block_write_begin() could have dirtied some buffers. Clean
901 * the dirty bit as jbd2_journal_get_write_access() could complain
902 * otherwise about fs integrity issues. Setting of the dirty bit
903 * by __block_write_begin() isn't a real problem here as we clear
904 * the bit before releasing a page lock and thus writeback cannot
905 * ever write the buffer.
908 clear_buffer_dirty(bh);
909 BUFFER_TRACE(bh, "get write access");
910 ret = ext4_journal_get_write_access(handle, bh);
912 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
916 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
917 struct buffer_head *bh_result, int create);
919 #ifdef CONFIG_EXT4_FS_ENCRYPTION
920 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
921 get_block_t *get_block)
923 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
924 unsigned to = from + len;
925 struct inode *inode = page->mapping->host;
926 unsigned block_start, block_end;
929 unsigned blocksize = inode->i_sb->s_blocksize;
931 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
932 bool decrypt = false;
934 BUG_ON(!PageLocked(page));
935 BUG_ON(from > PAGE_CACHE_SIZE);
936 BUG_ON(to > PAGE_CACHE_SIZE);
939 if (!page_has_buffers(page))
940 create_empty_buffers(page, blocksize, 0);
941 head = page_buffers(page);
942 bbits = ilog2(blocksize);
943 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
945 for (bh = head, block_start = 0; bh != head || !block_start;
946 block++, block_start = block_end, bh = bh->b_this_page) {
947 block_end = block_start + blocksize;
948 if (block_end <= from || block_start >= to) {
949 if (PageUptodate(page)) {
950 if (!buffer_uptodate(bh))
951 set_buffer_uptodate(bh);
956 clear_buffer_new(bh);
957 if (!buffer_mapped(bh)) {
958 WARN_ON(bh->b_size != blocksize);
959 err = get_block(inode, block, bh, 1);
962 if (buffer_new(bh)) {
963 unmap_underlying_metadata(bh->b_bdev,
965 if (PageUptodate(page)) {
966 clear_buffer_new(bh);
967 set_buffer_uptodate(bh);
968 mark_buffer_dirty(bh);
971 if (block_end > to || block_start < from)
972 zero_user_segments(page, to, block_end,
977 if (PageUptodate(page)) {
978 if (!buffer_uptodate(bh))
979 set_buffer_uptodate(bh);
982 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
983 !buffer_unwritten(bh) &&
984 (block_start < from || block_end > to)) {
985 ll_rw_block(READ, 1, &bh);
987 decrypt = ext4_encrypted_inode(inode) &&
988 S_ISREG(inode->i_mode);
992 * If we issued read requests, let them complete.
994 while (wait_bh > wait) {
995 wait_on_buffer(*--wait_bh);
996 if (!buffer_uptodate(*wait_bh))
1000 page_zero_new_buffers(page, from, to);
1002 err = ext4_decrypt(page);
1007 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1008 loff_t pos, unsigned len, unsigned flags,
1009 struct page **pagep, void **fsdata)
1011 struct inode *inode = mapping->host;
1012 int ret, needed_blocks;
1019 trace_ext4_write_begin(inode, pos, len, flags);
1021 * Reserve one block more for addition to orphan list in case
1022 * we allocate blocks but write fails for some reason
1024 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1025 index = pos >> PAGE_CACHE_SHIFT;
1026 from = pos & (PAGE_CACHE_SIZE - 1);
1029 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1030 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1039 * grab_cache_page_write_begin() can take a long time if the
1040 * system is thrashing due to memory pressure, or if the page
1041 * is being written back. So grab it first before we start
1042 * the transaction handle. This also allows us to allocate
1043 * the page (if needed) without using GFP_NOFS.
1046 page = grab_cache_page_write_begin(mapping, index, flags);
1052 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1053 if (IS_ERR(handle)) {
1054 page_cache_release(page);
1055 return PTR_ERR(handle);
1059 if (page->mapping != mapping) {
1060 /* The page got truncated from under us */
1062 page_cache_release(page);
1063 ext4_journal_stop(handle);
1066 /* In case writeback began while the page was unlocked */
1067 wait_for_stable_page(page);
1069 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1070 if (ext4_should_dioread_nolock(inode))
1071 ret = ext4_block_write_begin(page, pos, len,
1072 ext4_get_block_write);
1074 ret = ext4_block_write_begin(page, pos, len,
1077 if (ext4_should_dioread_nolock(inode))
1078 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1080 ret = __block_write_begin(page, pos, len, ext4_get_block);
1082 if (!ret && ext4_should_journal_data(inode)) {
1083 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1085 do_journal_get_write_access);
1091 * __block_write_begin may have instantiated a few blocks
1092 * outside i_size. Trim these off again. Don't need
1093 * i_size_read because we hold i_mutex.
1095 * Add inode to orphan list in case we crash before
1098 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1099 ext4_orphan_add(handle, inode);
1101 ext4_journal_stop(handle);
1102 if (pos + len > inode->i_size) {
1103 ext4_truncate_failed_write(inode);
1105 * If truncate failed early the inode might
1106 * still be on the orphan list; we need to
1107 * make sure the inode is removed from the
1108 * orphan list in that case.
1111 ext4_orphan_del(NULL, inode);
1114 if (ret == -ENOSPC &&
1115 ext4_should_retry_alloc(inode->i_sb, &retries))
1117 page_cache_release(page);
1124 /* For write_end() in data=journal mode */
1125 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1128 if (!buffer_mapped(bh) || buffer_freed(bh))
1130 set_buffer_uptodate(bh);
1131 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1132 clear_buffer_meta(bh);
1133 clear_buffer_prio(bh);
1138 * We need to pick up the new inode size which generic_commit_write gave us
1139 * `file' can be NULL - eg, when called from page_symlink().
1141 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1142 * buffers are managed internally.
1144 static int ext4_write_end(struct file *file,
1145 struct address_space *mapping,
1146 loff_t pos, unsigned len, unsigned copied,
1147 struct page *page, void *fsdata)
1149 handle_t *handle = ext4_journal_current_handle();
1150 struct inode *inode = mapping->host;
1151 loff_t old_size = inode->i_size;
1153 int i_size_changed = 0;
1155 trace_ext4_write_end(inode, pos, len, copied);
1156 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1157 ret = ext4_jbd2_file_inode(handle, inode);
1160 page_cache_release(page);
1165 if (ext4_has_inline_data(inode)) {
1166 ret = ext4_write_inline_data_end(inode, pos, len,
1172 copied = block_write_end(file, mapping, pos,
1173 len, copied, page, fsdata);
1175 * it's important to update i_size while still holding page lock:
1176 * page writeout could otherwise come in and zero beyond i_size.
1178 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1180 page_cache_release(page);
1183 pagecache_isize_extended(inode, old_size, pos);
1185 * Don't mark the inode dirty under page lock. First, it unnecessarily
1186 * makes the holding time of page lock longer. Second, it forces lock
1187 * ordering of page lock and transaction start for journaling
1191 ext4_mark_inode_dirty(handle, inode);
1193 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1194 /* if we have allocated more blocks and copied
1195 * less. We will have blocks allocated outside
1196 * inode->i_size. So truncate them
1198 ext4_orphan_add(handle, inode);
1200 ret2 = ext4_journal_stop(handle);
1204 if (pos + len > inode->i_size) {
1205 ext4_truncate_failed_write(inode);
1207 * If truncate failed early the inode might still be
1208 * on the orphan list; we need to make sure the inode
1209 * is removed from the orphan list in that case.
1212 ext4_orphan_del(NULL, inode);
1215 return ret ? ret : copied;
1219 * This is a private version of page_zero_new_buffers() which doesn't
1220 * set the buffer to be dirty, since in data=journalled mode we need
1221 * to call ext4_handle_dirty_metadata() instead.
1223 static void zero_new_buffers(struct page *page, unsigned from, unsigned to)
1225 unsigned int block_start = 0, block_end;
1226 struct buffer_head *head, *bh;
1228 bh = head = page_buffers(page);
1230 block_end = block_start + bh->b_size;
1231 if (buffer_new(bh)) {
1232 if (block_end > from && block_start < to) {
1233 if (!PageUptodate(page)) {
1234 unsigned start, size;
1236 start = max(from, block_start);
1237 size = min(to, block_end) - start;
1239 zero_user(page, start, size);
1240 set_buffer_uptodate(bh);
1242 clear_buffer_new(bh);
1245 block_start = block_end;
1246 bh = bh->b_this_page;
1247 } while (bh != head);
1250 static int ext4_journalled_write_end(struct file *file,
1251 struct address_space *mapping,
1252 loff_t pos, unsigned len, unsigned copied,
1253 struct page *page, void *fsdata)
1255 handle_t *handle = ext4_journal_current_handle();
1256 struct inode *inode = mapping->host;
1257 loff_t old_size = inode->i_size;
1261 int size_changed = 0;
1263 trace_ext4_journalled_write_end(inode, pos, len, copied);
1264 from = pos & (PAGE_CACHE_SIZE - 1);
1267 BUG_ON(!ext4_handle_valid(handle));
1269 if (ext4_has_inline_data(inode))
1270 copied = ext4_write_inline_data_end(inode, pos, len,
1274 if (!PageUptodate(page))
1276 zero_new_buffers(page, from+copied, to);
1279 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1280 to, &partial, write_end_fn);
1282 SetPageUptodate(page);
1284 size_changed = ext4_update_inode_size(inode, pos + copied);
1285 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1286 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1288 page_cache_release(page);
1291 pagecache_isize_extended(inode, old_size, pos);
1294 ret2 = ext4_mark_inode_dirty(handle, inode);
1299 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1300 /* if we have allocated more blocks and copied
1301 * less. We will have blocks allocated outside
1302 * inode->i_size. So truncate them
1304 ext4_orphan_add(handle, inode);
1306 ret2 = ext4_journal_stop(handle);
1309 if (pos + len > inode->i_size) {
1310 ext4_truncate_failed_write(inode);
1312 * If truncate failed early the inode might still be
1313 * on the orphan list; we need to make sure the inode
1314 * is removed from the orphan list in that case.
1317 ext4_orphan_del(NULL, inode);
1320 return ret ? ret : copied;
1324 * Reserve space for a single cluster
1326 static int ext4_da_reserve_space(struct inode *inode)
1328 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1329 struct ext4_inode_info *ei = EXT4_I(inode);
1333 * We will charge metadata quota at writeout time; this saves
1334 * us from metadata over-estimation, though we may go over by
1335 * a small amount in the end. Here we just reserve for data.
1337 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1341 spin_lock(&ei->i_block_reservation_lock);
1342 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1343 spin_unlock(&ei->i_block_reservation_lock);
1344 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1347 ei->i_reserved_data_blocks++;
1348 trace_ext4_da_reserve_space(inode);
1349 spin_unlock(&ei->i_block_reservation_lock);
1351 return 0; /* success */
1354 static void ext4_da_release_space(struct inode *inode, int to_free)
1356 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1357 struct ext4_inode_info *ei = EXT4_I(inode);
1360 return; /* Nothing to release, exit */
1362 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1364 trace_ext4_da_release_space(inode, to_free);
1365 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1367 * if there aren't enough reserved blocks, then the
1368 * counter is messed up somewhere. Since this
1369 * function is called from invalidate page, it's
1370 * harmless to return without any action.
1372 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1373 "ino %lu, to_free %d with only %d reserved "
1374 "data blocks", inode->i_ino, to_free,
1375 ei->i_reserved_data_blocks);
1377 to_free = ei->i_reserved_data_blocks;
1379 ei->i_reserved_data_blocks -= to_free;
1381 /* update fs dirty data blocks counter */
1382 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1384 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1386 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1389 static void ext4_da_page_release_reservation(struct page *page,
1390 unsigned int offset,
1391 unsigned int length)
1393 int to_release = 0, contiguous_blks = 0;
1394 struct buffer_head *head, *bh;
1395 unsigned int curr_off = 0;
1396 struct inode *inode = page->mapping->host;
1397 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1398 unsigned int stop = offset + length;
1402 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1404 head = page_buffers(page);
1407 unsigned int next_off = curr_off + bh->b_size;
1409 if (next_off > stop)
1412 if ((offset <= curr_off) && (buffer_delay(bh))) {
1415 clear_buffer_delay(bh);
1416 } else if (contiguous_blks) {
1417 lblk = page->index <<
1418 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1419 lblk += (curr_off >> inode->i_blkbits) -
1421 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1422 contiguous_blks = 0;
1424 curr_off = next_off;
1425 } while ((bh = bh->b_this_page) != head);
1427 if (contiguous_blks) {
1428 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1429 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1430 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1433 /* If we have released all the blocks belonging to a cluster, then we
1434 * need to release the reserved space for that cluster. */
1435 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1436 while (num_clusters > 0) {
1437 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1438 ((num_clusters - 1) << sbi->s_cluster_bits);
1439 if (sbi->s_cluster_ratio == 1 ||
1440 !ext4_find_delalloc_cluster(inode, lblk))
1441 ext4_da_release_space(inode, 1);
1448 * Delayed allocation stuff
1451 struct mpage_da_data {
1452 struct inode *inode;
1453 struct writeback_control *wbc;
1455 pgoff_t first_page; /* The first page to write */
1456 pgoff_t next_page; /* Current page to examine */
1457 pgoff_t last_page; /* Last page to examine */
1459 * Extent to map - this can be after first_page because that can be
1460 * fully mapped. We somewhat abuse m_flags to store whether the extent
1461 * is delalloc or unwritten.
1463 struct ext4_map_blocks map;
1464 struct ext4_io_submit io_submit; /* IO submission data */
1467 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1472 struct pagevec pvec;
1473 struct inode *inode = mpd->inode;
1474 struct address_space *mapping = inode->i_mapping;
1476 /* This is necessary when next_page == 0. */
1477 if (mpd->first_page >= mpd->next_page)
1480 index = mpd->first_page;
1481 end = mpd->next_page - 1;
1483 ext4_lblk_t start, last;
1484 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1485 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1486 ext4_es_remove_extent(inode, start, last - start + 1);
1489 pagevec_init(&pvec, 0);
1490 while (index <= end) {
1491 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1494 for (i = 0; i < nr_pages; i++) {
1495 struct page *page = pvec.pages[i];
1496 if (page->index > end)
1498 BUG_ON(!PageLocked(page));
1499 BUG_ON(PageWriteback(page));
1501 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1502 ClearPageUptodate(page);
1506 index = pvec.pages[nr_pages - 1]->index + 1;
1507 pagevec_release(&pvec);
1511 static void ext4_print_free_blocks(struct inode *inode)
1513 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1514 struct super_block *sb = inode->i_sb;
1515 struct ext4_inode_info *ei = EXT4_I(inode);
1517 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1518 EXT4_C2B(EXT4_SB(inode->i_sb),
1519 ext4_count_free_clusters(sb)));
1520 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1521 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1522 (long long) EXT4_C2B(EXT4_SB(sb),
1523 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1524 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1525 (long long) EXT4_C2B(EXT4_SB(sb),
1526 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1527 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1528 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1529 ei->i_reserved_data_blocks);
1533 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1535 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1539 * This function is grabs code from the very beginning of
1540 * ext4_map_blocks, but assumes that the caller is from delayed write
1541 * time. This function looks up the requested blocks and sets the
1542 * buffer delay bit under the protection of i_data_sem.
1544 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1545 struct ext4_map_blocks *map,
1546 struct buffer_head *bh)
1548 struct extent_status es;
1550 sector_t invalid_block = ~((sector_t) 0xffff);
1551 #ifdef ES_AGGRESSIVE_TEST
1552 struct ext4_map_blocks orig_map;
1554 memcpy(&orig_map, map, sizeof(*map));
1557 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1561 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1562 "logical block %lu\n", inode->i_ino, map->m_len,
1563 (unsigned long) map->m_lblk);
1565 /* Lookup extent status tree firstly */
1566 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1567 if (ext4_es_is_hole(&es)) {
1569 down_read(&EXT4_I(inode)->i_data_sem);
1574 * Delayed extent could be allocated by fallocate.
1575 * So we need to check it.
1577 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1578 map_bh(bh, inode->i_sb, invalid_block);
1580 set_buffer_delay(bh);
1584 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1585 retval = es.es_len - (iblock - es.es_lblk);
1586 if (retval > map->m_len)
1587 retval = map->m_len;
1588 map->m_len = retval;
1589 if (ext4_es_is_written(&es))
1590 map->m_flags |= EXT4_MAP_MAPPED;
1591 else if (ext4_es_is_unwritten(&es))
1592 map->m_flags |= EXT4_MAP_UNWRITTEN;
1596 #ifdef ES_AGGRESSIVE_TEST
1597 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1603 * Try to see if we can get the block without requesting a new
1604 * file system block.
1606 down_read(&EXT4_I(inode)->i_data_sem);
1607 if (ext4_has_inline_data(inode))
1609 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1610 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1612 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1618 * XXX: __block_prepare_write() unmaps passed block,
1622 * If the block was allocated from previously allocated cluster,
1623 * then we don't need to reserve it again. However we still need
1624 * to reserve metadata for every block we're going to write.
1626 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1627 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1628 ret = ext4_da_reserve_space(inode);
1630 /* not enough space to reserve */
1636 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1637 ~0, EXTENT_STATUS_DELAYED);
1643 map_bh(bh, inode->i_sb, invalid_block);
1645 set_buffer_delay(bh);
1646 } else if (retval > 0) {
1648 unsigned int status;
1650 if (unlikely(retval != map->m_len)) {
1651 ext4_warning(inode->i_sb,
1652 "ES len assertion failed for inode "
1653 "%lu: retval %d != map->m_len %d",
1654 inode->i_ino, retval, map->m_len);
1658 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1659 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1660 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1661 map->m_pblk, status);
1667 up_read((&EXT4_I(inode)->i_data_sem));
1673 * This is a special get_block_t callback which is used by
1674 * ext4_da_write_begin(). It will either return mapped block or
1675 * reserve space for a single block.
1677 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1678 * We also have b_blocknr = -1 and b_bdev initialized properly
1680 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1681 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1682 * initialized properly.
1684 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1685 struct buffer_head *bh, int create)
1687 struct ext4_map_blocks map;
1690 BUG_ON(create == 0);
1691 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1693 map.m_lblk = iblock;
1697 * first, we need to know whether the block is allocated already
1698 * preallocated blocks are unmapped but should treated
1699 * the same as allocated blocks.
1701 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1705 map_bh(bh, inode->i_sb, map.m_pblk);
1706 ext4_update_bh_state(bh, map.m_flags);
1708 if (buffer_unwritten(bh)) {
1709 /* A delayed write to unwritten bh should be marked
1710 * new and mapped. Mapped ensures that we don't do
1711 * get_block multiple times when we write to the same
1712 * offset and new ensures that we do proper zero out
1713 * for partial write.
1716 set_buffer_mapped(bh);
1721 static int bget_one(handle_t *handle, struct buffer_head *bh)
1727 static int bput_one(handle_t *handle, struct buffer_head *bh)
1733 static int __ext4_journalled_writepage(struct page *page,
1736 struct address_space *mapping = page->mapping;
1737 struct inode *inode = mapping->host;
1738 struct buffer_head *page_bufs = NULL;
1739 handle_t *handle = NULL;
1740 int ret = 0, err = 0;
1741 int inline_data = ext4_has_inline_data(inode);
1742 struct buffer_head *inode_bh = NULL;
1744 ClearPageChecked(page);
1747 BUG_ON(page->index != 0);
1748 BUG_ON(len > ext4_get_max_inline_size(inode));
1749 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1750 if (inode_bh == NULL)
1753 page_bufs = page_buffers(page);
1758 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1762 * We need to release the page lock before we start the
1763 * journal, so grab a reference so the page won't disappear
1764 * out from under us.
1769 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1770 ext4_writepage_trans_blocks(inode));
1771 if (IS_ERR(handle)) {
1772 ret = PTR_ERR(handle);
1774 goto out_no_pagelock;
1776 BUG_ON(!ext4_handle_valid(handle));
1780 if (page->mapping != mapping) {
1781 /* The page got truncated from under us */
1782 ext4_journal_stop(handle);
1788 BUFFER_TRACE(inode_bh, "get write access");
1789 ret = ext4_journal_get_write_access(handle, inode_bh);
1791 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1794 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1795 do_journal_get_write_access);
1797 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1802 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1803 err = ext4_journal_stop(handle);
1807 if (!ext4_has_inline_data(inode))
1808 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1810 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1819 * Note that we don't need to start a transaction unless we're journaling data
1820 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1821 * need to file the inode to the transaction's list in ordered mode because if
1822 * we are writing back data added by write(), the inode is already there and if
1823 * we are writing back data modified via mmap(), no one guarantees in which
1824 * transaction the data will hit the disk. In case we are journaling data, we
1825 * cannot start transaction directly because transaction start ranks above page
1826 * lock so we have to do some magic.
1828 * This function can get called via...
1829 * - ext4_writepages after taking page lock (have journal handle)
1830 * - journal_submit_inode_data_buffers (no journal handle)
1831 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1832 * - grab_page_cache when doing write_begin (have journal handle)
1834 * We don't do any block allocation in this function. If we have page with
1835 * multiple blocks we need to write those buffer_heads that are mapped. This
1836 * is important for mmaped based write. So if we do with blocksize 1K
1837 * truncate(f, 1024);
1838 * a = mmap(f, 0, 4096);
1840 * truncate(f, 4096);
1841 * we have in the page first buffer_head mapped via page_mkwrite call back
1842 * but other buffer_heads would be unmapped but dirty (dirty done via the
1843 * do_wp_page). So writepage should write the first block. If we modify
1844 * the mmap area beyond 1024 we will again get a page_fault and the
1845 * page_mkwrite callback will do the block allocation and mark the
1846 * buffer_heads mapped.
1848 * We redirty the page if we have any buffer_heads that is either delay or
1849 * unwritten in the page.
1851 * We can get recursively called as show below.
1853 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1856 * But since we don't do any block allocation we should not deadlock.
1857 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1859 static int ext4_writepage(struct page *page,
1860 struct writeback_control *wbc)
1865 struct buffer_head *page_bufs = NULL;
1866 struct inode *inode = page->mapping->host;
1867 struct ext4_io_submit io_submit;
1868 bool keep_towrite = false;
1870 trace_ext4_writepage(page);
1871 size = i_size_read(inode);
1872 if (page->index == size >> PAGE_CACHE_SHIFT)
1873 len = size & ~PAGE_CACHE_MASK;
1875 len = PAGE_CACHE_SIZE;
1877 page_bufs = page_buffers(page);
1879 * We cannot do block allocation or other extent handling in this
1880 * function. If there are buffers needing that, we have to redirty
1881 * the page. But we may reach here when we do a journal commit via
1882 * journal_submit_inode_data_buffers() and in that case we must write
1883 * allocated buffers to achieve data=ordered mode guarantees.
1885 * Also, if there is only one buffer per page (the fs block
1886 * size == the page size), if one buffer needs block
1887 * allocation or needs to modify the extent tree to clear the
1888 * unwritten flag, we know that the page can't be written at
1889 * all, so we might as well refuse the write immediately.
1890 * Unfortunately if the block size != page size, we can't as
1891 * easily detect this case using ext4_walk_page_buffers(), but
1892 * for the extremely common case, this is an optimization that
1893 * skips a useless round trip through ext4_bio_write_page().
1895 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1896 ext4_bh_delay_or_unwritten)) {
1897 redirty_page_for_writepage(wbc, page);
1898 if ((current->flags & PF_MEMALLOC) ||
1899 (inode->i_sb->s_blocksize == PAGE_CACHE_SIZE)) {
1901 * For memory cleaning there's no point in writing only
1902 * some buffers. So just bail out. Warn if we came here
1903 * from direct reclaim.
1905 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1910 keep_towrite = true;
1913 if (PageChecked(page) && ext4_should_journal_data(inode))
1915 * It's mmapped pagecache. Add buffers and journal it. There
1916 * doesn't seem much point in redirtying the page here.
1918 return __ext4_journalled_writepage(page, len);
1920 ext4_io_submit_init(&io_submit, wbc);
1921 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1922 if (!io_submit.io_end) {
1923 redirty_page_for_writepage(wbc, page);
1927 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1928 ext4_io_submit(&io_submit);
1929 /* Drop io_end reference we got from init */
1930 ext4_put_io_end_defer(io_submit.io_end);
1934 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1937 loff_t size = i_size_read(mpd->inode);
1940 BUG_ON(page->index != mpd->first_page);
1941 if (page->index == size >> PAGE_CACHE_SHIFT)
1942 len = size & ~PAGE_CACHE_MASK;
1944 len = PAGE_CACHE_SIZE;
1945 clear_page_dirty_for_io(page);
1946 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1948 mpd->wbc->nr_to_write--;
1954 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1957 * mballoc gives us at most this number of blocks...
1958 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1959 * The rest of mballoc seems to handle chunks up to full group size.
1961 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1964 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1966 * @mpd - extent of blocks
1967 * @lblk - logical number of the block in the file
1968 * @bh - buffer head we want to add to the extent
1970 * The function is used to collect contig. blocks in the same state. If the
1971 * buffer doesn't require mapping for writeback and we haven't started the
1972 * extent of buffers to map yet, the function returns 'true' immediately - the
1973 * caller can write the buffer right away. Otherwise the function returns true
1974 * if the block has been added to the extent, false if the block couldn't be
1977 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1978 struct buffer_head *bh)
1980 struct ext4_map_blocks *map = &mpd->map;
1982 /* Buffer that doesn't need mapping for writeback? */
1983 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1984 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1985 /* So far no extent to map => we write the buffer right away */
1986 if (map->m_len == 0)
1991 /* First block in the extent? */
1992 if (map->m_len == 0) {
1995 map->m_flags = bh->b_state & BH_FLAGS;
1999 /* Don't go larger than mballoc is willing to allocate */
2000 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2003 /* Can we merge the block to our big extent? */
2004 if (lblk == map->m_lblk + map->m_len &&
2005 (bh->b_state & BH_FLAGS) == map->m_flags) {
2013 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2015 * @mpd - extent of blocks for mapping
2016 * @head - the first buffer in the page
2017 * @bh - buffer we should start processing from
2018 * @lblk - logical number of the block in the file corresponding to @bh
2020 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2021 * the page for IO if all buffers in this page were mapped and there's no
2022 * accumulated extent of buffers to map or add buffers in the page to the
2023 * extent of buffers to map. The function returns 1 if the caller can continue
2024 * by processing the next page, 0 if it should stop adding buffers to the
2025 * extent to map because we cannot extend it anymore. It can also return value
2026 * < 0 in case of error during IO submission.
2028 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2029 struct buffer_head *head,
2030 struct buffer_head *bh,
2033 struct inode *inode = mpd->inode;
2035 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2036 >> inode->i_blkbits;
2039 BUG_ON(buffer_locked(bh));
2041 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2042 /* Found extent to map? */
2045 /* Everything mapped so far and we hit EOF */
2048 } while (lblk++, (bh = bh->b_this_page) != head);
2049 /* So far everything mapped? Submit the page for IO. */
2050 if (mpd->map.m_len == 0) {
2051 err = mpage_submit_page(mpd, head->b_page);
2055 return lblk < blocks;
2059 * mpage_map_buffers - update buffers corresponding to changed extent and
2060 * submit fully mapped pages for IO
2062 * @mpd - description of extent to map, on return next extent to map
2064 * Scan buffers corresponding to changed extent (we expect corresponding pages
2065 * to be already locked) and update buffer state according to new extent state.
2066 * We map delalloc buffers to their physical location, clear unwritten bits,
2067 * and mark buffers as uninit when we perform writes to unwritten extents
2068 * and do extent conversion after IO is finished. If the last page is not fully
2069 * mapped, we update @map to the next extent in the last page that needs
2070 * mapping. Otherwise we submit the page for IO.
2072 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2074 struct pagevec pvec;
2076 struct inode *inode = mpd->inode;
2077 struct buffer_head *head, *bh;
2078 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2084 start = mpd->map.m_lblk >> bpp_bits;
2085 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2086 lblk = start << bpp_bits;
2087 pblock = mpd->map.m_pblk;
2089 pagevec_init(&pvec, 0);
2090 while (start <= end) {
2091 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2095 for (i = 0; i < nr_pages; i++) {
2096 struct page *page = pvec.pages[i];
2098 if (page->index > end)
2100 /* Up to 'end' pages must be contiguous */
2101 BUG_ON(page->index != start);
2102 bh = head = page_buffers(page);
2104 if (lblk < mpd->map.m_lblk)
2106 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2108 * Buffer after end of mapped extent.
2109 * Find next buffer in the page to map.
2112 mpd->map.m_flags = 0;
2114 * FIXME: If dioread_nolock supports
2115 * blocksize < pagesize, we need to make
2116 * sure we add size mapped so far to
2117 * io_end->size as the following call
2118 * can submit the page for IO.
2120 err = mpage_process_page_bufs(mpd, head,
2122 pagevec_release(&pvec);
2127 if (buffer_delay(bh)) {
2128 clear_buffer_delay(bh);
2129 bh->b_blocknr = pblock++;
2131 clear_buffer_unwritten(bh);
2132 } while (lblk++, (bh = bh->b_this_page) != head);
2135 * FIXME: This is going to break if dioread_nolock
2136 * supports blocksize < pagesize as we will try to
2137 * convert potentially unmapped parts of inode.
2139 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2140 /* Page fully mapped - let IO run! */
2141 err = mpage_submit_page(mpd, page);
2143 pagevec_release(&pvec);
2148 pagevec_release(&pvec);
2150 /* Extent fully mapped and matches with page boundary. We are done. */
2152 mpd->map.m_flags = 0;
2156 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2158 struct inode *inode = mpd->inode;
2159 struct ext4_map_blocks *map = &mpd->map;
2160 int get_blocks_flags;
2161 int err, dioread_nolock;
2163 trace_ext4_da_write_pages_extent(inode, map);
2165 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2166 * to convert an unwritten extent to be initialized (in the case
2167 * where we have written into one or more preallocated blocks). It is
2168 * possible that we're going to need more metadata blocks than
2169 * previously reserved. However we must not fail because we're in
2170 * writeback and there is nothing we can do about it so it might result
2171 * in data loss. So use reserved blocks to allocate metadata if
2174 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2175 * the blocks in question are delalloc blocks. This indicates
2176 * that the blocks and quotas has already been checked when
2177 * the data was copied into the page cache.
2179 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2180 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2181 dioread_nolock = ext4_should_dioread_nolock(inode);
2183 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2184 if (map->m_flags & (1 << BH_Delay))
2185 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2187 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2190 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2191 if (!mpd->io_submit.io_end->handle &&
2192 ext4_handle_valid(handle)) {
2193 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2194 handle->h_rsv_handle = NULL;
2196 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2199 BUG_ON(map->m_len == 0);
2200 if (map->m_flags & EXT4_MAP_NEW) {
2201 struct block_device *bdev = inode->i_sb->s_bdev;
2204 for (i = 0; i < map->m_len; i++)
2205 unmap_underlying_metadata(bdev, map->m_pblk + i);
2211 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2212 * mpd->len and submit pages underlying it for IO
2214 * @handle - handle for journal operations
2215 * @mpd - extent to map
2216 * @give_up_on_write - we set this to true iff there is a fatal error and there
2217 * is no hope of writing the data. The caller should discard
2218 * dirty pages to avoid infinite loops.
2220 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2221 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2222 * them to initialized or split the described range from larger unwritten
2223 * extent. Note that we need not map all the described range since allocation
2224 * can return less blocks or the range is covered by more unwritten extents. We
2225 * cannot map more because we are limited by reserved transaction credits. On
2226 * the other hand we always make sure that the last touched page is fully
2227 * mapped so that it can be written out (and thus forward progress is
2228 * guaranteed). After mapping we submit all mapped pages for IO.
2230 static int mpage_map_and_submit_extent(handle_t *handle,
2231 struct mpage_da_data *mpd,
2232 bool *give_up_on_write)
2234 struct inode *inode = mpd->inode;
2235 struct ext4_map_blocks *map = &mpd->map;
2240 mpd->io_submit.io_end->offset =
2241 ((loff_t)map->m_lblk) << inode->i_blkbits;
2243 err = mpage_map_one_extent(handle, mpd);
2245 struct super_block *sb = inode->i_sb;
2247 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2248 goto invalidate_dirty_pages;
2250 * Let the uper layers retry transient errors.
2251 * In the case of ENOSPC, if ext4_count_free_blocks()
2252 * is non-zero, a commit should free up blocks.
2254 if ((err == -ENOMEM) ||
2255 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2257 goto update_disksize;
2260 ext4_msg(sb, KERN_CRIT,
2261 "Delayed block allocation failed for "
2262 "inode %lu at logical offset %llu with"
2263 " max blocks %u with error %d",
2265 (unsigned long long)map->m_lblk,
2266 (unsigned)map->m_len, -err);
2267 ext4_msg(sb, KERN_CRIT,
2268 "This should not happen!! Data will "
2271 ext4_print_free_blocks(inode);
2272 invalidate_dirty_pages:
2273 *give_up_on_write = true;
2278 * Update buffer state, submit mapped pages, and get us new
2281 err = mpage_map_and_submit_buffers(mpd);
2283 goto update_disksize;
2284 } while (map->m_len);
2288 * Update on-disk size after IO is submitted. Races with
2289 * truncate are avoided by checking i_size under i_data_sem.
2291 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2292 if (disksize > EXT4_I(inode)->i_disksize) {
2296 down_write(&EXT4_I(inode)->i_data_sem);
2297 i_size = i_size_read(inode);
2298 if (disksize > i_size)
2300 if (disksize > EXT4_I(inode)->i_disksize)
2301 EXT4_I(inode)->i_disksize = disksize;
2302 err2 = ext4_mark_inode_dirty(handle, inode);
2303 up_write(&EXT4_I(inode)->i_data_sem);
2305 ext4_error(inode->i_sb,
2306 "Failed to mark inode %lu dirty",
2315 * Calculate the total number of credits to reserve for one writepages
2316 * iteration. This is called from ext4_writepages(). We map an extent of
2317 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2318 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2319 * bpp - 1 blocks in bpp different extents.
2321 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2323 int bpp = ext4_journal_blocks_per_page(inode);
2325 return ext4_meta_trans_blocks(inode,
2326 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2330 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2331 * and underlying extent to map
2333 * @mpd - where to look for pages
2335 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2336 * IO immediately. When we find a page which isn't mapped we start accumulating
2337 * extent of buffers underlying these pages that needs mapping (formed by
2338 * either delayed or unwritten buffers). We also lock the pages containing
2339 * these buffers. The extent found is returned in @mpd structure (starting at
2340 * mpd->lblk with length mpd->len blocks).
2342 * Note that this function can attach bios to one io_end structure which are
2343 * neither logically nor physically contiguous. Although it may seem as an
2344 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2345 * case as we need to track IO to all buffers underlying a page in one io_end.
2347 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2349 struct address_space *mapping = mpd->inode->i_mapping;
2350 struct pagevec pvec;
2351 unsigned int nr_pages;
2352 long left = mpd->wbc->nr_to_write;
2353 pgoff_t index = mpd->first_page;
2354 pgoff_t end = mpd->last_page;
2357 int blkbits = mpd->inode->i_blkbits;
2359 struct buffer_head *head;
2361 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2362 tag = PAGECACHE_TAG_TOWRITE;
2364 tag = PAGECACHE_TAG_DIRTY;
2366 pagevec_init(&pvec, 0);
2368 mpd->next_page = index;
2369 while (index <= end) {
2370 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2371 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2375 for (i = 0; i < nr_pages; i++) {
2376 struct page *page = pvec.pages[i];
2379 * At this point, the page may be truncated or
2380 * invalidated (changing page->mapping to NULL), or
2381 * even swizzled back from swapper_space to tmpfs file
2382 * mapping. However, page->index will not change
2383 * because we have a reference on the page.
2385 if (page->index > end)
2389 * Accumulated enough dirty pages? This doesn't apply
2390 * to WB_SYNC_ALL mode. For integrity sync we have to
2391 * keep going because someone may be concurrently
2392 * dirtying pages, and we might have synced a lot of
2393 * newly appeared dirty pages, but have not synced all
2394 * of the old dirty pages.
2396 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2399 /* If we can't merge this page, we are done. */
2400 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2405 * If the page is no longer dirty, or its mapping no
2406 * longer corresponds to inode we are writing (which
2407 * means it has been truncated or invalidated), or the
2408 * page is already under writeback and we are not doing
2409 * a data integrity writeback, skip the page
2411 if (!PageDirty(page) ||
2412 (PageWriteback(page) &&
2413 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2414 unlikely(page->mapping != mapping)) {
2419 wait_on_page_writeback(page);
2420 BUG_ON(PageWriteback(page));
2422 if (mpd->map.m_len == 0)
2423 mpd->first_page = page->index;
2424 mpd->next_page = page->index + 1;
2425 /* Add all dirty buffers to mpd */
2426 lblk = ((ext4_lblk_t)page->index) <<
2427 (PAGE_CACHE_SHIFT - blkbits);
2428 head = page_buffers(page);
2429 err = mpage_process_page_bufs(mpd, head, head, lblk);
2435 pagevec_release(&pvec);
2440 pagevec_release(&pvec);
2444 static int __writepage(struct page *page, struct writeback_control *wbc,
2447 struct address_space *mapping = data;
2448 int ret = ext4_writepage(page, wbc);
2449 mapping_set_error(mapping, ret);
2453 static int ext4_writepages(struct address_space *mapping,
2454 struct writeback_control *wbc)
2456 pgoff_t writeback_index = 0;
2457 long nr_to_write = wbc->nr_to_write;
2458 int range_whole = 0;
2460 handle_t *handle = NULL;
2461 struct mpage_da_data mpd;
2462 struct inode *inode = mapping->host;
2463 int needed_blocks, rsv_blocks = 0, ret = 0;
2464 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2466 struct blk_plug plug;
2467 bool give_up_on_write = false;
2469 trace_ext4_writepages(inode, wbc);
2472 * No pages to write? This is mainly a kludge to avoid starting
2473 * a transaction for special inodes like journal inode on last iput()
2474 * because that could violate lock ordering on umount
2476 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2477 goto out_writepages;
2479 if (ext4_should_journal_data(inode)) {
2480 struct blk_plug plug;
2482 blk_start_plug(&plug);
2483 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2484 blk_finish_plug(&plug);
2485 goto out_writepages;
2489 * If the filesystem has aborted, it is read-only, so return
2490 * right away instead of dumping stack traces later on that
2491 * will obscure the real source of the problem. We test
2492 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2493 * the latter could be true if the filesystem is mounted
2494 * read-only, and in that case, ext4_writepages should
2495 * *never* be called, so if that ever happens, we would want
2498 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2500 goto out_writepages;
2503 if (ext4_should_dioread_nolock(inode)) {
2505 * We may need to convert up to one extent per block in
2506 * the page and we may dirty the inode.
2508 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2512 * If we have inline data and arrive here, it means that
2513 * we will soon create the block for the 1st page, so
2514 * we'd better clear the inline data here.
2516 if (ext4_has_inline_data(inode)) {
2517 /* Just inode will be modified... */
2518 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2519 if (IS_ERR(handle)) {
2520 ret = PTR_ERR(handle);
2521 goto out_writepages;
2523 BUG_ON(ext4_test_inode_state(inode,
2524 EXT4_STATE_MAY_INLINE_DATA));
2525 ext4_destroy_inline_data(handle, inode);
2526 ext4_journal_stop(handle);
2529 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2532 if (wbc->range_cyclic) {
2533 writeback_index = mapping->writeback_index;
2534 if (writeback_index)
2536 mpd.first_page = writeback_index;
2539 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2540 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2545 ext4_io_submit_init(&mpd.io_submit, wbc);
2547 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2548 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2550 blk_start_plug(&plug);
2551 while (!done && mpd.first_page <= mpd.last_page) {
2552 /* For each extent of pages we use new io_end */
2553 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2554 if (!mpd.io_submit.io_end) {
2560 * We have two constraints: We find one extent to map and we
2561 * must always write out whole page (makes a difference when
2562 * blocksize < pagesize) so that we don't block on IO when we
2563 * try to write out the rest of the page. Journalled mode is
2564 * not supported by delalloc.
2566 BUG_ON(ext4_should_journal_data(inode));
2567 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2569 /* start a new transaction */
2570 handle = ext4_journal_start_with_reserve(inode,
2571 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2572 if (IS_ERR(handle)) {
2573 ret = PTR_ERR(handle);
2574 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2575 "%ld pages, ino %lu; err %d", __func__,
2576 wbc->nr_to_write, inode->i_ino, ret);
2577 /* Release allocated io_end */
2578 ext4_put_io_end(mpd.io_submit.io_end);
2582 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2583 ret = mpage_prepare_extent_to_map(&mpd);
2586 ret = mpage_map_and_submit_extent(handle, &mpd,
2590 * We scanned the whole range (or exhausted
2591 * nr_to_write), submitted what was mapped and
2592 * didn't find anything needing mapping. We are
2599 * Caution: If the handle is synchronous,
2600 * ext4_journal_stop() can wait for transaction commit
2601 * to finish which may depend on writeback of pages to
2602 * complete or on page lock to be released. In that
2603 * case, we have to wait until after after we have
2604 * submitted all the IO, released page locks we hold,
2605 * and dropped io_end reference (for extent conversion
2606 * to be able to complete) before stopping the handle.
2608 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2609 ext4_journal_stop(handle);
2612 /* Submit prepared bio */
2613 ext4_io_submit(&mpd.io_submit);
2614 /* Unlock pages we didn't use */
2615 mpage_release_unused_pages(&mpd, give_up_on_write);
2617 * Drop our io_end reference we got from init. We have
2618 * to be careful and use deferred io_end finishing if
2619 * we are still holding the transaction as we can
2620 * release the last reference to io_end which may end
2621 * up doing unwritten extent conversion.
2624 ext4_put_io_end_defer(mpd.io_submit.io_end);
2625 ext4_journal_stop(handle);
2627 ext4_put_io_end(mpd.io_submit.io_end);
2629 if (ret == -ENOSPC && sbi->s_journal) {
2631 * Commit the transaction which would
2632 * free blocks released in the transaction
2635 jbd2_journal_force_commit_nested(sbi->s_journal);
2639 /* Fatal error - ENOMEM, EIO... */
2643 blk_finish_plug(&plug);
2644 if (!ret && !cycled && wbc->nr_to_write > 0) {
2646 mpd.last_page = writeback_index - 1;
2652 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2654 * Set the writeback_index so that range_cyclic
2655 * mode will write it back later
2657 mapping->writeback_index = mpd.first_page;
2660 trace_ext4_writepages_result(inode, wbc, ret,
2661 nr_to_write - wbc->nr_to_write);
2665 static int ext4_nonda_switch(struct super_block *sb)
2667 s64 free_clusters, dirty_clusters;
2668 struct ext4_sb_info *sbi = EXT4_SB(sb);
2671 * switch to non delalloc mode if we are running low
2672 * on free block. The free block accounting via percpu
2673 * counters can get slightly wrong with percpu_counter_batch getting
2674 * accumulated on each CPU without updating global counters
2675 * Delalloc need an accurate free block accounting. So switch
2676 * to non delalloc when we are near to error range.
2679 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2681 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2683 * Start pushing delalloc when 1/2 of free blocks are dirty.
2685 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2686 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2688 if (2 * free_clusters < 3 * dirty_clusters ||
2689 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2691 * free block count is less than 150% of dirty blocks
2692 * or free blocks is less than watermark
2699 /* We always reserve for an inode update; the superblock could be there too */
2700 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2702 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2705 if (pos + len <= 0x7fffffffULL)
2708 /* We might need to update the superblock to set LARGE_FILE */
2712 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2713 loff_t pos, unsigned len, unsigned flags,
2714 struct page **pagep, void **fsdata)
2716 int ret, retries = 0;
2719 struct inode *inode = mapping->host;
2722 index = pos >> PAGE_CACHE_SHIFT;
2724 if (ext4_nonda_switch(inode->i_sb)) {
2725 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2726 return ext4_write_begin(file, mapping, pos,
2727 len, flags, pagep, fsdata);
2729 *fsdata = (void *)0;
2730 trace_ext4_da_write_begin(inode, pos, len, flags);
2732 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2733 ret = ext4_da_write_inline_data_begin(mapping, inode,
2743 * grab_cache_page_write_begin() can take a long time if the
2744 * system is thrashing due to memory pressure, or if the page
2745 * is being written back. So grab it first before we start
2746 * the transaction handle. This also allows us to allocate
2747 * the page (if needed) without using GFP_NOFS.
2750 page = grab_cache_page_write_begin(mapping, index, flags);
2756 * With delayed allocation, we don't log the i_disksize update
2757 * if there is delayed block allocation. But we still need
2758 * to journalling the i_disksize update if writes to the end
2759 * of file which has an already mapped buffer.
2762 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2763 ext4_da_write_credits(inode, pos, len));
2764 if (IS_ERR(handle)) {
2765 page_cache_release(page);
2766 return PTR_ERR(handle);
2770 if (page->mapping != mapping) {
2771 /* The page got truncated from under us */
2773 page_cache_release(page);
2774 ext4_journal_stop(handle);
2777 /* In case writeback began while the page was unlocked */
2778 wait_for_stable_page(page);
2780 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2781 ret = ext4_block_write_begin(page, pos, len,
2782 ext4_da_get_block_prep);
2784 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2788 ext4_journal_stop(handle);
2790 * block_write_begin may have instantiated a few blocks
2791 * outside i_size. Trim these off again. Don't need
2792 * i_size_read because we hold i_mutex.
2794 if (pos + len > inode->i_size)
2795 ext4_truncate_failed_write(inode);
2797 if (ret == -ENOSPC &&
2798 ext4_should_retry_alloc(inode->i_sb, &retries))
2801 page_cache_release(page);
2810 * Check if we should update i_disksize
2811 * when write to the end of file but not require block allocation
2813 static int ext4_da_should_update_i_disksize(struct page *page,
2814 unsigned long offset)
2816 struct buffer_head *bh;
2817 struct inode *inode = page->mapping->host;
2821 bh = page_buffers(page);
2822 idx = offset >> inode->i_blkbits;
2824 for (i = 0; i < idx; i++)
2825 bh = bh->b_this_page;
2827 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2832 static int ext4_da_write_end(struct file *file,
2833 struct address_space *mapping,
2834 loff_t pos, unsigned len, unsigned copied,
2835 struct page *page, void *fsdata)
2837 struct inode *inode = mapping->host;
2839 handle_t *handle = ext4_journal_current_handle();
2841 unsigned long start, end;
2842 int write_mode = (int)(unsigned long)fsdata;
2844 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2845 return ext4_write_end(file, mapping, pos,
2846 len, copied, page, fsdata);
2848 trace_ext4_da_write_end(inode, pos, len, copied);
2849 start = pos & (PAGE_CACHE_SIZE - 1);
2850 end = start + copied - 1;
2853 * generic_write_end() will run mark_inode_dirty() if i_size
2854 * changes. So let's piggyback the i_disksize mark_inode_dirty
2857 new_i_size = pos + copied;
2858 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2859 if (ext4_has_inline_data(inode) ||
2860 ext4_da_should_update_i_disksize(page, end)) {
2861 ext4_update_i_disksize(inode, new_i_size);
2862 /* We need to mark inode dirty even if
2863 * new_i_size is less that inode->i_size
2864 * bu greater than i_disksize.(hint delalloc)
2866 ext4_mark_inode_dirty(handle, inode);
2870 if (write_mode != CONVERT_INLINE_DATA &&
2871 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2872 ext4_has_inline_data(inode))
2873 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2876 ret2 = generic_write_end(file, mapping, pos, len, copied,
2882 ret2 = ext4_journal_stop(handle);
2886 return ret ? ret : copied;
2889 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2890 unsigned int length)
2893 * Drop reserved blocks
2895 BUG_ON(!PageLocked(page));
2896 if (!page_has_buffers(page))
2899 ext4_da_page_release_reservation(page, offset, length);
2902 ext4_invalidatepage(page, offset, length);
2908 * Force all delayed allocation blocks to be allocated for a given inode.
2910 int ext4_alloc_da_blocks(struct inode *inode)
2912 trace_ext4_alloc_da_blocks(inode);
2914 if (!EXT4_I(inode)->i_reserved_data_blocks)
2918 * We do something simple for now. The filemap_flush() will
2919 * also start triggering a write of the data blocks, which is
2920 * not strictly speaking necessary (and for users of
2921 * laptop_mode, not even desirable). However, to do otherwise
2922 * would require replicating code paths in:
2924 * ext4_writepages() ->
2925 * write_cache_pages() ---> (via passed in callback function)
2926 * __mpage_da_writepage() -->
2927 * mpage_add_bh_to_extent()
2928 * mpage_da_map_blocks()
2930 * The problem is that write_cache_pages(), located in
2931 * mm/page-writeback.c, marks pages clean in preparation for
2932 * doing I/O, which is not desirable if we're not planning on
2935 * We could call write_cache_pages(), and then redirty all of
2936 * the pages by calling redirty_page_for_writepage() but that
2937 * would be ugly in the extreme. So instead we would need to
2938 * replicate parts of the code in the above functions,
2939 * simplifying them because we wouldn't actually intend to
2940 * write out the pages, but rather only collect contiguous
2941 * logical block extents, call the multi-block allocator, and
2942 * then update the buffer heads with the block allocations.
2944 * For now, though, we'll cheat by calling filemap_flush(),
2945 * which will map the blocks, and start the I/O, but not
2946 * actually wait for the I/O to complete.
2948 return filemap_flush(inode->i_mapping);
2952 * bmap() is special. It gets used by applications such as lilo and by
2953 * the swapper to find the on-disk block of a specific piece of data.
2955 * Naturally, this is dangerous if the block concerned is still in the
2956 * journal. If somebody makes a swapfile on an ext4 data-journaling
2957 * filesystem and enables swap, then they may get a nasty shock when the
2958 * data getting swapped to that swapfile suddenly gets overwritten by
2959 * the original zero's written out previously to the journal and
2960 * awaiting writeback in the kernel's buffer cache.
2962 * So, if we see any bmap calls here on a modified, data-journaled file,
2963 * take extra steps to flush any blocks which might be in the cache.
2965 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2967 struct inode *inode = mapping->host;
2972 * We can get here for an inline file via the FIBMAP ioctl
2974 if (ext4_has_inline_data(inode))
2977 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2978 test_opt(inode->i_sb, DELALLOC)) {
2980 * With delalloc we want to sync the file
2981 * so that we can make sure we allocate
2984 filemap_write_and_wait(mapping);
2987 if (EXT4_JOURNAL(inode) &&
2988 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2990 * This is a REALLY heavyweight approach, but the use of
2991 * bmap on dirty files is expected to be extremely rare:
2992 * only if we run lilo or swapon on a freshly made file
2993 * do we expect this to happen.
2995 * (bmap requires CAP_SYS_RAWIO so this does not
2996 * represent an unprivileged user DOS attack --- we'd be
2997 * in trouble if mortal users could trigger this path at
3000 * NB. EXT4_STATE_JDATA is not set on files other than
3001 * regular files. If somebody wants to bmap a directory
3002 * or symlink and gets confused because the buffer
3003 * hasn't yet been flushed to disk, they deserve
3004 * everything they get.
3007 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3008 journal = EXT4_JOURNAL(inode);
3009 jbd2_journal_lock_updates(journal);
3010 err = jbd2_journal_flush(journal);
3011 jbd2_journal_unlock_updates(journal);
3017 return generic_block_bmap(mapping, block, ext4_get_block);
3020 static int ext4_readpage(struct file *file, struct page *page)
3023 struct inode *inode = page->mapping->host;
3025 trace_ext4_readpage(page);
3027 if (ext4_has_inline_data(inode))
3028 ret = ext4_readpage_inline(inode, page);
3031 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3037 ext4_readpages(struct file *file, struct address_space *mapping,
3038 struct list_head *pages, unsigned nr_pages)
3040 struct inode *inode = mapping->host;
3042 /* If the file has inline data, no need to do readpages. */
3043 if (ext4_has_inline_data(inode))
3046 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3049 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3050 unsigned int length)
3052 trace_ext4_invalidatepage(page, offset, length);
3054 /* No journalling happens on data buffers when this function is used */
3055 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3057 block_invalidatepage(page, offset, length);
3060 static int __ext4_journalled_invalidatepage(struct page *page,
3061 unsigned int offset,
3062 unsigned int length)
3064 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3066 trace_ext4_journalled_invalidatepage(page, offset, length);
3069 * If it's a full truncate we just forget about the pending dirtying
3071 if (offset == 0 && length == PAGE_CACHE_SIZE)
3072 ClearPageChecked(page);
3074 return jbd2_journal_invalidatepage(journal, page, offset, length);
3077 /* Wrapper for aops... */
3078 static void ext4_journalled_invalidatepage(struct page *page,
3079 unsigned int offset,
3080 unsigned int length)
3082 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3085 static int ext4_releasepage(struct page *page, gfp_t wait)
3087 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3089 trace_ext4_releasepage(page);
3091 /* Page has dirty journalled data -> cannot release */
3092 if (PageChecked(page))
3095 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3097 return try_to_free_buffers(page);
3101 * ext4_get_block used when preparing for a DIO write or buffer write.
3102 * We allocate an uinitialized extent if blocks haven't been allocated.
3103 * The extent will be converted to initialized after the IO is complete.
3105 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3106 struct buffer_head *bh_result, int create)
3108 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3109 inode->i_ino, create);
3110 return _ext4_get_block(inode, iblock, bh_result,
3111 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3114 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3115 struct buffer_head *bh_result, int create)
3117 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3118 inode->i_ino, create);
3119 return _ext4_get_block(inode, iblock, bh_result,
3120 EXT4_GET_BLOCKS_NO_LOCK);
3123 int ext4_get_block_dax(struct inode *inode, sector_t iblock,
3124 struct buffer_head *bh_result, int create)
3126 int flags = EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_UNWRIT_EXT;
3128 flags |= EXT4_GET_BLOCKS_CREATE;
3129 ext4_debug("ext4_get_block_dax: inode %lu, create flag %d\n",
3130 inode->i_ino, create);
3131 return _ext4_get_block(inode, iblock, bh_result, flags);
3134 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3135 ssize_t size, void *private)
3137 ext4_io_end_t *io_end = iocb->private;
3139 /* if not async direct IO just return */
3143 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3144 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3145 iocb->private, io_end->inode->i_ino, iocb, offset,
3148 iocb->private = NULL;
3149 io_end->offset = offset;
3150 io_end->size = size;
3151 ext4_put_io_end(io_end);
3155 * For ext4 extent files, ext4 will do direct-io write to holes,
3156 * preallocated extents, and those write extend the file, no need to
3157 * fall back to buffered IO.
3159 * For holes, we fallocate those blocks, mark them as unwritten
3160 * If those blocks were preallocated, we mark sure they are split, but
3161 * still keep the range to write as unwritten.
3163 * The unwritten extents will be converted to written when DIO is completed.
3164 * For async direct IO, since the IO may still pending when return, we
3165 * set up an end_io call back function, which will do the conversion
3166 * when async direct IO completed.
3168 * If the O_DIRECT write will extend the file then add this inode to the
3169 * orphan list. So recovery will truncate it back to the original size
3170 * if the machine crashes during the write.
3173 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3176 struct file *file = iocb->ki_filp;
3177 struct inode *inode = file->f_mapping->host;
3179 size_t count = iov_iter_count(iter);
3181 get_block_t *get_block_func = NULL;
3183 loff_t final_size = offset + count;
3184 ext4_io_end_t *io_end = NULL;
3186 /* Use the old path for reads and writes beyond i_size. */
3187 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3188 return ext4_ind_direct_IO(iocb, iter, offset);
3190 BUG_ON(iocb->private == NULL);
3193 * Make all waiters for direct IO properly wait also for extent
3194 * conversion. This also disallows race between truncate() and
3195 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3197 if (iov_iter_rw(iter) == WRITE)
3198 inode_dio_begin(inode);
3200 /* If we do a overwrite dio, i_mutex locking can be released */
3201 overwrite = *((int *)iocb->private);
3204 down_read(&EXT4_I(inode)->i_data_sem);
3205 mutex_unlock(&inode->i_mutex);
3209 * We could direct write to holes and fallocate.
3211 * Allocated blocks to fill the hole are marked as
3212 * unwritten to prevent parallel buffered read to expose
3213 * the stale data before DIO complete the data IO.
3215 * As to previously fallocated extents, ext4 get_block will
3216 * just simply mark the buffer mapped but still keep the
3217 * extents unwritten.
3219 * For non AIO case, we will convert those unwritten extents
3220 * to written after return back from blockdev_direct_IO.
3222 * For async DIO, the conversion needs to be deferred when the
3223 * IO is completed. The ext4 end_io callback function will be
3224 * called to take care of the conversion work. Here for async
3225 * case, we allocate an io_end structure to hook to the iocb.
3227 iocb->private = NULL;
3228 ext4_inode_aio_set(inode, NULL);
3229 if (!is_sync_kiocb(iocb)) {
3230 io_end = ext4_init_io_end(inode, GFP_NOFS);
3236 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3238 iocb->private = ext4_get_io_end(io_end);
3240 * we save the io structure for current async direct
3241 * IO, so that later ext4_map_blocks() could flag the
3242 * io structure whether there is a unwritten extents
3243 * needs to be converted when IO is completed.
3245 ext4_inode_aio_set(inode, io_end);
3249 get_block_func = ext4_get_block_write_nolock;
3251 get_block_func = ext4_get_block_write;
3252 dio_flags = DIO_LOCKING;
3254 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3255 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3258 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3259 ext4_end_io_dio, dio_flags);
3261 ret = __blockdev_direct_IO(iocb, inode,
3262 inode->i_sb->s_bdev, iter, offset,
3264 ext4_end_io_dio, NULL, dio_flags);
3267 * Put our reference to io_end. This can free the io_end structure e.g.
3268 * in sync IO case or in case of error. It can even perform extent
3269 * conversion if all bios we submitted finished before we got here.
3270 * Note that in that case iocb->private can be already set to NULL
3274 ext4_inode_aio_set(inode, NULL);
3275 ext4_put_io_end(io_end);
3277 * When no IO was submitted ext4_end_io_dio() was not
3278 * called so we have to put iocb's reference.
3280 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3281 WARN_ON(iocb->private != io_end);
3282 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3283 ext4_put_io_end(io_end);
3284 iocb->private = NULL;
3287 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3288 EXT4_STATE_DIO_UNWRITTEN)) {
3291 * for non AIO case, since the IO is already
3292 * completed, we could do the conversion right here
3294 err = ext4_convert_unwritten_extents(NULL, inode,
3298 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3302 if (iov_iter_rw(iter) == WRITE)
3303 inode_dio_end(inode);
3304 /* take i_mutex locking again if we do a ovewrite dio */
3306 up_read(&EXT4_I(inode)->i_data_sem);
3307 mutex_lock(&inode->i_mutex);
3313 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3316 struct file *file = iocb->ki_filp;
3317 struct inode *inode = file->f_mapping->host;
3318 size_t count = iov_iter_count(iter);
3321 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3322 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3327 * If we are doing data journalling we don't support O_DIRECT
3329 if (ext4_should_journal_data(inode))
3332 /* Let buffer I/O handle the inline data case. */
3333 if (ext4_has_inline_data(inode))
3336 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3337 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3338 ret = ext4_ext_direct_IO(iocb, iter, offset);
3340 ret = ext4_ind_direct_IO(iocb, iter, offset);
3341 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3346 * Pages can be marked dirty completely asynchronously from ext4's journalling
3347 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3348 * much here because ->set_page_dirty is called under VFS locks. The page is
3349 * not necessarily locked.
3351 * We cannot just dirty the page and leave attached buffers clean, because the
3352 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3353 * or jbddirty because all the journalling code will explode.
3355 * So what we do is to mark the page "pending dirty" and next time writepage
3356 * is called, propagate that into the buffers appropriately.
3358 static int ext4_journalled_set_page_dirty(struct page *page)
3360 SetPageChecked(page);
3361 return __set_page_dirty_nobuffers(page);
3364 static const struct address_space_operations ext4_aops = {
3365 .readpage = ext4_readpage,
3366 .readpages = ext4_readpages,
3367 .writepage = ext4_writepage,
3368 .writepages = ext4_writepages,
3369 .write_begin = ext4_write_begin,
3370 .write_end = ext4_write_end,
3372 .invalidatepage = ext4_invalidatepage,
3373 .releasepage = ext4_releasepage,
3374 .direct_IO = ext4_direct_IO,
3375 .migratepage = buffer_migrate_page,
3376 .is_partially_uptodate = block_is_partially_uptodate,
3377 .error_remove_page = generic_error_remove_page,
3380 static const struct address_space_operations ext4_journalled_aops = {
3381 .readpage = ext4_readpage,
3382 .readpages = ext4_readpages,
3383 .writepage = ext4_writepage,
3384 .writepages = ext4_writepages,
3385 .write_begin = ext4_write_begin,
3386 .write_end = ext4_journalled_write_end,
3387 .set_page_dirty = ext4_journalled_set_page_dirty,
3389 .invalidatepage = ext4_journalled_invalidatepage,
3390 .releasepage = ext4_releasepage,
3391 .direct_IO = ext4_direct_IO,
3392 .is_partially_uptodate = block_is_partially_uptodate,
3393 .error_remove_page = generic_error_remove_page,
3396 static const struct address_space_operations ext4_da_aops = {
3397 .readpage = ext4_readpage,
3398 .readpages = ext4_readpages,
3399 .writepage = ext4_writepage,
3400 .writepages = ext4_writepages,
3401 .write_begin = ext4_da_write_begin,
3402 .write_end = ext4_da_write_end,
3404 .invalidatepage = ext4_da_invalidatepage,
3405 .releasepage = ext4_releasepage,
3406 .direct_IO = ext4_direct_IO,
3407 .migratepage = buffer_migrate_page,
3408 .is_partially_uptodate = block_is_partially_uptodate,
3409 .error_remove_page = generic_error_remove_page,
3412 void ext4_set_aops(struct inode *inode)
3414 switch (ext4_inode_journal_mode(inode)) {
3415 case EXT4_INODE_ORDERED_DATA_MODE:
3416 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3418 case EXT4_INODE_WRITEBACK_DATA_MODE:
3419 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3421 case EXT4_INODE_JOURNAL_DATA_MODE:
3422 inode->i_mapping->a_ops = &ext4_journalled_aops;
3427 if (test_opt(inode->i_sb, DELALLOC))
3428 inode->i_mapping->a_ops = &ext4_da_aops;
3430 inode->i_mapping->a_ops = &ext4_aops;
3433 static int __ext4_block_zero_page_range(handle_t *handle,
3434 struct address_space *mapping, loff_t from, loff_t length)
3436 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3437 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3438 unsigned blocksize, pos;
3440 struct inode *inode = mapping->host;
3441 struct buffer_head *bh;
3445 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3446 mapping_gfp_constraint(mapping, ~__GFP_FS));
3450 blocksize = inode->i_sb->s_blocksize;
3452 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3454 if (!page_has_buffers(page))
3455 create_empty_buffers(page, blocksize, 0);
3457 /* Find the buffer that contains "offset" */
3458 bh = page_buffers(page);
3460 while (offset >= pos) {
3461 bh = bh->b_this_page;
3465 if (buffer_freed(bh)) {
3466 BUFFER_TRACE(bh, "freed: skip");
3469 if (!buffer_mapped(bh)) {
3470 BUFFER_TRACE(bh, "unmapped");
3471 ext4_get_block(inode, iblock, bh, 0);
3472 /* unmapped? It's a hole - nothing to do */
3473 if (!buffer_mapped(bh)) {
3474 BUFFER_TRACE(bh, "still unmapped");
3479 /* Ok, it's mapped. Make sure it's up-to-date */
3480 if (PageUptodate(page))
3481 set_buffer_uptodate(bh);
3483 if (!buffer_uptodate(bh)) {
3485 ll_rw_block(READ, 1, &bh);
3487 /* Uhhuh. Read error. Complain and punt. */
3488 if (!buffer_uptodate(bh))
3490 if (S_ISREG(inode->i_mode) &&
3491 ext4_encrypted_inode(inode)) {
3492 /* We expect the key to be set. */
3493 BUG_ON(!ext4_has_encryption_key(inode));
3494 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3495 WARN_ON_ONCE(ext4_decrypt(page));
3498 if (ext4_should_journal_data(inode)) {
3499 BUFFER_TRACE(bh, "get write access");
3500 err = ext4_journal_get_write_access(handle, bh);
3504 zero_user(page, offset, length);
3505 BUFFER_TRACE(bh, "zeroed end of block");
3507 if (ext4_should_journal_data(inode)) {
3508 err = ext4_handle_dirty_metadata(handle, inode, bh);
3511 mark_buffer_dirty(bh);
3512 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3513 err = ext4_jbd2_file_inode(handle, inode);
3518 page_cache_release(page);
3523 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3524 * starting from file offset 'from'. The range to be zero'd must
3525 * be contained with in one block. If the specified range exceeds
3526 * the end of the block it will be shortened to end of the block
3527 * that cooresponds to 'from'
3529 static int ext4_block_zero_page_range(handle_t *handle,
3530 struct address_space *mapping, loff_t from, loff_t length)
3532 struct inode *inode = mapping->host;
3533 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3534 unsigned blocksize = inode->i_sb->s_blocksize;
3535 unsigned max = blocksize - (offset & (blocksize - 1));
3538 * correct length if it does not fall between
3539 * 'from' and the end of the block
3541 if (length > max || length < 0)
3545 return dax_zero_page_range(inode, from, length, ext4_get_block);
3546 return __ext4_block_zero_page_range(handle, mapping, from, length);
3550 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3551 * up to the end of the block which corresponds to `from'.
3552 * This required during truncate. We need to physically zero the tail end
3553 * of that block so it doesn't yield old data if the file is later grown.
3555 static int ext4_block_truncate_page(handle_t *handle,
3556 struct address_space *mapping, loff_t from)
3558 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3561 struct inode *inode = mapping->host;
3563 blocksize = inode->i_sb->s_blocksize;
3564 length = blocksize - (offset & (blocksize - 1));
3566 return ext4_block_zero_page_range(handle, mapping, from, length);
3569 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3570 loff_t lstart, loff_t length)
3572 struct super_block *sb = inode->i_sb;
3573 struct address_space *mapping = inode->i_mapping;
3574 unsigned partial_start, partial_end;
3575 ext4_fsblk_t start, end;
3576 loff_t byte_end = (lstart + length - 1);
3579 partial_start = lstart & (sb->s_blocksize - 1);
3580 partial_end = byte_end & (sb->s_blocksize - 1);
3582 start = lstart >> sb->s_blocksize_bits;
3583 end = byte_end >> sb->s_blocksize_bits;
3585 /* Handle partial zero within the single block */
3587 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3588 err = ext4_block_zero_page_range(handle, mapping,
3592 /* Handle partial zero out on the start of the range */
3593 if (partial_start) {
3594 err = ext4_block_zero_page_range(handle, mapping,
3595 lstart, sb->s_blocksize);
3599 /* Handle partial zero out on the end of the range */
3600 if (partial_end != sb->s_blocksize - 1)
3601 err = ext4_block_zero_page_range(handle, mapping,
3602 byte_end - partial_end,
3607 int ext4_can_truncate(struct inode *inode)
3609 if (S_ISREG(inode->i_mode))
3611 if (S_ISDIR(inode->i_mode))
3613 if (S_ISLNK(inode->i_mode))
3614 return !ext4_inode_is_fast_symlink(inode);
3619 * We have to make sure i_disksize gets properly updated before we truncate
3620 * page cache due to hole punching or zero range. Otherwise i_disksize update
3621 * can get lost as it may have been postponed to submission of writeback but
3622 * that will never happen after we truncate page cache.
3624 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3628 loff_t size = i_size_read(inode);
3630 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3631 if (offset > size || offset + len < size)
3634 if (EXT4_I(inode)->i_disksize >= size)
3637 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3639 return PTR_ERR(handle);
3640 ext4_update_i_disksize(inode, size);
3641 ext4_mark_inode_dirty(handle, inode);
3642 ext4_journal_stop(handle);
3648 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3649 * associated with the given offset and length
3651 * @inode: File inode
3652 * @offset: The offset where the hole will begin
3653 * @len: The length of the hole
3655 * Returns: 0 on success or negative on failure
3658 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3660 struct super_block *sb = inode->i_sb;
3661 ext4_lblk_t first_block, stop_block;
3662 struct address_space *mapping = inode->i_mapping;
3663 loff_t first_block_offset, last_block_offset;
3665 unsigned int credits;
3668 if (!S_ISREG(inode->i_mode))
3671 trace_ext4_punch_hole(inode, offset, length, 0);
3674 * Write out all dirty pages to avoid race conditions
3675 * Then release them.
3677 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3678 ret = filemap_write_and_wait_range(mapping, offset,
3679 offset + length - 1);
3684 mutex_lock(&inode->i_mutex);
3686 /* No need to punch hole beyond i_size */
3687 if (offset >= inode->i_size)
3691 * If the hole extends beyond i_size, set the hole
3692 * to end after the page that contains i_size
3694 if (offset + length > inode->i_size) {
3695 length = inode->i_size +
3696 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3700 if (offset & (sb->s_blocksize - 1) ||
3701 (offset + length) & (sb->s_blocksize - 1)) {
3703 * Attach jinode to inode for jbd2 if we do any zeroing of
3706 ret = ext4_inode_attach_jinode(inode);
3712 /* Wait all existing dio workers, newcomers will block on i_mutex */
3713 ext4_inode_block_unlocked_dio(inode);
3714 inode_dio_wait(inode);
3717 * Prevent page faults from reinstantiating pages we have released from
3720 down_write(&EXT4_I(inode)->i_mmap_sem);
3721 first_block_offset = round_up(offset, sb->s_blocksize);
3722 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3724 /* Now release the pages and zero block aligned part of pages*/
3725 if (last_block_offset > first_block_offset) {
3726 ret = ext4_update_disksize_before_punch(inode, offset, length);
3729 truncate_pagecache_range(inode, first_block_offset,
3733 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3734 credits = ext4_writepage_trans_blocks(inode);
3736 credits = ext4_blocks_for_truncate(inode);
3737 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3738 if (IS_ERR(handle)) {
3739 ret = PTR_ERR(handle);
3740 ext4_std_error(sb, ret);
3744 ret = ext4_zero_partial_blocks(handle, inode, offset,
3749 first_block = (offset + sb->s_blocksize - 1) >>
3750 EXT4_BLOCK_SIZE_BITS(sb);
3751 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3753 /* If there are no blocks to remove, return now */
3754 if (first_block >= stop_block)
3757 down_write(&EXT4_I(inode)->i_data_sem);
3758 ext4_discard_preallocations(inode);
3760 ret = ext4_es_remove_extent(inode, first_block,
3761 stop_block - first_block);
3763 up_write(&EXT4_I(inode)->i_data_sem);
3767 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3768 ret = ext4_ext_remove_space(inode, first_block,
3771 ret = ext4_ind_remove_space(handle, inode, first_block,
3774 up_write(&EXT4_I(inode)->i_data_sem);
3776 ext4_handle_sync(handle);
3778 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3779 ext4_mark_inode_dirty(handle, inode);
3781 ext4_journal_stop(handle);
3783 up_write(&EXT4_I(inode)->i_mmap_sem);
3784 ext4_inode_resume_unlocked_dio(inode);
3786 mutex_unlock(&inode->i_mutex);
3790 int ext4_inode_attach_jinode(struct inode *inode)
3792 struct ext4_inode_info *ei = EXT4_I(inode);
3793 struct jbd2_inode *jinode;
3795 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3798 jinode = jbd2_alloc_inode(GFP_KERNEL);
3799 spin_lock(&inode->i_lock);
3802 spin_unlock(&inode->i_lock);
3805 ei->jinode = jinode;
3806 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3809 spin_unlock(&inode->i_lock);
3810 if (unlikely(jinode != NULL))
3811 jbd2_free_inode(jinode);
3818 * We block out ext4_get_block() block instantiations across the entire
3819 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3820 * simultaneously on behalf of the same inode.
3822 * As we work through the truncate and commit bits of it to the journal there
3823 * is one core, guiding principle: the file's tree must always be consistent on
3824 * disk. We must be able to restart the truncate after a crash.
3826 * The file's tree may be transiently inconsistent in memory (although it
3827 * probably isn't), but whenever we close off and commit a journal transaction,
3828 * the contents of (the filesystem + the journal) must be consistent and
3829 * restartable. It's pretty simple, really: bottom up, right to left (although
3830 * left-to-right works OK too).
3832 * Note that at recovery time, journal replay occurs *before* the restart of
3833 * truncate against the orphan inode list.
3835 * The committed inode has the new, desired i_size (which is the same as
3836 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3837 * that this inode's truncate did not complete and it will again call
3838 * ext4_truncate() to have another go. So there will be instantiated blocks
3839 * to the right of the truncation point in a crashed ext4 filesystem. But
3840 * that's fine - as long as they are linked from the inode, the post-crash
3841 * ext4_truncate() run will find them and release them.
3843 void ext4_truncate(struct inode *inode)
3845 struct ext4_inode_info *ei = EXT4_I(inode);
3846 unsigned int credits;
3848 struct address_space *mapping = inode->i_mapping;
3851 * There is a possibility that we're either freeing the inode
3852 * or it's a completely new inode. In those cases we might not
3853 * have i_mutex locked because it's not necessary.
3855 if (!(inode->i_state & (I_NEW|I_FREEING)))
3856 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3857 trace_ext4_truncate_enter(inode);
3859 if (!ext4_can_truncate(inode))
3862 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3864 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3865 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3867 if (ext4_has_inline_data(inode)) {
3870 ext4_inline_data_truncate(inode, &has_inline);
3875 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3876 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3877 if (ext4_inode_attach_jinode(inode) < 0)
3881 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3882 credits = ext4_writepage_trans_blocks(inode);
3884 credits = ext4_blocks_for_truncate(inode);
3886 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3887 if (IS_ERR(handle)) {
3888 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3892 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3893 ext4_block_truncate_page(handle, mapping, inode->i_size);
3896 * We add the inode to the orphan list, so that if this
3897 * truncate spans multiple transactions, and we crash, we will
3898 * resume the truncate when the filesystem recovers. It also
3899 * marks the inode dirty, to catch the new size.
3901 * Implication: the file must always be in a sane, consistent
3902 * truncatable state while each transaction commits.
3904 if (ext4_orphan_add(handle, inode))
3907 down_write(&EXT4_I(inode)->i_data_sem);
3909 ext4_discard_preallocations(inode);
3911 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3912 ext4_ext_truncate(handle, inode);
3914 ext4_ind_truncate(handle, inode);
3916 up_write(&ei->i_data_sem);
3919 ext4_handle_sync(handle);
3923 * If this was a simple ftruncate() and the file will remain alive,
3924 * then we need to clear up the orphan record which we created above.
3925 * However, if this was a real unlink then we were called by
3926 * ext4_evict_inode(), and we allow that function to clean up the
3927 * orphan info for us.
3930 ext4_orphan_del(handle, inode);
3932 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3933 ext4_mark_inode_dirty(handle, inode);
3934 ext4_journal_stop(handle);
3936 trace_ext4_truncate_exit(inode);
3940 * ext4_get_inode_loc returns with an extra refcount against the inode's
3941 * underlying buffer_head on success. If 'in_mem' is true, we have all
3942 * data in memory that is needed to recreate the on-disk version of this
3945 static int __ext4_get_inode_loc(struct inode *inode,
3946 struct ext4_iloc *iloc, int in_mem)
3948 struct ext4_group_desc *gdp;
3949 struct buffer_head *bh;
3950 struct super_block *sb = inode->i_sb;
3952 int inodes_per_block, inode_offset;
3955 if (!ext4_valid_inum(sb, inode->i_ino))
3956 return -EFSCORRUPTED;
3958 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3959 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3964 * Figure out the offset within the block group inode table
3966 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3967 inode_offset = ((inode->i_ino - 1) %
3968 EXT4_INODES_PER_GROUP(sb));
3969 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3970 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3972 bh = sb_getblk(sb, block);
3975 if (!buffer_uptodate(bh)) {
3979 * If the buffer has the write error flag, we have failed
3980 * to write out another inode in the same block. In this
3981 * case, we don't have to read the block because we may
3982 * read the old inode data successfully.
3984 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3985 set_buffer_uptodate(bh);
3987 if (buffer_uptodate(bh)) {
3988 /* someone brought it uptodate while we waited */
3994 * If we have all information of the inode in memory and this
3995 * is the only valid inode in the block, we need not read the
3999 struct buffer_head *bitmap_bh;
4002 start = inode_offset & ~(inodes_per_block - 1);
4004 /* Is the inode bitmap in cache? */
4005 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4006 if (unlikely(!bitmap_bh))
4010 * If the inode bitmap isn't in cache then the
4011 * optimisation may end up performing two reads instead
4012 * of one, so skip it.
4014 if (!buffer_uptodate(bitmap_bh)) {
4018 for (i = start; i < start + inodes_per_block; i++) {
4019 if (i == inode_offset)
4021 if (ext4_test_bit(i, bitmap_bh->b_data))
4025 if (i == start + inodes_per_block) {
4026 /* all other inodes are free, so skip I/O */
4027 memset(bh->b_data, 0, bh->b_size);
4028 set_buffer_uptodate(bh);
4036 * If we need to do any I/O, try to pre-readahead extra
4037 * blocks from the inode table.
4039 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4040 ext4_fsblk_t b, end, table;
4042 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4044 table = ext4_inode_table(sb, gdp);
4045 /* s_inode_readahead_blks is always a power of 2 */
4046 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4050 num = EXT4_INODES_PER_GROUP(sb);
4051 if (ext4_has_group_desc_csum(sb))
4052 num -= ext4_itable_unused_count(sb, gdp);
4053 table += num / inodes_per_block;
4057 sb_breadahead(sb, b++);
4061 * There are other valid inodes in the buffer, this inode
4062 * has in-inode xattrs, or we don't have this inode in memory.
4063 * Read the block from disk.
4065 trace_ext4_load_inode(inode);
4067 bh->b_end_io = end_buffer_read_sync;
4068 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4070 if (!buffer_uptodate(bh)) {
4071 EXT4_ERROR_INODE_BLOCK(inode, block,
4072 "unable to read itable block");
4082 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4084 /* We have all inode data except xattrs in memory here. */
4085 return __ext4_get_inode_loc(inode, iloc,
4086 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4089 void ext4_set_inode_flags(struct inode *inode)
4091 unsigned int flags = EXT4_I(inode)->i_flags;
4092 unsigned int new_fl = 0;
4094 if (flags & EXT4_SYNC_FL)
4096 if (flags & EXT4_APPEND_FL)
4098 if (flags & EXT4_IMMUTABLE_FL)
4099 new_fl |= S_IMMUTABLE;
4100 if (flags & EXT4_NOATIME_FL)
4101 new_fl |= S_NOATIME;
4102 if (flags & EXT4_DIRSYNC_FL)
4103 new_fl |= S_DIRSYNC;
4104 if (test_opt(inode->i_sb, DAX))
4106 inode_set_flags(inode, new_fl,
4107 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4110 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4111 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4113 unsigned int vfs_fl;
4114 unsigned long old_fl, new_fl;
4117 vfs_fl = ei->vfs_inode.i_flags;
4118 old_fl = ei->i_flags;
4119 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4120 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4122 if (vfs_fl & S_SYNC)
4123 new_fl |= EXT4_SYNC_FL;
4124 if (vfs_fl & S_APPEND)
4125 new_fl |= EXT4_APPEND_FL;
4126 if (vfs_fl & S_IMMUTABLE)
4127 new_fl |= EXT4_IMMUTABLE_FL;
4128 if (vfs_fl & S_NOATIME)
4129 new_fl |= EXT4_NOATIME_FL;
4130 if (vfs_fl & S_DIRSYNC)
4131 new_fl |= EXT4_DIRSYNC_FL;
4132 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4135 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4136 struct ext4_inode_info *ei)
4139 struct inode *inode = &(ei->vfs_inode);
4140 struct super_block *sb = inode->i_sb;
4142 if (ext4_has_feature_huge_file(sb)) {
4143 /* we are using combined 48 bit field */
4144 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4145 le32_to_cpu(raw_inode->i_blocks_lo);
4146 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4147 /* i_blocks represent file system block size */
4148 return i_blocks << (inode->i_blkbits - 9);
4153 return le32_to_cpu(raw_inode->i_blocks_lo);
4157 static inline void ext4_iget_extra_inode(struct inode *inode,
4158 struct ext4_inode *raw_inode,
4159 struct ext4_inode_info *ei)
4161 __le32 *magic = (void *)raw_inode +
4162 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4163 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4164 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4165 ext4_find_inline_data_nolock(inode);
4167 EXT4_I(inode)->i_inline_off = 0;
4170 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4172 struct ext4_iloc iloc;
4173 struct ext4_inode *raw_inode;
4174 struct ext4_inode_info *ei;
4175 struct inode *inode;
4176 journal_t *journal = EXT4_SB(sb)->s_journal;
4182 inode = iget_locked(sb, ino);
4184 return ERR_PTR(-ENOMEM);
4185 if (!(inode->i_state & I_NEW))
4191 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4194 raw_inode = ext4_raw_inode(&iloc);
4196 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4197 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4198 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4199 EXT4_INODE_SIZE(inode->i_sb)) {
4200 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4201 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4202 EXT4_INODE_SIZE(inode->i_sb));
4203 ret = -EFSCORRUPTED;
4207 ei->i_extra_isize = 0;
4209 /* Precompute checksum seed for inode metadata */
4210 if (ext4_has_metadata_csum(sb)) {
4211 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4213 __le32 inum = cpu_to_le32(inode->i_ino);
4214 __le32 gen = raw_inode->i_generation;
4215 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4217 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4221 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4222 EXT4_ERROR_INODE(inode, "checksum invalid");
4227 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4228 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4229 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4230 if (!(test_opt(inode->i_sb, NO_UID32))) {
4231 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4232 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4234 i_uid_write(inode, i_uid);
4235 i_gid_write(inode, i_gid);
4236 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4238 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4239 ei->i_inline_off = 0;
4240 ei->i_dir_start_lookup = 0;
4241 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4242 /* We now have enough fields to check if the inode was active or not.
4243 * This is needed because nfsd might try to access dead inodes
4244 * the test is that same one that e2fsck uses
4245 * NeilBrown 1999oct15
4247 if (inode->i_nlink == 0) {
4248 if ((inode->i_mode == 0 ||
4249 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4250 ino != EXT4_BOOT_LOADER_INO) {
4251 /* this inode is deleted */
4255 /* The only unlinked inodes we let through here have
4256 * valid i_mode and are being read by the orphan
4257 * recovery code: that's fine, we're about to complete
4258 * the process of deleting those.
4259 * OR it is the EXT4_BOOT_LOADER_INO which is
4260 * not initialized on a new filesystem. */
4262 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4263 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4264 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4265 if (ext4_has_feature_64bit(sb))
4267 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4268 inode->i_size = ext4_isize(raw_inode);
4269 ei->i_disksize = inode->i_size;
4271 ei->i_reserved_quota = 0;
4273 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4274 ei->i_block_group = iloc.block_group;
4275 ei->i_last_alloc_group = ~0;
4277 * NOTE! The in-memory inode i_data array is in little-endian order
4278 * even on big-endian machines: we do NOT byteswap the block numbers!
4280 for (block = 0; block < EXT4_N_BLOCKS; block++)
4281 ei->i_data[block] = raw_inode->i_block[block];
4282 INIT_LIST_HEAD(&ei->i_orphan);
4285 * Set transaction id's of transactions that have to be committed
4286 * to finish f[data]sync. We set them to currently running transaction
4287 * as we cannot be sure that the inode or some of its metadata isn't
4288 * part of the transaction - the inode could have been reclaimed and
4289 * now it is reread from disk.
4292 transaction_t *transaction;
4295 read_lock(&journal->j_state_lock);
4296 if (journal->j_running_transaction)
4297 transaction = journal->j_running_transaction;
4299 transaction = journal->j_committing_transaction;
4301 tid = transaction->t_tid;
4303 tid = journal->j_commit_sequence;
4304 read_unlock(&journal->j_state_lock);
4305 ei->i_sync_tid = tid;
4306 ei->i_datasync_tid = tid;
4309 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4310 if (ei->i_extra_isize == 0) {
4311 /* The extra space is currently unused. Use it. */
4312 ei->i_extra_isize = sizeof(struct ext4_inode) -
4313 EXT4_GOOD_OLD_INODE_SIZE;
4315 ext4_iget_extra_inode(inode, raw_inode, ei);
4319 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4320 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4321 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4322 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4324 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4325 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4326 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4327 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4329 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4334 if (ei->i_file_acl &&
4335 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4336 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4338 ret = -EFSCORRUPTED;
4340 } else if (!ext4_has_inline_data(inode)) {
4341 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4342 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4343 (S_ISLNK(inode->i_mode) &&
4344 !ext4_inode_is_fast_symlink(inode))))
4345 /* Validate extent which is part of inode */
4346 ret = ext4_ext_check_inode(inode);
4347 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4348 (S_ISLNK(inode->i_mode) &&
4349 !ext4_inode_is_fast_symlink(inode))) {
4350 /* Validate block references which are part of inode */
4351 ret = ext4_ind_check_inode(inode);
4357 if (S_ISREG(inode->i_mode)) {
4358 inode->i_op = &ext4_file_inode_operations;
4359 inode->i_fop = &ext4_file_operations;
4360 ext4_set_aops(inode);
4361 } else if (S_ISDIR(inode->i_mode)) {
4362 inode->i_op = &ext4_dir_inode_operations;
4363 inode->i_fop = &ext4_dir_operations;
4364 } else if (S_ISLNK(inode->i_mode)) {
4365 if (ext4_encrypted_inode(inode)) {
4366 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4367 ext4_set_aops(inode);
4368 } else if (ext4_inode_is_fast_symlink(inode)) {
4369 inode->i_link = (char *)ei->i_data;
4370 inode->i_op = &ext4_fast_symlink_inode_operations;
4371 nd_terminate_link(ei->i_data, inode->i_size,
4372 sizeof(ei->i_data) - 1);
4374 inode->i_op = &ext4_symlink_inode_operations;
4375 ext4_set_aops(inode);
4377 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4378 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4379 inode->i_op = &ext4_special_inode_operations;
4380 if (raw_inode->i_block[0])
4381 init_special_inode(inode, inode->i_mode,
4382 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4384 init_special_inode(inode, inode->i_mode,
4385 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4386 } else if (ino == EXT4_BOOT_LOADER_INO) {
4387 make_bad_inode(inode);
4389 ret = -EFSCORRUPTED;
4390 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4394 ext4_set_inode_flags(inode);
4395 unlock_new_inode(inode);
4401 return ERR_PTR(ret);
4404 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4406 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4407 return ERR_PTR(-EFSCORRUPTED);
4408 return ext4_iget(sb, ino);
4411 static int ext4_inode_blocks_set(handle_t *handle,
4412 struct ext4_inode *raw_inode,
4413 struct ext4_inode_info *ei)
4415 struct inode *inode = &(ei->vfs_inode);
4416 u64 i_blocks = inode->i_blocks;
4417 struct super_block *sb = inode->i_sb;
4419 if (i_blocks <= ~0U) {
4421 * i_blocks can be represented in a 32 bit variable
4422 * as multiple of 512 bytes
4424 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4425 raw_inode->i_blocks_high = 0;
4426 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4429 if (!ext4_has_feature_huge_file(sb))
4432 if (i_blocks <= 0xffffffffffffULL) {
4434 * i_blocks can be represented in a 48 bit variable
4435 * as multiple of 512 bytes
4437 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4438 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4439 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4441 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4442 /* i_block is stored in file system block size */
4443 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4444 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4445 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4450 struct other_inode {
4451 unsigned long orig_ino;
4452 struct ext4_inode *raw_inode;
4455 static int other_inode_match(struct inode * inode, unsigned long ino,
4458 struct other_inode *oi = (struct other_inode *) data;
4460 if ((inode->i_ino != ino) ||
4461 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4462 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4463 ((inode->i_state & I_DIRTY_TIME) == 0))
4465 spin_lock(&inode->i_lock);
4466 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4467 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4468 (inode->i_state & I_DIRTY_TIME)) {
4469 struct ext4_inode_info *ei = EXT4_I(inode);
4471 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4472 spin_unlock(&inode->i_lock);
4474 spin_lock(&ei->i_raw_lock);
4475 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4476 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4477 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4478 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4479 spin_unlock(&ei->i_raw_lock);
4480 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4483 spin_unlock(&inode->i_lock);
4488 * Opportunistically update the other time fields for other inodes in
4489 * the same inode table block.
4491 static void ext4_update_other_inodes_time(struct super_block *sb,
4492 unsigned long orig_ino, char *buf)
4494 struct other_inode oi;
4496 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4497 int inode_size = EXT4_INODE_SIZE(sb);
4499 oi.orig_ino = orig_ino;
4501 * Calculate the first inode in the inode table block. Inode
4502 * numbers are one-based. That is, the first inode in a block
4503 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4505 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4506 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4507 if (ino == orig_ino)
4509 oi.raw_inode = (struct ext4_inode *) buf;
4510 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4515 * Post the struct inode info into an on-disk inode location in the
4516 * buffer-cache. This gobbles the caller's reference to the
4517 * buffer_head in the inode location struct.
4519 * The caller must have write access to iloc->bh.
4521 static int ext4_do_update_inode(handle_t *handle,
4522 struct inode *inode,
4523 struct ext4_iloc *iloc)
4525 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4526 struct ext4_inode_info *ei = EXT4_I(inode);
4527 struct buffer_head *bh = iloc->bh;
4528 struct super_block *sb = inode->i_sb;
4529 int err = 0, rc, block;
4530 int need_datasync = 0, set_large_file = 0;
4534 spin_lock(&ei->i_raw_lock);
4536 /* For fields not tracked in the in-memory inode,
4537 * initialise them to zero for new inodes. */
4538 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4539 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4541 ext4_get_inode_flags(ei);
4542 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4543 i_uid = i_uid_read(inode);
4544 i_gid = i_gid_read(inode);
4545 if (!(test_opt(inode->i_sb, NO_UID32))) {
4546 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4547 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4549 * Fix up interoperability with old kernels. Otherwise, old inodes get
4550 * re-used with the upper 16 bits of the uid/gid intact
4552 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4553 raw_inode->i_uid_high = 0;
4554 raw_inode->i_gid_high = 0;
4556 raw_inode->i_uid_high =
4557 cpu_to_le16(high_16_bits(i_uid));
4558 raw_inode->i_gid_high =
4559 cpu_to_le16(high_16_bits(i_gid));
4562 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4563 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4564 raw_inode->i_uid_high = 0;
4565 raw_inode->i_gid_high = 0;
4567 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4569 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4570 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4571 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4572 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4574 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4576 spin_unlock(&ei->i_raw_lock);
4579 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4580 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4581 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4582 raw_inode->i_file_acl_high =
4583 cpu_to_le16(ei->i_file_acl >> 32);
4584 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4585 if (ei->i_disksize != ext4_isize(raw_inode)) {
4586 ext4_isize_set(raw_inode, ei->i_disksize);
4589 if (ei->i_disksize > 0x7fffffffULL) {
4590 if (!ext4_has_feature_large_file(sb) ||
4591 EXT4_SB(sb)->s_es->s_rev_level ==
4592 cpu_to_le32(EXT4_GOOD_OLD_REV))
4595 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4596 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4597 if (old_valid_dev(inode->i_rdev)) {
4598 raw_inode->i_block[0] =
4599 cpu_to_le32(old_encode_dev(inode->i_rdev));
4600 raw_inode->i_block[1] = 0;
4602 raw_inode->i_block[0] = 0;
4603 raw_inode->i_block[1] =
4604 cpu_to_le32(new_encode_dev(inode->i_rdev));
4605 raw_inode->i_block[2] = 0;
4607 } else if (!ext4_has_inline_data(inode)) {
4608 for (block = 0; block < EXT4_N_BLOCKS; block++)
4609 raw_inode->i_block[block] = ei->i_data[block];
4612 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4613 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4614 if (ei->i_extra_isize) {
4615 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4616 raw_inode->i_version_hi =
4617 cpu_to_le32(inode->i_version >> 32);
4618 raw_inode->i_extra_isize =
4619 cpu_to_le16(ei->i_extra_isize);
4622 ext4_inode_csum_set(inode, raw_inode, ei);
4623 spin_unlock(&ei->i_raw_lock);
4624 if (inode->i_sb->s_flags & MS_LAZYTIME)
4625 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4628 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4629 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4632 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4633 if (set_large_file) {
4634 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4635 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4638 ext4_update_dynamic_rev(sb);
4639 ext4_set_feature_large_file(sb);
4640 ext4_handle_sync(handle);
4641 err = ext4_handle_dirty_super(handle, sb);
4643 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4646 ext4_std_error(inode->i_sb, err);
4651 * ext4_write_inode()
4653 * We are called from a few places:
4655 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4656 * Here, there will be no transaction running. We wait for any running
4657 * transaction to commit.
4659 * - Within flush work (sys_sync(), kupdate and such).
4660 * We wait on commit, if told to.
4662 * - Within iput_final() -> write_inode_now()
4663 * We wait on commit, if told to.
4665 * In all cases it is actually safe for us to return without doing anything,
4666 * because the inode has been copied into a raw inode buffer in
4667 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4670 * Note that we are absolutely dependent upon all inode dirtiers doing the
4671 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4672 * which we are interested.
4674 * It would be a bug for them to not do this. The code:
4676 * mark_inode_dirty(inode)
4678 * inode->i_size = expr;
4680 * is in error because write_inode() could occur while `stuff()' is running,
4681 * and the new i_size will be lost. Plus the inode will no longer be on the
4682 * superblock's dirty inode list.
4684 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4688 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4691 if (EXT4_SB(inode->i_sb)->s_journal) {
4692 if (ext4_journal_current_handle()) {
4693 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4699 * No need to force transaction in WB_SYNC_NONE mode. Also
4700 * ext4_sync_fs() will force the commit after everything is
4703 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4706 err = ext4_force_commit(inode->i_sb);
4708 struct ext4_iloc iloc;
4710 err = __ext4_get_inode_loc(inode, &iloc, 0);
4714 * sync(2) will flush the whole buffer cache. No need to do
4715 * it here separately for each inode.
4717 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4718 sync_dirty_buffer(iloc.bh);
4719 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4720 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4721 "IO error syncing inode");
4730 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4731 * buffers that are attached to a page stradding i_size and are undergoing
4732 * commit. In that case we have to wait for commit to finish and try again.
4734 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4738 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4739 tid_t commit_tid = 0;
4742 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4744 * All buffers in the last page remain valid? Then there's nothing to
4745 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4748 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4751 page = find_lock_page(inode->i_mapping,
4752 inode->i_size >> PAGE_CACHE_SHIFT);
4755 ret = __ext4_journalled_invalidatepage(page, offset,
4756 PAGE_CACHE_SIZE - offset);
4758 page_cache_release(page);
4762 read_lock(&journal->j_state_lock);
4763 if (journal->j_committing_transaction)
4764 commit_tid = journal->j_committing_transaction->t_tid;
4765 read_unlock(&journal->j_state_lock);
4767 jbd2_log_wait_commit(journal, commit_tid);
4774 * Called from notify_change.
4776 * We want to trap VFS attempts to truncate the file as soon as
4777 * possible. In particular, we want to make sure that when the VFS
4778 * shrinks i_size, we put the inode on the orphan list and modify
4779 * i_disksize immediately, so that during the subsequent flushing of
4780 * dirty pages and freeing of disk blocks, we can guarantee that any
4781 * commit will leave the blocks being flushed in an unused state on
4782 * disk. (On recovery, the inode will get truncated and the blocks will
4783 * be freed, so we have a strong guarantee that no future commit will
4784 * leave these blocks visible to the user.)
4786 * Another thing we have to assure is that if we are in ordered mode
4787 * and inode is still attached to the committing transaction, we must
4788 * we start writeout of all the dirty pages which are being truncated.
4789 * This way we are sure that all the data written in the previous
4790 * transaction are already on disk (truncate waits for pages under
4793 * Called with inode->i_mutex down.
4795 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4797 struct inode *inode = d_inode(dentry);
4800 const unsigned int ia_valid = attr->ia_valid;
4802 error = inode_change_ok(inode, attr);
4806 if (is_quota_modification(inode, attr)) {
4807 error = dquot_initialize(inode);
4811 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4812 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4815 /* (user+group)*(old+new) structure, inode write (sb,
4816 * inode block, ? - but truncate inode update has it) */
4817 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4818 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4819 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4820 if (IS_ERR(handle)) {
4821 error = PTR_ERR(handle);
4824 error = dquot_transfer(inode, attr);
4826 ext4_journal_stop(handle);
4829 /* Update corresponding info in inode so that everything is in
4830 * one transaction */
4831 if (attr->ia_valid & ATTR_UID)
4832 inode->i_uid = attr->ia_uid;
4833 if (attr->ia_valid & ATTR_GID)
4834 inode->i_gid = attr->ia_gid;
4835 error = ext4_mark_inode_dirty(handle, inode);
4836 ext4_journal_stop(handle);
4839 if (attr->ia_valid & ATTR_SIZE) {
4841 loff_t oldsize = inode->i_size;
4842 int shrink = (attr->ia_size <= inode->i_size);
4844 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4845 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4847 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4850 if (!S_ISREG(inode->i_mode))
4853 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4854 inode_inc_iversion(inode);
4856 if (ext4_should_order_data(inode) &&
4857 (attr->ia_size < inode->i_size)) {
4858 error = ext4_begin_ordered_truncate(inode,
4863 if (attr->ia_size != inode->i_size) {
4864 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4865 if (IS_ERR(handle)) {
4866 error = PTR_ERR(handle);
4869 if (ext4_handle_valid(handle) && shrink) {
4870 error = ext4_orphan_add(handle, inode);
4874 * Update c/mtime on truncate up, ext4_truncate() will
4875 * update c/mtime in shrink case below
4878 inode->i_mtime = ext4_current_time(inode);
4879 inode->i_ctime = inode->i_mtime;
4881 down_write(&EXT4_I(inode)->i_data_sem);
4882 EXT4_I(inode)->i_disksize = attr->ia_size;
4883 rc = ext4_mark_inode_dirty(handle, inode);
4887 * We have to update i_size under i_data_sem together
4888 * with i_disksize to avoid races with writeback code
4889 * running ext4_wb_update_i_disksize().
4892 i_size_write(inode, attr->ia_size);
4893 up_write(&EXT4_I(inode)->i_data_sem);
4894 ext4_journal_stop(handle);
4897 ext4_orphan_del(NULL, inode);
4902 pagecache_isize_extended(inode, oldsize, inode->i_size);
4905 * Blocks are going to be removed from the inode. Wait
4906 * for dio in flight. Temporarily disable
4907 * dioread_nolock to prevent livelock.
4910 if (!ext4_should_journal_data(inode)) {
4911 ext4_inode_block_unlocked_dio(inode);
4912 inode_dio_wait(inode);
4913 ext4_inode_resume_unlocked_dio(inode);
4915 ext4_wait_for_tail_page_commit(inode);
4917 down_write(&EXT4_I(inode)->i_mmap_sem);
4919 * Truncate pagecache after we've waited for commit
4920 * in data=journal mode to make pages freeable.
4922 truncate_pagecache(inode, inode->i_size);
4924 ext4_truncate(inode);
4925 up_write(&EXT4_I(inode)->i_mmap_sem);
4929 setattr_copy(inode, attr);
4930 mark_inode_dirty(inode);
4934 * If the call to ext4_truncate failed to get a transaction handle at
4935 * all, we need to clean up the in-core orphan list manually.
4937 if (orphan && inode->i_nlink)
4938 ext4_orphan_del(NULL, inode);
4940 if (!rc && (ia_valid & ATTR_MODE))
4941 rc = posix_acl_chmod(inode, inode->i_mode);
4944 ext4_std_error(inode->i_sb, error);
4950 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4953 struct inode *inode;
4954 unsigned long long delalloc_blocks;
4956 inode = d_inode(dentry);
4957 generic_fillattr(inode, stat);
4960 * If there is inline data in the inode, the inode will normally not
4961 * have data blocks allocated (it may have an external xattr block).
4962 * Report at least one sector for such files, so tools like tar, rsync,
4963 * others doen't incorrectly think the file is completely sparse.
4965 if (unlikely(ext4_has_inline_data(inode)))
4966 stat->blocks += (stat->size + 511) >> 9;
4969 * We can't update i_blocks if the block allocation is delayed
4970 * otherwise in the case of system crash before the real block
4971 * allocation is done, we will have i_blocks inconsistent with
4972 * on-disk file blocks.
4973 * We always keep i_blocks updated together with real
4974 * allocation. But to not confuse with user, stat
4975 * will return the blocks that include the delayed allocation
4976 * blocks for this file.
4978 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4979 EXT4_I(inode)->i_reserved_data_blocks);
4980 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4984 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4987 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4988 return ext4_ind_trans_blocks(inode, lblocks);
4989 return ext4_ext_index_trans_blocks(inode, pextents);
4993 * Account for index blocks, block groups bitmaps and block group
4994 * descriptor blocks if modify datablocks and index blocks
4995 * worse case, the indexs blocks spread over different block groups
4997 * If datablocks are discontiguous, they are possible to spread over
4998 * different block groups too. If they are contiguous, with flexbg,
4999 * they could still across block group boundary.
5001 * Also account for superblock, inode, quota and xattr blocks
5003 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5006 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5012 * How many index blocks need to touch to map @lblocks logical blocks
5013 * to @pextents physical extents?
5015 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5020 * Now let's see how many group bitmaps and group descriptors need
5023 groups = idxblocks + pextents;
5025 if (groups > ngroups)
5027 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5028 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5030 /* bitmaps and block group descriptor blocks */
5031 ret += groups + gdpblocks;
5033 /* Blocks for super block, inode, quota and xattr blocks */
5034 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5040 * Calculate the total number of credits to reserve to fit
5041 * the modification of a single pages into a single transaction,
5042 * which may include multiple chunks of block allocations.
5044 * This could be called via ext4_write_begin()
5046 * We need to consider the worse case, when
5047 * one new block per extent.
5049 int ext4_writepage_trans_blocks(struct inode *inode)
5051 int bpp = ext4_journal_blocks_per_page(inode);
5054 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5056 /* Account for data blocks for journalled mode */
5057 if (ext4_should_journal_data(inode))
5063 * Calculate the journal credits for a chunk of data modification.
5065 * This is called from DIO, fallocate or whoever calling
5066 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5068 * journal buffers for data blocks are not included here, as DIO
5069 * and fallocate do no need to journal data buffers.
5071 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5073 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5077 * The caller must have previously called ext4_reserve_inode_write().
5078 * Give this, we know that the caller already has write access to iloc->bh.
5080 int ext4_mark_iloc_dirty(handle_t *handle,
5081 struct inode *inode, struct ext4_iloc *iloc)
5085 if (IS_I_VERSION(inode))
5086 inode_inc_iversion(inode);
5088 /* the do_update_inode consumes one bh->b_count */
5091 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5092 err = ext4_do_update_inode(handle, inode, iloc);
5098 * On success, We end up with an outstanding reference count against
5099 * iloc->bh. This _must_ be cleaned up later.
5103 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5104 struct ext4_iloc *iloc)
5108 err = ext4_get_inode_loc(inode, iloc);
5110 BUFFER_TRACE(iloc->bh, "get_write_access");
5111 err = ext4_journal_get_write_access(handle, iloc->bh);
5117 ext4_std_error(inode->i_sb, err);
5122 * Expand an inode by new_extra_isize bytes.
5123 * Returns 0 on success or negative error number on failure.
5125 static int ext4_expand_extra_isize(struct inode *inode,
5126 unsigned int new_extra_isize,
5127 struct ext4_iloc iloc,
5130 struct ext4_inode *raw_inode;
5131 struct ext4_xattr_ibody_header *header;
5133 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5136 raw_inode = ext4_raw_inode(&iloc);
5138 header = IHDR(inode, raw_inode);
5140 /* No extended attributes present */
5141 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5142 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5143 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5145 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5149 /* try to expand with EAs present */
5150 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5155 * What we do here is to mark the in-core inode as clean with respect to inode
5156 * dirtiness (it may still be data-dirty).
5157 * This means that the in-core inode may be reaped by prune_icache
5158 * without having to perform any I/O. This is a very good thing,
5159 * because *any* task may call prune_icache - even ones which
5160 * have a transaction open against a different journal.
5162 * Is this cheating? Not really. Sure, we haven't written the
5163 * inode out, but prune_icache isn't a user-visible syncing function.
5164 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5165 * we start and wait on commits.
5167 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5169 struct ext4_iloc iloc;
5170 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5171 static unsigned int mnt_count;
5175 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5176 err = ext4_reserve_inode_write(handle, inode, &iloc);
5179 if (ext4_handle_valid(handle) &&
5180 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5181 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5183 * We need extra buffer credits since we may write into EA block
5184 * with this same handle. If journal_extend fails, then it will
5185 * only result in a minor loss of functionality for that inode.
5186 * If this is felt to be critical, then e2fsck should be run to
5187 * force a large enough s_min_extra_isize.
5189 if ((jbd2_journal_extend(handle,
5190 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5191 ret = ext4_expand_extra_isize(inode,
5192 sbi->s_want_extra_isize,
5196 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5197 ext4_warning(inode->i_sb,
5198 "Unable to expand inode %lu. Delete"
5199 " some EAs or run e2fsck.",
5202 le16_to_cpu(sbi->s_es->s_mnt_count);
5207 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5211 * ext4_dirty_inode() is called from __mark_inode_dirty()
5213 * We're really interested in the case where a file is being extended.
5214 * i_size has been changed by generic_commit_write() and we thus need
5215 * to include the updated inode in the current transaction.
5217 * Also, dquot_alloc_block() will always dirty the inode when blocks
5218 * are allocated to the file.
5220 * If the inode is marked synchronous, we don't honour that here - doing
5221 * so would cause a commit on atime updates, which we don't bother doing.
5222 * We handle synchronous inodes at the highest possible level.
5224 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5225 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5226 * to copy into the on-disk inode structure are the timestamp files.
5228 void ext4_dirty_inode(struct inode *inode, int flags)
5232 if (flags == I_DIRTY_TIME)
5234 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5238 ext4_mark_inode_dirty(handle, inode);
5240 ext4_journal_stop(handle);
5247 * Bind an inode's backing buffer_head into this transaction, to prevent
5248 * it from being flushed to disk early. Unlike
5249 * ext4_reserve_inode_write, this leaves behind no bh reference and
5250 * returns no iloc structure, so the caller needs to repeat the iloc
5251 * lookup to mark the inode dirty later.
5253 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5255 struct ext4_iloc iloc;
5259 err = ext4_get_inode_loc(inode, &iloc);
5261 BUFFER_TRACE(iloc.bh, "get_write_access");
5262 err = jbd2_journal_get_write_access(handle, iloc.bh);
5264 err = ext4_handle_dirty_metadata(handle,
5270 ext4_std_error(inode->i_sb, err);
5275 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5282 * We have to be very careful here: changing a data block's
5283 * journaling status dynamically is dangerous. If we write a
5284 * data block to the journal, change the status and then delete
5285 * that block, we risk forgetting to revoke the old log record
5286 * from the journal and so a subsequent replay can corrupt data.
5287 * So, first we make sure that the journal is empty and that
5288 * nobody is changing anything.
5291 journal = EXT4_JOURNAL(inode);
5294 if (is_journal_aborted(journal))
5296 /* We have to allocate physical blocks for delalloc blocks
5297 * before flushing journal. otherwise delalloc blocks can not
5298 * be allocated any more. even more truncate on delalloc blocks
5299 * could trigger BUG by flushing delalloc blocks in journal.
5300 * There is no delalloc block in non-journal data mode.
5302 if (val && test_opt(inode->i_sb, DELALLOC)) {
5303 err = ext4_alloc_da_blocks(inode);
5308 /* Wait for all existing dio workers */
5309 ext4_inode_block_unlocked_dio(inode);
5310 inode_dio_wait(inode);
5312 jbd2_journal_lock_updates(journal);
5315 * OK, there are no updates running now, and all cached data is
5316 * synced to disk. We are now in a completely consistent state
5317 * which doesn't have anything in the journal, and we know that
5318 * no filesystem updates are running, so it is safe to modify
5319 * the inode's in-core data-journaling state flag now.
5323 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5325 err = jbd2_journal_flush(journal);
5327 jbd2_journal_unlock_updates(journal);
5328 ext4_inode_resume_unlocked_dio(inode);
5331 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5333 ext4_set_aops(inode);
5335 jbd2_journal_unlock_updates(journal);
5336 ext4_inode_resume_unlocked_dio(inode);
5338 /* Finally we can mark the inode as dirty. */
5340 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5342 return PTR_ERR(handle);
5344 err = ext4_mark_inode_dirty(handle, inode);
5345 ext4_handle_sync(handle);
5346 ext4_journal_stop(handle);
5347 ext4_std_error(inode->i_sb, err);
5352 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5354 return !buffer_mapped(bh);
5357 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5359 struct page *page = vmf->page;
5363 struct file *file = vma->vm_file;
5364 struct inode *inode = file_inode(file);
5365 struct address_space *mapping = inode->i_mapping;
5367 get_block_t *get_block;
5370 sb_start_pagefault(inode->i_sb);
5371 file_update_time(vma->vm_file);
5373 down_read(&EXT4_I(inode)->i_mmap_sem);
5374 /* Delalloc case is easy... */
5375 if (test_opt(inode->i_sb, DELALLOC) &&
5376 !ext4_should_journal_data(inode) &&
5377 !ext4_nonda_switch(inode->i_sb)) {
5379 ret = block_page_mkwrite(vma, vmf,
5380 ext4_da_get_block_prep);
5381 } while (ret == -ENOSPC &&
5382 ext4_should_retry_alloc(inode->i_sb, &retries));
5387 size = i_size_read(inode);
5388 /* Page got truncated from under us? */
5389 if (page->mapping != mapping || page_offset(page) > size) {
5391 ret = VM_FAULT_NOPAGE;
5395 if (page->index == size >> PAGE_CACHE_SHIFT)
5396 len = size & ~PAGE_CACHE_MASK;
5398 len = PAGE_CACHE_SIZE;
5400 * Return if we have all the buffers mapped. This avoids the need to do
5401 * journal_start/journal_stop which can block and take a long time
5403 if (page_has_buffers(page)) {
5404 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5406 ext4_bh_unmapped)) {
5407 /* Wait so that we don't change page under IO */
5408 wait_for_stable_page(page);
5409 ret = VM_FAULT_LOCKED;
5414 /* OK, we need to fill the hole... */
5415 if (ext4_should_dioread_nolock(inode))
5416 get_block = ext4_get_block_write;
5418 get_block = ext4_get_block;
5420 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5421 ext4_writepage_trans_blocks(inode));
5422 if (IS_ERR(handle)) {
5423 ret = VM_FAULT_SIGBUS;
5426 ret = block_page_mkwrite(vma, vmf, get_block);
5427 if (!ret && ext4_should_journal_data(inode)) {
5428 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5429 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5431 ret = VM_FAULT_SIGBUS;
5432 ext4_journal_stop(handle);
5435 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5437 ext4_journal_stop(handle);
5438 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5441 ret = block_page_mkwrite_return(ret);
5443 up_read(&EXT4_I(inode)->i_mmap_sem);
5444 sb_end_pagefault(inode->i_sb);
5448 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5450 struct inode *inode = file_inode(vma->vm_file);
5453 down_read(&EXT4_I(inode)->i_mmap_sem);
5454 err = filemap_fault(vma, vmf);
5455 up_read(&EXT4_I(inode)->i_mmap_sem);