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>
47 #include <trace/events/android_fs.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 struct ext4_inode_info *ei)
54 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 int offset = offsetof(struct ext4_inode, i_checksum_lo);
58 unsigned int csum_size = sizeof(dummy_csum);
60 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
61 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
64 EXT4_GOOD_OLD_INODE_SIZE - offset);
66 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
67 offset = offsetof(struct ext4_inode, i_checksum_hi);
68 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
69 EXT4_GOOD_OLD_INODE_SIZE,
70 offset - EXT4_GOOD_OLD_INODE_SIZE);
71 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
72 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
75 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
76 EXT4_INODE_SIZE(inode->i_sb) -
84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 struct ext4_inode_info *ei)
87 __u32 provided, calculated;
89 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 cpu_to_le32(EXT4_OS_LINUX) ||
91 !ext4_has_metadata_csum(inode->i_sb))
94 provided = le16_to_cpu(raw->i_checksum_lo);
95 calculated = ext4_inode_csum(inode, raw, ei);
96 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100 calculated &= 0xFFFF;
102 return provided == calculated;
105 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 struct ext4_inode_info *ei)
110 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 cpu_to_le32(EXT4_OS_LINUX) ||
112 !ext4_has_metadata_csum(inode->i_sb))
115 csum = ext4_inode_csum(inode, raw, ei);
116 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
125 trace_ext4_begin_ordered_truncate(inode, new_size);
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
132 if (!EXT4_I(inode)->jinode)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 EXT4_I(inode)->jinode,
139 static void ext4_invalidatepage(struct page *page, unsigned int offset,
140 unsigned int length);
141 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
142 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
143 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
147 * Test whether an inode is a fast symlink.
149 int ext4_inode_is_fast_symlink(struct inode *inode)
151 int ea_blocks = EXT4_I(inode)->i_file_acl ?
152 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
154 if (ext4_has_inline_data(inode))
157 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 * Restart the transaction associated with *handle. This does a commit,
162 * so before we call here everything must be consistently dirtied against
165 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
171 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
172 * moment, get_block can be called only for blocks inside i_size since
173 * page cache has been already dropped and writes are blocked by
174 * i_mutex. So we can safely drop the i_data_sem here.
176 BUG_ON(EXT4_JOURNAL(inode) == NULL);
177 jbd_debug(2, "restarting handle %p\n", handle);
178 up_write(&EXT4_I(inode)->i_data_sem);
179 ret = ext4_journal_restart(handle, nblocks);
180 down_write(&EXT4_I(inode)->i_data_sem);
181 ext4_discard_preallocations(inode);
187 * Called at the last iput() if i_nlink is zero.
189 void ext4_evict_inode(struct inode *inode)
194 trace_ext4_evict_inode(inode);
196 if (inode->i_nlink) {
198 * When journalling data dirty buffers are tracked only in the
199 * journal. So although mm thinks everything is clean and
200 * ready for reaping the inode might still have some pages to
201 * write in the running transaction or waiting to be
202 * checkpointed. Thus calling jbd2_journal_invalidatepage()
203 * (via truncate_inode_pages()) to discard these buffers can
204 * cause data loss. Also even if we did not discard these
205 * buffers, we would have no way to find them after the inode
206 * is reaped and thus user could see stale data if he tries to
207 * read them before the transaction is checkpointed. So be
208 * careful and force everything to disk here... We use
209 * ei->i_datasync_tid to store the newest transaction
210 * containing inode's data.
212 * Note that directories do not have this problem because they
213 * don't use page cache.
215 if (inode->i_ino != EXT4_JOURNAL_INO &&
216 ext4_should_journal_data(inode) &&
217 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
218 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
219 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
221 jbd2_complete_transaction(journal, commit_tid);
222 filemap_write_and_wait(&inode->i_data);
224 truncate_inode_pages_final(&inode->i_data);
226 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
230 if (is_bad_inode(inode))
232 dquot_initialize(inode);
234 if (ext4_should_order_data(inode))
235 ext4_begin_ordered_truncate(inode, 0);
236 truncate_inode_pages_final(&inode->i_data);
238 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
241 * Protect us against freezing - iput() caller didn't have to have any
242 * protection against it
244 sb_start_intwrite(inode->i_sb);
245 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
246 ext4_blocks_for_truncate(inode)+3);
247 if (IS_ERR(handle)) {
248 ext4_std_error(inode->i_sb, PTR_ERR(handle));
250 * If we're going to skip the normal cleanup, we still need to
251 * make sure that the in-core orphan linked list is properly
254 ext4_orphan_del(NULL, inode);
255 sb_end_intwrite(inode->i_sb);
260 ext4_handle_sync(handle);
262 err = ext4_mark_inode_dirty(handle, inode);
264 ext4_warning(inode->i_sb,
265 "couldn't mark inode dirty (err %d)", err);
269 ext4_truncate(inode);
272 * ext4_ext_truncate() doesn't reserve any slop when it
273 * restarts journal transactions; therefore there may not be
274 * enough credits left in the handle to remove the inode from
275 * the orphan list and set the dtime field.
277 if (!ext4_handle_has_enough_credits(handle, 3)) {
278 err = ext4_journal_extend(handle, 3);
280 err = ext4_journal_restart(handle, 3);
282 ext4_warning(inode->i_sb,
283 "couldn't extend journal (err %d)", err);
285 ext4_journal_stop(handle);
286 ext4_orphan_del(NULL, inode);
287 sb_end_intwrite(inode->i_sb);
293 * Kill off the orphan record which ext4_truncate created.
294 * AKPM: I think this can be inside the above `if'.
295 * Note that ext4_orphan_del() has to be able to cope with the
296 * deletion of a non-existent orphan - this is because we don't
297 * know if ext4_truncate() actually created an orphan record.
298 * (Well, we could do this if we need to, but heck - it works)
300 ext4_orphan_del(handle, inode);
301 EXT4_I(inode)->i_dtime = get_seconds();
304 * One subtle ordering requirement: if anything has gone wrong
305 * (transaction abort, IO errors, whatever), then we can still
306 * do these next steps (the fs will already have been marked as
307 * having errors), but we can't free the inode if the mark_dirty
310 if (ext4_mark_inode_dirty(handle, inode))
311 /* If that failed, just do the required in-core inode clear. */
312 ext4_clear_inode(inode);
314 ext4_free_inode(handle, inode);
315 ext4_journal_stop(handle);
316 sb_end_intwrite(inode->i_sb);
319 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
323 qsize_t *ext4_get_reserved_space(struct inode *inode)
325 return &EXT4_I(inode)->i_reserved_quota;
330 * Called with i_data_sem down, which is important since we can call
331 * ext4_discard_preallocations() from here.
333 void ext4_da_update_reserve_space(struct inode *inode,
334 int used, int quota_claim)
336 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
337 struct ext4_inode_info *ei = EXT4_I(inode);
339 spin_lock(&ei->i_block_reservation_lock);
340 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
341 if (unlikely(used > ei->i_reserved_data_blocks)) {
342 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
343 "with only %d reserved data blocks",
344 __func__, inode->i_ino, used,
345 ei->i_reserved_data_blocks);
347 used = ei->i_reserved_data_blocks;
350 /* Update per-inode reservations */
351 ei->i_reserved_data_blocks -= used;
352 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
354 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
356 /* Update quota subsystem for data blocks */
358 dquot_claim_block(inode, EXT4_C2B(sbi, used));
361 * We did fallocate with an offset that is already delayed
362 * allocated. So on delayed allocated writeback we should
363 * not re-claim the quota for fallocated blocks.
365 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
369 * If we have done all the pending block allocations and if
370 * there aren't any writers on the inode, we can discard the
371 * inode's preallocations.
373 if ((ei->i_reserved_data_blocks == 0) &&
374 (atomic_read(&inode->i_writecount) == 0))
375 ext4_discard_preallocations(inode);
378 static int __check_block_validity(struct inode *inode, const char *func,
380 struct ext4_map_blocks *map)
382 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
384 ext4_error_inode(inode, func, line, map->m_pblk,
385 "lblock %lu mapped to illegal pblock "
386 "(length %d)", (unsigned long) map->m_lblk,
388 return -EFSCORRUPTED;
393 #define check_block_validity(inode, map) \
394 __check_block_validity((inode), __func__, __LINE__, (map))
396 #ifdef ES_AGGRESSIVE_TEST
397 static void ext4_map_blocks_es_recheck(handle_t *handle,
399 struct ext4_map_blocks *es_map,
400 struct ext4_map_blocks *map,
407 * There is a race window that the result is not the same.
408 * e.g. xfstests #223 when dioread_nolock enables. The reason
409 * is that we lookup a block mapping in extent status tree with
410 * out taking i_data_sem. So at the time the unwritten extent
411 * could be converted.
413 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
414 down_read(&EXT4_I(inode)->i_data_sem);
415 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
416 retval = ext4_ext_map_blocks(handle, inode, map, flags &
417 EXT4_GET_BLOCKS_KEEP_SIZE);
419 retval = ext4_ind_map_blocks(handle, inode, map, flags &
420 EXT4_GET_BLOCKS_KEEP_SIZE);
422 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
423 up_read((&EXT4_I(inode)->i_data_sem));
426 * We don't check m_len because extent will be collpased in status
427 * tree. So the m_len might not equal.
429 if (es_map->m_lblk != map->m_lblk ||
430 es_map->m_flags != map->m_flags ||
431 es_map->m_pblk != map->m_pblk) {
432 printk("ES cache assertion failed for inode: %lu "
433 "es_cached ex [%d/%d/%llu/%x] != "
434 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
435 inode->i_ino, es_map->m_lblk, es_map->m_len,
436 es_map->m_pblk, es_map->m_flags, map->m_lblk,
437 map->m_len, map->m_pblk, map->m_flags,
441 #endif /* ES_AGGRESSIVE_TEST */
444 * The ext4_map_blocks() function tries to look up the requested blocks,
445 * and returns if the blocks are already mapped.
447 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
448 * and store the allocated blocks in the result buffer head and mark it
451 * If file type is extents based, it will call ext4_ext_map_blocks(),
452 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
455 * On success, it returns the number of blocks being mapped or allocated.
456 * if create==0 and the blocks are pre-allocated and unwritten block,
457 * the result buffer head is unmapped. If the create ==1, it will make sure
458 * the buffer head is mapped.
460 * It returns 0 if plain look up failed (blocks have not been allocated), in
461 * that case, buffer head is unmapped
463 * It returns the error in case of allocation failure.
465 int ext4_map_blocks(handle_t *handle, struct inode *inode,
466 struct ext4_map_blocks *map, int flags)
468 struct extent_status es;
471 #ifdef ES_AGGRESSIVE_TEST
472 struct ext4_map_blocks orig_map;
474 memcpy(&orig_map, map, sizeof(*map));
478 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
479 "logical block %lu\n", inode->i_ino, flags, map->m_len,
480 (unsigned long) map->m_lblk);
483 * ext4_map_blocks returns an int, and m_len is an unsigned int
485 if (unlikely(map->m_len > INT_MAX))
486 map->m_len = INT_MAX;
488 /* We can handle the block number less than EXT_MAX_BLOCKS */
489 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
490 return -EFSCORRUPTED;
492 /* Lookup extent status tree firstly */
493 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
494 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
495 map->m_pblk = ext4_es_pblock(&es) +
496 map->m_lblk - es.es_lblk;
497 map->m_flags |= ext4_es_is_written(&es) ?
498 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
499 retval = es.es_len - (map->m_lblk - es.es_lblk);
500 if (retval > map->m_len)
503 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
508 #ifdef ES_AGGRESSIVE_TEST
509 ext4_map_blocks_es_recheck(handle, inode, map,
516 * Try to see if we can get the block without requesting a new
519 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
520 down_read(&EXT4_I(inode)->i_data_sem);
521 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
522 retval = ext4_ext_map_blocks(handle, inode, map, flags &
523 EXT4_GET_BLOCKS_KEEP_SIZE);
525 retval = ext4_ind_map_blocks(handle, inode, map, flags &
526 EXT4_GET_BLOCKS_KEEP_SIZE);
531 if (unlikely(retval != map->m_len)) {
532 ext4_warning(inode->i_sb,
533 "ES len assertion failed for inode "
534 "%lu: retval %d != map->m_len %d",
535 inode->i_ino, retval, map->m_len);
539 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
540 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
541 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
542 !(status & EXTENT_STATUS_WRITTEN) &&
543 ext4_find_delalloc_range(inode, map->m_lblk,
544 map->m_lblk + map->m_len - 1))
545 status |= EXTENT_STATUS_DELAYED;
546 ret = ext4_es_insert_extent(inode, map->m_lblk,
547 map->m_len, map->m_pblk, status);
551 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
552 up_read((&EXT4_I(inode)->i_data_sem));
555 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
556 ret = check_block_validity(inode, map);
561 /* If it is only a block(s) look up */
562 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
566 * Returns if the blocks have already allocated
568 * Note that if blocks have been preallocated
569 * ext4_ext_get_block() returns the create = 0
570 * with buffer head unmapped.
572 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
574 * If we need to convert extent to unwritten
575 * we continue and do the actual work in
576 * ext4_ext_map_blocks()
578 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
582 * Here we clear m_flags because after allocating an new extent,
583 * it will be set again.
585 map->m_flags &= ~EXT4_MAP_FLAGS;
588 * New blocks allocate and/or writing to unwritten extent
589 * will possibly result in updating i_data, so we take
590 * the write lock of i_data_sem, and call get_block()
591 * with create == 1 flag.
593 down_write(&EXT4_I(inode)->i_data_sem);
596 * We need to check for EXT4 here because migrate
597 * could have changed the inode type in between
599 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
600 retval = ext4_ext_map_blocks(handle, inode, map, flags);
602 retval = ext4_ind_map_blocks(handle, inode, map, flags);
604 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
606 * We allocated new blocks which will result in
607 * i_data's format changing. Force the migrate
608 * to fail by clearing migrate flags
610 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
614 * Update reserved blocks/metadata blocks after successful
615 * block allocation which had been deferred till now. We don't
616 * support fallocate for non extent files. So we can update
617 * reserve space here.
620 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
621 ext4_da_update_reserve_space(inode, retval, 1);
627 if (unlikely(retval != map->m_len)) {
628 ext4_warning(inode->i_sb,
629 "ES len assertion failed for inode "
630 "%lu: retval %d != map->m_len %d",
631 inode->i_ino, retval, map->m_len);
636 * If the extent has been zeroed out, we don't need to update
637 * extent status tree.
639 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
640 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
641 if (ext4_es_is_written(&es))
644 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
645 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
646 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
647 !(status & EXTENT_STATUS_WRITTEN) &&
648 ext4_find_delalloc_range(inode, map->m_lblk,
649 map->m_lblk + map->m_len - 1))
650 status |= EXTENT_STATUS_DELAYED;
651 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
652 map->m_pblk, status);
658 up_write((&EXT4_I(inode)->i_data_sem));
659 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
660 ret = check_block_validity(inode, map);
668 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
669 * we have to be careful as someone else may be manipulating b_state as well.
671 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
673 unsigned long old_state;
674 unsigned long new_state;
676 flags &= EXT4_MAP_FLAGS;
678 /* Dummy buffer_head? Set non-atomically. */
680 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
684 * Someone else may be modifying b_state. Be careful! This is ugly but
685 * once we get rid of using bh as a container for mapping information
686 * to pass to / from get_block functions, this can go away.
689 old_state = READ_ONCE(bh->b_state);
690 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
692 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
695 /* Maximum number of blocks we map for direct IO at once. */
696 #define DIO_MAX_BLOCKS 4096
698 static int _ext4_get_block(struct inode *inode, sector_t iblock,
699 struct buffer_head *bh, int flags)
701 handle_t *handle = ext4_journal_current_handle();
702 struct ext4_map_blocks map;
703 int ret = 0, started = 0;
706 if (ext4_has_inline_data(inode))
710 map.m_len = bh->b_size >> inode->i_blkbits;
712 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
713 /* Direct IO write... */
714 if (map.m_len > DIO_MAX_BLOCKS)
715 map.m_len = DIO_MAX_BLOCKS;
716 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
717 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
719 if (IS_ERR(handle)) {
720 ret = PTR_ERR(handle);
726 ret = ext4_map_blocks(handle, inode, &map, flags);
728 ext4_io_end_t *io_end = ext4_inode_aio(inode);
730 map_bh(bh, inode->i_sb, map.m_pblk);
731 ext4_update_bh_state(bh, map.m_flags);
732 if (IS_DAX(inode) && buffer_unwritten(bh)) {
734 * dgc: I suspect unwritten conversion on ext4+DAX is
735 * fundamentally broken here when there are concurrent
736 * read/write in progress on this inode.
738 WARN_ON_ONCE(io_end);
739 bh->b_assoc_map = inode->i_mapping;
740 bh->b_private = (void *)(unsigned long)iblock;
742 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
743 set_buffer_defer_completion(bh);
744 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
748 ext4_journal_stop(handle);
752 int ext4_get_block(struct inode *inode, sector_t iblock,
753 struct buffer_head *bh, int create)
755 return _ext4_get_block(inode, iblock, bh,
756 create ? EXT4_GET_BLOCKS_CREATE : 0);
760 * `handle' can be NULL if create is zero
762 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
763 ext4_lblk_t block, int map_flags)
765 struct ext4_map_blocks map;
766 struct buffer_head *bh;
767 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
770 J_ASSERT(handle != NULL || create == 0);
774 err = ext4_map_blocks(handle, inode, &map, map_flags);
777 return create ? ERR_PTR(-ENOSPC) : NULL;
781 bh = sb_getblk(inode->i_sb, map.m_pblk);
783 return ERR_PTR(-ENOMEM);
784 if (map.m_flags & EXT4_MAP_NEW) {
785 J_ASSERT(create != 0);
786 J_ASSERT(handle != NULL);
789 * Now that we do not always journal data, we should
790 * keep in mind whether this should always journal the
791 * new buffer as metadata. For now, regular file
792 * writes use ext4_get_block instead, so it's not a
796 BUFFER_TRACE(bh, "call get_create_access");
797 err = ext4_journal_get_create_access(handle, bh);
802 if (!buffer_uptodate(bh)) {
803 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
804 set_buffer_uptodate(bh);
807 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
808 err = ext4_handle_dirty_metadata(handle, inode, bh);
812 BUFFER_TRACE(bh, "not a new buffer");
819 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
820 ext4_lblk_t block, int map_flags)
822 struct buffer_head *bh;
824 bh = ext4_getblk(handle, inode, block, map_flags);
827 if (!bh || buffer_uptodate(bh))
829 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
831 if (buffer_uptodate(bh))
834 return ERR_PTR(-EIO);
837 int ext4_walk_page_buffers(handle_t *handle,
838 struct buffer_head *head,
842 int (*fn)(handle_t *handle,
843 struct buffer_head *bh))
845 struct buffer_head *bh;
846 unsigned block_start, block_end;
847 unsigned blocksize = head->b_size;
849 struct buffer_head *next;
851 for (bh = head, block_start = 0;
852 ret == 0 && (bh != head || !block_start);
853 block_start = block_end, bh = next) {
854 next = bh->b_this_page;
855 block_end = block_start + blocksize;
856 if (block_end <= from || block_start >= to) {
857 if (partial && !buffer_uptodate(bh))
861 err = (*fn)(handle, bh);
869 * To preserve ordering, it is essential that the hole instantiation and
870 * the data write be encapsulated in a single transaction. We cannot
871 * close off a transaction and start a new one between the ext4_get_block()
872 * and the commit_write(). So doing the jbd2_journal_start at the start of
873 * prepare_write() is the right place.
875 * Also, this function can nest inside ext4_writepage(). In that case, we
876 * *know* that ext4_writepage() has generated enough buffer credits to do the
877 * whole page. So we won't block on the journal in that case, which is good,
878 * because the caller may be PF_MEMALLOC.
880 * By accident, ext4 can be reentered when a transaction is open via
881 * quota file writes. If we were to commit the transaction while thus
882 * reentered, there can be a deadlock - we would be holding a quota
883 * lock, and the commit would never complete if another thread had a
884 * transaction open and was blocking on the quota lock - a ranking
887 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
888 * will _not_ run commit under these circumstances because handle->h_ref
889 * is elevated. We'll still have enough credits for the tiny quotafile
892 int do_journal_get_write_access(handle_t *handle,
893 struct buffer_head *bh)
895 int dirty = buffer_dirty(bh);
898 if (!buffer_mapped(bh) || buffer_freed(bh))
901 * __block_write_begin() could have dirtied some buffers. Clean
902 * the dirty bit as jbd2_journal_get_write_access() could complain
903 * otherwise about fs integrity issues. Setting of the dirty bit
904 * by __block_write_begin() isn't a real problem here as we clear
905 * the bit before releasing a page lock and thus writeback cannot
906 * ever write the buffer.
909 clear_buffer_dirty(bh);
910 BUFFER_TRACE(bh, "get write access");
911 ret = ext4_journal_get_write_access(handle, bh);
913 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
917 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
918 struct buffer_head *bh_result, int create);
920 #ifdef CONFIG_EXT4_FS_ENCRYPTION
921 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
922 get_block_t *get_block)
924 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
925 unsigned to = from + len;
926 struct inode *inode = page->mapping->host;
927 unsigned block_start, block_end;
930 unsigned blocksize = inode->i_sb->s_blocksize;
932 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
933 bool decrypt = false;
935 BUG_ON(!PageLocked(page));
936 BUG_ON(from > PAGE_CACHE_SIZE);
937 BUG_ON(to > PAGE_CACHE_SIZE);
940 if (!page_has_buffers(page))
941 create_empty_buffers(page, blocksize, 0);
942 head = page_buffers(page);
943 bbits = ilog2(blocksize);
944 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
946 for (bh = head, block_start = 0; bh != head || !block_start;
947 block++, block_start = block_end, bh = bh->b_this_page) {
948 block_end = block_start + blocksize;
949 if (block_end <= from || block_start >= to) {
950 if (PageUptodate(page)) {
951 if (!buffer_uptodate(bh))
952 set_buffer_uptodate(bh);
957 clear_buffer_new(bh);
958 if (!buffer_mapped(bh)) {
959 WARN_ON(bh->b_size != blocksize);
960 err = get_block(inode, block, bh, 1);
963 if (buffer_new(bh)) {
964 unmap_underlying_metadata(bh->b_bdev,
966 if (PageUptodate(page)) {
967 clear_buffer_new(bh);
968 set_buffer_uptodate(bh);
969 mark_buffer_dirty(bh);
972 if (block_end > to || block_start < from)
973 zero_user_segments(page, to, block_end,
978 if (PageUptodate(page)) {
979 if (!buffer_uptodate(bh))
980 set_buffer_uptodate(bh);
983 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
984 !buffer_unwritten(bh) &&
985 (block_start < from || block_end > to)) {
986 ll_rw_block(READ, 1, &bh);
988 decrypt = ext4_encrypted_inode(inode) &&
989 S_ISREG(inode->i_mode);
993 * If we issued read requests, let them complete.
995 while (wait_bh > wait) {
996 wait_on_buffer(*--wait_bh);
997 if (!buffer_uptodate(*wait_bh))
1001 page_zero_new_buffers(page, from, to);
1003 err = ext4_decrypt(page);
1008 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1009 loff_t pos, unsigned len, unsigned flags,
1010 struct page **pagep, void **fsdata)
1012 struct inode *inode = mapping->host;
1013 int ret, needed_blocks;
1020 trace_android_fs_datawrite_start(inode, pos, len,
1021 current->pid, current->comm);
1022 trace_ext4_write_begin(inode, pos, len, flags);
1024 * Reserve one block more for addition to orphan list in case
1025 * we allocate blocks but write fails for some reason
1027 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1028 index = pos >> PAGE_CACHE_SHIFT;
1029 from = pos & (PAGE_CACHE_SIZE - 1);
1032 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1033 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1042 * grab_cache_page_write_begin() can take a long time if the
1043 * system is thrashing due to memory pressure, or if the page
1044 * is being written back. So grab it first before we start
1045 * the transaction handle. This also allows us to allocate
1046 * the page (if needed) without using GFP_NOFS.
1049 page = grab_cache_page_write_begin(mapping, index, flags);
1055 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1056 if (IS_ERR(handle)) {
1057 page_cache_release(page);
1058 return PTR_ERR(handle);
1062 if (page->mapping != mapping) {
1063 /* The page got truncated from under us */
1065 page_cache_release(page);
1066 ext4_journal_stop(handle);
1069 /* In case writeback began while the page was unlocked */
1070 wait_for_stable_page(page);
1072 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1073 if (ext4_should_dioread_nolock(inode))
1074 ret = ext4_block_write_begin(page, pos, len,
1075 ext4_get_block_write);
1077 ret = ext4_block_write_begin(page, pos, len,
1080 if (ext4_should_dioread_nolock(inode))
1081 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1083 ret = __block_write_begin(page, pos, len, ext4_get_block);
1085 if (!ret && ext4_should_journal_data(inode)) {
1086 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1088 do_journal_get_write_access);
1094 * __block_write_begin may have instantiated a few blocks
1095 * outside i_size. Trim these off again. Don't need
1096 * i_size_read because we hold i_mutex.
1098 * Add inode to orphan list in case we crash before
1101 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1102 ext4_orphan_add(handle, inode);
1104 ext4_journal_stop(handle);
1105 if (pos + len > inode->i_size) {
1106 ext4_truncate_failed_write(inode);
1108 * If truncate failed early the inode might
1109 * still be on the orphan list; we need to
1110 * make sure the inode is removed from the
1111 * orphan list in that case.
1114 ext4_orphan_del(NULL, inode);
1117 if (ret == -ENOSPC &&
1118 ext4_should_retry_alloc(inode->i_sb, &retries))
1120 page_cache_release(page);
1127 /* For write_end() in data=journal mode */
1128 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1131 if (!buffer_mapped(bh) || buffer_freed(bh))
1133 set_buffer_uptodate(bh);
1134 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1135 clear_buffer_meta(bh);
1136 clear_buffer_prio(bh);
1141 * We need to pick up the new inode size which generic_commit_write gave us
1142 * `file' can be NULL - eg, when called from page_symlink().
1144 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1145 * buffers are managed internally.
1147 static int ext4_write_end(struct file *file,
1148 struct address_space *mapping,
1149 loff_t pos, unsigned len, unsigned copied,
1150 struct page *page, void *fsdata)
1152 handle_t *handle = ext4_journal_current_handle();
1153 struct inode *inode = mapping->host;
1154 loff_t old_size = inode->i_size;
1156 int i_size_changed = 0;
1158 trace_android_fs_datawrite_end(inode, pos, len);
1159 trace_ext4_write_end(inode, pos, len, copied);
1160 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1161 ret = ext4_jbd2_file_inode(handle, inode);
1164 page_cache_release(page);
1169 if (ext4_has_inline_data(inode)) {
1170 ret = ext4_write_inline_data_end(inode, pos, len,
1179 copied = block_write_end(file, mapping, pos,
1180 len, copied, page, fsdata);
1182 * it's important to update i_size while still holding page lock:
1183 * page writeout could otherwise come in and zero beyond i_size.
1185 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1187 page_cache_release(page);
1190 pagecache_isize_extended(inode, old_size, pos);
1192 * Don't mark the inode dirty under page lock. First, it unnecessarily
1193 * makes the holding time of page lock longer. Second, it forces lock
1194 * ordering of page lock and transaction start for journaling
1198 ext4_mark_inode_dirty(handle, inode);
1200 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1201 /* if we have allocated more blocks and copied
1202 * less. We will have blocks allocated outside
1203 * inode->i_size. So truncate them
1205 ext4_orphan_add(handle, inode);
1207 ret2 = ext4_journal_stop(handle);
1211 if (pos + len > inode->i_size) {
1212 ext4_truncate_failed_write(inode);
1214 * If truncate failed early the inode might still be
1215 * on the orphan list; we need to make sure the inode
1216 * is removed from the orphan list in that case.
1219 ext4_orphan_del(NULL, inode);
1222 return ret ? ret : copied;
1226 * This is a private version of page_zero_new_buffers() which doesn't
1227 * set the buffer to be dirty, since in data=journalled mode we need
1228 * to call ext4_handle_dirty_metadata() instead.
1230 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1232 unsigned from, unsigned to)
1234 unsigned int block_start = 0, block_end;
1235 struct buffer_head *head, *bh;
1237 bh = head = page_buffers(page);
1239 block_end = block_start + bh->b_size;
1240 if (buffer_new(bh)) {
1241 if (block_end > from && block_start < to) {
1242 if (!PageUptodate(page)) {
1243 unsigned start, size;
1245 start = max(from, block_start);
1246 size = min(to, block_end) - start;
1248 zero_user(page, start, size);
1249 write_end_fn(handle, bh);
1251 clear_buffer_new(bh);
1254 block_start = block_end;
1255 bh = bh->b_this_page;
1256 } while (bh != head);
1259 static int ext4_journalled_write_end(struct file *file,
1260 struct address_space *mapping,
1261 loff_t pos, unsigned len, unsigned copied,
1262 struct page *page, void *fsdata)
1264 handle_t *handle = ext4_journal_current_handle();
1265 struct inode *inode = mapping->host;
1266 loff_t old_size = inode->i_size;
1270 int size_changed = 0;
1272 trace_android_fs_datawrite_end(inode, pos, len);
1273 trace_ext4_journalled_write_end(inode, pos, len, copied);
1274 from = pos & (PAGE_CACHE_SIZE - 1);
1277 BUG_ON(!ext4_handle_valid(handle));
1279 if (ext4_has_inline_data(inode)) {
1280 ret = ext4_write_inline_data_end(inode, pos, len,
1288 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1290 ext4_journalled_zero_new_buffers(handle, page, from, to);
1292 if (unlikely(copied < len))
1293 ext4_journalled_zero_new_buffers(handle, page,
1295 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1296 from + copied, &partial,
1299 SetPageUptodate(page);
1301 size_changed = ext4_update_inode_size(inode, pos + copied);
1302 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1303 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1305 page_cache_release(page);
1308 pagecache_isize_extended(inode, old_size, pos);
1311 ret2 = ext4_mark_inode_dirty(handle, inode);
1316 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1317 /* if we have allocated more blocks and copied
1318 * less. We will have blocks allocated outside
1319 * inode->i_size. So truncate them
1321 ext4_orphan_add(handle, inode);
1324 ret2 = ext4_journal_stop(handle);
1327 if (pos + len > inode->i_size) {
1328 ext4_truncate_failed_write(inode);
1330 * If truncate failed early the inode might still be
1331 * on the orphan list; we need to make sure the inode
1332 * is removed from the orphan list in that case.
1335 ext4_orphan_del(NULL, inode);
1338 return ret ? ret : copied;
1342 * Reserve space for a single cluster
1344 static int ext4_da_reserve_space(struct inode *inode)
1346 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1347 struct ext4_inode_info *ei = EXT4_I(inode);
1351 * We will charge metadata quota at writeout time; this saves
1352 * us from metadata over-estimation, though we may go over by
1353 * a small amount in the end. Here we just reserve for data.
1355 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1359 spin_lock(&ei->i_block_reservation_lock);
1360 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1361 spin_unlock(&ei->i_block_reservation_lock);
1362 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1365 ei->i_reserved_data_blocks++;
1366 trace_ext4_da_reserve_space(inode);
1367 spin_unlock(&ei->i_block_reservation_lock);
1369 return 0; /* success */
1372 static void ext4_da_release_space(struct inode *inode, int to_free)
1374 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1375 struct ext4_inode_info *ei = EXT4_I(inode);
1378 return; /* Nothing to release, exit */
1380 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1382 trace_ext4_da_release_space(inode, to_free);
1383 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1385 * if there aren't enough reserved blocks, then the
1386 * counter is messed up somewhere. Since this
1387 * function is called from invalidate page, it's
1388 * harmless to return without any action.
1390 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1391 "ino %lu, to_free %d with only %d reserved "
1392 "data blocks", inode->i_ino, to_free,
1393 ei->i_reserved_data_blocks);
1395 to_free = ei->i_reserved_data_blocks;
1397 ei->i_reserved_data_blocks -= to_free;
1399 /* update fs dirty data blocks counter */
1400 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1402 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1404 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1407 static void ext4_da_page_release_reservation(struct page *page,
1408 unsigned int offset,
1409 unsigned int length)
1411 int to_release = 0, contiguous_blks = 0;
1412 struct buffer_head *head, *bh;
1413 unsigned int curr_off = 0;
1414 struct inode *inode = page->mapping->host;
1415 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1416 unsigned int stop = offset + length;
1420 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1422 head = page_buffers(page);
1425 unsigned int next_off = curr_off + bh->b_size;
1427 if (next_off > stop)
1430 if ((offset <= curr_off) && (buffer_delay(bh))) {
1433 clear_buffer_delay(bh);
1434 } else if (contiguous_blks) {
1435 lblk = page->index <<
1436 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1437 lblk += (curr_off >> inode->i_blkbits) -
1439 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1440 contiguous_blks = 0;
1442 curr_off = next_off;
1443 } while ((bh = bh->b_this_page) != head);
1445 if (contiguous_blks) {
1446 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1447 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1448 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1451 /* If we have released all the blocks belonging to a cluster, then we
1452 * need to release the reserved space for that cluster. */
1453 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1454 while (num_clusters > 0) {
1455 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1456 ((num_clusters - 1) << sbi->s_cluster_bits);
1457 if (sbi->s_cluster_ratio == 1 ||
1458 !ext4_find_delalloc_cluster(inode, lblk))
1459 ext4_da_release_space(inode, 1);
1466 * Delayed allocation stuff
1469 struct mpage_da_data {
1470 struct inode *inode;
1471 struct writeback_control *wbc;
1473 pgoff_t first_page; /* The first page to write */
1474 pgoff_t next_page; /* Current page to examine */
1475 pgoff_t last_page; /* Last page to examine */
1477 * Extent to map - this can be after first_page because that can be
1478 * fully mapped. We somewhat abuse m_flags to store whether the extent
1479 * is delalloc or unwritten.
1481 struct ext4_map_blocks map;
1482 struct ext4_io_submit io_submit; /* IO submission data */
1485 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1490 struct pagevec pvec;
1491 struct inode *inode = mpd->inode;
1492 struct address_space *mapping = inode->i_mapping;
1494 /* This is necessary when next_page == 0. */
1495 if (mpd->first_page >= mpd->next_page)
1498 index = mpd->first_page;
1499 end = mpd->next_page - 1;
1501 ext4_lblk_t start, last;
1502 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1503 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1504 ext4_es_remove_extent(inode, start, last - start + 1);
1507 pagevec_init(&pvec, 0);
1508 while (index <= end) {
1509 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1512 for (i = 0; i < nr_pages; i++) {
1513 struct page *page = pvec.pages[i];
1514 if (page->index > end)
1516 BUG_ON(!PageLocked(page));
1517 BUG_ON(PageWriteback(page));
1519 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1520 ClearPageUptodate(page);
1524 index = pvec.pages[nr_pages - 1]->index + 1;
1525 pagevec_release(&pvec);
1529 static void ext4_print_free_blocks(struct inode *inode)
1531 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1532 struct super_block *sb = inode->i_sb;
1533 struct ext4_inode_info *ei = EXT4_I(inode);
1535 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1536 EXT4_C2B(EXT4_SB(inode->i_sb),
1537 ext4_count_free_clusters(sb)));
1538 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1539 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1540 (long long) EXT4_C2B(EXT4_SB(sb),
1541 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1542 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1543 (long long) EXT4_C2B(EXT4_SB(sb),
1544 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1545 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1546 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1547 ei->i_reserved_data_blocks);
1551 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1553 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1557 * This function is grabs code from the very beginning of
1558 * ext4_map_blocks, but assumes that the caller is from delayed write
1559 * time. This function looks up the requested blocks and sets the
1560 * buffer delay bit under the protection of i_data_sem.
1562 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1563 struct ext4_map_blocks *map,
1564 struct buffer_head *bh)
1566 struct extent_status es;
1568 sector_t invalid_block = ~((sector_t) 0xffff);
1569 #ifdef ES_AGGRESSIVE_TEST
1570 struct ext4_map_blocks orig_map;
1572 memcpy(&orig_map, map, sizeof(*map));
1575 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1579 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1580 "logical block %lu\n", inode->i_ino, map->m_len,
1581 (unsigned long) map->m_lblk);
1583 /* Lookup extent status tree firstly */
1584 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1585 if (ext4_es_is_hole(&es)) {
1587 down_read(&EXT4_I(inode)->i_data_sem);
1592 * Delayed extent could be allocated by fallocate.
1593 * So we need to check it.
1595 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1596 map_bh(bh, inode->i_sb, invalid_block);
1598 set_buffer_delay(bh);
1602 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1603 retval = es.es_len - (iblock - es.es_lblk);
1604 if (retval > map->m_len)
1605 retval = map->m_len;
1606 map->m_len = retval;
1607 if (ext4_es_is_written(&es))
1608 map->m_flags |= EXT4_MAP_MAPPED;
1609 else if (ext4_es_is_unwritten(&es))
1610 map->m_flags |= EXT4_MAP_UNWRITTEN;
1614 #ifdef ES_AGGRESSIVE_TEST
1615 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1621 * Try to see if we can get the block without requesting a new
1622 * file system block.
1624 down_read(&EXT4_I(inode)->i_data_sem);
1625 if (ext4_has_inline_data(inode))
1627 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1628 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1630 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1636 * XXX: __block_prepare_write() unmaps passed block,
1640 * If the block was allocated from previously allocated cluster,
1641 * then we don't need to reserve it again. However we still need
1642 * to reserve metadata for every block we're going to write.
1644 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1645 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1646 ret = ext4_da_reserve_space(inode);
1648 /* not enough space to reserve */
1654 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1655 ~0, EXTENT_STATUS_DELAYED);
1661 map_bh(bh, inode->i_sb, invalid_block);
1663 set_buffer_delay(bh);
1664 } else if (retval > 0) {
1666 unsigned int status;
1668 if (unlikely(retval != map->m_len)) {
1669 ext4_warning(inode->i_sb,
1670 "ES len assertion failed for inode "
1671 "%lu: retval %d != map->m_len %d",
1672 inode->i_ino, retval, map->m_len);
1676 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1677 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1678 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1679 map->m_pblk, status);
1685 up_read((&EXT4_I(inode)->i_data_sem));
1691 * This is a special get_block_t callback which is used by
1692 * ext4_da_write_begin(). It will either return mapped block or
1693 * reserve space for a single block.
1695 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1696 * We also have b_blocknr = -1 and b_bdev initialized properly
1698 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1699 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1700 * initialized properly.
1702 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1703 struct buffer_head *bh, int create)
1705 struct ext4_map_blocks map;
1708 BUG_ON(create == 0);
1709 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1711 map.m_lblk = iblock;
1715 * first, we need to know whether the block is allocated already
1716 * preallocated blocks are unmapped but should treated
1717 * the same as allocated blocks.
1719 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1723 map_bh(bh, inode->i_sb, map.m_pblk);
1724 ext4_update_bh_state(bh, map.m_flags);
1726 if (buffer_unwritten(bh)) {
1727 /* A delayed write to unwritten bh should be marked
1728 * new and mapped. Mapped ensures that we don't do
1729 * get_block multiple times when we write to the same
1730 * offset and new ensures that we do proper zero out
1731 * for partial write.
1734 set_buffer_mapped(bh);
1739 static int bget_one(handle_t *handle, struct buffer_head *bh)
1745 static int bput_one(handle_t *handle, struct buffer_head *bh)
1751 static int __ext4_journalled_writepage(struct page *page,
1754 struct address_space *mapping = page->mapping;
1755 struct inode *inode = mapping->host;
1756 struct buffer_head *page_bufs = NULL;
1757 handle_t *handle = NULL;
1758 int ret = 0, err = 0;
1759 int inline_data = ext4_has_inline_data(inode);
1760 struct buffer_head *inode_bh = NULL;
1762 ClearPageChecked(page);
1765 BUG_ON(page->index != 0);
1766 BUG_ON(len > ext4_get_max_inline_size(inode));
1767 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1768 if (inode_bh == NULL)
1771 page_bufs = page_buffers(page);
1776 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1780 * We need to release the page lock before we start the
1781 * journal, so grab a reference so the page won't disappear
1782 * out from under us.
1787 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1788 ext4_writepage_trans_blocks(inode));
1789 if (IS_ERR(handle)) {
1790 ret = PTR_ERR(handle);
1792 goto out_no_pagelock;
1794 BUG_ON(!ext4_handle_valid(handle));
1798 if (page->mapping != mapping) {
1799 /* The page got truncated from under us */
1800 ext4_journal_stop(handle);
1806 BUFFER_TRACE(inode_bh, "get write access");
1807 ret = ext4_journal_get_write_access(handle, inode_bh);
1809 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1812 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1813 do_journal_get_write_access);
1815 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1820 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1821 err = ext4_journal_stop(handle);
1825 if (!ext4_has_inline_data(inode))
1826 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1828 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1837 * Note that we don't need to start a transaction unless we're journaling data
1838 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1839 * need to file the inode to the transaction's list in ordered mode because if
1840 * we are writing back data added by write(), the inode is already there and if
1841 * we are writing back data modified via mmap(), no one guarantees in which
1842 * transaction the data will hit the disk. In case we are journaling data, we
1843 * cannot start transaction directly because transaction start ranks above page
1844 * lock so we have to do some magic.
1846 * This function can get called via...
1847 * - ext4_writepages after taking page lock (have journal handle)
1848 * - journal_submit_inode_data_buffers (no journal handle)
1849 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1850 * - grab_page_cache when doing write_begin (have journal handle)
1852 * We don't do any block allocation in this function. If we have page with
1853 * multiple blocks we need to write those buffer_heads that are mapped. This
1854 * is important for mmaped based write. So if we do with blocksize 1K
1855 * truncate(f, 1024);
1856 * a = mmap(f, 0, 4096);
1858 * truncate(f, 4096);
1859 * we have in the page first buffer_head mapped via page_mkwrite call back
1860 * but other buffer_heads would be unmapped but dirty (dirty done via the
1861 * do_wp_page). So writepage should write the first block. If we modify
1862 * the mmap area beyond 1024 we will again get a page_fault and the
1863 * page_mkwrite callback will do the block allocation and mark the
1864 * buffer_heads mapped.
1866 * We redirty the page if we have any buffer_heads that is either delay or
1867 * unwritten in the page.
1869 * We can get recursively called as show below.
1871 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1874 * But since we don't do any block allocation we should not deadlock.
1875 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1877 static int ext4_writepage(struct page *page,
1878 struct writeback_control *wbc)
1883 struct buffer_head *page_bufs = NULL;
1884 struct inode *inode = page->mapping->host;
1885 struct ext4_io_submit io_submit;
1886 bool keep_towrite = false;
1888 trace_ext4_writepage(page);
1889 size = i_size_read(inode);
1890 if (page->index == size >> PAGE_CACHE_SHIFT)
1891 len = size & ~PAGE_CACHE_MASK;
1893 len = PAGE_CACHE_SIZE;
1895 page_bufs = page_buffers(page);
1897 * We cannot do block allocation or other extent handling in this
1898 * function. If there are buffers needing that, we have to redirty
1899 * the page. But we may reach here when we do a journal commit via
1900 * journal_submit_inode_data_buffers() and in that case we must write
1901 * allocated buffers to achieve data=ordered mode guarantees.
1903 * Also, if there is only one buffer per page (the fs block
1904 * size == the page size), if one buffer needs block
1905 * allocation or needs to modify the extent tree to clear the
1906 * unwritten flag, we know that the page can't be written at
1907 * all, so we might as well refuse the write immediately.
1908 * Unfortunately if the block size != page size, we can't as
1909 * easily detect this case using ext4_walk_page_buffers(), but
1910 * for the extremely common case, this is an optimization that
1911 * skips a useless round trip through ext4_bio_write_page().
1913 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1914 ext4_bh_delay_or_unwritten)) {
1915 redirty_page_for_writepage(wbc, page);
1916 if ((current->flags & PF_MEMALLOC) ||
1917 (inode->i_sb->s_blocksize == PAGE_CACHE_SIZE)) {
1919 * For memory cleaning there's no point in writing only
1920 * some buffers. So just bail out. Warn if we came here
1921 * from direct reclaim.
1923 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1928 keep_towrite = true;
1931 if (PageChecked(page) && ext4_should_journal_data(inode))
1933 * It's mmapped pagecache. Add buffers and journal it. There
1934 * doesn't seem much point in redirtying the page here.
1936 return __ext4_journalled_writepage(page, len);
1938 ext4_io_submit_init(&io_submit, wbc);
1939 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1940 if (!io_submit.io_end) {
1941 redirty_page_for_writepage(wbc, page);
1945 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1946 ext4_io_submit(&io_submit);
1947 /* Drop io_end reference we got from init */
1948 ext4_put_io_end_defer(io_submit.io_end);
1952 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1955 loff_t size = i_size_read(mpd->inode);
1958 BUG_ON(page->index != mpd->first_page);
1959 if (page->index == size >> PAGE_CACHE_SHIFT)
1960 len = size & ~PAGE_CACHE_MASK;
1962 len = PAGE_CACHE_SIZE;
1963 clear_page_dirty_for_io(page);
1964 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1966 mpd->wbc->nr_to_write--;
1972 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1975 * mballoc gives us at most this number of blocks...
1976 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1977 * The rest of mballoc seems to handle chunks up to full group size.
1979 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1982 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1984 * @mpd - extent of blocks
1985 * @lblk - logical number of the block in the file
1986 * @bh - buffer head we want to add to the extent
1988 * The function is used to collect contig. blocks in the same state. If the
1989 * buffer doesn't require mapping for writeback and we haven't started the
1990 * extent of buffers to map yet, the function returns 'true' immediately - the
1991 * caller can write the buffer right away. Otherwise the function returns true
1992 * if the block has been added to the extent, false if the block couldn't be
1995 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1996 struct buffer_head *bh)
1998 struct ext4_map_blocks *map = &mpd->map;
2000 /* Buffer that doesn't need mapping for writeback? */
2001 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2002 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2003 /* So far no extent to map => we write the buffer right away */
2004 if (map->m_len == 0)
2009 /* First block in the extent? */
2010 if (map->m_len == 0) {
2013 map->m_flags = bh->b_state & BH_FLAGS;
2017 /* Don't go larger than mballoc is willing to allocate */
2018 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2021 /* Can we merge the block to our big extent? */
2022 if (lblk == map->m_lblk + map->m_len &&
2023 (bh->b_state & BH_FLAGS) == map->m_flags) {
2031 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2033 * @mpd - extent of blocks for mapping
2034 * @head - the first buffer in the page
2035 * @bh - buffer we should start processing from
2036 * @lblk - logical number of the block in the file corresponding to @bh
2038 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2039 * the page for IO if all buffers in this page were mapped and there's no
2040 * accumulated extent of buffers to map or add buffers in the page to the
2041 * extent of buffers to map. The function returns 1 if the caller can continue
2042 * by processing the next page, 0 if it should stop adding buffers to the
2043 * extent to map because we cannot extend it anymore. It can also return value
2044 * < 0 in case of error during IO submission.
2046 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2047 struct buffer_head *head,
2048 struct buffer_head *bh,
2051 struct inode *inode = mpd->inode;
2053 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2054 >> inode->i_blkbits;
2057 BUG_ON(buffer_locked(bh));
2059 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2060 /* Found extent to map? */
2063 /* Everything mapped so far and we hit EOF */
2066 } while (lblk++, (bh = bh->b_this_page) != head);
2067 /* So far everything mapped? Submit the page for IO. */
2068 if (mpd->map.m_len == 0) {
2069 err = mpage_submit_page(mpd, head->b_page);
2073 return lblk < blocks;
2077 * mpage_map_buffers - update buffers corresponding to changed extent and
2078 * submit fully mapped pages for IO
2080 * @mpd - description of extent to map, on return next extent to map
2082 * Scan buffers corresponding to changed extent (we expect corresponding pages
2083 * to be already locked) and update buffer state according to new extent state.
2084 * We map delalloc buffers to their physical location, clear unwritten bits,
2085 * and mark buffers as uninit when we perform writes to unwritten extents
2086 * and do extent conversion after IO is finished. If the last page is not fully
2087 * mapped, we update @map to the next extent in the last page that needs
2088 * mapping. Otherwise we submit the page for IO.
2090 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2092 struct pagevec pvec;
2094 struct inode *inode = mpd->inode;
2095 struct buffer_head *head, *bh;
2096 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2102 start = mpd->map.m_lblk >> bpp_bits;
2103 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2104 lblk = start << bpp_bits;
2105 pblock = mpd->map.m_pblk;
2107 pagevec_init(&pvec, 0);
2108 while (start <= end) {
2109 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2113 for (i = 0; i < nr_pages; i++) {
2114 struct page *page = pvec.pages[i];
2116 if (page->index > end)
2118 /* Up to 'end' pages must be contiguous */
2119 BUG_ON(page->index != start);
2120 bh = head = page_buffers(page);
2122 if (lblk < mpd->map.m_lblk)
2124 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2126 * Buffer after end of mapped extent.
2127 * Find next buffer in the page to map.
2130 mpd->map.m_flags = 0;
2132 * FIXME: If dioread_nolock supports
2133 * blocksize < pagesize, we need to make
2134 * sure we add size mapped so far to
2135 * io_end->size as the following call
2136 * can submit the page for IO.
2138 err = mpage_process_page_bufs(mpd, head,
2140 pagevec_release(&pvec);
2145 if (buffer_delay(bh)) {
2146 clear_buffer_delay(bh);
2147 bh->b_blocknr = pblock++;
2149 clear_buffer_unwritten(bh);
2150 } while (lblk++, (bh = bh->b_this_page) != head);
2153 * FIXME: This is going to break if dioread_nolock
2154 * supports blocksize < pagesize as we will try to
2155 * convert potentially unmapped parts of inode.
2157 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2158 /* Page fully mapped - let IO run! */
2159 err = mpage_submit_page(mpd, page);
2161 pagevec_release(&pvec);
2166 pagevec_release(&pvec);
2168 /* Extent fully mapped and matches with page boundary. We are done. */
2170 mpd->map.m_flags = 0;
2174 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2176 struct inode *inode = mpd->inode;
2177 struct ext4_map_blocks *map = &mpd->map;
2178 int get_blocks_flags;
2179 int err, dioread_nolock;
2181 trace_ext4_da_write_pages_extent(inode, map);
2183 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2184 * to convert an unwritten extent to be initialized (in the case
2185 * where we have written into one or more preallocated blocks). It is
2186 * possible that we're going to need more metadata blocks than
2187 * previously reserved. However we must not fail because we're in
2188 * writeback and there is nothing we can do about it so it might result
2189 * in data loss. So use reserved blocks to allocate metadata if
2192 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2193 * the blocks in question are delalloc blocks. This indicates
2194 * that the blocks and quotas has already been checked when
2195 * the data was copied into the page cache.
2197 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2198 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2199 dioread_nolock = ext4_should_dioread_nolock(inode);
2201 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2202 if (map->m_flags & (1 << BH_Delay))
2203 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2205 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2208 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2209 if (!mpd->io_submit.io_end->handle &&
2210 ext4_handle_valid(handle)) {
2211 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2212 handle->h_rsv_handle = NULL;
2214 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2217 BUG_ON(map->m_len == 0);
2218 if (map->m_flags & EXT4_MAP_NEW) {
2219 struct block_device *bdev = inode->i_sb->s_bdev;
2222 for (i = 0; i < map->m_len; i++)
2223 unmap_underlying_metadata(bdev, map->m_pblk + i);
2229 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2230 * mpd->len and submit pages underlying it for IO
2232 * @handle - handle for journal operations
2233 * @mpd - extent to map
2234 * @give_up_on_write - we set this to true iff there is a fatal error and there
2235 * is no hope of writing the data. The caller should discard
2236 * dirty pages to avoid infinite loops.
2238 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2239 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2240 * them to initialized or split the described range from larger unwritten
2241 * extent. Note that we need not map all the described range since allocation
2242 * can return less blocks or the range is covered by more unwritten extents. We
2243 * cannot map more because we are limited by reserved transaction credits. On
2244 * the other hand we always make sure that the last touched page is fully
2245 * mapped so that it can be written out (and thus forward progress is
2246 * guaranteed). After mapping we submit all mapped pages for IO.
2248 static int mpage_map_and_submit_extent(handle_t *handle,
2249 struct mpage_da_data *mpd,
2250 bool *give_up_on_write)
2252 struct inode *inode = mpd->inode;
2253 struct ext4_map_blocks *map = &mpd->map;
2258 mpd->io_submit.io_end->offset =
2259 ((loff_t)map->m_lblk) << inode->i_blkbits;
2261 err = mpage_map_one_extent(handle, mpd);
2263 struct super_block *sb = inode->i_sb;
2265 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2266 goto invalidate_dirty_pages;
2268 * Let the uper layers retry transient errors.
2269 * In the case of ENOSPC, if ext4_count_free_blocks()
2270 * is non-zero, a commit should free up blocks.
2272 if ((err == -ENOMEM) ||
2273 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2275 goto update_disksize;
2278 ext4_msg(sb, KERN_CRIT,
2279 "Delayed block allocation failed for "
2280 "inode %lu at logical offset %llu with"
2281 " max blocks %u with error %d",
2283 (unsigned long long)map->m_lblk,
2284 (unsigned)map->m_len, -err);
2285 ext4_msg(sb, KERN_CRIT,
2286 "This should not happen!! Data will "
2289 ext4_print_free_blocks(inode);
2290 invalidate_dirty_pages:
2291 *give_up_on_write = true;
2296 * Update buffer state, submit mapped pages, and get us new
2299 err = mpage_map_and_submit_buffers(mpd);
2301 goto update_disksize;
2302 } while (map->m_len);
2306 * Update on-disk size after IO is submitted. Races with
2307 * truncate are avoided by checking i_size under i_data_sem.
2309 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2310 if (disksize > EXT4_I(inode)->i_disksize) {
2314 down_write(&EXT4_I(inode)->i_data_sem);
2315 i_size = i_size_read(inode);
2316 if (disksize > i_size)
2318 if (disksize > EXT4_I(inode)->i_disksize)
2319 EXT4_I(inode)->i_disksize = disksize;
2320 err2 = ext4_mark_inode_dirty(handle, inode);
2321 up_write(&EXT4_I(inode)->i_data_sem);
2323 ext4_error(inode->i_sb,
2324 "Failed to mark inode %lu dirty",
2333 * Calculate the total number of credits to reserve for one writepages
2334 * iteration. This is called from ext4_writepages(). We map an extent of
2335 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2336 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2337 * bpp - 1 blocks in bpp different extents.
2339 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2341 int bpp = ext4_journal_blocks_per_page(inode);
2343 return ext4_meta_trans_blocks(inode,
2344 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2348 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2349 * and underlying extent to map
2351 * @mpd - where to look for pages
2353 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2354 * IO immediately. When we find a page which isn't mapped we start accumulating
2355 * extent of buffers underlying these pages that needs mapping (formed by
2356 * either delayed or unwritten buffers). We also lock the pages containing
2357 * these buffers. The extent found is returned in @mpd structure (starting at
2358 * mpd->lblk with length mpd->len blocks).
2360 * Note that this function can attach bios to one io_end structure which are
2361 * neither logically nor physically contiguous. Although it may seem as an
2362 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2363 * case as we need to track IO to all buffers underlying a page in one io_end.
2365 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2367 struct address_space *mapping = mpd->inode->i_mapping;
2368 struct pagevec pvec;
2369 unsigned int nr_pages;
2370 long left = mpd->wbc->nr_to_write;
2371 pgoff_t index = mpd->first_page;
2372 pgoff_t end = mpd->last_page;
2375 int blkbits = mpd->inode->i_blkbits;
2377 struct buffer_head *head;
2379 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2380 tag = PAGECACHE_TAG_TOWRITE;
2382 tag = PAGECACHE_TAG_DIRTY;
2384 pagevec_init(&pvec, 0);
2386 mpd->next_page = index;
2387 while (index <= end) {
2388 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2389 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2393 for (i = 0; i < nr_pages; i++) {
2394 struct page *page = pvec.pages[i];
2397 * At this point, the page may be truncated or
2398 * invalidated (changing page->mapping to NULL), or
2399 * even swizzled back from swapper_space to tmpfs file
2400 * mapping. However, page->index will not change
2401 * because we have a reference on the page.
2403 if (page->index > end)
2407 * Accumulated enough dirty pages? This doesn't apply
2408 * to WB_SYNC_ALL mode. For integrity sync we have to
2409 * keep going because someone may be concurrently
2410 * dirtying pages, and we might have synced a lot of
2411 * newly appeared dirty pages, but have not synced all
2412 * of the old dirty pages.
2414 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2417 /* If we can't merge this page, we are done. */
2418 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2423 * If the page is no longer dirty, or its mapping no
2424 * longer corresponds to inode we are writing (which
2425 * means it has been truncated or invalidated), or the
2426 * page is already under writeback and we are not doing
2427 * a data integrity writeback, skip the page
2429 if (!PageDirty(page) ||
2430 (PageWriteback(page) &&
2431 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2432 unlikely(page->mapping != mapping)) {
2437 wait_on_page_writeback(page);
2438 BUG_ON(PageWriteback(page));
2440 if (mpd->map.m_len == 0)
2441 mpd->first_page = page->index;
2442 mpd->next_page = page->index + 1;
2443 /* Add all dirty buffers to mpd */
2444 lblk = ((ext4_lblk_t)page->index) <<
2445 (PAGE_CACHE_SHIFT - blkbits);
2446 head = page_buffers(page);
2447 err = mpage_process_page_bufs(mpd, head, head, lblk);
2453 pagevec_release(&pvec);
2458 pagevec_release(&pvec);
2462 static int __writepage(struct page *page, struct writeback_control *wbc,
2465 struct address_space *mapping = data;
2466 int ret = ext4_writepage(page, wbc);
2467 mapping_set_error(mapping, ret);
2471 static int ext4_writepages(struct address_space *mapping,
2472 struct writeback_control *wbc)
2474 pgoff_t writeback_index = 0;
2475 long nr_to_write = wbc->nr_to_write;
2476 int range_whole = 0;
2478 handle_t *handle = NULL;
2479 struct mpage_da_data mpd;
2480 struct inode *inode = mapping->host;
2481 int needed_blocks, rsv_blocks = 0, ret = 0;
2482 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2484 struct blk_plug plug;
2485 bool give_up_on_write = false;
2487 trace_ext4_writepages(inode, wbc);
2490 * No pages to write? This is mainly a kludge to avoid starting
2491 * a transaction for special inodes like journal inode on last iput()
2492 * because that could violate lock ordering on umount
2494 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2495 goto out_writepages;
2497 if (ext4_should_journal_data(inode)) {
2498 struct blk_plug plug;
2500 blk_start_plug(&plug);
2501 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2502 blk_finish_plug(&plug);
2503 goto out_writepages;
2507 * If the filesystem has aborted, it is read-only, so return
2508 * right away instead of dumping stack traces later on that
2509 * will obscure the real source of the problem. We test
2510 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2511 * the latter could be true if the filesystem is mounted
2512 * read-only, and in that case, ext4_writepages should
2513 * *never* be called, so if that ever happens, we would want
2516 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2518 goto out_writepages;
2521 if (ext4_should_dioread_nolock(inode)) {
2523 * We may need to convert up to one extent per block in
2524 * the page and we may dirty the inode.
2526 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2530 * If we have inline data and arrive here, it means that
2531 * we will soon create the block for the 1st page, so
2532 * we'd better clear the inline data here.
2534 if (ext4_has_inline_data(inode)) {
2535 /* Just inode will be modified... */
2536 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2537 if (IS_ERR(handle)) {
2538 ret = PTR_ERR(handle);
2539 goto out_writepages;
2541 BUG_ON(ext4_test_inode_state(inode,
2542 EXT4_STATE_MAY_INLINE_DATA));
2543 ext4_destroy_inline_data(handle, inode);
2544 ext4_journal_stop(handle);
2547 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2550 if (wbc->range_cyclic) {
2551 writeback_index = mapping->writeback_index;
2552 if (writeback_index)
2554 mpd.first_page = writeback_index;
2557 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2558 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2563 ext4_io_submit_init(&mpd.io_submit, wbc);
2565 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2566 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2568 blk_start_plug(&plug);
2569 while (!done && mpd.first_page <= mpd.last_page) {
2570 /* For each extent of pages we use new io_end */
2571 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2572 if (!mpd.io_submit.io_end) {
2578 * We have two constraints: We find one extent to map and we
2579 * must always write out whole page (makes a difference when
2580 * blocksize < pagesize) so that we don't block on IO when we
2581 * try to write out the rest of the page. Journalled mode is
2582 * not supported by delalloc.
2584 BUG_ON(ext4_should_journal_data(inode));
2585 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2587 /* start a new transaction */
2588 handle = ext4_journal_start_with_reserve(inode,
2589 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2590 if (IS_ERR(handle)) {
2591 ret = PTR_ERR(handle);
2592 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2593 "%ld pages, ino %lu; err %d", __func__,
2594 wbc->nr_to_write, inode->i_ino, ret);
2595 /* Release allocated io_end */
2596 ext4_put_io_end(mpd.io_submit.io_end);
2600 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2601 ret = mpage_prepare_extent_to_map(&mpd);
2604 ret = mpage_map_and_submit_extent(handle, &mpd,
2608 * We scanned the whole range (or exhausted
2609 * nr_to_write), submitted what was mapped and
2610 * didn't find anything needing mapping. We are
2617 * Caution: If the handle is synchronous,
2618 * ext4_journal_stop() can wait for transaction commit
2619 * to finish which may depend on writeback of pages to
2620 * complete or on page lock to be released. In that
2621 * case, we have to wait until after after we have
2622 * submitted all the IO, released page locks we hold,
2623 * and dropped io_end reference (for extent conversion
2624 * to be able to complete) before stopping the handle.
2626 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2627 ext4_journal_stop(handle);
2630 /* Submit prepared bio */
2631 ext4_io_submit(&mpd.io_submit);
2632 /* Unlock pages we didn't use */
2633 mpage_release_unused_pages(&mpd, give_up_on_write);
2635 * Drop our io_end reference we got from init. We have
2636 * to be careful and use deferred io_end finishing if
2637 * we are still holding the transaction as we can
2638 * release the last reference to io_end which may end
2639 * up doing unwritten extent conversion.
2642 ext4_put_io_end_defer(mpd.io_submit.io_end);
2643 ext4_journal_stop(handle);
2645 ext4_put_io_end(mpd.io_submit.io_end);
2647 if (ret == -ENOSPC && sbi->s_journal) {
2649 * Commit the transaction which would
2650 * free blocks released in the transaction
2653 jbd2_journal_force_commit_nested(sbi->s_journal);
2657 /* Fatal error - ENOMEM, EIO... */
2661 blk_finish_plug(&plug);
2662 if (!ret && !cycled && wbc->nr_to_write > 0) {
2664 mpd.last_page = writeback_index - 1;
2670 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2672 * Set the writeback_index so that range_cyclic
2673 * mode will write it back later
2675 mapping->writeback_index = mpd.first_page;
2678 trace_ext4_writepages_result(inode, wbc, ret,
2679 nr_to_write - wbc->nr_to_write);
2683 static int ext4_nonda_switch(struct super_block *sb)
2685 s64 free_clusters, dirty_clusters;
2686 struct ext4_sb_info *sbi = EXT4_SB(sb);
2689 * switch to non delalloc mode if we are running low
2690 * on free block. The free block accounting via percpu
2691 * counters can get slightly wrong with percpu_counter_batch getting
2692 * accumulated on each CPU without updating global counters
2693 * Delalloc need an accurate free block accounting. So switch
2694 * to non delalloc when we are near to error range.
2697 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2699 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2701 * Start pushing delalloc when 1/2 of free blocks are dirty.
2703 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2704 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2706 if (2 * free_clusters < 3 * dirty_clusters ||
2707 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2709 * free block count is less than 150% of dirty blocks
2710 * or free blocks is less than watermark
2717 /* We always reserve for an inode update; the superblock could be there too */
2718 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2720 if (likely(ext4_has_feature_large_file(inode->i_sb)))
2723 if (pos + len <= 0x7fffffffULL)
2726 /* We might need to update the superblock to set LARGE_FILE */
2730 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2731 loff_t pos, unsigned len, unsigned flags,
2732 struct page **pagep, void **fsdata)
2734 int ret, retries = 0;
2737 struct inode *inode = mapping->host;
2740 index = pos >> PAGE_CACHE_SHIFT;
2742 if (ext4_nonda_switch(inode->i_sb)) {
2743 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2744 return ext4_write_begin(file, mapping, pos,
2745 len, flags, pagep, fsdata);
2747 *fsdata = (void *)0;
2748 trace_android_fs_datawrite_start(inode, pos, len,
2749 current->pid, current->comm);
2750 trace_ext4_da_write_begin(inode, pos, len, flags);
2752 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2753 ret = ext4_da_write_inline_data_begin(mapping, inode,
2763 * grab_cache_page_write_begin() can take a long time if the
2764 * system is thrashing due to memory pressure, or if the page
2765 * is being written back. So grab it first before we start
2766 * the transaction handle. This also allows us to allocate
2767 * the page (if needed) without using GFP_NOFS.
2770 page = grab_cache_page_write_begin(mapping, index, flags);
2776 * With delayed allocation, we don't log the i_disksize update
2777 * if there is delayed block allocation. But we still need
2778 * to journalling the i_disksize update if writes to the end
2779 * of file which has an already mapped buffer.
2782 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2783 ext4_da_write_credits(inode, pos, len));
2784 if (IS_ERR(handle)) {
2785 page_cache_release(page);
2786 return PTR_ERR(handle);
2790 if (page->mapping != mapping) {
2791 /* The page got truncated from under us */
2793 page_cache_release(page);
2794 ext4_journal_stop(handle);
2797 /* In case writeback began while the page was unlocked */
2798 wait_for_stable_page(page);
2800 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2801 ret = ext4_block_write_begin(page, pos, len,
2802 ext4_da_get_block_prep);
2804 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2808 ext4_journal_stop(handle);
2810 * block_write_begin may have instantiated a few blocks
2811 * outside i_size. Trim these off again. Don't need
2812 * i_size_read because we hold i_mutex.
2814 if (pos + len > inode->i_size)
2815 ext4_truncate_failed_write(inode);
2817 if (ret == -ENOSPC &&
2818 ext4_should_retry_alloc(inode->i_sb, &retries))
2821 page_cache_release(page);
2830 * Check if we should update i_disksize
2831 * when write to the end of file but not require block allocation
2833 static int ext4_da_should_update_i_disksize(struct page *page,
2834 unsigned long offset)
2836 struct buffer_head *bh;
2837 struct inode *inode = page->mapping->host;
2841 bh = page_buffers(page);
2842 idx = offset >> inode->i_blkbits;
2844 for (i = 0; i < idx; i++)
2845 bh = bh->b_this_page;
2847 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2852 static int ext4_da_write_end(struct file *file,
2853 struct address_space *mapping,
2854 loff_t pos, unsigned len, unsigned copied,
2855 struct page *page, void *fsdata)
2857 struct inode *inode = mapping->host;
2859 handle_t *handle = ext4_journal_current_handle();
2861 unsigned long start, end;
2862 int write_mode = (int)(unsigned long)fsdata;
2864 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2865 return ext4_write_end(file, mapping, pos,
2866 len, copied, page, fsdata);
2868 trace_android_fs_datawrite_end(inode, pos, len);
2869 trace_ext4_da_write_end(inode, pos, len, copied);
2870 start = pos & (PAGE_CACHE_SIZE - 1);
2871 end = start + copied - 1;
2874 * generic_write_end() will run mark_inode_dirty() if i_size
2875 * changes. So let's piggyback the i_disksize mark_inode_dirty
2878 new_i_size = pos + copied;
2879 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2880 if (ext4_has_inline_data(inode) ||
2881 ext4_da_should_update_i_disksize(page, end)) {
2882 ext4_update_i_disksize(inode, new_i_size);
2883 /* We need to mark inode dirty even if
2884 * new_i_size is less that inode->i_size
2885 * bu greater than i_disksize.(hint delalloc)
2887 ext4_mark_inode_dirty(handle, inode);
2891 if (write_mode != CONVERT_INLINE_DATA &&
2892 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2893 ext4_has_inline_data(inode))
2894 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2897 ret2 = generic_write_end(file, mapping, pos, len, copied,
2903 ret2 = ext4_journal_stop(handle);
2907 return ret ? ret : copied;
2910 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2911 unsigned int length)
2914 * Drop reserved blocks
2916 BUG_ON(!PageLocked(page));
2917 if (!page_has_buffers(page))
2920 ext4_da_page_release_reservation(page, offset, length);
2923 ext4_invalidatepage(page, offset, length);
2929 * Force all delayed allocation blocks to be allocated for a given inode.
2931 int ext4_alloc_da_blocks(struct inode *inode)
2933 trace_ext4_alloc_da_blocks(inode);
2935 if (!EXT4_I(inode)->i_reserved_data_blocks)
2939 * We do something simple for now. The filemap_flush() will
2940 * also start triggering a write of the data blocks, which is
2941 * not strictly speaking necessary (and for users of
2942 * laptop_mode, not even desirable). However, to do otherwise
2943 * would require replicating code paths in:
2945 * ext4_writepages() ->
2946 * write_cache_pages() ---> (via passed in callback function)
2947 * __mpage_da_writepage() -->
2948 * mpage_add_bh_to_extent()
2949 * mpage_da_map_blocks()
2951 * The problem is that write_cache_pages(), located in
2952 * mm/page-writeback.c, marks pages clean in preparation for
2953 * doing I/O, which is not desirable if we're not planning on
2956 * We could call write_cache_pages(), and then redirty all of
2957 * the pages by calling redirty_page_for_writepage() but that
2958 * would be ugly in the extreme. So instead we would need to
2959 * replicate parts of the code in the above functions,
2960 * simplifying them because we wouldn't actually intend to
2961 * write out the pages, but rather only collect contiguous
2962 * logical block extents, call the multi-block allocator, and
2963 * then update the buffer heads with the block allocations.
2965 * For now, though, we'll cheat by calling filemap_flush(),
2966 * which will map the blocks, and start the I/O, but not
2967 * actually wait for the I/O to complete.
2969 return filemap_flush(inode->i_mapping);
2973 * bmap() is special. It gets used by applications such as lilo and by
2974 * the swapper to find the on-disk block of a specific piece of data.
2976 * Naturally, this is dangerous if the block concerned is still in the
2977 * journal. If somebody makes a swapfile on an ext4 data-journaling
2978 * filesystem and enables swap, then they may get a nasty shock when the
2979 * data getting swapped to that swapfile suddenly gets overwritten by
2980 * the original zero's written out previously to the journal and
2981 * awaiting writeback in the kernel's buffer cache.
2983 * So, if we see any bmap calls here on a modified, data-journaled file,
2984 * take extra steps to flush any blocks which might be in the cache.
2986 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2988 struct inode *inode = mapping->host;
2993 * We can get here for an inline file via the FIBMAP ioctl
2995 if (ext4_has_inline_data(inode))
2998 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2999 test_opt(inode->i_sb, DELALLOC)) {
3001 * With delalloc we want to sync the file
3002 * so that we can make sure we allocate
3005 filemap_write_and_wait(mapping);
3008 if (EXT4_JOURNAL(inode) &&
3009 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3011 * This is a REALLY heavyweight approach, but the use of
3012 * bmap on dirty files is expected to be extremely rare:
3013 * only if we run lilo or swapon on a freshly made file
3014 * do we expect this to happen.
3016 * (bmap requires CAP_SYS_RAWIO so this does not
3017 * represent an unprivileged user DOS attack --- we'd be
3018 * in trouble if mortal users could trigger this path at
3021 * NB. EXT4_STATE_JDATA is not set on files other than
3022 * regular files. If somebody wants to bmap a directory
3023 * or symlink and gets confused because the buffer
3024 * hasn't yet been flushed to disk, they deserve
3025 * everything they get.
3028 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3029 journal = EXT4_JOURNAL(inode);
3030 jbd2_journal_lock_updates(journal);
3031 err = jbd2_journal_flush(journal);
3032 jbd2_journal_unlock_updates(journal);
3038 return generic_block_bmap(mapping, block, ext4_get_block);
3041 static int ext4_readpage(struct file *file, struct page *page)
3044 struct inode *inode = page->mapping->host;
3046 trace_ext4_readpage(page);
3048 if (ext4_has_inline_data(inode))
3049 ret = ext4_readpage_inline(inode, page);
3052 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3058 ext4_readpages(struct file *file, struct address_space *mapping,
3059 struct list_head *pages, unsigned nr_pages)
3061 struct inode *inode = mapping->host;
3063 /* If the file has inline data, no need to do readpages. */
3064 if (ext4_has_inline_data(inode))
3067 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3070 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3071 unsigned int length)
3073 trace_ext4_invalidatepage(page, offset, length);
3075 /* No journalling happens on data buffers when this function is used */
3076 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3078 block_invalidatepage(page, offset, length);
3081 static int __ext4_journalled_invalidatepage(struct page *page,
3082 unsigned int offset,
3083 unsigned int length)
3085 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3087 trace_ext4_journalled_invalidatepage(page, offset, length);
3090 * If it's a full truncate we just forget about the pending dirtying
3092 if (offset == 0 && length == PAGE_CACHE_SIZE)
3093 ClearPageChecked(page);
3095 return jbd2_journal_invalidatepage(journal, page, offset, length);
3098 /* Wrapper for aops... */
3099 static void ext4_journalled_invalidatepage(struct page *page,
3100 unsigned int offset,
3101 unsigned int length)
3103 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3106 static int ext4_releasepage(struct page *page, gfp_t wait)
3108 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3110 trace_ext4_releasepage(page);
3112 /* Page has dirty journalled data -> cannot release */
3113 if (PageChecked(page))
3116 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3118 return try_to_free_buffers(page);
3122 * ext4_get_block used when preparing for a DIO write or buffer write.
3123 * We allocate an uinitialized extent if blocks haven't been allocated.
3124 * The extent will be converted to initialized after the IO is complete.
3126 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3127 struct buffer_head *bh_result, int create)
3129 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3130 inode->i_ino, create);
3131 return _ext4_get_block(inode, iblock, bh_result,
3132 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3135 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3136 struct buffer_head *bh_result, int create)
3138 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3139 inode->i_ino, create);
3140 return _ext4_get_block(inode, iblock, bh_result,
3141 EXT4_GET_BLOCKS_NO_LOCK);
3144 int ext4_get_block_dax(struct inode *inode, sector_t iblock,
3145 struct buffer_head *bh_result, int create)
3147 int flags = EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_UNWRIT_EXT;
3149 flags |= EXT4_GET_BLOCKS_CREATE;
3150 ext4_debug("ext4_get_block_dax: inode %lu, create flag %d\n",
3151 inode->i_ino, create);
3152 return _ext4_get_block(inode, iblock, bh_result, flags);
3155 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3156 ssize_t size, void *private)
3158 ext4_io_end_t *io_end = iocb->private;
3160 /* if not async direct IO just return */
3164 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3165 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3166 iocb->private, io_end->inode->i_ino, iocb, offset,
3169 iocb->private = NULL;
3170 io_end->offset = offset;
3171 io_end->size = size;
3172 ext4_put_io_end(io_end);
3176 * For ext4 extent files, ext4 will do direct-io write to holes,
3177 * preallocated extents, and those write extend the file, no need to
3178 * fall back to buffered IO.
3180 * For holes, we fallocate those blocks, mark them as unwritten
3181 * If those blocks were preallocated, we mark sure they are split, but
3182 * still keep the range to write as unwritten.
3184 * The unwritten extents will be converted to written when DIO is completed.
3185 * For async direct IO, since the IO may still pending when return, we
3186 * set up an end_io call back function, which will do the conversion
3187 * when async direct IO completed.
3189 * If the O_DIRECT write will extend the file then add this inode to the
3190 * orphan list. So recovery will truncate it back to the original size
3191 * if the machine crashes during the write.
3194 static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3197 struct file *file = iocb->ki_filp;
3198 struct inode *inode = file->f_mapping->host;
3200 size_t count = iov_iter_count(iter);
3202 get_block_t *get_block_func = NULL;
3204 loff_t final_size = offset + count;
3205 ext4_io_end_t *io_end = NULL;
3207 /* Use the old path for reads and writes beyond i_size. */
3208 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3209 return ext4_ind_direct_IO(iocb, iter, offset);
3211 BUG_ON(iocb->private == NULL);
3214 * Make all waiters for direct IO properly wait also for extent
3215 * conversion. This also disallows race between truncate() and
3216 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3218 if (iov_iter_rw(iter) == WRITE)
3219 inode_dio_begin(inode);
3221 /* If we do a overwrite dio, i_mutex locking can be released */
3222 overwrite = *((int *)iocb->private);
3225 down_read(&EXT4_I(inode)->i_data_sem);
3226 mutex_unlock(&inode->i_mutex);
3230 * We could direct write to holes and fallocate.
3232 * Allocated blocks to fill the hole are marked as
3233 * unwritten to prevent parallel buffered read to expose
3234 * the stale data before DIO complete the data IO.
3236 * As to previously fallocated extents, ext4 get_block will
3237 * just simply mark the buffer mapped but still keep the
3238 * extents unwritten.
3240 * For non AIO case, we will convert those unwritten extents
3241 * to written after return back from blockdev_direct_IO.
3243 * For async DIO, the conversion needs to be deferred when the
3244 * IO is completed. The ext4 end_io callback function will be
3245 * called to take care of the conversion work. Here for async
3246 * case, we allocate an io_end structure to hook to the iocb.
3248 iocb->private = NULL;
3249 ext4_inode_aio_set(inode, NULL);
3250 if (!is_sync_kiocb(iocb)) {
3251 io_end = ext4_init_io_end(inode, GFP_NOFS);
3257 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3259 iocb->private = ext4_get_io_end(io_end);
3261 * we save the io structure for current async direct
3262 * IO, so that later ext4_map_blocks() could flag the
3263 * io structure whether there is a unwritten extents
3264 * needs to be converted when IO is completed.
3266 ext4_inode_aio_set(inode, io_end);
3270 get_block_func = ext4_get_block_write_nolock;
3272 get_block_func = ext4_get_block_write;
3273 dio_flags = DIO_LOCKING;
3275 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3276 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3279 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
3280 ext4_end_io_dio, dio_flags);
3282 ret = __blockdev_direct_IO(iocb, inode,
3283 inode->i_sb->s_bdev, iter, offset,
3285 ext4_end_io_dio, NULL, dio_flags);
3288 * Put our reference to io_end. This can free the io_end structure e.g.
3289 * in sync IO case or in case of error. It can even perform extent
3290 * conversion if all bios we submitted finished before we got here.
3291 * Note that in that case iocb->private can be already set to NULL
3295 ext4_inode_aio_set(inode, NULL);
3296 ext4_put_io_end(io_end);
3298 * When no IO was submitted ext4_end_io_dio() was not
3299 * called so we have to put iocb's reference.
3301 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3302 WARN_ON(iocb->private != io_end);
3303 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3304 ext4_put_io_end(io_end);
3305 iocb->private = NULL;
3308 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3309 EXT4_STATE_DIO_UNWRITTEN)) {
3312 * for non AIO case, since the IO is already
3313 * completed, we could do the conversion right here
3315 err = ext4_convert_unwritten_extents(NULL, inode,
3319 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3323 if (iov_iter_rw(iter) == WRITE)
3324 inode_dio_end(inode);
3325 /* take i_mutex locking again if we do a ovewrite dio */
3327 up_read(&EXT4_I(inode)->i_data_sem);
3328 mutex_lock(&inode->i_mutex);
3334 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3337 struct file *file = iocb->ki_filp;
3338 struct inode *inode = file->f_mapping->host;
3339 size_t count = iov_iter_count(iter);
3342 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3343 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3348 * If we are doing data journalling we don't support O_DIRECT
3350 if (ext4_should_journal_data(inode))
3353 /* Let buffer I/O handle the inline data case. */
3354 if (ext4_has_inline_data(inode))
3357 if (trace_android_fs_dataread_start_enabled() &&
3358 (iov_iter_rw(iter) == READ))
3359 trace_android_fs_dataread_start(inode, offset, count,
3362 if (trace_android_fs_datawrite_start_enabled() &&
3363 (iov_iter_rw(iter) == WRITE))
3364 trace_android_fs_datawrite_start(inode, offset, count,
3368 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3369 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3370 ret = ext4_ext_direct_IO(iocb, iter, offset);
3372 ret = ext4_ind_direct_IO(iocb, iter, offset);
3373 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3375 if (trace_android_fs_dataread_start_enabled() &&
3376 (iov_iter_rw(iter) == READ))
3377 trace_android_fs_dataread_end(inode, offset, count);
3378 if (trace_android_fs_datawrite_start_enabled() &&
3379 (iov_iter_rw(iter) == WRITE))
3380 trace_android_fs_datawrite_end(inode, offset, count);
3386 * Pages can be marked dirty completely asynchronously from ext4's journalling
3387 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3388 * much here because ->set_page_dirty is called under VFS locks. The page is
3389 * not necessarily locked.
3391 * We cannot just dirty the page and leave attached buffers clean, because the
3392 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3393 * or jbddirty because all the journalling code will explode.
3395 * So what we do is to mark the page "pending dirty" and next time writepage
3396 * is called, propagate that into the buffers appropriately.
3398 static int ext4_journalled_set_page_dirty(struct page *page)
3400 SetPageChecked(page);
3401 return __set_page_dirty_nobuffers(page);
3404 static const struct address_space_operations ext4_aops = {
3405 .readpage = ext4_readpage,
3406 .readpages = ext4_readpages,
3407 .writepage = ext4_writepage,
3408 .writepages = ext4_writepages,
3409 .write_begin = ext4_write_begin,
3410 .write_end = ext4_write_end,
3412 .invalidatepage = ext4_invalidatepage,
3413 .releasepage = ext4_releasepage,
3414 .direct_IO = ext4_direct_IO,
3415 .migratepage = buffer_migrate_page,
3416 .is_partially_uptodate = block_is_partially_uptodate,
3417 .error_remove_page = generic_error_remove_page,
3420 static const struct address_space_operations ext4_journalled_aops = {
3421 .readpage = ext4_readpage,
3422 .readpages = ext4_readpages,
3423 .writepage = ext4_writepage,
3424 .writepages = ext4_writepages,
3425 .write_begin = ext4_write_begin,
3426 .write_end = ext4_journalled_write_end,
3427 .set_page_dirty = ext4_journalled_set_page_dirty,
3429 .invalidatepage = ext4_journalled_invalidatepage,
3430 .releasepage = ext4_releasepage,
3431 .direct_IO = ext4_direct_IO,
3432 .is_partially_uptodate = block_is_partially_uptodate,
3433 .error_remove_page = generic_error_remove_page,
3436 static const struct address_space_operations ext4_da_aops = {
3437 .readpage = ext4_readpage,
3438 .readpages = ext4_readpages,
3439 .writepage = ext4_writepage,
3440 .writepages = ext4_writepages,
3441 .write_begin = ext4_da_write_begin,
3442 .write_end = ext4_da_write_end,
3444 .invalidatepage = ext4_da_invalidatepage,
3445 .releasepage = ext4_releasepage,
3446 .direct_IO = ext4_direct_IO,
3447 .migratepage = buffer_migrate_page,
3448 .is_partially_uptodate = block_is_partially_uptodate,
3449 .error_remove_page = generic_error_remove_page,
3452 void ext4_set_aops(struct inode *inode)
3454 switch (ext4_inode_journal_mode(inode)) {
3455 case EXT4_INODE_ORDERED_DATA_MODE:
3456 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3458 case EXT4_INODE_WRITEBACK_DATA_MODE:
3459 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3461 case EXT4_INODE_JOURNAL_DATA_MODE:
3462 inode->i_mapping->a_ops = &ext4_journalled_aops;
3467 if (test_opt(inode->i_sb, DELALLOC))
3468 inode->i_mapping->a_ops = &ext4_da_aops;
3470 inode->i_mapping->a_ops = &ext4_aops;
3473 static int __ext4_block_zero_page_range(handle_t *handle,
3474 struct address_space *mapping, loff_t from, loff_t length)
3476 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3477 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3478 unsigned blocksize, pos;
3480 struct inode *inode = mapping->host;
3481 struct buffer_head *bh;
3485 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3486 mapping_gfp_constraint(mapping, ~__GFP_FS));
3490 blocksize = inode->i_sb->s_blocksize;
3492 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3494 if (!page_has_buffers(page))
3495 create_empty_buffers(page, blocksize, 0);
3497 /* Find the buffer that contains "offset" */
3498 bh = page_buffers(page);
3500 while (offset >= pos) {
3501 bh = bh->b_this_page;
3505 if (buffer_freed(bh)) {
3506 BUFFER_TRACE(bh, "freed: skip");
3509 if (!buffer_mapped(bh)) {
3510 BUFFER_TRACE(bh, "unmapped");
3511 ext4_get_block(inode, iblock, bh, 0);
3512 /* unmapped? It's a hole - nothing to do */
3513 if (!buffer_mapped(bh)) {
3514 BUFFER_TRACE(bh, "still unmapped");
3519 /* Ok, it's mapped. Make sure it's up-to-date */
3520 if (PageUptodate(page))
3521 set_buffer_uptodate(bh);
3523 if (!buffer_uptodate(bh)) {
3525 ll_rw_block(READ, 1, &bh);
3527 /* Uhhuh. Read error. Complain and punt. */
3528 if (!buffer_uptodate(bh))
3530 if (S_ISREG(inode->i_mode) &&
3531 ext4_encrypted_inode(inode)) {
3532 /* We expect the key to be set. */
3533 BUG_ON(!ext4_has_encryption_key(inode));
3534 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3535 WARN_ON_ONCE(ext4_decrypt(page));
3538 if (ext4_should_journal_data(inode)) {
3539 BUFFER_TRACE(bh, "get write access");
3540 err = ext4_journal_get_write_access(handle, bh);
3544 zero_user(page, offset, length);
3545 BUFFER_TRACE(bh, "zeroed end of block");
3547 if (ext4_should_journal_data(inode)) {
3548 err = ext4_handle_dirty_metadata(handle, inode, bh);
3551 mark_buffer_dirty(bh);
3552 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3553 err = ext4_jbd2_file_inode(handle, inode);
3558 page_cache_release(page);
3563 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3564 * starting from file offset 'from'. The range to be zero'd must
3565 * be contained with in one block. If the specified range exceeds
3566 * the end of the block it will be shortened to end of the block
3567 * that cooresponds to 'from'
3569 static int ext4_block_zero_page_range(handle_t *handle,
3570 struct address_space *mapping, loff_t from, loff_t length)
3572 struct inode *inode = mapping->host;
3573 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3574 unsigned blocksize = inode->i_sb->s_blocksize;
3575 unsigned max = blocksize - (offset & (blocksize - 1));
3578 * correct length if it does not fall between
3579 * 'from' and the end of the block
3581 if (length > max || length < 0)
3585 return dax_zero_page_range(inode, from, length, ext4_get_block);
3586 return __ext4_block_zero_page_range(handle, mapping, from, length);
3590 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3591 * up to the end of the block which corresponds to `from'.
3592 * This required during truncate. We need to physically zero the tail end
3593 * of that block so it doesn't yield old data if the file is later grown.
3595 static int ext4_block_truncate_page(handle_t *handle,
3596 struct address_space *mapping, loff_t from)
3598 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3601 struct inode *inode = mapping->host;
3603 /* If we are processing an encrypted inode during orphan list handling */
3604 if (ext4_encrypted_inode(inode) && !ext4_has_encryption_key(inode))
3607 blocksize = inode->i_sb->s_blocksize;
3608 length = blocksize - (offset & (blocksize - 1));
3610 return ext4_block_zero_page_range(handle, mapping, from, length);
3613 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3614 loff_t lstart, loff_t length)
3616 struct super_block *sb = inode->i_sb;
3617 struct address_space *mapping = inode->i_mapping;
3618 unsigned partial_start, partial_end;
3619 ext4_fsblk_t start, end;
3620 loff_t byte_end = (lstart + length - 1);
3623 partial_start = lstart & (sb->s_blocksize - 1);
3624 partial_end = byte_end & (sb->s_blocksize - 1);
3626 start = lstart >> sb->s_blocksize_bits;
3627 end = byte_end >> sb->s_blocksize_bits;
3629 /* Handle partial zero within the single block */
3631 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3632 err = ext4_block_zero_page_range(handle, mapping,
3636 /* Handle partial zero out on the start of the range */
3637 if (partial_start) {
3638 err = ext4_block_zero_page_range(handle, mapping,
3639 lstart, sb->s_blocksize);
3643 /* Handle partial zero out on the end of the range */
3644 if (partial_end != sb->s_blocksize - 1)
3645 err = ext4_block_zero_page_range(handle, mapping,
3646 byte_end - partial_end,
3651 int ext4_can_truncate(struct inode *inode)
3653 if (S_ISREG(inode->i_mode))
3655 if (S_ISDIR(inode->i_mode))
3657 if (S_ISLNK(inode->i_mode))
3658 return !ext4_inode_is_fast_symlink(inode);
3663 * We have to make sure i_disksize gets properly updated before we truncate
3664 * page cache due to hole punching or zero range. Otherwise i_disksize update
3665 * can get lost as it may have been postponed to submission of writeback but
3666 * that will never happen after we truncate page cache.
3668 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3672 loff_t size = i_size_read(inode);
3674 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3675 if (offset > size || offset + len < size)
3678 if (EXT4_I(inode)->i_disksize >= size)
3681 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3683 return PTR_ERR(handle);
3684 ext4_update_i_disksize(inode, size);
3685 ext4_mark_inode_dirty(handle, inode);
3686 ext4_journal_stop(handle);
3692 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3693 * associated with the given offset and length
3695 * @inode: File inode
3696 * @offset: The offset where the hole will begin
3697 * @len: The length of the hole
3699 * Returns: 0 on success or negative on failure
3702 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3704 struct super_block *sb = inode->i_sb;
3705 ext4_lblk_t first_block, stop_block;
3706 struct address_space *mapping = inode->i_mapping;
3707 loff_t first_block_offset, last_block_offset;
3709 unsigned int credits;
3712 if (!S_ISREG(inode->i_mode))
3715 trace_ext4_punch_hole(inode, offset, length, 0);
3718 * Write out all dirty pages to avoid race conditions
3719 * Then release them.
3721 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3722 ret = filemap_write_and_wait_range(mapping, offset,
3723 offset + length - 1);
3728 mutex_lock(&inode->i_mutex);
3730 /* No need to punch hole beyond i_size */
3731 if (offset >= inode->i_size)
3735 * If the hole extends beyond i_size, set the hole
3736 * to end after the page that contains i_size
3738 if (offset + length > inode->i_size) {
3739 length = inode->i_size +
3740 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3744 if (offset & (sb->s_blocksize - 1) ||
3745 (offset + length) & (sb->s_blocksize - 1)) {
3747 * Attach jinode to inode for jbd2 if we do any zeroing of
3750 ret = ext4_inode_attach_jinode(inode);
3756 /* Wait all existing dio workers, newcomers will block on i_mutex */
3757 ext4_inode_block_unlocked_dio(inode);
3758 inode_dio_wait(inode);
3761 * Prevent page faults from reinstantiating pages we have released from
3764 down_write(&EXT4_I(inode)->i_mmap_sem);
3765 first_block_offset = round_up(offset, sb->s_blocksize);
3766 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3768 /* Now release the pages and zero block aligned part of pages*/
3769 if (last_block_offset > first_block_offset) {
3770 ret = ext4_update_disksize_before_punch(inode, offset, length);
3773 truncate_pagecache_range(inode, first_block_offset,
3777 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3778 credits = ext4_writepage_trans_blocks(inode);
3780 credits = ext4_blocks_for_truncate(inode);
3781 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3782 if (IS_ERR(handle)) {
3783 ret = PTR_ERR(handle);
3784 ext4_std_error(sb, ret);
3788 ret = ext4_zero_partial_blocks(handle, inode, offset,
3793 first_block = (offset + sb->s_blocksize - 1) >>
3794 EXT4_BLOCK_SIZE_BITS(sb);
3795 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3797 /* If there are no blocks to remove, return now */
3798 if (first_block >= stop_block)
3801 down_write(&EXT4_I(inode)->i_data_sem);
3802 ext4_discard_preallocations(inode);
3804 ret = ext4_es_remove_extent(inode, first_block,
3805 stop_block - first_block);
3807 up_write(&EXT4_I(inode)->i_data_sem);
3811 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3812 ret = ext4_ext_remove_space(inode, first_block,
3815 ret = ext4_ind_remove_space(handle, inode, first_block,
3818 up_write(&EXT4_I(inode)->i_data_sem);
3820 ext4_handle_sync(handle);
3822 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3823 ext4_mark_inode_dirty(handle, inode);
3825 ext4_journal_stop(handle);
3827 up_write(&EXT4_I(inode)->i_mmap_sem);
3828 ext4_inode_resume_unlocked_dio(inode);
3830 mutex_unlock(&inode->i_mutex);
3834 int ext4_inode_attach_jinode(struct inode *inode)
3836 struct ext4_inode_info *ei = EXT4_I(inode);
3837 struct jbd2_inode *jinode;
3839 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3842 jinode = jbd2_alloc_inode(GFP_KERNEL);
3843 spin_lock(&inode->i_lock);
3846 spin_unlock(&inode->i_lock);
3849 ei->jinode = jinode;
3850 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3853 spin_unlock(&inode->i_lock);
3854 if (unlikely(jinode != NULL))
3855 jbd2_free_inode(jinode);
3862 * We block out ext4_get_block() block instantiations across the entire
3863 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3864 * simultaneously on behalf of the same inode.
3866 * As we work through the truncate and commit bits of it to the journal there
3867 * is one core, guiding principle: the file's tree must always be consistent on
3868 * disk. We must be able to restart the truncate after a crash.
3870 * The file's tree may be transiently inconsistent in memory (although it
3871 * probably isn't), but whenever we close off and commit a journal transaction,
3872 * the contents of (the filesystem + the journal) must be consistent and
3873 * restartable. It's pretty simple, really: bottom up, right to left (although
3874 * left-to-right works OK too).
3876 * Note that at recovery time, journal replay occurs *before* the restart of
3877 * truncate against the orphan inode list.
3879 * The committed inode has the new, desired i_size (which is the same as
3880 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3881 * that this inode's truncate did not complete and it will again call
3882 * ext4_truncate() to have another go. So there will be instantiated blocks
3883 * to the right of the truncation point in a crashed ext4 filesystem. But
3884 * that's fine - as long as they are linked from the inode, the post-crash
3885 * ext4_truncate() run will find them and release them.
3887 void ext4_truncate(struct inode *inode)
3889 struct ext4_inode_info *ei = EXT4_I(inode);
3890 unsigned int credits;
3892 struct address_space *mapping = inode->i_mapping;
3895 * There is a possibility that we're either freeing the inode
3896 * or it's a completely new inode. In those cases we might not
3897 * have i_mutex locked because it's not necessary.
3899 if (!(inode->i_state & (I_NEW|I_FREEING)))
3900 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3901 trace_ext4_truncate_enter(inode);
3903 if (!ext4_can_truncate(inode))
3906 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3908 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3909 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3911 if (ext4_has_inline_data(inode)) {
3914 ext4_inline_data_truncate(inode, &has_inline);
3919 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3920 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3921 if (ext4_inode_attach_jinode(inode) < 0)
3925 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3926 credits = ext4_writepage_trans_blocks(inode);
3928 credits = ext4_blocks_for_truncate(inode);
3930 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3931 if (IS_ERR(handle)) {
3932 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3936 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3937 ext4_block_truncate_page(handle, mapping, inode->i_size);
3940 * We add the inode to the orphan list, so that if this
3941 * truncate spans multiple transactions, and we crash, we will
3942 * resume the truncate when the filesystem recovers. It also
3943 * marks the inode dirty, to catch the new size.
3945 * Implication: the file must always be in a sane, consistent
3946 * truncatable state while each transaction commits.
3948 if (ext4_orphan_add(handle, inode))
3951 down_write(&EXT4_I(inode)->i_data_sem);
3953 ext4_discard_preallocations(inode);
3955 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3956 ext4_ext_truncate(handle, inode);
3958 ext4_ind_truncate(handle, inode);
3960 up_write(&ei->i_data_sem);
3963 ext4_handle_sync(handle);
3967 * If this was a simple ftruncate() and the file will remain alive,
3968 * then we need to clear up the orphan record which we created above.
3969 * However, if this was a real unlink then we were called by
3970 * ext4_evict_inode(), and we allow that function to clean up the
3971 * orphan info for us.
3974 ext4_orphan_del(handle, inode);
3976 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3977 ext4_mark_inode_dirty(handle, inode);
3978 ext4_journal_stop(handle);
3980 trace_ext4_truncate_exit(inode);
3984 * ext4_get_inode_loc returns with an extra refcount against the inode's
3985 * underlying buffer_head on success. If 'in_mem' is true, we have all
3986 * data in memory that is needed to recreate the on-disk version of this
3989 static int __ext4_get_inode_loc(struct inode *inode,
3990 struct ext4_iloc *iloc, int in_mem)
3992 struct ext4_group_desc *gdp;
3993 struct buffer_head *bh;
3994 struct super_block *sb = inode->i_sb;
3996 int inodes_per_block, inode_offset;
3999 if (!ext4_valid_inum(sb, inode->i_ino))
4000 return -EFSCORRUPTED;
4002 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4003 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4008 * Figure out the offset within the block group inode table
4010 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4011 inode_offset = ((inode->i_ino - 1) %
4012 EXT4_INODES_PER_GROUP(sb));
4013 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4014 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4016 bh = sb_getblk(sb, block);
4019 if (!buffer_uptodate(bh)) {
4023 * If the buffer has the write error flag, we have failed
4024 * to write out another inode in the same block. In this
4025 * case, we don't have to read the block because we may
4026 * read the old inode data successfully.
4028 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4029 set_buffer_uptodate(bh);
4031 if (buffer_uptodate(bh)) {
4032 /* someone brought it uptodate while we waited */
4038 * If we have all information of the inode in memory and this
4039 * is the only valid inode in the block, we need not read the
4043 struct buffer_head *bitmap_bh;
4046 start = inode_offset & ~(inodes_per_block - 1);
4048 /* Is the inode bitmap in cache? */
4049 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4050 if (unlikely(!bitmap_bh))
4054 * If the inode bitmap isn't in cache then the
4055 * optimisation may end up performing two reads instead
4056 * of one, so skip it.
4058 if (!buffer_uptodate(bitmap_bh)) {
4062 for (i = start; i < start + inodes_per_block; i++) {
4063 if (i == inode_offset)
4065 if (ext4_test_bit(i, bitmap_bh->b_data))
4069 if (i == start + inodes_per_block) {
4070 /* all other inodes are free, so skip I/O */
4071 memset(bh->b_data, 0, bh->b_size);
4072 set_buffer_uptodate(bh);
4080 * If we need to do any I/O, try to pre-readahead extra
4081 * blocks from the inode table.
4083 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4084 ext4_fsblk_t b, end, table;
4086 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4088 table = ext4_inode_table(sb, gdp);
4089 /* s_inode_readahead_blks is always a power of 2 */
4090 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4094 num = EXT4_INODES_PER_GROUP(sb);
4095 if (ext4_has_group_desc_csum(sb))
4096 num -= ext4_itable_unused_count(sb, gdp);
4097 table += num / inodes_per_block;
4101 sb_breadahead(sb, b++);
4105 * There are other valid inodes in the buffer, this inode
4106 * has in-inode xattrs, or we don't have this inode in memory.
4107 * Read the block from disk.
4109 trace_ext4_load_inode(inode);
4111 bh->b_end_io = end_buffer_read_sync;
4112 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4114 if (!buffer_uptodate(bh)) {
4115 EXT4_ERROR_INODE_BLOCK(inode, block,
4116 "unable to read itable block");
4126 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4128 /* We have all inode data except xattrs in memory here. */
4129 return __ext4_get_inode_loc(inode, iloc,
4130 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4133 void ext4_set_inode_flags(struct inode *inode)
4135 unsigned int flags = EXT4_I(inode)->i_flags;
4136 unsigned int new_fl = 0;
4138 if (flags & EXT4_SYNC_FL)
4140 if (flags & EXT4_APPEND_FL)
4142 if (flags & EXT4_IMMUTABLE_FL)
4143 new_fl |= S_IMMUTABLE;
4144 if (flags & EXT4_NOATIME_FL)
4145 new_fl |= S_NOATIME;
4146 if (flags & EXT4_DIRSYNC_FL)
4147 new_fl |= S_DIRSYNC;
4148 if (test_opt(inode->i_sb, DAX))
4150 inode_set_flags(inode, new_fl,
4151 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4154 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4155 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4157 unsigned int vfs_fl;
4158 unsigned long old_fl, new_fl;
4161 vfs_fl = ei->vfs_inode.i_flags;
4162 old_fl = ei->i_flags;
4163 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4164 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4166 if (vfs_fl & S_SYNC)
4167 new_fl |= EXT4_SYNC_FL;
4168 if (vfs_fl & S_APPEND)
4169 new_fl |= EXT4_APPEND_FL;
4170 if (vfs_fl & S_IMMUTABLE)
4171 new_fl |= EXT4_IMMUTABLE_FL;
4172 if (vfs_fl & S_NOATIME)
4173 new_fl |= EXT4_NOATIME_FL;
4174 if (vfs_fl & S_DIRSYNC)
4175 new_fl |= EXT4_DIRSYNC_FL;
4176 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4179 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4180 struct ext4_inode_info *ei)
4183 struct inode *inode = &(ei->vfs_inode);
4184 struct super_block *sb = inode->i_sb;
4186 if (ext4_has_feature_huge_file(sb)) {
4187 /* we are using combined 48 bit field */
4188 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4189 le32_to_cpu(raw_inode->i_blocks_lo);
4190 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4191 /* i_blocks represent file system block size */
4192 return i_blocks << (inode->i_blkbits - 9);
4197 return le32_to_cpu(raw_inode->i_blocks_lo);
4201 static inline void ext4_iget_extra_inode(struct inode *inode,
4202 struct ext4_inode *raw_inode,
4203 struct ext4_inode_info *ei)
4205 __le32 *magic = (void *)raw_inode +
4206 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4207 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4208 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4209 ext4_find_inline_data_nolock(inode);
4211 EXT4_I(inode)->i_inline_off = 0;
4214 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4216 struct ext4_iloc iloc;
4217 struct ext4_inode *raw_inode;
4218 struct ext4_inode_info *ei;
4219 struct inode *inode;
4220 journal_t *journal = EXT4_SB(sb)->s_journal;
4227 inode = iget_locked(sb, ino);
4229 return ERR_PTR(-ENOMEM);
4230 if (!(inode->i_state & I_NEW))
4236 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4239 raw_inode = ext4_raw_inode(&iloc);
4241 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4242 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4243 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4244 EXT4_INODE_SIZE(inode->i_sb)) {
4245 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4246 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4247 EXT4_INODE_SIZE(inode->i_sb));
4248 ret = -EFSCORRUPTED;
4252 ei->i_extra_isize = 0;
4254 /* Precompute checksum seed for inode metadata */
4255 if (ext4_has_metadata_csum(sb)) {
4256 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4258 __le32 inum = cpu_to_le32(inode->i_ino);
4259 __le32 gen = raw_inode->i_generation;
4260 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4262 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4266 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4267 EXT4_ERROR_INODE(inode, "checksum invalid");
4272 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4273 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4274 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4275 if (!(test_opt(inode->i_sb, NO_UID32))) {
4276 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4277 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4279 i_uid_write(inode, i_uid);
4280 i_gid_write(inode, i_gid);
4281 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4283 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4284 ei->i_inline_off = 0;
4285 ei->i_dir_start_lookup = 0;
4286 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4287 /* We now have enough fields to check if the inode was active or not.
4288 * This is needed because nfsd might try to access dead inodes
4289 * the test is that same one that e2fsck uses
4290 * NeilBrown 1999oct15
4292 if (inode->i_nlink == 0) {
4293 if ((inode->i_mode == 0 ||
4294 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4295 ino != EXT4_BOOT_LOADER_INO) {
4296 /* this inode is deleted */
4300 /* The only unlinked inodes we let through here have
4301 * valid i_mode and are being read by the orphan
4302 * recovery code: that's fine, we're about to complete
4303 * the process of deleting those.
4304 * OR it is the EXT4_BOOT_LOADER_INO which is
4305 * not initialized on a new filesystem. */
4307 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4308 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4309 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4310 if (ext4_has_feature_64bit(sb))
4312 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4313 inode->i_size = ext4_isize(raw_inode);
4314 if ((size = i_size_read(inode)) < 0) {
4315 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4316 ret = -EFSCORRUPTED;
4319 ei->i_disksize = inode->i_size;
4321 ei->i_reserved_quota = 0;
4323 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4324 ei->i_block_group = iloc.block_group;
4325 ei->i_last_alloc_group = ~0;
4327 * NOTE! The in-memory inode i_data array is in little-endian order
4328 * even on big-endian machines: we do NOT byteswap the block numbers!
4330 for (block = 0; block < EXT4_N_BLOCKS; block++)
4331 ei->i_data[block] = raw_inode->i_block[block];
4332 INIT_LIST_HEAD(&ei->i_orphan);
4335 * Set transaction id's of transactions that have to be committed
4336 * to finish f[data]sync. We set them to currently running transaction
4337 * as we cannot be sure that the inode or some of its metadata isn't
4338 * part of the transaction - the inode could have been reclaimed and
4339 * now it is reread from disk.
4342 transaction_t *transaction;
4345 read_lock(&journal->j_state_lock);
4346 if (journal->j_running_transaction)
4347 transaction = journal->j_running_transaction;
4349 transaction = journal->j_committing_transaction;
4351 tid = transaction->t_tid;
4353 tid = journal->j_commit_sequence;
4354 read_unlock(&journal->j_state_lock);
4355 ei->i_sync_tid = tid;
4356 ei->i_datasync_tid = tid;
4359 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4360 if (ei->i_extra_isize == 0) {
4361 /* The extra space is currently unused. Use it. */
4362 ei->i_extra_isize = sizeof(struct ext4_inode) -
4363 EXT4_GOOD_OLD_INODE_SIZE;
4365 ext4_iget_extra_inode(inode, raw_inode, ei);
4369 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4370 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4371 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4372 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4374 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4375 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4376 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4377 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4379 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4384 if (ei->i_file_acl &&
4385 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4386 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4388 ret = -EFSCORRUPTED;
4390 } else if (!ext4_has_inline_data(inode)) {
4391 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4392 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4393 (S_ISLNK(inode->i_mode) &&
4394 !ext4_inode_is_fast_symlink(inode))))
4395 /* Validate extent which is part of inode */
4396 ret = ext4_ext_check_inode(inode);
4397 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4398 (S_ISLNK(inode->i_mode) &&
4399 !ext4_inode_is_fast_symlink(inode))) {
4400 /* Validate block references which are part of inode */
4401 ret = ext4_ind_check_inode(inode);
4407 if (S_ISREG(inode->i_mode)) {
4408 inode->i_op = &ext4_file_inode_operations;
4409 inode->i_fop = &ext4_file_operations;
4410 ext4_set_aops(inode);
4411 } else if (S_ISDIR(inode->i_mode)) {
4412 inode->i_op = &ext4_dir_inode_operations;
4413 inode->i_fop = &ext4_dir_operations;
4414 } else if (S_ISLNK(inode->i_mode)) {
4415 if (ext4_encrypted_inode(inode)) {
4416 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4417 ext4_set_aops(inode);
4418 } else if (ext4_inode_is_fast_symlink(inode)) {
4419 inode->i_link = (char *)ei->i_data;
4420 inode->i_op = &ext4_fast_symlink_inode_operations;
4421 nd_terminate_link(ei->i_data, inode->i_size,
4422 sizeof(ei->i_data) - 1);
4424 inode->i_op = &ext4_symlink_inode_operations;
4425 ext4_set_aops(inode);
4427 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4428 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4429 inode->i_op = &ext4_special_inode_operations;
4430 if (raw_inode->i_block[0])
4431 init_special_inode(inode, inode->i_mode,
4432 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4434 init_special_inode(inode, inode->i_mode,
4435 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4436 } else if (ino == EXT4_BOOT_LOADER_INO) {
4437 make_bad_inode(inode);
4439 ret = -EFSCORRUPTED;
4440 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4444 ext4_set_inode_flags(inode);
4445 unlock_new_inode(inode);
4451 return ERR_PTR(ret);
4454 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4456 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4457 return ERR_PTR(-EFSCORRUPTED);
4458 return ext4_iget(sb, ino);
4461 static int ext4_inode_blocks_set(handle_t *handle,
4462 struct ext4_inode *raw_inode,
4463 struct ext4_inode_info *ei)
4465 struct inode *inode = &(ei->vfs_inode);
4466 u64 i_blocks = inode->i_blocks;
4467 struct super_block *sb = inode->i_sb;
4469 if (i_blocks <= ~0U) {
4471 * i_blocks can be represented in a 32 bit variable
4472 * as multiple of 512 bytes
4474 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4475 raw_inode->i_blocks_high = 0;
4476 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4479 if (!ext4_has_feature_huge_file(sb))
4482 if (i_blocks <= 0xffffffffffffULL) {
4484 * i_blocks can be represented in a 48 bit variable
4485 * as multiple of 512 bytes
4487 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4488 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4489 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4491 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4492 /* i_block is stored in file system block size */
4493 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4494 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4495 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4500 struct other_inode {
4501 unsigned long orig_ino;
4502 struct ext4_inode *raw_inode;
4505 static int other_inode_match(struct inode * inode, unsigned long ino,
4508 struct other_inode *oi = (struct other_inode *) data;
4510 if ((inode->i_ino != ino) ||
4511 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4512 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4513 ((inode->i_state & I_DIRTY_TIME) == 0))
4515 spin_lock(&inode->i_lock);
4516 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4517 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4518 (inode->i_state & I_DIRTY_TIME)) {
4519 struct ext4_inode_info *ei = EXT4_I(inode);
4521 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4522 spin_unlock(&inode->i_lock);
4524 spin_lock(&ei->i_raw_lock);
4525 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4526 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4527 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4528 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4529 spin_unlock(&ei->i_raw_lock);
4530 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4533 spin_unlock(&inode->i_lock);
4538 * Opportunistically update the other time fields for other inodes in
4539 * the same inode table block.
4541 static void ext4_update_other_inodes_time(struct super_block *sb,
4542 unsigned long orig_ino, char *buf)
4544 struct other_inode oi;
4546 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4547 int inode_size = EXT4_INODE_SIZE(sb);
4549 oi.orig_ino = orig_ino;
4551 * Calculate the first inode in the inode table block. Inode
4552 * numbers are one-based. That is, the first inode in a block
4553 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4555 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4556 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4557 if (ino == orig_ino)
4559 oi.raw_inode = (struct ext4_inode *) buf;
4560 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4565 * Post the struct inode info into an on-disk inode location in the
4566 * buffer-cache. This gobbles the caller's reference to the
4567 * buffer_head in the inode location struct.
4569 * The caller must have write access to iloc->bh.
4571 static int ext4_do_update_inode(handle_t *handle,
4572 struct inode *inode,
4573 struct ext4_iloc *iloc)
4575 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4576 struct ext4_inode_info *ei = EXT4_I(inode);
4577 struct buffer_head *bh = iloc->bh;
4578 struct super_block *sb = inode->i_sb;
4579 int err = 0, rc, block;
4580 int need_datasync = 0, set_large_file = 0;
4584 spin_lock(&ei->i_raw_lock);
4586 /* For fields not tracked in the in-memory inode,
4587 * initialise them to zero for new inodes. */
4588 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4589 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4591 ext4_get_inode_flags(ei);
4592 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4593 i_uid = i_uid_read(inode);
4594 i_gid = i_gid_read(inode);
4595 if (!(test_opt(inode->i_sb, NO_UID32))) {
4596 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4597 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4599 * Fix up interoperability with old kernels. Otherwise, old inodes get
4600 * re-used with the upper 16 bits of the uid/gid intact
4602 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4603 raw_inode->i_uid_high = 0;
4604 raw_inode->i_gid_high = 0;
4606 raw_inode->i_uid_high =
4607 cpu_to_le16(high_16_bits(i_uid));
4608 raw_inode->i_gid_high =
4609 cpu_to_le16(high_16_bits(i_gid));
4612 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4613 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4614 raw_inode->i_uid_high = 0;
4615 raw_inode->i_gid_high = 0;
4617 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4619 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4620 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4621 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4622 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4624 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4626 spin_unlock(&ei->i_raw_lock);
4629 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4630 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4631 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4632 raw_inode->i_file_acl_high =
4633 cpu_to_le16(ei->i_file_acl >> 32);
4634 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4635 if (ei->i_disksize != ext4_isize(raw_inode)) {
4636 ext4_isize_set(raw_inode, ei->i_disksize);
4639 if (ei->i_disksize > 0x7fffffffULL) {
4640 if (!ext4_has_feature_large_file(sb) ||
4641 EXT4_SB(sb)->s_es->s_rev_level ==
4642 cpu_to_le32(EXT4_GOOD_OLD_REV))
4645 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4646 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4647 if (old_valid_dev(inode->i_rdev)) {
4648 raw_inode->i_block[0] =
4649 cpu_to_le32(old_encode_dev(inode->i_rdev));
4650 raw_inode->i_block[1] = 0;
4652 raw_inode->i_block[0] = 0;
4653 raw_inode->i_block[1] =
4654 cpu_to_le32(new_encode_dev(inode->i_rdev));
4655 raw_inode->i_block[2] = 0;
4657 } else if (!ext4_has_inline_data(inode)) {
4658 for (block = 0; block < EXT4_N_BLOCKS; block++)
4659 raw_inode->i_block[block] = ei->i_data[block];
4662 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4663 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4664 if (ei->i_extra_isize) {
4665 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4666 raw_inode->i_version_hi =
4667 cpu_to_le32(inode->i_version >> 32);
4668 raw_inode->i_extra_isize =
4669 cpu_to_le16(ei->i_extra_isize);
4672 ext4_inode_csum_set(inode, raw_inode, ei);
4673 spin_unlock(&ei->i_raw_lock);
4674 if (inode->i_sb->s_flags & MS_LAZYTIME)
4675 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4678 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4679 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4682 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4683 if (set_large_file) {
4684 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4685 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4688 ext4_update_dynamic_rev(sb);
4689 ext4_set_feature_large_file(sb);
4690 ext4_handle_sync(handle);
4691 err = ext4_handle_dirty_super(handle, sb);
4693 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4696 ext4_std_error(inode->i_sb, err);
4701 * ext4_write_inode()
4703 * We are called from a few places:
4705 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4706 * Here, there will be no transaction running. We wait for any running
4707 * transaction to commit.
4709 * - Within flush work (sys_sync(), kupdate and such).
4710 * We wait on commit, if told to.
4712 * - Within iput_final() -> write_inode_now()
4713 * We wait on commit, if told to.
4715 * In all cases it is actually safe for us to return without doing anything,
4716 * because the inode has been copied into a raw inode buffer in
4717 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4720 * Note that we are absolutely dependent upon all inode dirtiers doing the
4721 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4722 * which we are interested.
4724 * It would be a bug for them to not do this. The code:
4726 * mark_inode_dirty(inode)
4728 * inode->i_size = expr;
4730 * is in error because write_inode() could occur while `stuff()' is running,
4731 * and the new i_size will be lost. Plus the inode will no longer be on the
4732 * superblock's dirty inode list.
4734 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4738 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4741 if (EXT4_SB(inode->i_sb)->s_journal) {
4742 if (ext4_journal_current_handle()) {
4743 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4749 * No need to force transaction in WB_SYNC_NONE mode. Also
4750 * ext4_sync_fs() will force the commit after everything is
4753 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4756 err = ext4_force_commit(inode->i_sb);
4758 struct ext4_iloc iloc;
4760 err = __ext4_get_inode_loc(inode, &iloc, 0);
4764 * sync(2) will flush the whole buffer cache. No need to do
4765 * it here separately for each inode.
4767 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4768 sync_dirty_buffer(iloc.bh);
4769 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4770 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4771 "IO error syncing inode");
4780 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4781 * buffers that are attached to a page stradding i_size and are undergoing
4782 * commit. In that case we have to wait for commit to finish and try again.
4784 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4788 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4789 tid_t commit_tid = 0;
4792 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4794 * All buffers in the last page remain valid? Then there's nothing to
4795 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4798 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4801 page = find_lock_page(inode->i_mapping,
4802 inode->i_size >> PAGE_CACHE_SHIFT);
4805 ret = __ext4_journalled_invalidatepage(page, offset,
4806 PAGE_CACHE_SIZE - offset);
4808 page_cache_release(page);
4812 read_lock(&journal->j_state_lock);
4813 if (journal->j_committing_transaction)
4814 commit_tid = journal->j_committing_transaction->t_tid;
4815 read_unlock(&journal->j_state_lock);
4817 jbd2_log_wait_commit(journal, commit_tid);
4824 * Called from notify_change.
4826 * We want to trap VFS attempts to truncate the file as soon as
4827 * possible. In particular, we want to make sure that when the VFS
4828 * shrinks i_size, we put the inode on the orphan list and modify
4829 * i_disksize immediately, so that during the subsequent flushing of
4830 * dirty pages and freeing of disk blocks, we can guarantee that any
4831 * commit will leave the blocks being flushed in an unused state on
4832 * disk. (On recovery, the inode will get truncated and the blocks will
4833 * be freed, so we have a strong guarantee that no future commit will
4834 * leave these blocks visible to the user.)
4836 * Another thing we have to assure is that if we are in ordered mode
4837 * and inode is still attached to the committing transaction, we must
4838 * we start writeout of all the dirty pages which are being truncated.
4839 * This way we are sure that all the data written in the previous
4840 * transaction are already on disk (truncate waits for pages under
4843 * Called with inode->i_mutex down.
4845 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4847 struct inode *inode = d_inode(dentry);
4850 const unsigned int ia_valid = attr->ia_valid;
4852 error = inode_change_ok(inode, attr);
4856 if (is_quota_modification(inode, attr)) {
4857 error = dquot_initialize(inode);
4861 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4862 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4865 /* (user+group)*(old+new) structure, inode write (sb,
4866 * inode block, ? - but truncate inode update has it) */
4867 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4868 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4869 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4870 if (IS_ERR(handle)) {
4871 error = PTR_ERR(handle);
4874 error = dquot_transfer(inode, attr);
4876 ext4_journal_stop(handle);
4879 /* Update corresponding info in inode so that everything is in
4880 * one transaction */
4881 if (attr->ia_valid & ATTR_UID)
4882 inode->i_uid = attr->ia_uid;
4883 if (attr->ia_valid & ATTR_GID)
4884 inode->i_gid = attr->ia_gid;
4885 error = ext4_mark_inode_dirty(handle, inode);
4886 ext4_journal_stop(handle);
4889 if (attr->ia_valid & ATTR_SIZE) {
4891 loff_t oldsize = inode->i_size;
4892 int shrink = (attr->ia_size <= inode->i_size);
4894 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4895 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4897 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4900 if (!S_ISREG(inode->i_mode))
4903 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4904 inode_inc_iversion(inode);
4906 if (ext4_should_order_data(inode) &&
4907 (attr->ia_size < inode->i_size)) {
4908 error = ext4_begin_ordered_truncate(inode,
4913 if (attr->ia_size != inode->i_size) {
4914 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4915 if (IS_ERR(handle)) {
4916 error = PTR_ERR(handle);
4919 if (ext4_handle_valid(handle) && shrink) {
4920 error = ext4_orphan_add(handle, inode);
4924 * Update c/mtime on truncate up, ext4_truncate() will
4925 * update c/mtime in shrink case below
4928 inode->i_mtime = ext4_current_time(inode);
4929 inode->i_ctime = inode->i_mtime;
4931 down_write(&EXT4_I(inode)->i_data_sem);
4932 EXT4_I(inode)->i_disksize = attr->ia_size;
4933 rc = ext4_mark_inode_dirty(handle, inode);
4937 * We have to update i_size under i_data_sem together
4938 * with i_disksize to avoid races with writeback code
4939 * running ext4_wb_update_i_disksize().
4942 i_size_write(inode, attr->ia_size);
4943 up_write(&EXT4_I(inode)->i_data_sem);
4944 ext4_journal_stop(handle);
4947 ext4_orphan_del(NULL, inode);
4952 pagecache_isize_extended(inode, oldsize, inode->i_size);
4955 * Blocks are going to be removed from the inode. Wait
4956 * for dio in flight. Temporarily disable
4957 * dioread_nolock to prevent livelock.
4960 if (!ext4_should_journal_data(inode)) {
4961 ext4_inode_block_unlocked_dio(inode);
4962 inode_dio_wait(inode);
4963 ext4_inode_resume_unlocked_dio(inode);
4965 ext4_wait_for_tail_page_commit(inode);
4967 down_write(&EXT4_I(inode)->i_mmap_sem);
4969 * Truncate pagecache after we've waited for commit
4970 * in data=journal mode to make pages freeable.
4972 truncate_pagecache(inode, inode->i_size);
4974 ext4_truncate(inode);
4975 up_write(&EXT4_I(inode)->i_mmap_sem);
4979 setattr_copy(inode, attr);
4980 mark_inode_dirty(inode);
4984 * If the call to ext4_truncate failed to get a transaction handle at
4985 * all, we need to clean up the in-core orphan list manually.
4987 if (orphan && inode->i_nlink)
4988 ext4_orphan_del(NULL, inode);
4990 if (!rc && (ia_valid & ATTR_MODE))
4991 rc = posix_acl_chmod(inode, inode->i_mode);
4994 ext4_std_error(inode->i_sb, error);
5000 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5003 struct inode *inode;
5004 unsigned long long delalloc_blocks;
5006 inode = d_inode(dentry);
5007 generic_fillattr(inode, stat);
5010 * If there is inline data in the inode, the inode will normally not
5011 * have data blocks allocated (it may have an external xattr block).
5012 * Report at least one sector for such files, so tools like tar, rsync,
5013 * others doen't incorrectly think the file is completely sparse.
5015 if (unlikely(ext4_has_inline_data(inode)))
5016 stat->blocks += (stat->size + 511) >> 9;
5019 * We can't update i_blocks if the block allocation is delayed
5020 * otherwise in the case of system crash before the real block
5021 * allocation is done, we will have i_blocks inconsistent with
5022 * on-disk file blocks.
5023 * We always keep i_blocks updated together with real
5024 * allocation. But to not confuse with user, stat
5025 * will return the blocks that include the delayed allocation
5026 * blocks for this file.
5028 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5029 EXT4_I(inode)->i_reserved_data_blocks);
5030 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5034 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5037 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5038 return ext4_ind_trans_blocks(inode, lblocks);
5039 return ext4_ext_index_trans_blocks(inode, pextents);
5043 * Account for index blocks, block groups bitmaps and block group
5044 * descriptor blocks if modify datablocks and index blocks
5045 * worse case, the indexs blocks spread over different block groups
5047 * If datablocks are discontiguous, they are possible to spread over
5048 * different block groups too. If they are contiguous, with flexbg,
5049 * they could still across block group boundary.
5051 * Also account for superblock, inode, quota and xattr blocks
5053 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5056 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5062 * How many index blocks need to touch to map @lblocks logical blocks
5063 * to @pextents physical extents?
5065 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5070 * Now let's see how many group bitmaps and group descriptors need
5073 groups = idxblocks + pextents;
5075 if (groups > ngroups)
5077 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5078 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5080 /* bitmaps and block group descriptor blocks */
5081 ret += groups + gdpblocks;
5083 /* Blocks for super block, inode, quota and xattr blocks */
5084 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5090 * Calculate the total number of credits to reserve to fit
5091 * the modification of a single pages into a single transaction,
5092 * which may include multiple chunks of block allocations.
5094 * This could be called via ext4_write_begin()
5096 * We need to consider the worse case, when
5097 * one new block per extent.
5099 int ext4_writepage_trans_blocks(struct inode *inode)
5101 int bpp = ext4_journal_blocks_per_page(inode);
5104 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5106 /* Account for data blocks for journalled mode */
5107 if (ext4_should_journal_data(inode))
5113 * Calculate the journal credits for a chunk of data modification.
5115 * This is called from DIO, fallocate or whoever calling
5116 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5118 * journal buffers for data blocks are not included here, as DIO
5119 * and fallocate do no need to journal data buffers.
5121 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5123 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5127 * The caller must have previously called ext4_reserve_inode_write().
5128 * Give this, we know that the caller already has write access to iloc->bh.
5130 int ext4_mark_iloc_dirty(handle_t *handle,
5131 struct inode *inode, struct ext4_iloc *iloc)
5135 if (IS_I_VERSION(inode))
5136 inode_inc_iversion(inode);
5138 /* the do_update_inode consumes one bh->b_count */
5141 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5142 err = ext4_do_update_inode(handle, inode, iloc);
5148 * On success, We end up with an outstanding reference count against
5149 * iloc->bh. This _must_ be cleaned up later.
5153 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5154 struct ext4_iloc *iloc)
5158 err = ext4_get_inode_loc(inode, iloc);
5160 BUFFER_TRACE(iloc->bh, "get_write_access");
5161 err = ext4_journal_get_write_access(handle, iloc->bh);
5167 ext4_std_error(inode->i_sb, err);
5172 * Expand an inode by new_extra_isize bytes.
5173 * Returns 0 on success or negative error number on failure.
5175 static int ext4_expand_extra_isize(struct inode *inode,
5176 unsigned int new_extra_isize,
5177 struct ext4_iloc iloc,
5180 struct ext4_inode *raw_inode;
5181 struct ext4_xattr_ibody_header *header;
5183 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5186 raw_inode = ext4_raw_inode(&iloc);
5188 header = IHDR(inode, raw_inode);
5190 /* No extended attributes present */
5191 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5192 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5193 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5195 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5199 /* try to expand with EAs present */
5200 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5205 * What we do here is to mark the in-core inode as clean with respect to inode
5206 * dirtiness (it may still be data-dirty).
5207 * This means that the in-core inode may be reaped by prune_icache
5208 * without having to perform any I/O. This is a very good thing,
5209 * because *any* task may call prune_icache - even ones which
5210 * have a transaction open against a different journal.
5212 * Is this cheating? Not really. Sure, we haven't written the
5213 * inode out, but prune_icache isn't a user-visible syncing function.
5214 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5215 * we start and wait on commits.
5217 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5219 struct ext4_iloc iloc;
5220 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5221 static unsigned int mnt_count;
5225 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5226 err = ext4_reserve_inode_write(handle, inode, &iloc);
5229 if (ext4_handle_valid(handle) &&
5230 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5231 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5233 * We need extra buffer credits since we may write into EA block
5234 * with this same handle. If journal_extend fails, then it will
5235 * only result in a minor loss of functionality for that inode.
5236 * If this is felt to be critical, then e2fsck should be run to
5237 * force a large enough s_min_extra_isize.
5239 if ((jbd2_journal_extend(handle,
5240 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5241 ret = ext4_expand_extra_isize(inode,
5242 sbi->s_want_extra_isize,
5246 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5247 ext4_warning(inode->i_sb,
5248 "Unable to expand inode %lu. Delete"
5249 " some EAs or run e2fsck.",
5252 le16_to_cpu(sbi->s_es->s_mnt_count);
5257 return ext4_mark_iloc_dirty(handle, inode, &iloc);
5261 * ext4_dirty_inode() is called from __mark_inode_dirty()
5263 * We're really interested in the case where a file is being extended.
5264 * i_size has been changed by generic_commit_write() and we thus need
5265 * to include the updated inode in the current transaction.
5267 * Also, dquot_alloc_block() will always dirty the inode when blocks
5268 * are allocated to the file.
5270 * If the inode is marked synchronous, we don't honour that here - doing
5271 * so would cause a commit on atime updates, which we don't bother doing.
5272 * We handle synchronous inodes at the highest possible level.
5274 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5275 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5276 * to copy into the on-disk inode structure are the timestamp files.
5278 void ext4_dirty_inode(struct inode *inode, int flags)
5282 if (flags == I_DIRTY_TIME)
5284 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5288 ext4_mark_inode_dirty(handle, inode);
5290 ext4_journal_stop(handle);
5297 * Bind an inode's backing buffer_head into this transaction, to prevent
5298 * it from being flushed to disk early. Unlike
5299 * ext4_reserve_inode_write, this leaves behind no bh reference and
5300 * returns no iloc structure, so the caller needs to repeat the iloc
5301 * lookup to mark the inode dirty later.
5303 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5305 struct ext4_iloc iloc;
5309 err = ext4_get_inode_loc(inode, &iloc);
5311 BUFFER_TRACE(iloc.bh, "get_write_access");
5312 err = jbd2_journal_get_write_access(handle, iloc.bh);
5314 err = ext4_handle_dirty_metadata(handle,
5320 ext4_std_error(inode->i_sb, err);
5325 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5332 * We have to be very careful here: changing a data block's
5333 * journaling status dynamically is dangerous. If we write a
5334 * data block to the journal, change the status and then delete
5335 * that block, we risk forgetting to revoke the old log record
5336 * from the journal and so a subsequent replay can corrupt data.
5337 * So, first we make sure that the journal is empty and that
5338 * nobody is changing anything.
5341 journal = EXT4_JOURNAL(inode);
5344 if (is_journal_aborted(journal))
5346 /* We have to allocate physical blocks for delalloc blocks
5347 * before flushing journal. otherwise delalloc blocks can not
5348 * be allocated any more. even more truncate on delalloc blocks
5349 * could trigger BUG by flushing delalloc blocks in journal.
5350 * There is no delalloc block in non-journal data mode.
5352 if (val && test_opt(inode->i_sb, DELALLOC)) {
5353 err = ext4_alloc_da_blocks(inode);
5358 /* Wait for all existing dio workers */
5359 ext4_inode_block_unlocked_dio(inode);
5360 inode_dio_wait(inode);
5362 jbd2_journal_lock_updates(journal);
5365 * OK, there are no updates running now, and all cached data is
5366 * synced to disk. We are now in a completely consistent state
5367 * which doesn't have anything in the journal, and we know that
5368 * no filesystem updates are running, so it is safe to modify
5369 * the inode's in-core data-journaling state flag now.
5373 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5375 err = jbd2_journal_flush(journal);
5377 jbd2_journal_unlock_updates(journal);
5378 ext4_inode_resume_unlocked_dio(inode);
5381 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5383 ext4_set_aops(inode);
5385 jbd2_journal_unlock_updates(journal);
5386 ext4_inode_resume_unlocked_dio(inode);
5388 /* Finally we can mark the inode as dirty. */
5390 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5392 return PTR_ERR(handle);
5394 err = ext4_mark_inode_dirty(handle, inode);
5395 ext4_handle_sync(handle);
5396 ext4_journal_stop(handle);
5397 ext4_std_error(inode->i_sb, err);
5402 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5404 return !buffer_mapped(bh);
5407 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5409 struct page *page = vmf->page;
5413 struct file *file = vma->vm_file;
5414 struct inode *inode = file_inode(file);
5415 struct address_space *mapping = inode->i_mapping;
5417 get_block_t *get_block;
5420 sb_start_pagefault(inode->i_sb);
5421 file_update_time(vma->vm_file);
5423 down_read(&EXT4_I(inode)->i_mmap_sem);
5424 /* Delalloc case is easy... */
5425 if (test_opt(inode->i_sb, DELALLOC) &&
5426 !ext4_should_journal_data(inode) &&
5427 !ext4_nonda_switch(inode->i_sb)) {
5429 ret = block_page_mkwrite(vma, vmf,
5430 ext4_da_get_block_prep);
5431 } while (ret == -ENOSPC &&
5432 ext4_should_retry_alloc(inode->i_sb, &retries));
5437 size = i_size_read(inode);
5438 /* Page got truncated from under us? */
5439 if (page->mapping != mapping || page_offset(page) > size) {
5441 ret = VM_FAULT_NOPAGE;
5445 if (page->index == size >> PAGE_CACHE_SHIFT)
5446 len = size & ~PAGE_CACHE_MASK;
5448 len = PAGE_CACHE_SIZE;
5450 * Return if we have all the buffers mapped. This avoids the need to do
5451 * journal_start/journal_stop which can block and take a long time
5453 if (page_has_buffers(page)) {
5454 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5456 ext4_bh_unmapped)) {
5457 /* Wait so that we don't change page under IO */
5458 wait_for_stable_page(page);
5459 ret = VM_FAULT_LOCKED;
5464 /* OK, we need to fill the hole... */
5465 if (ext4_should_dioread_nolock(inode))
5466 get_block = ext4_get_block_write;
5468 get_block = ext4_get_block;
5470 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5471 ext4_writepage_trans_blocks(inode));
5472 if (IS_ERR(handle)) {
5473 ret = VM_FAULT_SIGBUS;
5476 ret = block_page_mkwrite(vma, vmf, get_block);
5477 if (!ret && ext4_should_journal_data(inode)) {
5478 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5479 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5481 ret = VM_FAULT_SIGBUS;
5482 ext4_journal_stop(handle);
5485 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5487 ext4_journal_stop(handle);
5488 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5491 ret = block_page_mkwrite_return(ret);
5493 up_read(&EXT4_I(inode)->i_mmap_sem);
5494 sb_end_pagefault(inode->i_sb);
5498 int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5500 struct inode *inode = file_inode(vma->vm_file);
5503 down_read(&EXT4_I(inode)->i_mmap_sem);
5504 err = filemap_fault(vma, vmf);
5505 up_read(&EXT4_I(inode)->i_mmap_sem);