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/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.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/ratelimit.h>
40 #include <linux/bitops.h>
42 #include "ext4_jbd2.h"
47 #include <trace/events/ext4.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);
59 csum_lo = le16_to_cpu(raw->i_checksum_lo);
60 raw->i_checksum_lo = 0;
61 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
62 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
63 csum_hi = le16_to_cpu(raw->i_checksum_hi);
64 raw->i_checksum_hi = 0;
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
68 EXT4_INODE_SIZE(inode->i_sb));
70 raw->i_checksum_lo = cpu_to_le16(csum_lo);
71 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
72 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
73 raw->i_checksum_hi = cpu_to_le16(csum_hi);
78 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
79 struct ext4_inode_info *ei)
81 __u32 provided, calculated;
83 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
84 cpu_to_le32(EXT4_OS_LINUX) ||
85 !ext4_has_metadata_csum(inode->i_sb))
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
96 return provided == calculated;
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !ext4_has_metadata_csum(inode->i_sb))
109 csum = ext4_inode_csum(inode, raw, ei);
110 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
111 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
112 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
113 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
116 static inline int ext4_begin_ordered_truncate(struct inode *inode,
119 trace_ext4_begin_ordered_truncate(inode, new_size);
121 * If jinode is zero, then we never opened the file for
122 * writing, so there's no need to call
123 * jbd2_journal_begin_ordered_truncate() since there's no
124 * outstanding writes we need to flush.
126 if (!EXT4_I(inode)->jinode)
128 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
129 EXT4_I(inode)->jinode,
133 static void ext4_invalidatepage(struct page *page, unsigned int offset,
134 unsigned int length);
135 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
136 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
137 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
141 * Test whether an inode is a fast symlink.
143 static int ext4_inode_is_fast_symlink(struct inode *inode)
145 int ea_blocks = EXT4_I(inode)->i_file_acl ?
146 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
148 if (ext4_has_inline_data(inode))
151 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
155 * Restart the transaction associated with *handle. This does a commit,
156 * so before we call here everything must be consistently dirtied against
159 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
165 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
166 * moment, get_block can be called only for blocks inside i_size since
167 * page cache has been already dropped and writes are blocked by
168 * i_mutex. So we can safely drop the i_data_sem here.
170 BUG_ON(EXT4_JOURNAL(inode) == NULL);
171 jbd_debug(2, "restarting handle %p\n", handle);
172 up_write(&EXT4_I(inode)->i_data_sem);
173 ret = ext4_journal_restart(handle, nblocks);
174 down_write(&EXT4_I(inode)->i_data_sem);
175 ext4_discard_preallocations(inode);
181 * Called at the last iput() if i_nlink is zero.
183 void ext4_evict_inode(struct inode *inode)
188 trace_ext4_evict_inode(inode);
190 if (inode->i_nlink) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode) &&
210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
211 inode->i_ino != EXT4_JOURNAL_INO) {
212 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
213 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
215 jbd2_complete_transaction(journal, commit_tid);
216 filemap_write_and_wait(&inode->i_data);
218 truncate_inode_pages_final(&inode->i_data);
220 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
224 if (is_bad_inode(inode))
226 dquot_initialize(inode);
228 if (ext4_should_order_data(inode))
229 ext4_begin_ordered_truncate(inode, 0);
230 truncate_inode_pages_final(&inode->i_data);
232 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
235 * Protect us against freezing - iput() caller didn't have to have any
236 * protection against it
238 sb_start_intwrite(inode->i_sb);
239 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
240 ext4_blocks_for_truncate(inode)+3);
241 if (IS_ERR(handle)) {
242 ext4_std_error(inode->i_sb, PTR_ERR(handle));
244 * If we're going to skip the normal cleanup, we still need to
245 * make sure that the in-core orphan linked list is properly
248 ext4_orphan_del(NULL, inode);
249 sb_end_intwrite(inode->i_sb);
254 ext4_handle_sync(handle);
256 err = ext4_mark_inode_dirty(handle, inode);
258 ext4_warning(inode->i_sb,
259 "couldn't mark inode dirty (err %d)", err);
263 ext4_truncate(inode);
266 * ext4_ext_truncate() doesn't reserve any slop when it
267 * restarts journal transactions; therefore there may not be
268 * enough credits left in the handle to remove the inode from
269 * the orphan list and set the dtime field.
271 if (!ext4_handle_has_enough_credits(handle, 3)) {
272 err = ext4_journal_extend(handle, 3);
274 err = ext4_journal_restart(handle, 3);
276 ext4_warning(inode->i_sb,
277 "couldn't extend journal (err %d)", err);
279 ext4_journal_stop(handle);
280 ext4_orphan_del(NULL, inode);
281 sb_end_intwrite(inode->i_sb);
287 * Kill off the orphan record which ext4_truncate created.
288 * AKPM: I think this can be inside the above `if'.
289 * Note that ext4_orphan_del() has to be able to cope with the
290 * deletion of a non-existent orphan - this is because we don't
291 * know if ext4_truncate() actually created an orphan record.
292 * (Well, we could do this if we need to, but heck - it works)
294 ext4_orphan_del(handle, inode);
295 EXT4_I(inode)->i_dtime = get_seconds();
298 * One subtle ordering requirement: if anything has gone wrong
299 * (transaction abort, IO errors, whatever), then we can still
300 * do these next steps (the fs will already have been marked as
301 * having errors), but we can't free the inode if the mark_dirty
304 if (ext4_mark_inode_dirty(handle, inode))
305 /* If that failed, just do the required in-core inode clear. */
306 ext4_clear_inode(inode);
308 ext4_free_inode(handle, inode);
309 ext4_journal_stop(handle);
310 sb_end_intwrite(inode->i_sb);
313 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
317 qsize_t *ext4_get_reserved_space(struct inode *inode)
319 return &EXT4_I(inode)->i_reserved_quota;
324 * Called with i_data_sem down, which is important since we can call
325 * ext4_discard_preallocations() from here.
327 void ext4_da_update_reserve_space(struct inode *inode,
328 int used, int quota_claim)
330 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
331 struct ext4_inode_info *ei = EXT4_I(inode);
333 spin_lock(&ei->i_block_reservation_lock);
334 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
335 if (unlikely(used > ei->i_reserved_data_blocks)) {
336 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
337 "with only %d reserved data blocks",
338 __func__, inode->i_ino, used,
339 ei->i_reserved_data_blocks);
341 used = ei->i_reserved_data_blocks;
344 /* Update per-inode reservations */
345 ei->i_reserved_data_blocks -= used;
346 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
348 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
350 /* Update quota subsystem for data blocks */
352 dquot_claim_block(inode, EXT4_C2B(sbi, used));
355 * We did fallocate with an offset that is already delayed
356 * allocated. So on delayed allocated writeback we should
357 * not re-claim the quota for fallocated blocks.
359 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
363 * If we have done all the pending block allocations and if
364 * there aren't any writers on the inode, we can discard the
365 * inode's preallocations.
367 if ((ei->i_reserved_data_blocks == 0) &&
368 (atomic_read(&inode->i_writecount) == 0))
369 ext4_discard_preallocations(inode);
372 static int __check_block_validity(struct inode *inode, const char *func,
374 struct ext4_map_blocks *map)
376 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
378 ext4_error_inode(inode, func, line, map->m_pblk,
379 "lblock %lu mapped to illegal pblock "
380 "(length %d)", (unsigned long) map->m_lblk,
387 #define check_block_validity(inode, map) \
388 __check_block_validity((inode), __func__, __LINE__, (map))
390 #ifdef ES_AGGRESSIVE_TEST
391 static void ext4_map_blocks_es_recheck(handle_t *handle,
393 struct ext4_map_blocks *es_map,
394 struct ext4_map_blocks *map,
401 * There is a race window that the result is not the same.
402 * e.g. xfstests #223 when dioread_nolock enables. The reason
403 * is that we lookup a block mapping in extent status tree with
404 * out taking i_data_sem. So at the time the unwritten extent
405 * could be converted.
407 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
408 down_read(&EXT4_I(inode)->i_data_sem);
409 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
410 retval = ext4_ext_map_blocks(handle, inode, map, flags &
411 EXT4_GET_BLOCKS_KEEP_SIZE);
413 retval = ext4_ind_map_blocks(handle, inode, map, flags &
414 EXT4_GET_BLOCKS_KEEP_SIZE);
416 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
417 up_read((&EXT4_I(inode)->i_data_sem));
420 * We don't check m_len because extent will be collpased in status
421 * tree. So the m_len might not equal.
423 if (es_map->m_lblk != map->m_lblk ||
424 es_map->m_flags != map->m_flags ||
425 es_map->m_pblk != map->m_pblk) {
426 printk("ES cache assertion failed for inode: %lu "
427 "es_cached ex [%d/%d/%llu/%x] != "
428 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
429 inode->i_ino, es_map->m_lblk, es_map->m_len,
430 es_map->m_pblk, es_map->m_flags, map->m_lblk,
431 map->m_len, map->m_pblk, map->m_flags,
435 #endif /* ES_AGGRESSIVE_TEST */
438 * The ext4_map_blocks() function tries to look up the requested blocks,
439 * and returns if the blocks are already mapped.
441 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
442 * and store the allocated blocks in the result buffer head and mark it
445 * If file type is extents based, it will call ext4_ext_map_blocks(),
446 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
449 * On success, it returns the number of blocks being mapped or allocated.
450 * if create==0 and the blocks are pre-allocated and unwritten block,
451 * the result buffer head is unmapped. If the create ==1, it will make sure
452 * the buffer head is mapped.
454 * It returns 0 if plain look up failed (blocks have not been allocated), in
455 * that case, buffer head is unmapped
457 * It returns the error in case of allocation failure.
459 int ext4_map_blocks(handle_t *handle, struct inode *inode,
460 struct ext4_map_blocks *map, int flags)
462 struct extent_status es;
465 #ifdef ES_AGGRESSIVE_TEST
466 struct ext4_map_blocks orig_map;
468 memcpy(&orig_map, map, sizeof(*map));
472 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
473 "logical block %lu\n", inode->i_ino, flags, map->m_len,
474 (unsigned long) map->m_lblk);
477 * ext4_map_blocks returns an int, and m_len is an unsigned int
479 if (unlikely(map->m_len > INT_MAX))
480 map->m_len = INT_MAX;
482 /* We can handle the block number less than EXT_MAX_BLOCKS */
483 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
486 /* Lookup extent status tree firstly */
487 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
488 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
489 map->m_pblk = ext4_es_pblock(&es) +
490 map->m_lblk - es.es_lblk;
491 map->m_flags |= ext4_es_is_written(&es) ?
492 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
493 retval = es.es_len - (map->m_lblk - es.es_lblk);
494 if (retval > map->m_len)
497 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
502 #ifdef ES_AGGRESSIVE_TEST
503 ext4_map_blocks_es_recheck(handle, inode, map,
510 * Try to see if we can get the block without requesting a new
513 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
514 down_read(&EXT4_I(inode)->i_data_sem);
515 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
516 retval = ext4_ext_map_blocks(handle, inode, map, flags &
517 EXT4_GET_BLOCKS_KEEP_SIZE);
519 retval = ext4_ind_map_blocks(handle, inode, map, flags &
520 EXT4_GET_BLOCKS_KEEP_SIZE);
525 if (unlikely(retval != map->m_len)) {
526 ext4_warning(inode->i_sb,
527 "ES len assertion failed for inode "
528 "%lu: retval %d != map->m_len %d",
529 inode->i_ino, retval, map->m_len);
533 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
534 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
535 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
536 ext4_find_delalloc_range(inode, map->m_lblk,
537 map->m_lblk + map->m_len - 1))
538 status |= EXTENT_STATUS_DELAYED;
539 ret = ext4_es_insert_extent(inode, map->m_lblk,
540 map->m_len, map->m_pblk, status);
544 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
545 up_read((&EXT4_I(inode)->i_data_sem));
548 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
549 ret = check_block_validity(inode, map);
554 /* If it is only a block(s) look up */
555 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
559 * Returns if the blocks have already allocated
561 * Note that if blocks have been preallocated
562 * ext4_ext_get_block() returns the create = 0
563 * with buffer head unmapped.
565 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
567 * If we need to convert extent to unwritten
568 * we continue and do the actual work in
569 * ext4_ext_map_blocks()
571 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
575 * Here we clear m_flags because after allocating an new extent,
576 * it will be set again.
578 map->m_flags &= ~EXT4_MAP_FLAGS;
581 * New blocks allocate and/or writing to unwritten extent
582 * will possibly result in updating i_data, so we take
583 * the write lock of i_data_sem, and call get_block()
584 * with create == 1 flag.
586 down_write(&EXT4_I(inode)->i_data_sem);
589 * We need to check for EXT4 here because migrate
590 * could have changed the inode type in between
592 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
593 retval = ext4_ext_map_blocks(handle, inode, map, flags);
595 retval = ext4_ind_map_blocks(handle, inode, map, flags);
597 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
599 * We allocated new blocks which will result in
600 * i_data's format changing. Force the migrate
601 * to fail by clearing migrate flags
603 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
607 * Update reserved blocks/metadata blocks after successful
608 * block allocation which had been deferred till now. We don't
609 * support fallocate for non extent files. So we can update
610 * reserve space here.
613 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
614 ext4_da_update_reserve_space(inode, retval, 1);
620 if (unlikely(retval != map->m_len)) {
621 ext4_warning(inode->i_sb,
622 "ES len assertion failed for inode "
623 "%lu: retval %d != map->m_len %d",
624 inode->i_ino, retval, map->m_len);
629 * If the extent has been zeroed out, we don't need to update
630 * extent status tree.
632 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
633 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
634 if (ext4_es_is_written(&es))
637 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
638 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
639 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
640 ext4_find_delalloc_range(inode, map->m_lblk,
641 map->m_lblk + map->m_len - 1))
642 status |= EXTENT_STATUS_DELAYED;
643 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
644 map->m_pblk, status);
650 up_write((&EXT4_I(inode)->i_data_sem));
651 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
652 ret = check_block_validity(inode, map);
659 static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate)
661 struct inode *inode = bh->b_assoc_map->host;
662 /* XXX: breaks on 32-bit > 16GB. Is that even supported? */
663 loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits;
667 WARN_ON(!buffer_unwritten(bh));
668 err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size);
671 /* Maximum number of blocks we map for direct IO at once. */
672 #define DIO_MAX_BLOCKS 4096
674 static int _ext4_get_block(struct inode *inode, sector_t iblock,
675 struct buffer_head *bh, int flags)
677 handle_t *handle = ext4_journal_current_handle();
678 struct ext4_map_blocks map;
679 int ret = 0, started = 0;
682 if (ext4_has_inline_data(inode))
686 map.m_len = bh->b_size >> inode->i_blkbits;
688 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
689 /* Direct IO write... */
690 if (map.m_len > DIO_MAX_BLOCKS)
691 map.m_len = DIO_MAX_BLOCKS;
692 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
693 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
695 if (IS_ERR(handle)) {
696 ret = PTR_ERR(handle);
702 ret = ext4_map_blocks(handle, inode, &map, flags);
704 ext4_io_end_t *io_end = ext4_inode_aio(inode);
706 map_bh(bh, inode->i_sb, map.m_pblk);
707 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
708 if (IS_DAX(inode) && buffer_unwritten(bh) && !io_end) {
709 bh->b_assoc_map = inode->i_mapping;
710 bh->b_private = (void *)(unsigned long)iblock;
711 bh->b_end_io = ext4_end_io_unwritten;
713 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
714 set_buffer_defer_completion(bh);
715 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
719 ext4_journal_stop(handle);
723 int ext4_get_block(struct inode *inode, sector_t iblock,
724 struct buffer_head *bh, int create)
726 return _ext4_get_block(inode, iblock, bh,
727 create ? EXT4_GET_BLOCKS_CREATE : 0);
731 * `handle' can be NULL if create is zero
733 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
734 ext4_lblk_t block, int create)
736 struct ext4_map_blocks map;
737 struct buffer_head *bh;
740 J_ASSERT(handle != NULL || create == 0);
744 err = ext4_map_blocks(handle, inode, &map,
745 create ? EXT4_GET_BLOCKS_CREATE : 0);
748 return create ? ERR_PTR(-ENOSPC) : NULL;
752 bh = sb_getblk(inode->i_sb, map.m_pblk);
754 return ERR_PTR(-ENOMEM);
755 if (map.m_flags & EXT4_MAP_NEW) {
756 J_ASSERT(create != 0);
757 J_ASSERT(handle != NULL);
760 * Now that we do not always journal data, we should
761 * keep in mind whether this should always journal the
762 * new buffer as metadata. For now, regular file
763 * writes use ext4_get_block instead, so it's not a
767 BUFFER_TRACE(bh, "call get_create_access");
768 err = ext4_journal_get_create_access(handle, bh);
773 if (!buffer_uptodate(bh)) {
774 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
775 set_buffer_uptodate(bh);
778 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
779 err = ext4_handle_dirty_metadata(handle, inode, bh);
783 BUFFER_TRACE(bh, "not a new buffer");
790 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
791 ext4_lblk_t block, int create)
793 struct buffer_head *bh;
795 bh = ext4_getblk(handle, inode, block, create);
798 if (!bh || buffer_uptodate(bh))
800 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
802 if (buffer_uptodate(bh))
805 return ERR_PTR(-EIO);
808 int ext4_walk_page_buffers(handle_t *handle,
809 struct buffer_head *head,
813 int (*fn)(handle_t *handle,
814 struct buffer_head *bh))
816 struct buffer_head *bh;
817 unsigned block_start, block_end;
818 unsigned blocksize = head->b_size;
820 struct buffer_head *next;
822 for (bh = head, block_start = 0;
823 ret == 0 && (bh != head || !block_start);
824 block_start = block_end, bh = next) {
825 next = bh->b_this_page;
826 block_end = block_start + blocksize;
827 if (block_end <= from || block_start >= to) {
828 if (partial && !buffer_uptodate(bh))
832 err = (*fn)(handle, bh);
840 * To preserve ordering, it is essential that the hole instantiation and
841 * the data write be encapsulated in a single transaction. We cannot
842 * close off a transaction and start a new one between the ext4_get_block()
843 * and the commit_write(). So doing the jbd2_journal_start at the start of
844 * prepare_write() is the right place.
846 * Also, this function can nest inside ext4_writepage(). In that case, we
847 * *know* that ext4_writepage() has generated enough buffer credits to do the
848 * whole page. So we won't block on the journal in that case, which is good,
849 * because the caller may be PF_MEMALLOC.
851 * By accident, ext4 can be reentered when a transaction is open via
852 * quota file writes. If we were to commit the transaction while thus
853 * reentered, there can be a deadlock - we would be holding a quota
854 * lock, and the commit would never complete if another thread had a
855 * transaction open and was blocking on the quota lock - a ranking
858 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
859 * will _not_ run commit under these circumstances because handle->h_ref
860 * is elevated. We'll still have enough credits for the tiny quotafile
863 int do_journal_get_write_access(handle_t *handle,
864 struct buffer_head *bh)
866 int dirty = buffer_dirty(bh);
869 if (!buffer_mapped(bh) || buffer_freed(bh))
872 * __block_write_begin() could have dirtied some buffers. Clean
873 * the dirty bit as jbd2_journal_get_write_access() could complain
874 * otherwise about fs integrity issues. Setting of the dirty bit
875 * by __block_write_begin() isn't a real problem here as we clear
876 * the bit before releasing a page lock and thus writeback cannot
877 * ever write the buffer.
880 clear_buffer_dirty(bh);
881 BUFFER_TRACE(bh, "get write access");
882 ret = ext4_journal_get_write_access(handle, bh);
884 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
888 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
889 struct buffer_head *bh_result, int create);
890 static int ext4_write_begin(struct file *file, struct address_space *mapping,
891 loff_t pos, unsigned len, unsigned flags,
892 struct page **pagep, void **fsdata)
894 struct inode *inode = mapping->host;
895 int ret, needed_blocks;
902 trace_ext4_write_begin(inode, pos, len, flags);
904 * Reserve one block more for addition to orphan list in case
905 * we allocate blocks but write fails for some reason
907 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
908 index = pos >> PAGE_CACHE_SHIFT;
909 from = pos & (PAGE_CACHE_SIZE - 1);
912 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
913 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
922 * grab_cache_page_write_begin() can take a long time if the
923 * system is thrashing due to memory pressure, or if the page
924 * is being written back. So grab it first before we start
925 * the transaction handle. This also allows us to allocate
926 * the page (if needed) without using GFP_NOFS.
929 page = grab_cache_page_write_begin(mapping, index, flags);
935 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
936 if (IS_ERR(handle)) {
937 page_cache_release(page);
938 return PTR_ERR(handle);
942 if (page->mapping != mapping) {
943 /* The page got truncated from under us */
945 page_cache_release(page);
946 ext4_journal_stop(handle);
949 /* In case writeback began while the page was unlocked */
950 wait_for_stable_page(page);
952 if (ext4_should_dioread_nolock(inode))
953 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
955 ret = __block_write_begin(page, pos, len, ext4_get_block);
957 if (!ret && ext4_should_journal_data(inode)) {
958 ret = ext4_walk_page_buffers(handle, page_buffers(page),
960 do_journal_get_write_access);
966 * __block_write_begin may have instantiated a few blocks
967 * outside i_size. Trim these off again. Don't need
968 * i_size_read because we hold i_mutex.
970 * Add inode to orphan list in case we crash before
973 if (pos + len > inode->i_size && ext4_can_truncate(inode))
974 ext4_orphan_add(handle, inode);
976 ext4_journal_stop(handle);
977 if (pos + len > inode->i_size) {
978 ext4_truncate_failed_write(inode);
980 * If truncate failed early the inode might
981 * still be on the orphan list; we need to
982 * make sure the inode is removed from the
983 * orphan list in that case.
986 ext4_orphan_del(NULL, inode);
989 if (ret == -ENOSPC &&
990 ext4_should_retry_alloc(inode->i_sb, &retries))
992 page_cache_release(page);
999 /* For write_end() in data=journal mode */
1000 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1003 if (!buffer_mapped(bh) || buffer_freed(bh))
1005 set_buffer_uptodate(bh);
1006 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1007 clear_buffer_meta(bh);
1008 clear_buffer_prio(bh);
1013 * We need to pick up the new inode size which generic_commit_write gave us
1014 * `file' can be NULL - eg, when called from page_symlink().
1016 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1017 * buffers are managed internally.
1019 static int ext4_write_end(struct file *file,
1020 struct address_space *mapping,
1021 loff_t pos, unsigned len, unsigned copied,
1022 struct page *page, void *fsdata)
1024 handle_t *handle = ext4_journal_current_handle();
1025 struct inode *inode = mapping->host;
1027 int i_size_changed = 0;
1029 trace_ext4_write_end(inode, pos, len, copied);
1030 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1031 ret = ext4_jbd2_file_inode(handle, inode);
1034 page_cache_release(page);
1039 if (ext4_has_inline_data(inode)) {
1040 ret = ext4_write_inline_data_end(inode, pos, len,
1046 copied = block_write_end(file, mapping, pos,
1047 len, copied, page, fsdata);
1049 * it's important to update i_size while still holding page lock:
1050 * page writeout could otherwise come in and zero beyond i_size.
1052 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1054 page_cache_release(page);
1057 * Don't mark the inode dirty under page lock. First, it unnecessarily
1058 * makes the holding time of page lock longer. Second, it forces lock
1059 * ordering of page lock and transaction start for journaling
1063 ext4_mark_inode_dirty(handle, inode);
1065 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1066 /* if we have allocated more blocks and copied
1067 * less. We will have blocks allocated outside
1068 * inode->i_size. So truncate them
1070 ext4_orphan_add(handle, inode);
1072 ret2 = ext4_journal_stop(handle);
1076 if (pos + len > inode->i_size) {
1077 ext4_truncate_failed_write(inode);
1079 * If truncate failed early the inode might still be
1080 * on the orphan list; we need to make sure the inode
1081 * is removed from the orphan list in that case.
1084 ext4_orphan_del(NULL, inode);
1087 return ret ? ret : copied;
1090 static int ext4_journalled_write_end(struct file *file,
1091 struct address_space *mapping,
1092 loff_t pos, unsigned len, unsigned copied,
1093 struct page *page, void *fsdata)
1095 handle_t *handle = ext4_journal_current_handle();
1096 struct inode *inode = mapping->host;
1100 int size_changed = 0;
1102 trace_ext4_journalled_write_end(inode, pos, len, copied);
1103 from = pos & (PAGE_CACHE_SIZE - 1);
1106 BUG_ON(!ext4_handle_valid(handle));
1108 if (ext4_has_inline_data(inode))
1109 copied = ext4_write_inline_data_end(inode, pos, len,
1113 if (!PageUptodate(page))
1115 page_zero_new_buffers(page, from+copied, to);
1118 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1119 to, &partial, write_end_fn);
1121 SetPageUptodate(page);
1123 size_changed = ext4_update_inode_size(inode, pos + copied);
1124 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1125 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1127 page_cache_release(page);
1130 ret2 = ext4_mark_inode_dirty(handle, inode);
1135 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1136 /* if we have allocated more blocks and copied
1137 * less. We will have blocks allocated outside
1138 * inode->i_size. So truncate them
1140 ext4_orphan_add(handle, inode);
1142 ret2 = ext4_journal_stop(handle);
1145 if (pos + len > inode->i_size) {
1146 ext4_truncate_failed_write(inode);
1148 * If truncate failed early the inode might still be
1149 * on the orphan list; we need to make sure the inode
1150 * is removed from the orphan list in that case.
1153 ext4_orphan_del(NULL, inode);
1156 return ret ? ret : copied;
1160 * Reserve a single cluster located at lblock
1162 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1164 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1165 struct ext4_inode_info *ei = EXT4_I(inode);
1166 unsigned int md_needed;
1170 * We will charge metadata quota at writeout time; this saves
1171 * us from metadata over-estimation, though we may go over by
1172 * a small amount in the end. Here we just reserve for data.
1174 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1179 * recalculate the amount of metadata blocks to reserve
1180 * in order to allocate nrblocks
1181 * worse case is one extent per block
1183 spin_lock(&ei->i_block_reservation_lock);
1185 * ext4_calc_metadata_amount() has side effects, which we have
1186 * to be prepared undo if we fail to claim space.
1189 trace_ext4_da_reserve_space(inode, 0);
1191 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1192 spin_unlock(&ei->i_block_reservation_lock);
1193 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1196 ei->i_reserved_data_blocks++;
1197 spin_unlock(&ei->i_block_reservation_lock);
1199 return 0; /* success */
1202 static void ext4_da_release_space(struct inode *inode, int to_free)
1204 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1205 struct ext4_inode_info *ei = EXT4_I(inode);
1208 return; /* Nothing to release, exit */
1210 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1212 trace_ext4_da_release_space(inode, to_free);
1213 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1215 * if there aren't enough reserved blocks, then the
1216 * counter is messed up somewhere. Since this
1217 * function is called from invalidate page, it's
1218 * harmless to return without any action.
1220 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1221 "ino %lu, to_free %d with only %d reserved "
1222 "data blocks", inode->i_ino, to_free,
1223 ei->i_reserved_data_blocks);
1225 to_free = ei->i_reserved_data_blocks;
1227 ei->i_reserved_data_blocks -= to_free;
1229 /* update fs dirty data blocks counter */
1230 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1232 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1234 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1237 static void ext4_da_page_release_reservation(struct page *page,
1238 unsigned int offset,
1239 unsigned int length)
1242 struct buffer_head *head, *bh;
1243 unsigned int curr_off = 0;
1244 struct inode *inode = page->mapping->host;
1245 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1246 unsigned int stop = offset + length;
1250 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1252 head = page_buffers(page);
1255 unsigned int next_off = curr_off + bh->b_size;
1257 if (next_off > stop)
1260 if ((offset <= curr_off) && (buffer_delay(bh))) {
1262 clear_buffer_delay(bh);
1264 curr_off = next_off;
1265 } while ((bh = bh->b_this_page) != head);
1268 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1269 ext4_es_remove_extent(inode, lblk, to_release);
1272 /* If we have released all the blocks belonging to a cluster, then we
1273 * need to release the reserved space for that cluster. */
1274 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1275 while (num_clusters > 0) {
1276 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1277 ((num_clusters - 1) << sbi->s_cluster_bits);
1278 if (sbi->s_cluster_ratio == 1 ||
1279 !ext4_find_delalloc_cluster(inode, lblk))
1280 ext4_da_release_space(inode, 1);
1287 * Delayed allocation stuff
1290 struct mpage_da_data {
1291 struct inode *inode;
1292 struct writeback_control *wbc;
1294 pgoff_t first_page; /* The first page to write */
1295 pgoff_t next_page; /* Current page to examine */
1296 pgoff_t last_page; /* Last page to examine */
1298 * Extent to map - this can be after first_page because that can be
1299 * fully mapped. We somewhat abuse m_flags to store whether the extent
1300 * is delalloc or unwritten.
1302 struct ext4_map_blocks map;
1303 struct ext4_io_submit io_submit; /* IO submission data */
1306 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1311 struct pagevec pvec;
1312 struct inode *inode = mpd->inode;
1313 struct address_space *mapping = inode->i_mapping;
1315 /* This is necessary when next_page == 0. */
1316 if (mpd->first_page >= mpd->next_page)
1319 index = mpd->first_page;
1320 end = mpd->next_page - 1;
1322 ext4_lblk_t start, last;
1323 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1324 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1325 ext4_es_remove_extent(inode, start, last - start + 1);
1328 pagevec_init(&pvec, 0);
1329 while (index <= end) {
1330 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1333 for (i = 0; i < nr_pages; i++) {
1334 struct page *page = pvec.pages[i];
1335 if (page->index > end)
1337 BUG_ON(!PageLocked(page));
1338 BUG_ON(PageWriteback(page));
1340 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1341 ClearPageUptodate(page);
1345 index = pvec.pages[nr_pages - 1]->index + 1;
1346 pagevec_release(&pvec);
1350 static void ext4_print_free_blocks(struct inode *inode)
1352 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1353 struct super_block *sb = inode->i_sb;
1354 struct ext4_inode_info *ei = EXT4_I(inode);
1356 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1357 EXT4_C2B(EXT4_SB(inode->i_sb),
1358 ext4_count_free_clusters(sb)));
1359 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1360 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1361 (long long) EXT4_C2B(EXT4_SB(sb),
1362 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1363 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1364 (long long) EXT4_C2B(EXT4_SB(sb),
1365 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1366 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1367 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1368 ei->i_reserved_data_blocks);
1372 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1374 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1378 * This function is grabs code from the very beginning of
1379 * ext4_map_blocks, but assumes that the caller is from delayed write
1380 * time. This function looks up the requested blocks and sets the
1381 * buffer delay bit under the protection of i_data_sem.
1383 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1384 struct ext4_map_blocks *map,
1385 struct buffer_head *bh)
1387 struct extent_status es;
1389 sector_t invalid_block = ~((sector_t) 0xffff);
1390 #ifdef ES_AGGRESSIVE_TEST
1391 struct ext4_map_blocks orig_map;
1393 memcpy(&orig_map, map, sizeof(*map));
1396 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1400 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1401 "logical block %lu\n", inode->i_ino, map->m_len,
1402 (unsigned long) map->m_lblk);
1404 /* Lookup extent status tree firstly */
1405 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1406 if (ext4_es_is_hole(&es)) {
1408 down_read(&EXT4_I(inode)->i_data_sem);
1413 * Delayed extent could be allocated by fallocate.
1414 * So we need to check it.
1416 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1417 map_bh(bh, inode->i_sb, invalid_block);
1419 set_buffer_delay(bh);
1423 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1424 retval = es.es_len - (iblock - es.es_lblk);
1425 if (retval > map->m_len)
1426 retval = map->m_len;
1427 map->m_len = retval;
1428 if (ext4_es_is_written(&es))
1429 map->m_flags |= EXT4_MAP_MAPPED;
1430 else if (ext4_es_is_unwritten(&es))
1431 map->m_flags |= EXT4_MAP_UNWRITTEN;
1435 #ifdef ES_AGGRESSIVE_TEST
1436 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1442 * Try to see if we can get the block without requesting a new
1443 * file system block.
1445 down_read(&EXT4_I(inode)->i_data_sem);
1446 if (ext4_has_inline_data(inode))
1448 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1449 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1451 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1457 * XXX: __block_prepare_write() unmaps passed block,
1461 * If the block was allocated from previously allocated cluster,
1462 * then we don't need to reserve it again. However we still need
1463 * to reserve metadata for every block we're going to write.
1465 if (EXT4_SB(inode->i_sb)->s_cluster_ratio <= 1 ||
1466 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1467 ret = ext4_da_reserve_space(inode, iblock);
1469 /* not enough space to reserve */
1475 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1476 ~0, EXTENT_STATUS_DELAYED);
1482 map_bh(bh, inode->i_sb, invalid_block);
1484 set_buffer_delay(bh);
1485 } else if (retval > 0) {
1487 unsigned int status;
1489 if (unlikely(retval != map->m_len)) {
1490 ext4_warning(inode->i_sb,
1491 "ES len assertion failed for inode "
1492 "%lu: retval %d != map->m_len %d",
1493 inode->i_ino, retval, map->m_len);
1497 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1498 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1499 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1500 map->m_pblk, status);
1506 up_read((&EXT4_I(inode)->i_data_sem));
1512 * This is a special get_block_t callback which is used by
1513 * ext4_da_write_begin(). It will either return mapped block or
1514 * reserve space for a single block.
1516 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1517 * We also have b_blocknr = -1 and b_bdev initialized properly
1519 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1520 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1521 * initialized properly.
1523 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1524 struct buffer_head *bh, int create)
1526 struct ext4_map_blocks map;
1529 BUG_ON(create == 0);
1530 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1532 map.m_lblk = iblock;
1536 * first, we need to know whether the block is allocated already
1537 * preallocated blocks are unmapped but should treated
1538 * the same as allocated blocks.
1540 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1544 map_bh(bh, inode->i_sb, map.m_pblk);
1545 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1547 if (buffer_unwritten(bh)) {
1548 /* A delayed write to unwritten bh should be marked
1549 * new and mapped. Mapped ensures that we don't do
1550 * get_block multiple times when we write to the same
1551 * offset and new ensures that we do proper zero out
1552 * for partial write.
1555 set_buffer_mapped(bh);
1560 static int bget_one(handle_t *handle, struct buffer_head *bh)
1566 static int bput_one(handle_t *handle, struct buffer_head *bh)
1572 static int __ext4_journalled_writepage(struct page *page,
1575 struct address_space *mapping = page->mapping;
1576 struct inode *inode = mapping->host;
1577 struct buffer_head *page_bufs = NULL;
1578 handle_t *handle = NULL;
1579 int ret = 0, err = 0;
1580 int inline_data = ext4_has_inline_data(inode);
1581 struct buffer_head *inode_bh = NULL;
1583 ClearPageChecked(page);
1586 BUG_ON(page->index != 0);
1587 BUG_ON(len > ext4_get_max_inline_size(inode));
1588 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1589 if (inode_bh == NULL)
1592 page_bufs = page_buffers(page);
1597 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1600 /* As soon as we unlock the page, it can go away, but we have
1601 * references to buffers so we are safe */
1604 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1605 ext4_writepage_trans_blocks(inode));
1606 if (IS_ERR(handle)) {
1607 ret = PTR_ERR(handle);
1611 BUG_ON(!ext4_handle_valid(handle));
1614 BUFFER_TRACE(inode_bh, "get write access");
1615 ret = ext4_journal_get_write_access(handle, inode_bh);
1617 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1620 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1621 do_journal_get_write_access);
1623 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1628 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1629 err = ext4_journal_stop(handle);
1633 if (!ext4_has_inline_data(inode))
1634 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1636 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1643 * Note that we don't need to start a transaction unless we're journaling data
1644 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1645 * need to file the inode to the transaction's list in ordered mode because if
1646 * we are writing back data added by write(), the inode is already there and if
1647 * we are writing back data modified via mmap(), no one guarantees in which
1648 * transaction the data will hit the disk. In case we are journaling data, we
1649 * cannot start transaction directly because transaction start ranks above page
1650 * lock so we have to do some magic.
1652 * This function can get called via...
1653 * - ext4_writepages after taking page lock (have journal handle)
1654 * - journal_submit_inode_data_buffers (no journal handle)
1655 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1656 * - grab_page_cache when doing write_begin (have journal handle)
1658 * We don't do any block allocation in this function. If we have page with
1659 * multiple blocks we need to write those buffer_heads that are mapped. This
1660 * is important for mmaped based write. So if we do with blocksize 1K
1661 * truncate(f, 1024);
1662 * a = mmap(f, 0, 4096);
1664 * truncate(f, 4096);
1665 * we have in the page first buffer_head mapped via page_mkwrite call back
1666 * but other buffer_heads would be unmapped but dirty (dirty done via the
1667 * do_wp_page). So writepage should write the first block. If we modify
1668 * the mmap area beyond 1024 we will again get a page_fault and the
1669 * page_mkwrite callback will do the block allocation and mark the
1670 * buffer_heads mapped.
1672 * We redirty the page if we have any buffer_heads that is either delay or
1673 * unwritten in the page.
1675 * We can get recursively called as show below.
1677 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1680 * But since we don't do any block allocation we should not deadlock.
1681 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1683 static int ext4_writepage(struct page *page,
1684 struct writeback_control *wbc)
1689 struct buffer_head *page_bufs = NULL;
1690 struct inode *inode = page->mapping->host;
1691 struct ext4_io_submit io_submit;
1692 bool keep_towrite = false;
1694 trace_ext4_writepage(page);
1695 size = i_size_read(inode);
1696 if (page->index == size >> PAGE_CACHE_SHIFT)
1697 len = size & ~PAGE_CACHE_MASK;
1699 len = PAGE_CACHE_SIZE;
1701 page_bufs = page_buffers(page);
1703 * We cannot do block allocation or other extent handling in this
1704 * function. If there are buffers needing that, we have to redirty
1705 * the page. But we may reach here when we do a journal commit via
1706 * journal_submit_inode_data_buffers() and in that case we must write
1707 * allocated buffers to achieve data=ordered mode guarantees.
1709 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1710 ext4_bh_delay_or_unwritten)) {
1711 redirty_page_for_writepage(wbc, page);
1712 if (current->flags & PF_MEMALLOC) {
1714 * For memory cleaning there's no point in writing only
1715 * some buffers. So just bail out. Warn if we came here
1716 * from direct reclaim.
1718 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1723 keep_towrite = true;
1726 if (PageChecked(page) && ext4_should_journal_data(inode))
1728 * It's mmapped pagecache. Add buffers and journal it. There
1729 * doesn't seem much point in redirtying the page here.
1731 return __ext4_journalled_writepage(page, len);
1733 ext4_io_submit_init(&io_submit, wbc);
1734 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1735 if (!io_submit.io_end) {
1736 redirty_page_for_writepage(wbc, page);
1740 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1741 ext4_io_submit(&io_submit);
1742 /* Drop io_end reference we got from init */
1743 ext4_put_io_end_defer(io_submit.io_end);
1747 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1750 loff_t size = i_size_read(mpd->inode);
1753 BUG_ON(page->index != mpd->first_page);
1754 if (page->index == size >> PAGE_CACHE_SHIFT)
1755 len = size & ~PAGE_CACHE_MASK;
1757 len = PAGE_CACHE_SIZE;
1758 clear_page_dirty_for_io(page);
1759 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1761 mpd->wbc->nr_to_write--;
1767 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1770 * mballoc gives us at most this number of blocks...
1771 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1772 * The rest of mballoc seems to handle chunks up to full group size.
1774 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1777 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1779 * @mpd - extent of blocks
1780 * @lblk - logical number of the block in the file
1781 * @bh - buffer head we want to add to the extent
1783 * The function is used to collect contig. blocks in the same state. If the
1784 * buffer doesn't require mapping for writeback and we haven't started the
1785 * extent of buffers to map yet, the function returns 'true' immediately - the
1786 * caller can write the buffer right away. Otherwise the function returns true
1787 * if the block has been added to the extent, false if the block couldn't be
1790 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1791 struct buffer_head *bh)
1793 struct ext4_map_blocks *map = &mpd->map;
1795 /* Buffer that doesn't need mapping for writeback? */
1796 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1797 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1798 /* So far no extent to map => we write the buffer right away */
1799 if (map->m_len == 0)
1804 /* First block in the extent? */
1805 if (map->m_len == 0) {
1808 map->m_flags = bh->b_state & BH_FLAGS;
1812 /* Don't go larger than mballoc is willing to allocate */
1813 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1816 /* Can we merge the block to our big extent? */
1817 if (lblk == map->m_lblk + map->m_len &&
1818 (bh->b_state & BH_FLAGS) == map->m_flags) {
1826 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1828 * @mpd - extent of blocks for mapping
1829 * @head - the first buffer in the page
1830 * @bh - buffer we should start processing from
1831 * @lblk - logical number of the block in the file corresponding to @bh
1833 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1834 * the page for IO if all buffers in this page were mapped and there's no
1835 * accumulated extent of buffers to map or add buffers in the page to the
1836 * extent of buffers to map. The function returns 1 if the caller can continue
1837 * by processing the next page, 0 if it should stop adding buffers to the
1838 * extent to map because we cannot extend it anymore. It can also return value
1839 * < 0 in case of error during IO submission.
1841 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1842 struct buffer_head *head,
1843 struct buffer_head *bh,
1846 struct inode *inode = mpd->inode;
1848 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1849 >> inode->i_blkbits;
1852 BUG_ON(buffer_locked(bh));
1854 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1855 /* Found extent to map? */
1858 /* Everything mapped so far and we hit EOF */
1861 } while (lblk++, (bh = bh->b_this_page) != head);
1862 /* So far everything mapped? Submit the page for IO. */
1863 if (mpd->map.m_len == 0) {
1864 err = mpage_submit_page(mpd, head->b_page);
1868 return lblk < blocks;
1872 * mpage_map_buffers - update buffers corresponding to changed extent and
1873 * submit fully mapped pages for IO
1875 * @mpd - description of extent to map, on return next extent to map
1877 * Scan buffers corresponding to changed extent (we expect corresponding pages
1878 * to be already locked) and update buffer state according to new extent state.
1879 * We map delalloc buffers to their physical location, clear unwritten bits,
1880 * and mark buffers as uninit when we perform writes to unwritten extents
1881 * and do extent conversion after IO is finished. If the last page is not fully
1882 * mapped, we update @map to the next extent in the last page that needs
1883 * mapping. Otherwise we submit the page for IO.
1885 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1887 struct pagevec pvec;
1889 struct inode *inode = mpd->inode;
1890 struct buffer_head *head, *bh;
1891 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
1897 start = mpd->map.m_lblk >> bpp_bits;
1898 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
1899 lblk = start << bpp_bits;
1900 pblock = mpd->map.m_pblk;
1902 pagevec_init(&pvec, 0);
1903 while (start <= end) {
1904 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
1908 for (i = 0; i < nr_pages; i++) {
1909 struct page *page = pvec.pages[i];
1911 if (page->index > end)
1913 /* Up to 'end' pages must be contiguous */
1914 BUG_ON(page->index != start);
1915 bh = head = page_buffers(page);
1917 if (lblk < mpd->map.m_lblk)
1919 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
1921 * Buffer after end of mapped extent.
1922 * Find next buffer in the page to map.
1925 mpd->map.m_flags = 0;
1927 * FIXME: If dioread_nolock supports
1928 * blocksize < pagesize, we need to make
1929 * sure we add size mapped so far to
1930 * io_end->size as the following call
1931 * can submit the page for IO.
1933 err = mpage_process_page_bufs(mpd, head,
1935 pagevec_release(&pvec);
1940 if (buffer_delay(bh)) {
1941 clear_buffer_delay(bh);
1942 bh->b_blocknr = pblock++;
1944 clear_buffer_unwritten(bh);
1945 } while (lblk++, (bh = bh->b_this_page) != head);
1948 * FIXME: This is going to break if dioread_nolock
1949 * supports blocksize < pagesize as we will try to
1950 * convert potentially unmapped parts of inode.
1952 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
1953 /* Page fully mapped - let IO run! */
1954 err = mpage_submit_page(mpd, page);
1956 pagevec_release(&pvec);
1961 pagevec_release(&pvec);
1963 /* Extent fully mapped and matches with page boundary. We are done. */
1965 mpd->map.m_flags = 0;
1969 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
1971 struct inode *inode = mpd->inode;
1972 struct ext4_map_blocks *map = &mpd->map;
1973 int get_blocks_flags;
1974 int err, dioread_nolock;
1976 trace_ext4_da_write_pages_extent(inode, map);
1978 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
1979 * to convert an unwritten extent to be initialized (in the case
1980 * where we have written into one or more preallocated blocks). It is
1981 * possible that we're going to need more metadata blocks than
1982 * previously reserved. However we must not fail because we're in
1983 * writeback and there is nothing we can do about it so it might result
1984 * in data loss. So use reserved blocks to allocate metadata if
1987 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
1988 * the blocks in question are delalloc blocks. This indicates
1989 * that the blocks and quotas has already been checked when
1990 * the data was copied into the page cache.
1992 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1993 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1994 dioread_nolock = ext4_should_dioread_nolock(inode);
1996 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1997 if (map->m_flags & (1 << BH_Delay))
1998 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2000 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2003 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2004 if (!mpd->io_submit.io_end->handle &&
2005 ext4_handle_valid(handle)) {
2006 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2007 handle->h_rsv_handle = NULL;
2009 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2012 BUG_ON(map->m_len == 0);
2013 if (map->m_flags & EXT4_MAP_NEW) {
2014 struct block_device *bdev = inode->i_sb->s_bdev;
2017 for (i = 0; i < map->m_len; i++)
2018 unmap_underlying_metadata(bdev, map->m_pblk + i);
2024 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2025 * mpd->len and submit pages underlying it for IO
2027 * @handle - handle for journal operations
2028 * @mpd - extent to map
2029 * @give_up_on_write - we set this to true iff there is a fatal error and there
2030 * is no hope of writing the data. The caller should discard
2031 * dirty pages to avoid infinite loops.
2033 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2034 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2035 * them to initialized or split the described range from larger unwritten
2036 * extent. Note that we need not map all the described range since allocation
2037 * can return less blocks or the range is covered by more unwritten extents. We
2038 * cannot map more because we are limited by reserved transaction credits. On
2039 * the other hand we always make sure that the last touched page is fully
2040 * mapped so that it can be written out (and thus forward progress is
2041 * guaranteed). After mapping we submit all mapped pages for IO.
2043 static int mpage_map_and_submit_extent(handle_t *handle,
2044 struct mpage_da_data *mpd,
2045 bool *give_up_on_write)
2047 struct inode *inode = mpd->inode;
2048 struct ext4_map_blocks *map = &mpd->map;
2053 mpd->io_submit.io_end->offset =
2054 ((loff_t)map->m_lblk) << inode->i_blkbits;
2056 err = mpage_map_one_extent(handle, mpd);
2058 struct super_block *sb = inode->i_sb;
2060 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2061 goto invalidate_dirty_pages;
2063 * Let the uper layers retry transient errors.
2064 * In the case of ENOSPC, if ext4_count_free_blocks()
2065 * is non-zero, a commit should free up blocks.
2067 if ((err == -ENOMEM) ||
2068 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2070 goto update_disksize;
2073 ext4_msg(sb, KERN_CRIT,
2074 "Delayed block allocation failed for "
2075 "inode %lu at logical offset %llu with"
2076 " max blocks %u with error %d",
2078 (unsigned long long)map->m_lblk,
2079 (unsigned)map->m_len, -err);
2080 ext4_msg(sb, KERN_CRIT,
2081 "This should not happen!! Data will "
2084 ext4_print_free_blocks(inode);
2085 invalidate_dirty_pages:
2086 *give_up_on_write = true;
2091 * Update buffer state, submit mapped pages, and get us new
2094 err = mpage_map_and_submit_buffers(mpd);
2096 goto update_disksize;
2097 } while (map->m_len);
2101 * Update on-disk size after IO is submitted. Races with
2102 * truncate are avoided by checking i_size under i_data_sem.
2104 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2105 if (disksize > EXT4_I(inode)->i_disksize) {
2109 down_write(&EXT4_I(inode)->i_data_sem);
2110 i_size = i_size_read(inode);
2111 if (disksize > i_size)
2113 if (disksize > EXT4_I(inode)->i_disksize)
2114 EXT4_I(inode)->i_disksize = disksize;
2115 err2 = ext4_mark_inode_dirty(handle, inode);
2116 up_write(&EXT4_I(inode)->i_data_sem);
2118 ext4_error(inode->i_sb,
2119 "Failed to mark inode %lu dirty",
2128 * Calculate the total number of credits to reserve for one writepages
2129 * iteration. This is called from ext4_writepages(). We map an extent of
2130 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2131 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2132 * bpp - 1 blocks in bpp different extents.
2134 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2136 int bpp = ext4_journal_blocks_per_page(inode);
2138 return ext4_meta_trans_blocks(inode,
2139 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2143 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2144 * and underlying extent to map
2146 * @mpd - where to look for pages
2148 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2149 * IO immediately. When we find a page which isn't mapped we start accumulating
2150 * extent of buffers underlying these pages that needs mapping (formed by
2151 * either delayed or unwritten buffers). We also lock the pages containing
2152 * these buffers. The extent found is returned in @mpd structure (starting at
2153 * mpd->lblk with length mpd->len blocks).
2155 * Note that this function can attach bios to one io_end structure which are
2156 * neither logically nor physically contiguous. Although it may seem as an
2157 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2158 * case as we need to track IO to all buffers underlying a page in one io_end.
2160 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2162 struct address_space *mapping = mpd->inode->i_mapping;
2163 struct pagevec pvec;
2164 unsigned int nr_pages;
2165 long left = mpd->wbc->nr_to_write;
2166 pgoff_t index = mpd->first_page;
2167 pgoff_t end = mpd->last_page;
2170 int blkbits = mpd->inode->i_blkbits;
2172 struct buffer_head *head;
2174 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2175 tag = PAGECACHE_TAG_TOWRITE;
2177 tag = PAGECACHE_TAG_DIRTY;
2179 pagevec_init(&pvec, 0);
2181 mpd->next_page = index;
2182 while (index <= end) {
2183 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2184 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2188 for (i = 0; i < nr_pages; i++) {
2189 struct page *page = pvec.pages[i];
2192 * At this point, the page may be truncated or
2193 * invalidated (changing page->mapping to NULL), or
2194 * even swizzled back from swapper_space to tmpfs file
2195 * mapping. However, page->index will not change
2196 * because we have a reference on the page.
2198 if (page->index > end)
2202 * Accumulated enough dirty pages? This doesn't apply
2203 * to WB_SYNC_ALL mode. For integrity sync we have to
2204 * keep going because someone may be concurrently
2205 * dirtying pages, and we might have synced a lot of
2206 * newly appeared dirty pages, but have not synced all
2207 * of the old dirty pages.
2209 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2212 /* If we can't merge this page, we are done. */
2213 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2218 * If the page is no longer dirty, or its mapping no
2219 * longer corresponds to inode we are writing (which
2220 * means it has been truncated or invalidated), or the
2221 * page is already under writeback and we are not doing
2222 * a data integrity writeback, skip the page
2224 if (!PageDirty(page) ||
2225 (PageWriteback(page) &&
2226 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2227 unlikely(page->mapping != mapping)) {
2232 wait_on_page_writeback(page);
2233 BUG_ON(PageWriteback(page));
2235 if (mpd->map.m_len == 0)
2236 mpd->first_page = page->index;
2237 mpd->next_page = page->index + 1;
2238 /* Add all dirty buffers to mpd */
2239 lblk = ((ext4_lblk_t)page->index) <<
2240 (PAGE_CACHE_SHIFT - blkbits);
2241 head = page_buffers(page);
2242 err = mpage_process_page_bufs(mpd, head, head, lblk);
2248 pagevec_release(&pvec);
2253 pagevec_release(&pvec);
2257 static int __writepage(struct page *page, struct writeback_control *wbc,
2260 struct address_space *mapping = data;
2261 int ret = ext4_writepage(page, wbc);
2262 mapping_set_error(mapping, ret);
2266 static int ext4_writepages(struct address_space *mapping,
2267 struct writeback_control *wbc)
2269 pgoff_t writeback_index = 0;
2270 long nr_to_write = wbc->nr_to_write;
2271 int range_whole = 0;
2273 handle_t *handle = NULL;
2274 struct mpage_da_data mpd;
2275 struct inode *inode = mapping->host;
2276 int needed_blocks, rsv_blocks = 0, ret = 0;
2277 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2279 struct blk_plug plug;
2280 bool give_up_on_write = false;
2282 trace_ext4_writepages(inode, wbc);
2285 * No pages to write? This is mainly a kludge to avoid starting
2286 * a transaction for special inodes like journal inode on last iput()
2287 * because that could violate lock ordering on umount
2289 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2290 goto out_writepages;
2292 if (ext4_should_journal_data(inode)) {
2293 struct blk_plug plug;
2295 blk_start_plug(&plug);
2296 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2297 blk_finish_plug(&plug);
2298 goto out_writepages;
2302 * If the filesystem has aborted, it is read-only, so return
2303 * right away instead of dumping stack traces later on that
2304 * will obscure the real source of the problem. We test
2305 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2306 * the latter could be true if the filesystem is mounted
2307 * read-only, and in that case, ext4_writepages should
2308 * *never* be called, so if that ever happens, we would want
2311 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2313 goto out_writepages;
2316 if (ext4_should_dioread_nolock(inode)) {
2318 * We may need to convert up to one extent per block in
2319 * the page and we may dirty the inode.
2321 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2325 * If we have inline data and arrive here, it means that
2326 * we will soon create the block for the 1st page, so
2327 * we'd better clear the inline data here.
2329 if (ext4_has_inline_data(inode)) {
2330 /* Just inode will be modified... */
2331 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2332 if (IS_ERR(handle)) {
2333 ret = PTR_ERR(handle);
2334 goto out_writepages;
2336 BUG_ON(ext4_test_inode_state(inode,
2337 EXT4_STATE_MAY_INLINE_DATA));
2338 ext4_destroy_inline_data(handle, inode);
2339 ext4_journal_stop(handle);
2342 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2345 if (wbc->range_cyclic) {
2346 writeback_index = mapping->writeback_index;
2347 if (writeback_index)
2349 mpd.first_page = writeback_index;
2352 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2353 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2358 ext4_io_submit_init(&mpd.io_submit, wbc);
2360 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2361 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2363 blk_start_plug(&plug);
2364 while (!done && mpd.first_page <= mpd.last_page) {
2365 /* For each extent of pages we use new io_end */
2366 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2367 if (!mpd.io_submit.io_end) {
2373 * We have two constraints: We find one extent to map and we
2374 * must always write out whole page (makes a difference when
2375 * blocksize < pagesize) so that we don't block on IO when we
2376 * try to write out the rest of the page. Journalled mode is
2377 * not supported by delalloc.
2379 BUG_ON(ext4_should_journal_data(inode));
2380 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2382 /* start a new transaction */
2383 handle = ext4_journal_start_with_reserve(inode,
2384 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2385 if (IS_ERR(handle)) {
2386 ret = PTR_ERR(handle);
2387 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2388 "%ld pages, ino %lu; err %d", __func__,
2389 wbc->nr_to_write, inode->i_ino, ret);
2390 /* Release allocated io_end */
2391 ext4_put_io_end(mpd.io_submit.io_end);
2395 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2396 ret = mpage_prepare_extent_to_map(&mpd);
2399 ret = mpage_map_and_submit_extent(handle, &mpd,
2403 * We scanned the whole range (or exhausted
2404 * nr_to_write), submitted what was mapped and
2405 * didn't find anything needing mapping. We are
2411 ext4_journal_stop(handle);
2412 /* Submit prepared bio */
2413 ext4_io_submit(&mpd.io_submit);
2414 /* Unlock pages we didn't use */
2415 mpage_release_unused_pages(&mpd, give_up_on_write);
2416 /* Drop our io_end reference we got from init */
2417 ext4_put_io_end(mpd.io_submit.io_end);
2419 if (ret == -ENOSPC && sbi->s_journal) {
2421 * Commit the transaction which would
2422 * free blocks released in the transaction
2425 jbd2_journal_force_commit_nested(sbi->s_journal);
2429 /* Fatal error - ENOMEM, EIO... */
2433 blk_finish_plug(&plug);
2434 if (!ret && !cycled && wbc->nr_to_write > 0) {
2436 mpd.last_page = writeback_index - 1;
2442 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2444 * Set the writeback_index so that range_cyclic
2445 * mode will write it back later
2447 mapping->writeback_index = mpd.first_page;
2450 trace_ext4_writepages_result(inode, wbc, ret,
2451 nr_to_write - wbc->nr_to_write);
2455 static int ext4_nonda_switch(struct super_block *sb)
2457 s64 free_clusters, dirty_clusters;
2458 struct ext4_sb_info *sbi = EXT4_SB(sb);
2461 * switch to non delalloc mode if we are running low
2462 * on free block. The free block accounting via percpu
2463 * counters can get slightly wrong with percpu_counter_batch getting
2464 * accumulated on each CPU without updating global counters
2465 * Delalloc need an accurate free block accounting. So switch
2466 * to non delalloc when we are near to error range.
2469 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2471 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2473 * Start pushing delalloc when 1/2 of free blocks are dirty.
2475 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2476 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2478 if (2 * free_clusters < 3 * dirty_clusters ||
2479 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2481 * free block count is less than 150% of dirty blocks
2482 * or free blocks is less than watermark
2489 /* We always reserve for an inode update; the superblock could be there too */
2490 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2492 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2493 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2496 if (pos + len <= 0x7fffffffULL)
2499 /* We might need to update the superblock to set LARGE_FILE */
2503 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2504 loff_t pos, unsigned len, unsigned flags,
2505 struct page **pagep, void **fsdata)
2507 int ret, retries = 0;
2510 struct inode *inode = mapping->host;
2513 index = pos >> PAGE_CACHE_SHIFT;
2515 if (ext4_nonda_switch(inode->i_sb)) {
2516 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2517 return ext4_write_begin(file, mapping, pos,
2518 len, flags, pagep, fsdata);
2520 *fsdata = (void *)0;
2521 trace_ext4_da_write_begin(inode, pos, len, flags);
2523 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2524 ret = ext4_da_write_inline_data_begin(mapping, inode,
2534 * grab_cache_page_write_begin() can take a long time if the
2535 * system is thrashing due to memory pressure, or if the page
2536 * is being written back. So grab it first before we start
2537 * the transaction handle. This also allows us to allocate
2538 * the page (if needed) without using GFP_NOFS.
2541 page = grab_cache_page_write_begin(mapping, index, flags);
2547 * With delayed allocation, we don't log the i_disksize update
2548 * if there is delayed block allocation. But we still need
2549 * to journalling the i_disksize update if writes to the end
2550 * of file which has an already mapped buffer.
2553 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2554 ext4_da_write_credits(inode, pos, len));
2555 if (IS_ERR(handle)) {
2556 page_cache_release(page);
2557 return PTR_ERR(handle);
2561 if (page->mapping != mapping) {
2562 /* The page got truncated from under us */
2564 page_cache_release(page);
2565 ext4_journal_stop(handle);
2568 /* In case writeback began while the page was unlocked */
2569 wait_for_stable_page(page);
2571 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2574 ext4_journal_stop(handle);
2576 * block_write_begin may have instantiated a few blocks
2577 * outside i_size. Trim these off again. Don't need
2578 * i_size_read because we hold i_mutex.
2580 if (pos + len > inode->i_size)
2581 ext4_truncate_failed_write(inode);
2583 if (ret == -ENOSPC &&
2584 ext4_should_retry_alloc(inode->i_sb, &retries))
2587 page_cache_release(page);
2596 * Check if we should update i_disksize
2597 * when write to the end of file but not require block allocation
2599 static int ext4_da_should_update_i_disksize(struct page *page,
2600 unsigned long offset)
2602 struct buffer_head *bh;
2603 struct inode *inode = page->mapping->host;
2607 bh = page_buffers(page);
2608 idx = offset >> inode->i_blkbits;
2610 for (i = 0; i < idx; i++)
2611 bh = bh->b_this_page;
2613 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2618 static int ext4_da_write_end(struct file *file,
2619 struct address_space *mapping,
2620 loff_t pos, unsigned len, unsigned copied,
2621 struct page *page, void *fsdata)
2623 struct inode *inode = mapping->host;
2625 handle_t *handle = ext4_journal_current_handle();
2627 unsigned long start, end;
2628 int write_mode = (int)(unsigned long)fsdata;
2630 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2631 return ext4_write_end(file, mapping, pos,
2632 len, copied, page, fsdata);
2634 trace_ext4_da_write_end(inode, pos, len, copied);
2635 start = pos & (PAGE_CACHE_SIZE - 1);
2636 end = start + copied - 1;
2639 * generic_write_end() will run mark_inode_dirty() if i_size
2640 * changes. So let's piggyback the i_disksize mark_inode_dirty
2643 new_i_size = pos + copied;
2644 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2645 if (ext4_has_inline_data(inode) ||
2646 ext4_da_should_update_i_disksize(page, end)) {
2647 ext4_update_i_disksize(inode, new_i_size);
2648 /* We need to mark inode dirty even if
2649 * new_i_size is less that inode->i_size
2650 * bu greater than i_disksize.(hint delalloc)
2652 ext4_mark_inode_dirty(handle, inode);
2656 if (write_mode != CONVERT_INLINE_DATA &&
2657 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2658 ext4_has_inline_data(inode))
2659 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2662 ret2 = generic_write_end(file, mapping, pos, len, copied,
2668 ret2 = ext4_journal_stop(handle);
2672 return ret ? ret : copied;
2675 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2676 unsigned int length)
2679 * Drop reserved blocks
2681 BUG_ON(!PageLocked(page));
2682 if (!page_has_buffers(page))
2685 ext4_da_page_release_reservation(page, offset, length);
2688 ext4_invalidatepage(page, offset, length);
2694 * Force all delayed allocation blocks to be allocated for a given inode.
2696 int ext4_alloc_da_blocks(struct inode *inode)
2698 trace_ext4_alloc_da_blocks(inode);
2700 if (!EXT4_I(inode)->i_reserved_data_blocks)
2704 * We do something simple for now. The filemap_flush() will
2705 * also start triggering a write of the data blocks, which is
2706 * not strictly speaking necessary (and for users of
2707 * laptop_mode, not even desirable). However, to do otherwise
2708 * would require replicating code paths in:
2710 * ext4_writepages() ->
2711 * write_cache_pages() ---> (via passed in callback function)
2712 * __mpage_da_writepage() -->
2713 * mpage_add_bh_to_extent()
2714 * mpage_da_map_blocks()
2716 * The problem is that write_cache_pages(), located in
2717 * mm/page-writeback.c, marks pages clean in preparation for
2718 * doing I/O, which is not desirable if we're not planning on
2721 * We could call write_cache_pages(), and then redirty all of
2722 * the pages by calling redirty_page_for_writepage() but that
2723 * would be ugly in the extreme. So instead we would need to
2724 * replicate parts of the code in the above functions,
2725 * simplifying them because we wouldn't actually intend to
2726 * write out the pages, but rather only collect contiguous
2727 * logical block extents, call the multi-block allocator, and
2728 * then update the buffer heads with the block allocations.
2730 * For now, though, we'll cheat by calling filemap_flush(),
2731 * which will map the blocks, and start the I/O, but not
2732 * actually wait for the I/O to complete.
2734 return filemap_flush(inode->i_mapping);
2738 * bmap() is special. It gets used by applications such as lilo and by
2739 * the swapper to find the on-disk block of a specific piece of data.
2741 * Naturally, this is dangerous if the block concerned is still in the
2742 * journal. If somebody makes a swapfile on an ext4 data-journaling
2743 * filesystem and enables swap, then they may get a nasty shock when the
2744 * data getting swapped to that swapfile suddenly gets overwritten by
2745 * the original zero's written out previously to the journal and
2746 * awaiting writeback in the kernel's buffer cache.
2748 * So, if we see any bmap calls here on a modified, data-journaled file,
2749 * take extra steps to flush any blocks which might be in the cache.
2751 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2753 struct inode *inode = mapping->host;
2758 * We can get here for an inline file via the FIBMAP ioctl
2760 if (ext4_has_inline_data(inode))
2763 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2764 test_opt(inode->i_sb, DELALLOC)) {
2766 * With delalloc we want to sync the file
2767 * so that we can make sure we allocate
2770 filemap_write_and_wait(mapping);
2773 if (EXT4_JOURNAL(inode) &&
2774 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2776 * This is a REALLY heavyweight approach, but the use of
2777 * bmap on dirty files is expected to be extremely rare:
2778 * only if we run lilo or swapon on a freshly made file
2779 * do we expect this to happen.
2781 * (bmap requires CAP_SYS_RAWIO so this does not
2782 * represent an unprivileged user DOS attack --- we'd be
2783 * in trouble if mortal users could trigger this path at
2786 * NB. EXT4_STATE_JDATA is not set on files other than
2787 * regular files. If somebody wants to bmap a directory
2788 * or symlink and gets confused because the buffer
2789 * hasn't yet been flushed to disk, they deserve
2790 * everything they get.
2793 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2794 journal = EXT4_JOURNAL(inode);
2795 jbd2_journal_lock_updates(journal);
2796 err = jbd2_journal_flush(journal);
2797 jbd2_journal_unlock_updates(journal);
2803 return generic_block_bmap(mapping, block, ext4_get_block);
2806 static int ext4_readpage(struct file *file, struct page *page)
2809 struct inode *inode = page->mapping->host;
2811 trace_ext4_readpage(page);
2813 if (ext4_has_inline_data(inode))
2814 ret = ext4_readpage_inline(inode, page);
2817 return mpage_readpage(page, ext4_get_block);
2823 ext4_readpages(struct file *file, struct address_space *mapping,
2824 struct list_head *pages, unsigned nr_pages)
2826 struct inode *inode = mapping->host;
2828 /* If the file has inline data, no need to do readpages. */
2829 if (ext4_has_inline_data(inode))
2832 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2835 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2836 unsigned int length)
2838 trace_ext4_invalidatepage(page, offset, length);
2840 /* No journalling happens on data buffers when this function is used */
2841 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2843 block_invalidatepage(page, offset, length);
2846 static int __ext4_journalled_invalidatepage(struct page *page,
2847 unsigned int offset,
2848 unsigned int length)
2850 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2852 trace_ext4_journalled_invalidatepage(page, offset, length);
2855 * If it's a full truncate we just forget about the pending dirtying
2857 if (offset == 0 && length == PAGE_CACHE_SIZE)
2858 ClearPageChecked(page);
2860 return jbd2_journal_invalidatepage(journal, page, offset, length);
2863 /* Wrapper for aops... */
2864 static void ext4_journalled_invalidatepage(struct page *page,
2865 unsigned int offset,
2866 unsigned int length)
2868 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2871 static int ext4_releasepage(struct page *page, gfp_t wait)
2873 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2875 trace_ext4_releasepage(page);
2877 /* Page has dirty journalled data -> cannot release */
2878 if (PageChecked(page))
2881 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2883 return try_to_free_buffers(page);
2887 * ext4_get_block used when preparing for a DIO write or buffer write.
2888 * We allocate an uinitialized extent if blocks haven't been allocated.
2889 * The extent will be converted to initialized after the IO is complete.
2891 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2892 struct buffer_head *bh_result, int create)
2894 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2895 inode->i_ino, create);
2896 return _ext4_get_block(inode, iblock, bh_result,
2897 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2900 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2901 struct buffer_head *bh_result, int create)
2903 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2904 inode->i_ino, create);
2905 return _ext4_get_block(inode, iblock, bh_result,
2906 EXT4_GET_BLOCKS_NO_LOCK);
2909 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2910 ssize_t size, void *private)
2912 ext4_io_end_t *io_end = iocb->private;
2914 /* if not async direct IO just return */
2918 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2919 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2920 iocb->private, io_end->inode->i_ino, iocb, offset,
2923 iocb->private = NULL;
2924 io_end->offset = offset;
2925 io_end->size = size;
2926 ext4_put_io_end(io_end);
2930 * For ext4 extent files, ext4 will do direct-io write to holes,
2931 * preallocated extents, and those write extend the file, no need to
2932 * fall back to buffered IO.
2934 * For holes, we fallocate those blocks, mark them as unwritten
2935 * If those blocks were preallocated, we mark sure they are split, but
2936 * still keep the range to write as unwritten.
2938 * The unwritten extents will be converted to written when DIO is completed.
2939 * For async direct IO, since the IO may still pending when return, we
2940 * set up an end_io call back function, which will do the conversion
2941 * when async direct IO completed.
2943 * If the O_DIRECT write will extend the file then add this inode to the
2944 * orphan list. So recovery will truncate it back to the original size
2945 * if the machine crashes during the write.
2948 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2949 struct iov_iter *iter, loff_t offset)
2951 struct file *file = iocb->ki_filp;
2952 struct inode *inode = file->f_mapping->host;
2954 size_t count = iov_iter_count(iter);
2956 get_block_t *get_block_func = NULL;
2958 loff_t final_size = offset + count;
2959 ext4_io_end_t *io_end = NULL;
2961 /* Use the old path for reads and writes beyond i_size. */
2962 if (rw != WRITE || final_size > inode->i_size)
2963 return ext4_ind_direct_IO(rw, iocb, iter, offset);
2965 BUG_ON(iocb->private == NULL);
2968 * Make all waiters for direct IO properly wait also for extent
2969 * conversion. This also disallows race between truncate() and
2970 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
2973 atomic_inc(&inode->i_dio_count);
2975 /* If we do a overwrite dio, i_mutex locking can be released */
2976 overwrite = *((int *)iocb->private);
2979 down_read(&EXT4_I(inode)->i_data_sem);
2980 mutex_unlock(&inode->i_mutex);
2984 * We could direct write to holes and fallocate.
2986 * Allocated blocks to fill the hole are marked as
2987 * unwritten to prevent parallel buffered read to expose
2988 * the stale data before DIO complete the data IO.
2990 * As to previously fallocated extents, ext4 get_block will
2991 * just simply mark the buffer mapped but still keep the
2992 * extents unwritten.
2994 * For non AIO case, we will convert those unwritten extents
2995 * to written after return back from blockdev_direct_IO.
2997 * For async DIO, the conversion needs to be deferred when the
2998 * IO is completed. The ext4 end_io callback function will be
2999 * called to take care of the conversion work. Here for async
3000 * case, we allocate an io_end structure to hook to the iocb.
3002 iocb->private = NULL;
3003 ext4_inode_aio_set(inode, NULL);
3004 if (!is_sync_kiocb(iocb)) {
3005 io_end = ext4_init_io_end(inode, GFP_NOFS);
3011 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3013 iocb->private = ext4_get_io_end(io_end);
3015 * we save the io structure for current async direct
3016 * IO, so that later ext4_map_blocks() could flag the
3017 * io structure whether there is a unwritten extents
3018 * needs to be converted when IO is completed.
3020 ext4_inode_aio_set(inode, io_end);
3024 get_block_func = ext4_get_block_write_nolock;
3026 get_block_func = ext4_get_block_write;
3027 dio_flags = DIO_LOCKING;
3030 ret = dax_do_io(rw, iocb, inode, iter, offset, get_block_func,
3031 ext4_end_io_dio, dio_flags);
3033 ret = __blockdev_direct_IO(rw, iocb, inode,
3034 inode->i_sb->s_bdev, iter, offset,
3036 ext4_end_io_dio, NULL, dio_flags);
3039 * Put our reference to io_end. This can free the io_end structure e.g.
3040 * in sync IO case or in case of error. It can even perform extent
3041 * conversion if all bios we submitted finished before we got here.
3042 * Note that in that case iocb->private can be already set to NULL
3046 ext4_inode_aio_set(inode, NULL);
3047 ext4_put_io_end(io_end);
3049 * When no IO was submitted ext4_end_io_dio() was not
3050 * called so we have to put iocb's reference.
3052 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3053 WARN_ON(iocb->private != io_end);
3054 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3055 ext4_put_io_end(io_end);
3056 iocb->private = NULL;
3059 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3060 EXT4_STATE_DIO_UNWRITTEN)) {
3063 * for non AIO case, since the IO is already
3064 * completed, we could do the conversion right here
3066 err = ext4_convert_unwritten_extents(NULL, inode,
3070 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3075 inode_dio_done(inode);
3076 /* take i_mutex locking again if we do a ovewrite dio */
3078 up_read(&EXT4_I(inode)->i_data_sem);
3079 mutex_lock(&inode->i_mutex);
3085 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3086 struct iov_iter *iter, loff_t offset)
3088 struct file *file = iocb->ki_filp;
3089 struct inode *inode = file->f_mapping->host;
3090 size_t count = iov_iter_count(iter);
3094 * If we are doing data journalling we don't support O_DIRECT
3096 if (ext4_should_journal_data(inode))
3099 /* Let buffer I/O handle the inline data case. */
3100 if (ext4_has_inline_data(inode))
3103 trace_ext4_direct_IO_enter(inode, offset, count, rw);
3104 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3105 ret = ext4_ext_direct_IO(rw, iocb, iter, offset);
3107 ret = ext4_ind_direct_IO(rw, iocb, iter, offset);
3108 trace_ext4_direct_IO_exit(inode, offset, count, rw, ret);
3113 * Pages can be marked dirty completely asynchronously from ext4's journalling
3114 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3115 * much here because ->set_page_dirty is called under VFS locks. The page is
3116 * not necessarily locked.
3118 * We cannot just dirty the page and leave attached buffers clean, because the
3119 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3120 * or jbddirty because all the journalling code will explode.
3122 * So what we do is to mark the page "pending dirty" and next time writepage
3123 * is called, propagate that into the buffers appropriately.
3125 static int ext4_journalled_set_page_dirty(struct page *page)
3127 SetPageChecked(page);
3128 return __set_page_dirty_nobuffers(page);
3131 static const struct address_space_operations ext4_aops = {
3132 .readpage = ext4_readpage,
3133 .readpages = ext4_readpages,
3134 .writepage = ext4_writepage,
3135 .writepages = ext4_writepages,
3136 .write_begin = ext4_write_begin,
3137 .write_end = ext4_write_end,
3139 .invalidatepage = ext4_invalidatepage,
3140 .releasepage = ext4_releasepage,
3141 .direct_IO = ext4_direct_IO,
3142 .migratepage = buffer_migrate_page,
3143 .is_partially_uptodate = block_is_partially_uptodate,
3144 .error_remove_page = generic_error_remove_page,
3147 static const struct address_space_operations ext4_journalled_aops = {
3148 .readpage = ext4_readpage,
3149 .readpages = ext4_readpages,
3150 .writepage = ext4_writepage,
3151 .writepages = ext4_writepages,
3152 .write_begin = ext4_write_begin,
3153 .write_end = ext4_journalled_write_end,
3154 .set_page_dirty = ext4_journalled_set_page_dirty,
3156 .invalidatepage = ext4_journalled_invalidatepage,
3157 .releasepage = ext4_releasepage,
3158 .direct_IO = ext4_direct_IO,
3159 .is_partially_uptodate = block_is_partially_uptodate,
3160 .error_remove_page = generic_error_remove_page,
3163 static const struct address_space_operations ext4_da_aops = {
3164 .readpage = ext4_readpage,
3165 .readpages = ext4_readpages,
3166 .writepage = ext4_writepage,
3167 .writepages = ext4_writepages,
3168 .write_begin = ext4_da_write_begin,
3169 .write_end = ext4_da_write_end,
3171 .invalidatepage = ext4_da_invalidatepage,
3172 .releasepage = ext4_releasepage,
3173 .direct_IO = ext4_direct_IO,
3174 .migratepage = buffer_migrate_page,
3175 .is_partially_uptodate = block_is_partially_uptodate,
3176 .error_remove_page = generic_error_remove_page,
3179 void ext4_set_aops(struct inode *inode)
3181 switch (ext4_inode_journal_mode(inode)) {
3182 case EXT4_INODE_ORDERED_DATA_MODE:
3183 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3185 case EXT4_INODE_WRITEBACK_DATA_MODE:
3186 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3188 case EXT4_INODE_JOURNAL_DATA_MODE:
3189 inode->i_mapping->a_ops = &ext4_journalled_aops;
3194 if (test_opt(inode->i_sb, DELALLOC))
3195 inode->i_mapping->a_ops = &ext4_da_aops;
3197 inode->i_mapping->a_ops = &ext4_aops;
3200 static int __ext4_block_zero_page_range(handle_t *handle,
3201 struct address_space *mapping, loff_t from, loff_t length)
3203 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3204 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3205 unsigned blocksize, pos;
3207 struct inode *inode = mapping->host;
3208 struct buffer_head *bh;
3212 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3213 mapping_gfp_mask(mapping) & ~__GFP_FS);
3217 blocksize = inode->i_sb->s_blocksize;
3219 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3221 if (!page_has_buffers(page))
3222 create_empty_buffers(page, blocksize, 0);
3224 /* Find the buffer that contains "offset" */
3225 bh = page_buffers(page);
3227 while (offset >= pos) {
3228 bh = bh->b_this_page;
3232 if (buffer_freed(bh)) {
3233 BUFFER_TRACE(bh, "freed: skip");
3236 if (!buffer_mapped(bh)) {
3237 BUFFER_TRACE(bh, "unmapped");
3238 ext4_get_block(inode, iblock, bh, 0);
3239 /* unmapped? It's a hole - nothing to do */
3240 if (!buffer_mapped(bh)) {
3241 BUFFER_TRACE(bh, "still unmapped");
3246 /* Ok, it's mapped. Make sure it's up-to-date */
3247 if (PageUptodate(page))
3248 set_buffer_uptodate(bh);
3250 if (!buffer_uptodate(bh)) {
3252 ll_rw_block(READ, 1, &bh);
3254 /* Uhhuh. Read error. Complain and punt. */
3255 if (!buffer_uptodate(bh))
3258 if (ext4_should_journal_data(inode)) {
3259 BUFFER_TRACE(bh, "get write access");
3260 err = ext4_journal_get_write_access(handle, bh);
3264 zero_user(page, offset, length);
3265 BUFFER_TRACE(bh, "zeroed end of block");
3267 if (ext4_should_journal_data(inode)) {
3268 err = ext4_handle_dirty_metadata(handle, inode, bh);
3271 mark_buffer_dirty(bh);
3272 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3273 err = ext4_jbd2_file_inode(handle, inode);
3278 page_cache_release(page);
3283 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3284 * starting from file offset 'from'. The range to be zero'd must
3285 * be contained with in one block. If the specified range exceeds
3286 * the end of the block it will be shortened to end of the block
3287 * that cooresponds to 'from'
3289 static int ext4_block_zero_page_range(handle_t *handle,
3290 struct address_space *mapping, loff_t from, loff_t length)
3292 struct inode *inode = mapping->host;
3293 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3294 unsigned blocksize = inode->i_sb->s_blocksize;
3295 unsigned max = blocksize - (offset & (blocksize - 1));
3298 * correct length if it does not fall between
3299 * 'from' and the end of the block
3301 if (length > max || length < 0)
3305 return dax_zero_page_range(inode, from, length, ext4_get_block);
3306 return __ext4_block_zero_page_range(handle, mapping, from, length);
3310 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3311 * up to the end of the block which corresponds to `from'.
3312 * This required during truncate. We need to physically zero the tail end
3313 * of that block so it doesn't yield old data if the file is later grown.
3315 static int ext4_block_truncate_page(handle_t *handle,
3316 struct address_space *mapping, loff_t from)
3318 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3321 struct inode *inode = mapping->host;
3323 blocksize = inode->i_sb->s_blocksize;
3324 length = blocksize - (offset & (blocksize - 1));
3326 return ext4_block_zero_page_range(handle, mapping, from, length);
3329 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3330 loff_t lstart, loff_t length)
3332 struct super_block *sb = inode->i_sb;
3333 struct address_space *mapping = inode->i_mapping;
3334 unsigned partial_start, partial_end;
3335 ext4_fsblk_t start, end;
3336 loff_t byte_end = (lstart + length - 1);
3339 partial_start = lstart & (sb->s_blocksize - 1);
3340 partial_end = byte_end & (sb->s_blocksize - 1);
3342 start = lstart >> sb->s_blocksize_bits;
3343 end = byte_end >> sb->s_blocksize_bits;
3345 /* Handle partial zero within the single block */
3347 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3348 err = ext4_block_zero_page_range(handle, mapping,
3352 /* Handle partial zero out on the start of the range */
3353 if (partial_start) {
3354 err = ext4_block_zero_page_range(handle, mapping,
3355 lstart, sb->s_blocksize);
3359 /* Handle partial zero out on the end of the range */
3360 if (partial_end != sb->s_blocksize - 1)
3361 err = ext4_block_zero_page_range(handle, mapping,
3362 byte_end - partial_end,
3367 int ext4_can_truncate(struct inode *inode)
3369 if (S_ISREG(inode->i_mode))
3371 if (S_ISDIR(inode->i_mode))
3373 if (S_ISLNK(inode->i_mode))
3374 return !ext4_inode_is_fast_symlink(inode);
3379 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3380 * associated with the given offset and length
3382 * @inode: File inode
3383 * @offset: The offset where the hole will begin
3384 * @len: The length of the hole
3386 * Returns: 0 on success or negative on failure
3389 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3391 struct super_block *sb = inode->i_sb;
3392 ext4_lblk_t first_block, stop_block;
3393 struct address_space *mapping = inode->i_mapping;
3394 loff_t first_block_offset, last_block_offset;
3396 unsigned int credits;
3399 if (!S_ISREG(inode->i_mode))
3402 trace_ext4_punch_hole(inode, offset, length, 0);
3405 * Write out all dirty pages to avoid race conditions
3406 * Then release them.
3408 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3409 ret = filemap_write_and_wait_range(mapping, offset,
3410 offset + length - 1);
3415 mutex_lock(&inode->i_mutex);
3417 /* No need to punch hole beyond i_size */
3418 if (offset >= inode->i_size)
3422 * If the hole extends beyond i_size, set the hole
3423 * to end after the page that contains i_size
3425 if (offset + length > inode->i_size) {
3426 length = inode->i_size +
3427 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3431 if (offset & (sb->s_blocksize - 1) ||
3432 (offset + length) & (sb->s_blocksize - 1)) {
3434 * Attach jinode to inode for jbd2 if we do any zeroing of
3437 ret = ext4_inode_attach_jinode(inode);
3443 first_block_offset = round_up(offset, sb->s_blocksize);
3444 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3446 /* Now release the pages and zero block aligned part of pages*/
3447 if (last_block_offset > first_block_offset)
3448 truncate_pagecache_range(inode, first_block_offset,
3451 /* Wait all existing dio workers, newcomers will block on i_mutex */
3452 ext4_inode_block_unlocked_dio(inode);
3453 inode_dio_wait(inode);
3455 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3456 credits = ext4_writepage_trans_blocks(inode);
3458 credits = ext4_blocks_for_truncate(inode);
3459 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3460 if (IS_ERR(handle)) {
3461 ret = PTR_ERR(handle);
3462 ext4_std_error(sb, ret);
3466 ret = ext4_zero_partial_blocks(handle, inode, offset,
3471 first_block = (offset + sb->s_blocksize - 1) >>
3472 EXT4_BLOCK_SIZE_BITS(sb);
3473 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3475 /* If there are no blocks to remove, return now */
3476 if (first_block >= stop_block)
3479 down_write(&EXT4_I(inode)->i_data_sem);
3480 ext4_discard_preallocations(inode);
3482 ret = ext4_es_remove_extent(inode, first_block,
3483 stop_block - first_block);
3485 up_write(&EXT4_I(inode)->i_data_sem);
3489 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3490 ret = ext4_ext_remove_space(inode, first_block,
3493 ret = ext4_ind_remove_space(handle, inode, first_block,
3496 up_write(&EXT4_I(inode)->i_data_sem);
3498 ext4_handle_sync(handle);
3500 /* Now release the pages again to reduce race window */
3501 if (last_block_offset > first_block_offset)
3502 truncate_pagecache_range(inode, first_block_offset,
3505 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3506 ext4_mark_inode_dirty(handle, inode);
3508 ext4_journal_stop(handle);
3510 ext4_inode_resume_unlocked_dio(inode);
3512 mutex_unlock(&inode->i_mutex);
3516 int ext4_inode_attach_jinode(struct inode *inode)
3518 struct ext4_inode_info *ei = EXT4_I(inode);
3519 struct jbd2_inode *jinode;
3521 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3524 jinode = jbd2_alloc_inode(GFP_KERNEL);
3525 spin_lock(&inode->i_lock);
3528 spin_unlock(&inode->i_lock);
3531 ei->jinode = jinode;
3532 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3535 spin_unlock(&inode->i_lock);
3536 if (unlikely(jinode != NULL))
3537 jbd2_free_inode(jinode);
3544 * We block out ext4_get_block() block instantiations across the entire
3545 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3546 * simultaneously on behalf of the same inode.
3548 * As we work through the truncate and commit bits of it to the journal there
3549 * is one core, guiding principle: the file's tree must always be consistent on
3550 * disk. We must be able to restart the truncate after a crash.
3552 * The file's tree may be transiently inconsistent in memory (although it
3553 * probably isn't), but whenever we close off and commit a journal transaction,
3554 * the contents of (the filesystem + the journal) must be consistent and
3555 * restartable. It's pretty simple, really: bottom up, right to left (although
3556 * left-to-right works OK too).
3558 * Note that at recovery time, journal replay occurs *before* the restart of
3559 * truncate against the orphan inode list.
3561 * The committed inode has the new, desired i_size (which is the same as
3562 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3563 * that this inode's truncate did not complete and it will again call
3564 * ext4_truncate() to have another go. So there will be instantiated blocks
3565 * to the right of the truncation point in a crashed ext4 filesystem. But
3566 * that's fine - as long as they are linked from the inode, the post-crash
3567 * ext4_truncate() run will find them and release them.
3569 void ext4_truncate(struct inode *inode)
3571 struct ext4_inode_info *ei = EXT4_I(inode);
3572 unsigned int credits;
3574 struct address_space *mapping = inode->i_mapping;
3577 * There is a possibility that we're either freeing the inode
3578 * or it's a completely new inode. In those cases we might not
3579 * have i_mutex locked because it's not necessary.
3581 if (!(inode->i_state & (I_NEW|I_FREEING)))
3582 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3583 trace_ext4_truncate_enter(inode);
3585 if (!ext4_can_truncate(inode))
3588 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3590 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3591 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3593 if (ext4_has_inline_data(inode)) {
3596 ext4_inline_data_truncate(inode, &has_inline);
3601 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3602 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3603 if (ext4_inode_attach_jinode(inode) < 0)
3607 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3608 credits = ext4_writepage_trans_blocks(inode);
3610 credits = ext4_blocks_for_truncate(inode);
3612 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3613 if (IS_ERR(handle)) {
3614 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3618 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3619 ext4_block_truncate_page(handle, mapping, inode->i_size);
3622 * We add the inode to the orphan list, so that if this
3623 * truncate spans multiple transactions, and we crash, we will
3624 * resume the truncate when the filesystem recovers. It also
3625 * marks the inode dirty, to catch the new size.
3627 * Implication: the file must always be in a sane, consistent
3628 * truncatable state while each transaction commits.
3630 if (ext4_orphan_add(handle, inode))
3633 down_write(&EXT4_I(inode)->i_data_sem);
3635 ext4_discard_preallocations(inode);
3637 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3638 ext4_ext_truncate(handle, inode);
3640 ext4_ind_truncate(handle, inode);
3642 up_write(&ei->i_data_sem);
3645 ext4_handle_sync(handle);
3649 * If this was a simple ftruncate() and the file will remain alive,
3650 * then we need to clear up the orphan record which we created above.
3651 * However, if this was a real unlink then we were called by
3652 * ext4_evict_inode(), and we allow that function to clean up the
3653 * orphan info for us.
3656 ext4_orphan_del(handle, inode);
3658 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3659 ext4_mark_inode_dirty(handle, inode);
3660 ext4_journal_stop(handle);
3662 trace_ext4_truncate_exit(inode);
3666 * ext4_get_inode_loc returns with an extra refcount against the inode's
3667 * underlying buffer_head on success. If 'in_mem' is true, we have all
3668 * data in memory that is needed to recreate the on-disk version of this
3671 static int __ext4_get_inode_loc(struct inode *inode,
3672 struct ext4_iloc *iloc, int in_mem)
3674 struct ext4_group_desc *gdp;
3675 struct buffer_head *bh;
3676 struct super_block *sb = inode->i_sb;
3678 int inodes_per_block, inode_offset;
3681 if (!ext4_valid_inum(sb, inode->i_ino))
3684 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3685 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3690 * Figure out the offset within the block group inode table
3692 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3693 inode_offset = ((inode->i_ino - 1) %
3694 EXT4_INODES_PER_GROUP(sb));
3695 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3696 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3698 bh = sb_getblk(sb, block);
3701 if (!buffer_uptodate(bh)) {
3705 * If the buffer has the write error flag, we have failed
3706 * to write out another inode in the same block. In this
3707 * case, we don't have to read the block because we may
3708 * read the old inode data successfully.
3710 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3711 set_buffer_uptodate(bh);
3713 if (buffer_uptodate(bh)) {
3714 /* someone brought it uptodate while we waited */
3720 * If we have all information of the inode in memory and this
3721 * is the only valid inode in the block, we need not read the
3725 struct buffer_head *bitmap_bh;
3728 start = inode_offset & ~(inodes_per_block - 1);
3730 /* Is the inode bitmap in cache? */
3731 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3732 if (unlikely(!bitmap_bh))
3736 * If the inode bitmap isn't in cache then the
3737 * optimisation may end up performing two reads instead
3738 * of one, so skip it.
3740 if (!buffer_uptodate(bitmap_bh)) {
3744 for (i = start; i < start + inodes_per_block; i++) {
3745 if (i == inode_offset)
3747 if (ext4_test_bit(i, bitmap_bh->b_data))
3751 if (i == start + inodes_per_block) {
3752 /* all other inodes are free, so skip I/O */
3753 memset(bh->b_data, 0, bh->b_size);
3754 set_buffer_uptodate(bh);
3762 * If we need to do any I/O, try to pre-readahead extra
3763 * blocks from the inode table.
3765 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3766 ext4_fsblk_t b, end, table;
3768 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3770 table = ext4_inode_table(sb, gdp);
3771 /* s_inode_readahead_blks is always a power of 2 */
3772 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3776 num = EXT4_INODES_PER_GROUP(sb);
3777 if (ext4_has_group_desc_csum(sb))
3778 num -= ext4_itable_unused_count(sb, gdp);
3779 table += num / inodes_per_block;
3783 sb_breadahead(sb, b++);
3787 * There are other valid inodes in the buffer, this inode
3788 * has in-inode xattrs, or we don't have this inode in memory.
3789 * Read the block from disk.
3791 trace_ext4_load_inode(inode);
3793 bh->b_end_io = end_buffer_read_sync;
3794 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3796 if (!buffer_uptodate(bh)) {
3797 EXT4_ERROR_INODE_BLOCK(inode, block,
3798 "unable to read itable block");
3808 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3810 /* We have all inode data except xattrs in memory here. */
3811 return __ext4_get_inode_loc(inode, iloc,
3812 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3815 void ext4_set_inode_flags(struct inode *inode)
3817 unsigned int flags = EXT4_I(inode)->i_flags;
3818 unsigned int new_fl = 0;
3820 if (flags & EXT4_SYNC_FL)
3822 if (flags & EXT4_APPEND_FL)
3824 if (flags & EXT4_IMMUTABLE_FL)
3825 new_fl |= S_IMMUTABLE;
3826 if (flags & EXT4_NOATIME_FL)
3827 new_fl |= S_NOATIME;
3828 if (flags & EXT4_DIRSYNC_FL)
3829 new_fl |= S_DIRSYNC;
3830 if (test_opt(inode->i_sb, DAX))
3832 inode_set_flags(inode, new_fl,
3833 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
3836 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3837 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3839 unsigned int vfs_fl;
3840 unsigned long old_fl, new_fl;
3843 vfs_fl = ei->vfs_inode.i_flags;
3844 old_fl = ei->i_flags;
3845 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3846 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3848 if (vfs_fl & S_SYNC)
3849 new_fl |= EXT4_SYNC_FL;
3850 if (vfs_fl & S_APPEND)
3851 new_fl |= EXT4_APPEND_FL;
3852 if (vfs_fl & S_IMMUTABLE)
3853 new_fl |= EXT4_IMMUTABLE_FL;
3854 if (vfs_fl & S_NOATIME)
3855 new_fl |= EXT4_NOATIME_FL;
3856 if (vfs_fl & S_DIRSYNC)
3857 new_fl |= EXT4_DIRSYNC_FL;
3858 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3861 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3862 struct ext4_inode_info *ei)
3865 struct inode *inode = &(ei->vfs_inode);
3866 struct super_block *sb = inode->i_sb;
3868 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3869 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3870 /* we are using combined 48 bit field */
3871 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3872 le32_to_cpu(raw_inode->i_blocks_lo);
3873 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3874 /* i_blocks represent file system block size */
3875 return i_blocks << (inode->i_blkbits - 9);
3880 return le32_to_cpu(raw_inode->i_blocks_lo);
3884 static inline void ext4_iget_extra_inode(struct inode *inode,
3885 struct ext4_inode *raw_inode,
3886 struct ext4_inode_info *ei)
3888 __le32 *magic = (void *)raw_inode +
3889 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3890 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3891 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3892 ext4_find_inline_data_nolock(inode);
3894 EXT4_I(inode)->i_inline_off = 0;
3897 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3899 struct ext4_iloc iloc;
3900 struct ext4_inode *raw_inode;
3901 struct ext4_inode_info *ei;
3902 struct inode *inode;
3903 journal_t *journal = EXT4_SB(sb)->s_journal;
3909 inode = iget_locked(sb, ino);
3911 return ERR_PTR(-ENOMEM);
3912 if (!(inode->i_state & I_NEW))
3918 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3921 raw_inode = ext4_raw_inode(&iloc);
3923 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3924 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3925 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3926 EXT4_INODE_SIZE(inode->i_sb)) {
3927 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3928 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3929 EXT4_INODE_SIZE(inode->i_sb));
3934 ei->i_extra_isize = 0;
3936 /* Precompute checksum seed for inode metadata */
3937 if (ext4_has_metadata_csum(sb)) {
3938 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3940 __le32 inum = cpu_to_le32(inode->i_ino);
3941 __le32 gen = raw_inode->i_generation;
3942 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3944 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3948 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3949 EXT4_ERROR_INODE(inode, "checksum invalid");
3954 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3955 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3956 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3957 if (!(test_opt(inode->i_sb, NO_UID32))) {
3958 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3959 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3961 i_uid_write(inode, i_uid);
3962 i_gid_write(inode, i_gid);
3963 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3965 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3966 ei->i_inline_off = 0;
3967 ei->i_dir_start_lookup = 0;
3968 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3969 /* We now have enough fields to check if the inode was active or not.
3970 * This is needed because nfsd might try to access dead inodes
3971 * the test is that same one that e2fsck uses
3972 * NeilBrown 1999oct15
3974 if (inode->i_nlink == 0) {
3975 if ((inode->i_mode == 0 ||
3976 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
3977 ino != EXT4_BOOT_LOADER_INO) {
3978 /* this inode is deleted */
3982 /* The only unlinked inodes we let through here have
3983 * valid i_mode and are being read by the orphan
3984 * recovery code: that's fine, we're about to complete
3985 * the process of deleting those.
3986 * OR it is the EXT4_BOOT_LOADER_INO which is
3987 * not initialized on a new filesystem. */
3989 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3990 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3991 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3992 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3994 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3995 inode->i_size = ext4_isize(raw_inode);
3996 ei->i_disksize = inode->i_size;
3998 ei->i_reserved_quota = 0;
4000 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4001 ei->i_block_group = iloc.block_group;
4002 ei->i_last_alloc_group = ~0;
4004 * NOTE! The in-memory inode i_data array is in little-endian order
4005 * even on big-endian machines: we do NOT byteswap the block numbers!
4007 for (block = 0; block < EXT4_N_BLOCKS; block++)
4008 ei->i_data[block] = raw_inode->i_block[block];
4009 INIT_LIST_HEAD(&ei->i_orphan);
4012 * Set transaction id's of transactions that have to be committed
4013 * to finish f[data]sync. We set them to currently running transaction
4014 * as we cannot be sure that the inode or some of its metadata isn't
4015 * part of the transaction - the inode could have been reclaimed and
4016 * now it is reread from disk.
4019 transaction_t *transaction;
4022 read_lock(&journal->j_state_lock);
4023 if (journal->j_running_transaction)
4024 transaction = journal->j_running_transaction;
4026 transaction = journal->j_committing_transaction;
4028 tid = transaction->t_tid;
4030 tid = journal->j_commit_sequence;
4031 read_unlock(&journal->j_state_lock);
4032 ei->i_sync_tid = tid;
4033 ei->i_datasync_tid = tid;
4036 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4037 if (ei->i_extra_isize == 0) {
4038 /* The extra space is currently unused. Use it. */
4039 ei->i_extra_isize = sizeof(struct ext4_inode) -
4040 EXT4_GOOD_OLD_INODE_SIZE;
4042 ext4_iget_extra_inode(inode, raw_inode, ei);
4046 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4047 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4048 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4049 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4051 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4052 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4053 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4054 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4056 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4061 if (ei->i_file_acl &&
4062 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4063 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4067 } else if (!ext4_has_inline_data(inode)) {
4068 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4069 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4070 (S_ISLNK(inode->i_mode) &&
4071 !ext4_inode_is_fast_symlink(inode))))
4072 /* Validate extent which is part of inode */
4073 ret = ext4_ext_check_inode(inode);
4074 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4075 (S_ISLNK(inode->i_mode) &&
4076 !ext4_inode_is_fast_symlink(inode))) {
4077 /* Validate block references which are part of inode */
4078 ret = ext4_ind_check_inode(inode);
4084 if (S_ISREG(inode->i_mode)) {
4085 inode->i_op = &ext4_file_inode_operations;
4086 if (test_opt(inode->i_sb, DAX))
4087 inode->i_fop = &ext4_dax_file_operations;
4089 inode->i_fop = &ext4_file_operations;
4090 ext4_set_aops(inode);
4091 } else if (S_ISDIR(inode->i_mode)) {
4092 inode->i_op = &ext4_dir_inode_operations;
4093 inode->i_fop = &ext4_dir_operations;
4094 } else if (S_ISLNK(inode->i_mode)) {
4095 if (ext4_inode_is_fast_symlink(inode)) {
4096 inode->i_op = &ext4_fast_symlink_inode_operations;
4097 nd_terminate_link(ei->i_data, inode->i_size,
4098 sizeof(ei->i_data) - 1);
4100 inode->i_op = &ext4_symlink_inode_operations;
4101 ext4_set_aops(inode);
4103 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4104 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4105 inode->i_op = &ext4_special_inode_operations;
4106 if (raw_inode->i_block[0])
4107 init_special_inode(inode, inode->i_mode,
4108 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4110 init_special_inode(inode, inode->i_mode,
4111 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4112 } else if (ino == EXT4_BOOT_LOADER_INO) {
4113 make_bad_inode(inode);
4116 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4120 ext4_set_inode_flags(inode);
4121 unlock_new_inode(inode);
4127 return ERR_PTR(ret);
4130 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4132 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4133 return ERR_PTR(-EIO);
4134 return ext4_iget(sb, ino);
4137 static int ext4_inode_blocks_set(handle_t *handle,
4138 struct ext4_inode *raw_inode,
4139 struct ext4_inode_info *ei)
4141 struct inode *inode = &(ei->vfs_inode);
4142 u64 i_blocks = inode->i_blocks;
4143 struct super_block *sb = inode->i_sb;
4145 if (i_blocks <= ~0U) {
4147 * i_blocks can be represented in a 32 bit variable
4148 * as multiple of 512 bytes
4150 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4151 raw_inode->i_blocks_high = 0;
4152 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4155 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4158 if (i_blocks <= 0xffffffffffffULL) {
4160 * i_blocks can be represented in a 48 bit variable
4161 * as multiple of 512 bytes
4163 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4164 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4165 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4167 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4168 /* i_block is stored in file system block size */
4169 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4170 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4171 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4176 struct other_inode {
4177 unsigned long orig_ino;
4178 struct ext4_inode *raw_inode;
4181 static int other_inode_match(struct inode * inode, unsigned long ino,
4184 struct other_inode *oi = (struct other_inode *) data;
4186 if ((inode->i_ino != ino) ||
4187 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4188 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4189 ((inode->i_state & I_DIRTY_TIME) == 0))
4191 spin_lock(&inode->i_lock);
4192 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4193 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4194 (inode->i_state & I_DIRTY_TIME)) {
4195 struct ext4_inode_info *ei = EXT4_I(inode);
4197 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4198 spin_unlock(&inode->i_lock);
4200 spin_lock(&ei->i_raw_lock);
4201 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4202 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4203 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4204 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4205 spin_unlock(&ei->i_raw_lock);
4206 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4209 spin_unlock(&inode->i_lock);
4214 * Opportunistically update the other time fields for other inodes in
4215 * the same inode table block.
4217 static void ext4_update_other_inodes_time(struct super_block *sb,
4218 unsigned long orig_ino, char *buf)
4220 struct other_inode oi;
4222 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4223 int inode_size = EXT4_INODE_SIZE(sb);
4225 oi.orig_ino = orig_ino;
4226 ino = orig_ino & ~(inodes_per_block - 1);
4227 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4228 if (ino == orig_ino)
4230 oi.raw_inode = (struct ext4_inode *) buf;
4231 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4236 * Post the struct inode info into an on-disk inode location in the
4237 * buffer-cache. This gobbles the caller's reference to the
4238 * buffer_head in the inode location struct.
4240 * The caller must have write access to iloc->bh.
4242 static int ext4_do_update_inode(handle_t *handle,
4243 struct inode *inode,
4244 struct ext4_iloc *iloc)
4246 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4247 struct ext4_inode_info *ei = EXT4_I(inode);
4248 struct buffer_head *bh = iloc->bh;
4249 struct super_block *sb = inode->i_sb;
4250 int err = 0, rc, block;
4251 int need_datasync = 0, set_large_file = 0;
4255 spin_lock(&ei->i_raw_lock);
4257 /* For fields not tracked in the in-memory inode,
4258 * initialise them to zero for new inodes. */
4259 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4260 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4262 ext4_get_inode_flags(ei);
4263 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4264 i_uid = i_uid_read(inode);
4265 i_gid = i_gid_read(inode);
4266 if (!(test_opt(inode->i_sb, NO_UID32))) {
4267 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4268 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4270 * Fix up interoperability with old kernels. Otherwise, old inodes get
4271 * re-used with the upper 16 bits of the uid/gid intact
4274 raw_inode->i_uid_high =
4275 cpu_to_le16(high_16_bits(i_uid));
4276 raw_inode->i_gid_high =
4277 cpu_to_le16(high_16_bits(i_gid));
4279 raw_inode->i_uid_high = 0;
4280 raw_inode->i_gid_high = 0;
4283 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4284 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4285 raw_inode->i_uid_high = 0;
4286 raw_inode->i_gid_high = 0;
4288 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4290 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4291 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4292 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4293 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4295 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4297 spin_unlock(&ei->i_raw_lock);
4300 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4301 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4302 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4303 raw_inode->i_file_acl_high =
4304 cpu_to_le16(ei->i_file_acl >> 32);
4305 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4306 if (ei->i_disksize != ext4_isize(raw_inode)) {
4307 ext4_isize_set(raw_inode, ei->i_disksize);
4310 if (ei->i_disksize > 0x7fffffffULL) {
4311 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4312 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4313 EXT4_SB(sb)->s_es->s_rev_level ==
4314 cpu_to_le32(EXT4_GOOD_OLD_REV))
4317 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4318 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4319 if (old_valid_dev(inode->i_rdev)) {
4320 raw_inode->i_block[0] =
4321 cpu_to_le32(old_encode_dev(inode->i_rdev));
4322 raw_inode->i_block[1] = 0;
4324 raw_inode->i_block[0] = 0;
4325 raw_inode->i_block[1] =
4326 cpu_to_le32(new_encode_dev(inode->i_rdev));
4327 raw_inode->i_block[2] = 0;
4329 } else if (!ext4_has_inline_data(inode)) {
4330 for (block = 0; block < EXT4_N_BLOCKS; block++)
4331 raw_inode->i_block[block] = ei->i_data[block];
4334 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4335 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4336 if (ei->i_extra_isize) {
4337 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4338 raw_inode->i_version_hi =
4339 cpu_to_le32(inode->i_version >> 32);
4340 raw_inode->i_extra_isize =
4341 cpu_to_le16(ei->i_extra_isize);
4344 ext4_inode_csum_set(inode, raw_inode, ei);
4345 spin_unlock(&ei->i_raw_lock);
4346 if (inode->i_sb->s_flags & MS_LAZYTIME)
4347 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4350 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4351 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4354 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4355 if (set_large_file) {
4356 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4357 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4360 ext4_update_dynamic_rev(sb);
4361 EXT4_SET_RO_COMPAT_FEATURE(sb,
4362 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4363 ext4_handle_sync(handle);
4364 err = ext4_handle_dirty_super(handle, sb);
4366 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4369 ext4_std_error(inode->i_sb, err);
4374 * ext4_write_inode()
4376 * We are called from a few places:
4378 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4379 * Here, there will be no transaction running. We wait for any running
4380 * transaction to commit.
4382 * - Within flush work (sys_sync(), kupdate and such).
4383 * We wait on commit, if told to.
4385 * - Within iput_final() -> write_inode_now()
4386 * We wait on commit, if told to.
4388 * In all cases it is actually safe for us to return without doing anything,
4389 * because the inode has been copied into a raw inode buffer in
4390 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4393 * Note that we are absolutely dependent upon all inode dirtiers doing the
4394 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4395 * which we are interested.
4397 * It would be a bug for them to not do this. The code:
4399 * mark_inode_dirty(inode)
4401 * inode->i_size = expr;
4403 * is in error because write_inode() could occur while `stuff()' is running,
4404 * and the new i_size will be lost. Plus the inode will no longer be on the
4405 * superblock's dirty inode list.
4407 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4411 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4414 if (EXT4_SB(inode->i_sb)->s_journal) {
4415 if (ext4_journal_current_handle()) {
4416 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4422 * No need to force transaction in WB_SYNC_NONE mode. Also
4423 * ext4_sync_fs() will force the commit after everything is
4426 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4429 err = ext4_force_commit(inode->i_sb);
4431 struct ext4_iloc iloc;
4433 err = __ext4_get_inode_loc(inode, &iloc, 0);
4437 * sync(2) will flush the whole buffer cache. No need to do
4438 * it here separately for each inode.
4440 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4441 sync_dirty_buffer(iloc.bh);
4442 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4443 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4444 "IO error syncing inode");
4453 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4454 * buffers that are attached to a page stradding i_size and are undergoing
4455 * commit. In that case we have to wait for commit to finish and try again.
4457 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4461 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4462 tid_t commit_tid = 0;
4465 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4467 * All buffers in the last page remain valid? Then there's nothing to
4468 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4471 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4474 page = find_lock_page(inode->i_mapping,
4475 inode->i_size >> PAGE_CACHE_SHIFT);
4478 ret = __ext4_journalled_invalidatepage(page, offset,
4479 PAGE_CACHE_SIZE - offset);
4481 page_cache_release(page);
4485 read_lock(&journal->j_state_lock);
4486 if (journal->j_committing_transaction)
4487 commit_tid = journal->j_committing_transaction->t_tid;
4488 read_unlock(&journal->j_state_lock);
4490 jbd2_log_wait_commit(journal, commit_tid);
4497 * Called from notify_change.
4499 * We want to trap VFS attempts to truncate the file as soon as
4500 * possible. In particular, we want to make sure that when the VFS
4501 * shrinks i_size, we put the inode on the orphan list and modify
4502 * i_disksize immediately, so that during the subsequent flushing of
4503 * dirty pages and freeing of disk blocks, we can guarantee that any
4504 * commit will leave the blocks being flushed in an unused state on
4505 * disk. (On recovery, the inode will get truncated and the blocks will
4506 * be freed, so we have a strong guarantee that no future commit will
4507 * leave these blocks visible to the user.)
4509 * Another thing we have to assure is that if we are in ordered mode
4510 * and inode is still attached to the committing transaction, we must
4511 * we start writeout of all the dirty pages which are being truncated.
4512 * This way we are sure that all the data written in the previous
4513 * transaction are already on disk (truncate waits for pages under
4516 * Called with inode->i_mutex down.
4518 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4520 struct inode *inode = dentry->d_inode;
4523 const unsigned int ia_valid = attr->ia_valid;
4525 error = inode_change_ok(inode, attr);
4529 if (is_quota_modification(inode, attr))
4530 dquot_initialize(inode);
4531 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4532 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4535 /* (user+group)*(old+new) structure, inode write (sb,
4536 * inode block, ? - but truncate inode update has it) */
4537 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4538 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4539 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4540 if (IS_ERR(handle)) {
4541 error = PTR_ERR(handle);
4544 error = dquot_transfer(inode, attr);
4546 ext4_journal_stop(handle);
4549 /* Update corresponding info in inode so that everything is in
4550 * one transaction */
4551 if (attr->ia_valid & ATTR_UID)
4552 inode->i_uid = attr->ia_uid;
4553 if (attr->ia_valid & ATTR_GID)
4554 inode->i_gid = attr->ia_gid;
4555 error = ext4_mark_inode_dirty(handle, inode);
4556 ext4_journal_stop(handle);
4559 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4562 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4563 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4565 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4569 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4570 inode_inc_iversion(inode);
4572 if (S_ISREG(inode->i_mode) &&
4573 (attr->ia_size < inode->i_size)) {
4574 if (ext4_should_order_data(inode)) {
4575 error = ext4_begin_ordered_truncate(inode,
4580 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4581 if (IS_ERR(handle)) {
4582 error = PTR_ERR(handle);
4585 if (ext4_handle_valid(handle)) {
4586 error = ext4_orphan_add(handle, inode);
4589 down_write(&EXT4_I(inode)->i_data_sem);
4590 EXT4_I(inode)->i_disksize = attr->ia_size;
4591 rc = ext4_mark_inode_dirty(handle, inode);
4595 * We have to update i_size under i_data_sem together
4596 * with i_disksize to avoid races with writeback code
4597 * running ext4_wb_update_i_disksize().
4600 i_size_write(inode, attr->ia_size);
4601 up_write(&EXT4_I(inode)->i_data_sem);
4602 ext4_journal_stop(handle);
4604 ext4_orphan_del(NULL, inode);
4608 loff_t oldsize = inode->i_size;
4610 i_size_write(inode, attr->ia_size);
4611 pagecache_isize_extended(inode, oldsize, inode->i_size);
4615 * Blocks are going to be removed from the inode. Wait
4616 * for dio in flight. Temporarily disable
4617 * dioread_nolock to prevent livelock.
4620 if (!ext4_should_journal_data(inode)) {
4621 ext4_inode_block_unlocked_dio(inode);
4622 inode_dio_wait(inode);
4623 ext4_inode_resume_unlocked_dio(inode);
4625 ext4_wait_for_tail_page_commit(inode);
4628 * Truncate pagecache after we've waited for commit
4629 * in data=journal mode to make pages freeable.
4631 truncate_pagecache(inode, inode->i_size);
4634 * We want to call ext4_truncate() even if attr->ia_size ==
4635 * inode->i_size for cases like truncation of fallocated space
4637 if (attr->ia_valid & ATTR_SIZE)
4638 ext4_truncate(inode);
4641 setattr_copy(inode, attr);
4642 mark_inode_dirty(inode);
4646 * If the call to ext4_truncate failed to get a transaction handle at
4647 * all, we need to clean up the in-core orphan list manually.
4649 if (orphan && inode->i_nlink)
4650 ext4_orphan_del(NULL, inode);
4652 if (!rc && (ia_valid & ATTR_MODE))
4653 rc = posix_acl_chmod(inode, inode->i_mode);
4656 ext4_std_error(inode->i_sb, error);
4662 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4665 struct inode *inode;
4666 unsigned long long delalloc_blocks;
4668 inode = dentry->d_inode;
4669 generic_fillattr(inode, stat);
4672 * If there is inline data in the inode, the inode will normally not
4673 * have data blocks allocated (it may have an external xattr block).
4674 * Report at least one sector for such files, so tools like tar, rsync,
4675 * others doen't incorrectly think the file is completely sparse.
4677 if (unlikely(ext4_has_inline_data(inode)))
4678 stat->blocks += (stat->size + 511) >> 9;
4681 * We can't update i_blocks if the block allocation is delayed
4682 * otherwise in the case of system crash before the real block
4683 * allocation is done, we will have i_blocks inconsistent with
4684 * on-disk file blocks.
4685 * We always keep i_blocks updated together with real
4686 * allocation. But to not confuse with user, stat
4687 * will return the blocks that include the delayed allocation
4688 * blocks for this file.
4690 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4691 EXT4_I(inode)->i_reserved_data_blocks);
4692 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4696 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4699 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4700 return ext4_ind_trans_blocks(inode, lblocks);
4701 return ext4_ext_index_trans_blocks(inode, pextents);
4705 * Account for index blocks, block groups bitmaps and block group
4706 * descriptor blocks if modify datablocks and index blocks
4707 * worse case, the indexs blocks spread over different block groups
4709 * If datablocks are discontiguous, they are possible to spread over
4710 * different block groups too. If they are contiguous, with flexbg,
4711 * they could still across block group boundary.
4713 * Also account for superblock, inode, quota and xattr blocks
4715 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4718 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4724 * How many index blocks need to touch to map @lblocks logical blocks
4725 * to @pextents physical extents?
4727 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4732 * Now let's see how many group bitmaps and group descriptors need
4735 groups = idxblocks + pextents;
4737 if (groups > ngroups)
4739 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4740 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4742 /* bitmaps and block group descriptor blocks */
4743 ret += groups + gdpblocks;
4745 /* Blocks for super block, inode, quota and xattr blocks */
4746 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4752 * Calculate the total number of credits to reserve to fit
4753 * the modification of a single pages into a single transaction,
4754 * which may include multiple chunks of block allocations.
4756 * This could be called via ext4_write_begin()
4758 * We need to consider the worse case, when
4759 * one new block per extent.
4761 int ext4_writepage_trans_blocks(struct inode *inode)
4763 int bpp = ext4_journal_blocks_per_page(inode);
4766 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4768 /* Account for data blocks for journalled mode */
4769 if (ext4_should_journal_data(inode))
4775 * Calculate the journal credits for a chunk of data modification.
4777 * This is called from DIO, fallocate or whoever calling
4778 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4780 * journal buffers for data blocks are not included here, as DIO
4781 * and fallocate do no need to journal data buffers.
4783 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4785 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4789 * The caller must have previously called ext4_reserve_inode_write().
4790 * Give this, we know that the caller already has write access to iloc->bh.
4792 int ext4_mark_iloc_dirty(handle_t *handle,
4793 struct inode *inode, struct ext4_iloc *iloc)
4797 if (IS_I_VERSION(inode))
4798 inode_inc_iversion(inode);
4800 /* the do_update_inode consumes one bh->b_count */
4803 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4804 err = ext4_do_update_inode(handle, inode, iloc);
4810 * On success, We end up with an outstanding reference count against
4811 * iloc->bh. This _must_ be cleaned up later.
4815 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4816 struct ext4_iloc *iloc)
4820 err = ext4_get_inode_loc(inode, iloc);
4822 BUFFER_TRACE(iloc->bh, "get_write_access");
4823 err = ext4_journal_get_write_access(handle, iloc->bh);
4829 ext4_std_error(inode->i_sb, err);
4834 * Expand an inode by new_extra_isize bytes.
4835 * Returns 0 on success or negative error number on failure.
4837 static int ext4_expand_extra_isize(struct inode *inode,
4838 unsigned int new_extra_isize,
4839 struct ext4_iloc iloc,
4842 struct ext4_inode *raw_inode;
4843 struct ext4_xattr_ibody_header *header;
4845 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4848 raw_inode = ext4_raw_inode(&iloc);
4850 header = IHDR(inode, raw_inode);
4852 /* No extended attributes present */
4853 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4854 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4855 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4857 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4861 /* try to expand with EAs present */
4862 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4867 * What we do here is to mark the in-core inode as clean with respect to inode
4868 * dirtiness (it may still be data-dirty).
4869 * This means that the in-core inode may be reaped by prune_icache
4870 * without having to perform any I/O. This is a very good thing,
4871 * because *any* task may call prune_icache - even ones which
4872 * have a transaction open against a different journal.
4874 * Is this cheating? Not really. Sure, we haven't written the
4875 * inode out, but prune_icache isn't a user-visible syncing function.
4876 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4877 * we start and wait on commits.
4879 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4881 struct ext4_iloc iloc;
4882 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4883 static unsigned int mnt_count;
4887 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4888 err = ext4_reserve_inode_write(handle, inode, &iloc);
4889 if (ext4_handle_valid(handle) &&
4890 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4891 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4893 * We need extra buffer credits since we may write into EA block
4894 * with this same handle. If journal_extend fails, then it will
4895 * only result in a minor loss of functionality for that inode.
4896 * If this is felt to be critical, then e2fsck should be run to
4897 * force a large enough s_min_extra_isize.
4899 if ((jbd2_journal_extend(handle,
4900 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4901 ret = ext4_expand_extra_isize(inode,
4902 sbi->s_want_extra_isize,
4905 ext4_set_inode_state(inode,
4906 EXT4_STATE_NO_EXPAND);
4908 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4909 ext4_warning(inode->i_sb,
4910 "Unable to expand inode %lu. Delete"
4911 " some EAs or run e2fsck.",
4914 le16_to_cpu(sbi->s_es->s_mnt_count);
4920 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4925 * ext4_dirty_inode() is called from __mark_inode_dirty()
4927 * We're really interested in the case where a file is being extended.
4928 * i_size has been changed by generic_commit_write() and we thus need
4929 * to include the updated inode in the current transaction.
4931 * Also, dquot_alloc_block() will always dirty the inode when blocks
4932 * are allocated to the file.
4934 * If the inode is marked synchronous, we don't honour that here - doing
4935 * so would cause a commit on atime updates, which we don't bother doing.
4936 * We handle synchronous inodes at the highest possible level.
4938 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
4939 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
4940 * to copy into the on-disk inode structure are the timestamp files.
4942 void ext4_dirty_inode(struct inode *inode, int flags)
4946 if (flags == I_DIRTY_TIME)
4948 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4952 ext4_mark_inode_dirty(handle, inode);
4954 ext4_journal_stop(handle);
4961 * Bind an inode's backing buffer_head into this transaction, to prevent
4962 * it from being flushed to disk early. Unlike
4963 * ext4_reserve_inode_write, this leaves behind no bh reference and
4964 * returns no iloc structure, so the caller needs to repeat the iloc
4965 * lookup to mark the inode dirty later.
4967 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4969 struct ext4_iloc iloc;
4973 err = ext4_get_inode_loc(inode, &iloc);
4975 BUFFER_TRACE(iloc.bh, "get_write_access");
4976 err = jbd2_journal_get_write_access(handle, iloc.bh);
4978 err = ext4_handle_dirty_metadata(handle,
4984 ext4_std_error(inode->i_sb, err);
4989 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4996 * We have to be very careful here: changing a data block's
4997 * journaling status dynamically is dangerous. If we write a
4998 * data block to the journal, change the status and then delete
4999 * that block, we risk forgetting to revoke the old log record
5000 * from the journal and so a subsequent replay can corrupt data.
5001 * So, first we make sure that the journal is empty and that
5002 * nobody is changing anything.
5005 journal = EXT4_JOURNAL(inode);
5008 if (is_journal_aborted(journal))
5010 /* We have to allocate physical blocks for delalloc blocks
5011 * before flushing journal. otherwise delalloc blocks can not
5012 * be allocated any more. even more truncate on delalloc blocks
5013 * could trigger BUG by flushing delalloc blocks in journal.
5014 * There is no delalloc block in non-journal data mode.
5016 if (val && test_opt(inode->i_sb, DELALLOC)) {
5017 err = ext4_alloc_da_blocks(inode);
5022 /* Wait for all existing dio workers */
5023 ext4_inode_block_unlocked_dio(inode);
5024 inode_dio_wait(inode);
5026 jbd2_journal_lock_updates(journal);
5029 * OK, there are no updates running now, and all cached data is
5030 * synced to disk. We are now in a completely consistent state
5031 * which doesn't have anything in the journal, and we know that
5032 * no filesystem updates are running, so it is safe to modify
5033 * the inode's in-core data-journaling state flag now.
5037 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5039 err = jbd2_journal_flush(journal);
5041 jbd2_journal_unlock_updates(journal);
5042 ext4_inode_resume_unlocked_dio(inode);
5045 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5047 ext4_set_aops(inode);
5049 jbd2_journal_unlock_updates(journal);
5050 ext4_inode_resume_unlocked_dio(inode);
5052 /* Finally we can mark the inode as dirty. */
5054 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5056 return PTR_ERR(handle);
5058 err = ext4_mark_inode_dirty(handle, inode);
5059 ext4_handle_sync(handle);
5060 ext4_journal_stop(handle);
5061 ext4_std_error(inode->i_sb, err);
5066 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5068 return !buffer_mapped(bh);
5071 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5073 struct page *page = vmf->page;
5077 struct file *file = vma->vm_file;
5078 struct inode *inode = file_inode(file);
5079 struct address_space *mapping = inode->i_mapping;
5081 get_block_t *get_block;
5084 sb_start_pagefault(inode->i_sb);
5085 file_update_time(vma->vm_file);
5086 /* Delalloc case is easy... */
5087 if (test_opt(inode->i_sb, DELALLOC) &&
5088 !ext4_should_journal_data(inode) &&
5089 !ext4_nonda_switch(inode->i_sb)) {
5091 ret = __block_page_mkwrite(vma, vmf,
5092 ext4_da_get_block_prep);
5093 } while (ret == -ENOSPC &&
5094 ext4_should_retry_alloc(inode->i_sb, &retries));
5099 size = i_size_read(inode);
5100 /* Page got truncated from under us? */
5101 if (page->mapping != mapping || page_offset(page) > size) {
5103 ret = VM_FAULT_NOPAGE;
5107 if (page->index == size >> PAGE_CACHE_SHIFT)
5108 len = size & ~PAGE_CACHE_MASK;
5110 len = PAGE_CACHE_SIZE;
5112 * Return if we have all the buffers mapped. This avoids the need to do
5113 * journal_start/journal_stop which can block and take a long time
5115 if (page_has_buffers(page)) {
5116 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5118 ext4_bh_unmapped)) {
5119 /* Wait so that we don't change page under IO */
5120 wait_for_stable_page(page);
5121 ret = VM_FAULT_LOCKED;
5126 /* OK, we need to fill the hole... */
5127 if (ext4_should_dioread_nolock(inode))
5128 get_block = ext4_get_block_write;
5130 get_block = ext4_get_block;
5132 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5133 ext4_writepage_trans_blocks(inode));
5134 if (IS_ERR(handle)) {
5135 ret = VM_FAULT_SIGBUS;
5138 ret = __block_page_mkwrite(vma, vmf, get_block);
5139 if (!ret && ext4_should_journal_data(inode)) {
5140 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5141 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5143 ret = VM_FAULT_SIGBUS;
5144 ext4_journal_stop(handle);
5147 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5149 ext4_journal_stop(handle);
5150 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5153 ret = block_page_mkwrite_return(ret);
5155 sb_end_pagefault(inode->i_sb);