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/aio.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
60 csum_lo = le16_to_cpu(raw->i_checksum_lo);
61 raw->i_checksum_lo = 0;
62 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
63 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
64 csum_hi = le16_to_cpu(raw->i_checksum_hi);
65 raw->i_checksum_hi = 0;
68 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
69 EXT4_INODE_SIZE(inode->i_sb));
71 raw->i_checksum_lo = cpu_to_le16(csum_lo);
72 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
73 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
74 raw->i_checksum_hi = cpu_to_le16(csum_hi);
79 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
80 struct ext4_inode_info *ei)
82 __u32 provided, calculated;
84 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
85 cpu_to_le32(EXT4_OS_LINUX) ||
86 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
87 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
90 provided = le16_to_cpu(raw->i_checksum_lo);
91 calculated = ext4_inode_csum(inode, raw, ei);
92 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
93 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
94 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
98 return provided == calculated;
101 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
102 struct ext4_inode_info *ei)
106 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
107 cpu_to_le32(EXT4_OS_LINUX) ||
108 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
109 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
112 csum = ext4_inode_csum(inode, raw, ei);
113 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
114 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
115 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
116 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
119 static inline int ext4_begin_ordered_truncate(struct inode *inode,
122 trace_ext4_begin_ordered_truncate(inode, new_size);
124 * If jinode is zero, then we never opened the file for
125 * writing, so there's no need to call
126 * jbd2_journal_begin_ordered_truncate() since there's no
127 * outstanding writes we need to flush.
129 if (!EXT4_I(inode)->jinode)
131 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
132 EXT4_I(inode)->jinode,
136 static void ext4_invalidatepage(struct page *page, unsigned int offset,
137 unsigned int length);
138 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
139 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
140 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
144 * Test whether an inode is a fast symlink.
146 static int ext4_inode_is_fast_symlink(struct inode *inode)
148 int ea_blocks = EXT4_I(inode)->i_file_acl ?
149 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
151 if (ext4_has_inline_data(inode))
154 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
158 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against
162 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
168 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
169 * moment, get_block can be called only for blocks inside i_size since
170 * page cache has been already dropped and writes are blocked by
171 * i_mutex. So we can safely drop the i_data_sem here.
173 BUG_ON(EXT4_JOURNAL(inode) == NULL);
174 jbd_debug(2, "restarting handle %p\n", handle);
175 up_write(&EXT4_I(inode)->i_data_sem);
176 ret = ext4_journal_restart(handle, nblocks);
177 down_write(&EXT4_I(inode)->i_data_sem);
178 ext4_discard_preallocations(inode);
184 * Called at the last iput() if i_nlink is zero.
186 void ext4_evict_inode(struct inode *inode)
191 trace_ext4_evict_inode(inode);
193 if (inode->i_nlink) {
195 * When journalling data dirty buffers are tracked only in the
196 * journal. So although mm thinks everything is clean and
197 * ready for reaping the inode might still have some pages to
198 * write in the running transaction or waiting to be
199 * checkpointed. Thus calling jbd2_journal_invalidatepage()
200 * (via truncate_inode_pages()) to discard these buffers can
201 * cause data loss. Also even if we did not discard these
202 * buffers, we would have no way to find them after the inode
203 * is reaped and thus user could see stale data if he tries to
204 * read them before the transaction is checkpointed. So be
205 * careful and force everything to disk here... We use
206 * ei->i_datasync_tid to store the newest transaction
207 * containing inode's data.
209 * Note that directories do not have this problem because they
210 * don't use page cache.
212 if (ext4_should_journal_data(inode) &&
213 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
214 inode->i_ino != EXT4_JOURNAL_INO) {
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
218 jbd2_complete_transaction(journal, commit_tid);
219 filemap_write_and_wait(&inode->i_data);
221 truncate_inode_pages_final(&inode->i_data);
223 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
227 if (!is_bad_inode(inode))
228 dquot_initialize(inode);
230 if (ext4_should_order_data(inode))
231 ext4_begin_ordered_truncate(inode, 0);
232 truncate_inode_pages_final(&inode->i_data);
234 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
235 if (is_bad_inode(inode))
239 * Protect us against freezing - iput() caller didn't have to have any
240 * protection against it
242 sb_start_intwrite(inode->i_sb);
243 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
244 ext4_blocks_for_truncate(inode)+3);
245 if (IS_ERR(handle)) {
246 ext4_std_error(inode->i_sb, PTR_ERR(handle));
248 * If we're going to skip the normal cleanup, we still need to
249 * make sure that the in-core orphan linked list is properly
252 ext4_orphan_del(NULL, inode);
253 sb_end_intwrite(inode->i_sb);
258 ext4_handle_sync(handle);
260 err = ext4_mark_inode_dirty(handle, inode);
262 ext4_warning(inode->i_sb,
263 "couldn't mark inode dirty (err %d)", err);
267 ext4_truncate(inode);
270 * ext4_ext_truncate() doesn't reserve any slop when it
271 * restarts journal transactions; therefore there may not be
272 * enough credits left in the handle to remove the inode from
273 * the orphan list and set the dtime field.
275 if (!ext4_handle_has_enough_credits(handle, 3)) {
276 err = ext4_journal_extend(handle, 3);
278 err = ext4_journal_restart(handle, 3);
280 ext4_warning(inode->i_sb,
281 "couldn't extend journal (err %d)", err);
283 ext4_journal_stop(handle);
284 ext4_orphan_del(NULL, inode);
285 sb_end_intwrite(inode->i_sb);
291 * Kill off the orphan record which ext4_truncate created.
292 * AKPM: I think this can be inside the above `if'.
293 * Note that ext4_orphan_del() has to be able to cope with the
294 * deletion of a non-existent orphan - this is because we don't
295 * know if ext4_truncate() actually created an orphan record.
296 * (Well, we could do this if we need to, but heck - it works)
298 ext4_orphan_del(handle, inode);
299 EXT4_I(inode)->i_dtime = get_seconds();
302 * One subtle ordering requirement: if anything has gone wrong
303 * (transaction abort, IO errors, whatever), then we can still
304 * do these next steps (the fs will already have been marked as
305 * having errors), but we can't free the inode if the mark_dirty
308 if (ext4_mark_inode_dirty(handle, inode))
309 /* If that failed, just do the required in-core inode clear. */
310 ext4_clear_inode(inode);
312 ext4_free_inode(handle, inode);
313 ext4_journal_stop(handle);
314 sb_end_intwrite(inode->i_sb);
317 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
321 qsize_t *ext4_get_reserved_space(struct inode *inode)
323 return &EXT4_I(inode)->i_reserved_quota;
328 * Called with i_data_sem down, which is important since we can call
329 * ext4_discard_preallocations() from here.
331 void ext4_da_update_reserve_space(struct inode *inode,
332 int used, int quota_claim)
334 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
335 struct ext4_inode_info *ei = EXT4_I(inode);
337 spin_lock(&ei->i_block_reservation_lock);
338 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
339 if (unlikely(used > ei->i_reserved_data_blocks)) {
340 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
341 "with only %d reserved data blocks",
342 __func__, inode->i_ino, used,
343 ei->i_reserved_data_blocks);
345 used = ei->i_reserved_data_blocks;
348 /* Update per-inode reservations */
349 ei->i_reserved_data_blocks -= used;
350 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
352 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
354 /* Update quota subsystem for data blocks */
356 dquot_claim_block(inode, EXT4_C2B(sbi, used));
359 * We did fallocate with an offset that is already delayed
360 * allocated. So on delayed allocated writeback we should
361 * not re-claim the quota for fallocated blocks.
363 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
367 * If we have done all the pending block allocations and if
368 * there aren't any writers on the inode, we can discard the
369 * inode's preallocations.
371 if ((ei->i_reserved_data_blocks == 0) &&
372 (atomic_read(&inode->i_writecount) == 0))
373 ext4_discard_preallocations(inode);
376 static int __check_block_validity(struct inode *inode, const char *func,
378 struct ext4_map_blocks *map)
380 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
382 ext4_error_inode(inode, func, line, map->m_pblk,
383 "lblock %lu mapped to illegal pblock "
384 "(length %d)", (unsigned long) map->m_lblk,
391 #define check_block_validity(inode, map) \
392 __check_block_validity((inode), __func__, __LINE__, (map))
394 #ifdef ES_AGGRESSIVE_TEST
395 static void ext4_map_blocks_es_recheck(handle_t *handle,
397 struct ext4_map_blocks *es_map,
398 struct ext4_map_blocks *map,
405 * There is a race window that the result is not the same.
406 * e.g. xfstests #223 when dioread_nolock enables. The reason
407 * is that we lookup a block mapping in extent status tree with
408 * out taking i_data_sem. So at the time the unwritten extent
409 * could be converted.
411 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
412 down_read(&EXT4_I(inode)->i_data_sem);
413 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
414 retval = ext4_ext_map_blocks(handle, inode, map, flags &
415 EXT4_GET_BLOCKS_KEEP_SIZE);
417 retval = ext4_ind_map_blocks(handle, inode, map, flags &
418 EXT4_GET_BLOCKS_KEEP_SIZE);
420 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
421 up_read((&EXT4_I(inode)->i_data_sem));
423 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
424 * because it shouldn't be marked in es_map->m_flags.
426 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
429 * We don't check m_len because extent will be collpased in status
430 * tree. So the m_len might not equal.
432 if (es_map->m_lblk != map->m_lblk ||
433 es_map->m_flags != map->m_flags ||
434 es_map->m_pblk != map->m_pblk) {
435 printk("ES cache assertion failed for inode: %lu "
436 "es_cached ex [%d/%d/%llu/%x] != "
437 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
438 inode->i_ino, es_map->m_lblk, es_map->m_len,
439 es_map->m_pblk, es_map->m_flags, map->m_lblk,
440 map->m_len, map->m_pblk, map->m_flags,
444 #endif /* ES_AGGRESSIVE_TEST */
447 * The ext4_map_blocks() function tries to look up the requested blocks,
448 * and returns if the blocks are already mapped.
450 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
451 * and store the allocated blocks in the result buffer head and mark it
454 * If file type is extents based, it will call ext4_ext_map_blocks(),
455 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
458 * On success, it returns the number of blocks being mapped or allocated.
459 * if create==0 and the blocks are pre-allocated and unwritten block,
460 * the result buffer head is unmapped. If the create ==1, it will make sure
461 * the buffer head is mapped.
463 * It returns 0 if plain look up failed (blocks have not been allocated), in
464 * that case, buffer head is unmapped
466 * It returns the error in case of allocation failure.
468 int ext4_map_blocks(handle_t *handle, struct inode *inode,
469 struct ext4_map_blocks *map, int flags)
471 struct extent_status es;
474 #ifdef ES_AGGRESSIVE_TEST
475 struct ext4_map_blocks orig_map;
477 memcpy(&orig_map, map, sizeof(*map));
481 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
482 "logical block %lu\n", inode->i_ino, flags, map->m_len,
483 (unsigned long) map->m_lblk);
486 * ext4_map_blocks returns an int, and m_len is an unsigned int
488 if (unlikely(map->m_len > INT_MAX))
489 map->m_len = INT_MAX;
491 /* We can handle the block number less than EXT_MAX_BLOCKS */
492 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
495 /* Lookup extent status tree firstly */
496 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
497 ext4_es_lru_add(inode);
498 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
499 map->m_pblk = ext4_es_pblock(&es) +
500 map->m_lblk - es.es_lblk;
501 map->m_flags |= ext4_es_is_written(&es) ?
502 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
503 retval = es.es_len - (map->m_lblk - es.es_lblk);
504 if (retval > map->m_len)
507 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
512 #ifdef ES_AGGRESSIVE_TEST
513 ext4_map_blocks_es_recheck(handle, inode, map,
520 * Try to see if we can get the block without requesting a new
523 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
524 down_read(&EXT4_I(inode)->i_data_sem);
525 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
526 retval = ext4_ext_map_blocks(handle, inode, map, flags &
527 EXT4_GET_BLOCKS_KEEP_SIZE);
529 retval = ext4_ind_map_blocks(handle, inode, map, flags &
530 EXT4_GET_BLOCKS_KEEP_SIZE);
535 if (unlikely(retval != map->m_len)) {
536 ext4_warning(inode->i_sb,
537 "ES len assertion failed for inode "
538 "%lu: retval %d != map->m_len %d",
539 inode->i_ino, retval, map->m_len);
543 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
544 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
545 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
546 ext4_find_delalloc_range(inode, map->m_lblk,
547 map->m_lblk + map->m_len - 1))
548 status |= EXTENT_STATUS_DELAYED;
549 ret = ext4_es_insert_extent(inode, map->m_lblk,
550 map->m_len, map->m_pblk, status);
554 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
555 up_read((&EXT4_I(inode)->i_data_sem));
558 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
559 ret = check_block_validity(inode, map);
564 /* If it is only a block(s) look up */
565 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
569 * Returns if the blocks have already allocated
571 * Note that if blocks have been preallocated
572 * ext4_ext_get_block() returns the create = 0
573 * with buffer head unmapped.
575 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
577 * If we need to convert extent to unwritten
578 * we continue and do the actual work in
579 * ext4_ext_map_blocks()
581 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
585 * Here we clear m_flags because after allocating an new extent,
586 * it will be set again.
588 map->m_flags &= ~EXT4_MAP_FLAGS;
591 * New blocks allocate and/or writing to unwritten extent
592 * will possibly result in updating i_data, so we take
593 * the write lock of i_data_sem, and call get_blocks()
594 * with create == 1 flag.
596 down_write(&EXT4_I(inode)->i_data_sem);
599 * if the caller is from delayed allocation writeout path
600 * we have already reserved fs blocks for allocation
601 * let the underlying get_block() function know to
602 * avoid double accounting
604 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
605 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
607 * We need to check for EXT4 here because migrate
608 * could have changed the inode type in between
610 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
611 retval = ext4_ext_map_blocks(handle, inode, map, flags);
613 retval = ext4_ind_map_blocks(handle, inode, map, flags);
615 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
617 * We allocated new blocks which will result in
618 * i_data's format changing. Force the migrate
619 * to fail by clearing migrate flags
621 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
625 * Update reserved blocks/metadata blocks after successful
626 * block allocation which had been deferred till now. We don't
627 * support fallocate for non extent files. So we can update
628 * reserve space here.
631 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
632 ext4_da_update_reserve_space(inode, retval, 1);
634 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
635 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
640 if (unlikely(retval != map->m_len)) {
641 ext4_warning(inode->i_sb,
642 "ES len assertion failed for inode "
643 "%lu: retval %d != map->m_len %d",
644 inode->i_ino, retval, map->m_len);
649 * If the extent has been zeroed out, we don't need to update
650 * extent status tree.
652 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
653 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
654 if (ext4_es_is_written(&es))
657 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
658 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
659 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
660 ext4_find_delalloc_range(inode, map->m_lblk,
661 map->m_lblk + map->m_len - 1))
662 status |= EXTENT_STATUS_DELAYED;
663 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
664 map->m_pblk, status);
670 up_write((&EXT4_I(inode)->i_data_sem));
671 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
672 ret = check_block_validity(inode, map);
679 /* Maximum number of blocks we map for direct IO at once. */
680 #define DIO_MAX_BLOCKS 4096
682 static int _ext4_get_block(struct inode *inode, sector_t iblock,
683 struct buffer_head *bh, int flags)
685 handle_t *handle = ext4_journal_current_handle();
686 struct ext4_map_blocks map;
687 int ret = 0, started = 0;
690 if (ext4_has_inline_data(inode))
694 map.m_len = bh->b_size >> inode->i_blkbits;
696 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
697 /* Direct IO write... */
698 if (map.m_len > DIO_MAX_BLOCKS)
699 map.m_len = DIO_MAX_BLOCKS;
700 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
701 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
703 if (IS_ERR(handle)) {
704 ret = PTR_ERR(handle);
710 ret = ext4_map_blocks(handle, inode, &map, flags);
712 ext4_io_end_t *io_end = ext4_inode_aio(inode);
714 map_bh(bh, inode->i_sb, map.m_pblk);
715 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
716 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
717 set_buffer_defer_completion(bh);
718 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
722 ext4_journal_stop(handle);
726 int ext4_get_block(struct inode *inode, sector_t iblock,
727 struct buffer_head *bh, int create)
729 return _ext4_get_block(inode, iblock, bh,
730 create ? EXT4_GET_BLOCKS_CREATE : 0);
734 * `handle' can be NULL if create is zero
736 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
737 ext4_lblk_t block, int create)
739 struct ext4_map_blocks map;
740 struct buffer_head *bh;
743 J_ASSERT(handle != NULL || create == 0);
747 err = ext4_map_blocks(handle, inode, &map,
748 create ? EXT4_GET_BLOCKS_CREATE : 0);
751 return create ? ERR_PTR(-ENOSPC) : NULL;
755 bh = sb_getblk(inode->i_sb, map.m_pblk);
757 return ERR_PTR(-ENOMEM);
758 if (map.m_flags & EXT4_MAP_NEW) {
759 J_ASSERT(create != 0);
760 J_ASSERT(handle != NULL);
763 * Now that we do not always journal data, we should
764 * keep in mind whether this should always journal the
765 * new buffer as metadata. For now, regular file
766 * writes use ext4_get_block instead, so it's not a
770 BUFFER_TRACE(bh, "call get_create_access");
771 err = ext4_journal_get_create_access(handle, bh);
776 if (!buffer_uptodate(bh)) {
777 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
778 set_buffer_uptodate(bh);
781 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
782 err = ext4_handle_dirty_metadata(handle, inode, bh);
786 BUFFER_TRACE(bh, "not a new buffer");
793 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
794 ext4_lblk_t block, int create)
796 struct buffer_head *bh;
798 bh = ext4_getblk(handle, inode, block, create);
801 if (!bh || buffer_uptodate(bh))
803 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
805 if (buffer_uptodate(bh))
808 return ERR_PTR(-EIO);
811 int ext4_walk_page_buffers(handle_t *handle,
812 struct buffer_head *head,
816 int (*fn)(handle_t *handle,
817 struct buffer_head *bh))
819 struct buffer_head *bh;
820 unsigned block_start, block_end;
821 unsigned blocksize = head->b_size;
823 struct buffer_head *next;
825 for (bh = head, block_start = 0;
826 ret == 0 && (bh != head || !block_start);
827 block_start = block_end, bh = next) {
828 next = bh->b_this_page;
829 block_end = block_start + blocksize;
830 if (block_end <= from || block_start >= to) {
831 if (partial && !buffer_uptodate(bh))
835 err = (*fn)(handle, bh);
843 * To preserve ordering, it is essential that the hole instantiation and
844 * the data write be encapsulated in a single transaction. We cannot
845 * close off a transaction and start a new one between the ext4_get_block()
846 * and the commit_write(). So doing the jbd2_journal_start at the start of
847 * prepare_write() is the right place.
849 * Also, this function can nest inside ext4_writepage(). In that case, we
850 * *know* that ext4_writepage() has generated enough buffer credits to do the
851 * whole page. So we won't block on the journal in that case, which is good,
852 * because the caller may be PF_MEMALLOC.
854 * By accident, ext4 can be reentered when a transaction is open via
855 * quota file writes. If we were to commit the transaction while thus
856 * reentered, there can be a deadlock - we would be holding a quota
857 * lock, and the commit would never complete if another thread had a
858 * transaction open and was blocking on the quota lock - a ranking
861 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
862 * will _not_ run commit under these circumstances because handle->h_ref
863 * is elevated. We'll still have enough credits for the tiny quotafile
866 int do_journal_get_write_access(handle_t *handle,
867 struct buffer_head *bh)
869 int dirty = buffer_dirty(bh);
872 if (!buffer_mapped(bh) || buffer_freed(bh))
875 * __block_write_begin() could have dirtied some buffers. Clean
876 * the dirty bit as jbd2_journal_get_write_access() could complain
877 * otherwise about fs integrity issues. Setting of the dirty bit
878 * by __block_write_begin() isn't a real problem here as we clear
879 * the bit before releasing a page lock and thus writeback cannot
880 * ever write the buffer.
883 clear_buffer_dirty(bh);
884 BUFFER_TRACE(bh, "get write access");
885 ret = ext4_journal_get_write_access(handle, bh);
887 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
891 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
892 struct buffer_head *bh_result, int create);
893 static int ext4_write_begin(struct file *file, struct address_space *mapping,
894 loff_t pos, unsigned len, unsigned flags,
895 struct page **pagep, void **fsdata)
897 struct inode *inode = mapping->host;
898 int ret, needed_blocks;
905 trace_ext4_write_begin(inode, pos, len, flags);
907 * Reserve one block more for addition to orphan list in case
908 * we allocate blocks but write fails for some reason
910 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
911 index = pos >> PAGE_CACHE_SHIFT;
912 from = pos & (PAGE_CACHE_SIZE - 1);
915 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
916 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
925 * grab_cache_page_write_begin() can take a long time if the
926 * system is thrashing due to memory pressure, or if the page
927 * is being written back. So grab it first before we start
928 * the transaction handle. This also allows us to allocate
929 * the page (if needed) without using GFP_NOFS.
932 page = grab_cache_page_write_begin(mapping, index, flags);
938 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
939 if (IS_ERR(handle)) {
940 page_cache_release(page);
941 return PTR_ERR(handle);
945 if (page->mapping != mapping) {
946 /* The page got truncated from under us */
948 page_cache_release(page);
949 ext4_journal_stop(handle);
952 /* In case writeback began while the page was unlocked */
953 wait_for_stable_page(page);
955 if (ext4_should_dioread_nolock(inode))
956 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
958 ret = __block_write_begin(page, pos, len, ext4_get_block);
960 if (!ret && ext4_should_journal_data(inode)) {
961 ret = ext4_walk_page_buffers(handle, page_buffers(page),
963 do_journal_get_write_access);
969 * __block_write_begin may have instantiated a few blocks
970 * outside i_size. Trim these off again. Don't need
971 * i_size_read because we hold i_mutex.
973 * Add inode to orphan list in case we crash before
976 if (pos + len > inode->i_size && ext4_can_truncate(inode))
977 ext4_orphan_add(handle, inode);
979 ext4_journal_stop(handle);
980 if (pos + len > inode->i_size) {
981 ext4_truncate_failed_write(inode);
983 * If truncate failed early the inode might
984 * still be on the orphan list; we need to
985 * make sure the inode is removed from the
986 * orphan list in that case.
989 ext4_orphan_del(NULL, inode);
992 if (ret == -ENOSPC &&
993 ext4_should_retry_alloc(inode->i_sb, &retries))
995 page_cache_release(page);
1002 /* For write_end() in data=journal mode */
1003 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1006 if (!buffer_mapped(bh) || buffer_freed(bh))
1008 set_buffer_uptodate(bh);
1009 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1010 clear_buffer_meta(bh);
1011 clear_buffer_prio(bh);
1016 * We need to pick up the new inode size which generic_commit_write gave us
1017 * `file' can be NULL - eg, when called from page_symlink().
1019 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1020 * buffers are managed internally.
1022 static int ext4_write_end(struct file *file,
1023 struct address_space *mapping,
1024 loff_t pos, unsigned len, unsigned copied,
1025 struct page *page, void *fsdata)
1027 handle_t *handle = ext4_journal_current_handle();
1028 struct inode *inode = mapping->host;
1030 int i_size_changed = 0;
1032 trace_ext4_write_end(inode, pos, len, copied);
1033 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1034 ret = ext4_jbd2_file_inode(handle, inode);
1037 page_cache_release(page);
1042 if (ext4_has_inline_data(inode)) {
1043 ret = ext4_write_inline_data_end(inode, pos, len,
1049 copied = block_write_end(file, mapping, pos,
1050 len, copied, page, fsdata);
1052 * it's important to update i_size while still holding page lock:
1053 * page writeout could otherwise come in and zero beyond i_size.
1055 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1057 page_cache_release(page);
1060 * Don't mark the inode dirty under page lock. First, it unnecessarily
1061 * makes the holding time of page lock longer. Second, it forces lock
1062 * ordering of page lock and transaction start for journaling
1066 ext4_mark_inode_dirty(handle, inode);
1068 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1069 /* if we have allocated more blocks and copied
1070 * less. We will have blocks allocated outside
1071 * inode->i_size. So truncate them
1073 ext4_orphan_add(handle, inode);
1075 ret2 = ext4_journal_stop(handle);
1079 if (pos + len > inode->i_size) {
1080 ext4_truncate_failed_write(inode);
1082 * If truncate failed early the inode might still be
1083 * on the orphan list; we need to make sure the inode
1084 * is removed from the orphan list in that case.
1087 ext4_orphan_del(NULL, inode);
1090 return ret ? ret : copied;
1093 static int ext4_journalled_write_end(struct file *file,
1094 struct address_space *mapping,
1095 loff_t pos, unsigned len, unsigned copied,
1096 struct page *page, void *fsdata)
1098 handle_t *handle = ext4_journal_current_handle();
1099 struct inode *inode = mapping->host;
1103 int size_changed = 0;
1105 trace_ext4_journalled_write_end(inode, pos, len, copied);
1106 from = pos & (PAGE_CACHE_SIZE - 1);
1109 BUG_ON(!ext4_handle_valid(handle));
1111 if (ext4_has_inline_data(inode))
1112 copied = ext4_write_inline_data_end(inode, pos, len,
1116 if (!PageUptodate(page))
1118 page_zero_new_buffers(page, from+copied, to);
1121 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1122 to, &partial, write_end_fn);
1124 SetPageUptodate(page);
1126 size_changed = ext4_update_inode_size(inode, pos + copied);
1127 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1128 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1130 page_cache_release(page);
1133 ret2 = ext4_mark_inode_dirty(handle, inode);
1138 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1139 /* if we have allocated more blocks and copied
1140 * less. We will have blocks allocated outside
1141 * inode->i_size. So truncate them
1143 ext4_orphan_add(handle, inode);
1145 ret2 = ext4_journal_stop(handle);
1148 if (pos + len > inode->i_size) {
1149 ext4_truncate_failed_write(inode);
1151 * If truncate failed early the inode might still be
1152 * on the orphan list; we need to make sure the inode
1153 * is removed from the orphan list in that case.
1156 ext4_orphan_del(NULL, inode);
1159 return ret ? ret : copied;
1163 * Reserve a single cluster located at lblock
1165 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1167 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1168 struct ext4_inode_info *ei = EXT4_I(inode);
1169 unsigned int md_needed;
1173 * We will charge metadata quota at writeout time; this saves
1174 * us from metadata over-estimation, though we may go over by
1175 * a small amount in the end. Here we just reserve for data.
1177 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1182 * recalculate the amount of metadata blocks to reserve
1183 * in order to allocate nrblocks
1184 * worse case is one extent per block
1186 spin_lock(&ei->i_block_reservation_lock);
1188 * ext4_calc_metadata_amount() has side effects, which we have
1189 * to be prepared undo if we fail to claim space.
1192 trace_ext4_da_reserve_space(inode, 0);
1194 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1195 spin_unlock(&ei->i_block_reservation_lock);
1196 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1199 ei->i_reserved_data_blocks++;
1200 spin_unlock(&ei->i_block_reservation_lock);
1202 return 0; /* success */
1205 static void ext4_da_release_space(struct inode *inode, int to_free)
1207 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1208 struct ext4_inode_info *ei = EXT4_I(inode);
1211 return; /* Nothing to release, exit */
1213 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1215 trace_ext4_da_release_space(inode, to_free);
1216 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1218 * if there aren't enough reserved blocks, then the
1219 * counter is messed up somewhere. Since this
1220 * function is called from invalidate page, it's
1221 * harmless to return without any action.
1223 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1224 "ino %lu, to_free %d with only %d reserved "
1225 "data blocks", inode->i_ino, to_free,
1226 ei->i_reserved_data_blocks);
1228 to_free = ei->i_reserved_data_blocks;
1230 ei->i_reserved_data_blocks -= to_free;
1232 /* update fs dirty data blocks counter */
1233 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1235 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1237 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1240 static void ext4_da_page_release_reservation(struct page *page,
1241 unsigned int offset,
1242 unsigned int length)
1245 struct buffer_head *head, *bh;
1246 unsigned int curr_off = 0;
1247 struct inode *inode = page->mapping->host;
1248 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1249 unsigned int stop = offset + length;
1253 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1255 head = page_buffers(page);
1258 unsigned int next_off = curr_off + bh->b_size;
1260 if (next_off > stop)
1263 if ((offset <= curr_off) && (buffer_delay(bh))) {
1265 clear_buffer_delay(bh);
1267 curr_off = next_off;
1268 } while ((bh = bh->b_this_page) != head);
1271 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1272 ext4_es_remove_extent(inode, lblk, to_release);
1275 /* If we have released all the blocks belonging to a cluster, then we
1276 * need to release the reserved space for that cluster. */
1277 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1278 while (num_clusters > 0) {
1279 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1280 ((num_clusters - 1) << sbi->s_cluster_bits);
1281 if (sbi->s_cluster_ratio == 1 ||
1282 !ext4_find_delalloc_cluster(inode, lblk))
1283 ext4_da_release_space(inode, 1);
1290 * Delayed allocation stuff
1293 struct mpage_da_data {
1294 struct inode *inode;
1295 struct writeback_control *wbc;
1297 pgoff_t first_page; /* The first page to write */
1298 pgoff_t next_page; /* Current page to examine */
1299 pgoff_t last_page; /* Last page to examine */
1301 * Extent to map - this can be after first_page because that can be
1302 * fully mapped. We somewhat abuse m_flags to store whether the extent
1303 * is delalloc or unwritten.
1305 struct ext4_map_blocks map;
1306 struct ext4_io_submit io_submit; /* IO submission data */
1309 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1314 struct pagevec pvec;
1315 struct inode *inode = mpd->inode;
1316 struct address_space *mapping = inode->i_mapping;
1318 /* This is necessary when next_page == 0. */
1319 if (mpd->first_page >= mpd->next_page)
1322 index = mpd->first_page;
1323 end = mpd->next_page - 1;
1325 ext4_lblk_t start, last;
1326 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1327 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1328 ext4_es_remove_extent(inode, start, last - start + 1);
1331 pagevec_init(&pvec, 0);
1332 while (index <= end) {
1333 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1336 for (i = 0; i < nr_pages; i++) {
1337 struct page *page = pvec.pages[i];
1338 if (page->index > end)
1340 BUG_ON(!PageLocked(page));
1341 BUG_ON(PageWriteback(page));
1343 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1344 ClearPageUptodate(page);
1348 index = pvec.pages[nr_pages - 1]->index + 1;
1349 pagevec_release(&pvec);
1353 static void ext4_print_free_blocks(struct inode *inode)
1355 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1356 struct super_block *sb = inode->i_sb;
1357 struct ext4_inode_info *ei = EXT4_I(inode);
1359 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1360 EXT4_C2B(EXT4_SB(inode->i_sb),
1361 ext4_count_free_clusters(sb)));
1362 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1363 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1364 (long long) EXT4_C2B(EXT4_SB(sb),
1365 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1366 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1367 (long long) EXT4_C2B(EXT4_SB(sb),
1368 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1369 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1370 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1371 ei->i_reserved_data_blocks);
1375 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1377 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1381 * This function is grabs code from the very beginning of
1382 * ext4_map_blocks, but assumes that the caller is from delayed write
1383 * time. This function looks up the requested blocks and sets the
1384 * buffer delay bit under the protection of i_data_sem.
1386 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1387 struct ext4_map_blocks *map,
1388 struct buffer_head *bh)
1390 struct extent_status es;
1392 sector_t invalid_block = ~((sector_t) 0xffff);
1393 #ifdef ES_AGGRESSIVE_TEST
1394 struct ext4_map_blocks orig_map;
1396 memcpy(&orig_map, map, sizeof(*map));
1399 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1403 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1404 "logical block %lu\n", inode->i_ino, map->m_len,
1405 (unsigned long) map->m_lblk);
1407 /* Lookup extent status tree firstly */
1408 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1409 ext4_es_lru_add(inode);
1410 if (ext4_es_is_hole(&es)) {
1412 down_read(&EXT4_I(inode)->i_data_sem);
1417 * Delayed extent could be allocated by fallocate.
1418 * So we need to check it.
1420 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1421 map_bh(bh, inode->i_sb, invalid_block);
1423 set_buffer_delay(bh);
1427 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1428 retval = es.es_len - (iblock - es.es_lblk);
1429 if (retval > map->m_len)
1430 retval = map->m_len;
1431 map->m_len = retval;
1432 if (ext4_es_is_written(&es))
1433 map->m_flags |= EXT4_MAP_MAPPED;
1434 else if (ext4_es_is_unwritten(&es))
1435 map->m_flags |= EXT4_MAP_UNWRITTEN;
1439 #ifdef ES_AGGRESSIVE_TEST
1440 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1446 * Try to see if we can get the block without requesting a new
1447 * file system block.
1449 down_read(&EXT4_I(inode)->i_data_sem);
1450 if (ext4_has_inline_data(inode)) {
1452 * We will soon create blocks for this page, and let
1453 * us pretend as if the blocks aren't allocated yet.
1454 * In case of clusters, we have to handle the work
1455 * of mapping from cluster so that the reserved space
1456 * is calculated properly.
1458 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1459 ext4_find_delalloc_cluster(inode, map->m_lblk))
1460 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1462 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1463 retval = ext4_ext_map_blocks(NULL, inode, map,
1464 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1466 retval = ext4_ind_map_blocks(NULL, inode, map,
1467 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1473 * XXX: __block_prepare_write() unmaps passed block,
1477 * If the block was allocated from previously allocated cluster,
1478 * then we don't need to reserve it again. However we still need
1479 * to reserve metadata for every block we're going to write.
1481 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1482 ret = ext4_da_reserve_space(inode, iblock);
1484 /* not enough space to reserve */
1490 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1491 ~0, EXTENT_STATUS_DELAYED);
1497 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1498 * and it should not appear on the bh->b_state.
1500 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1502 map_bh(bh, inode->i_sb, invalid_block);
1504 set_buffer_delay(bh);
1505 } else if (retval > 0) {
1507 unsigned int status;
1509 if (unlikely(retval != map->m_len)) {
1510 ext4_warning(inode->i_sb,
1511 "ES len assertion failed for inode "
1512 "%lu: retval %d != map->m_len %d",
1513 inode->i_ino, retval, map->m_len);
1517 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1518 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1519 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1520 map->m_pblk, status);
1526 up_read((&EXT4_I(inode)->i_data_sem));
1532 * This is a special get_blocks_t callback which is used by
1533 * ext4_da_write_begin(). It will either return mapped block or
1534 * reserve space for a single block.
1536 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1537 * We also have b_blocknr = -1 and b_bdev initialized properly
1539 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1540 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1541 * initialized properly.
1543 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1544 struct buffer_head *bh, int create)
1546 struct ext4_map_blocks map;
1549 BUG_ON(create == 0);
1550 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1552 map.m_lblk = iblock;
1556 * first, we need to know whether the block is allocated already
1557 * preallocated blocks are unmapped but should treated
1558 * the same as allocated blocks.
1560 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1564 map_bh(bh, inode->i_sb, map.m_pblk);
1565 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1567 if (buffer_unwritten(bh)) {
1568 /* A delayed write to unwritten bh should be marked
1569 * new and mapped. Mapped ensures that we don't do
1570 * get_block multiple times when we write to the same
1571 * offset and new ensures that we do proper zero out
1572 * for partial write.
1575 set_buffer_mapped(bh);
1580 static int bget_one(handle_t *handle, struct buffer_head *bh)
1586 static int bput_one(handle_t *handle, struct buffer_head *bh)
1592 static int __ext4_journalled_writepage(struct page *page,
1595 struct address_space *mapping = page->mapping;
1596 struct inode *inode = mapping->host;
1597 struct buffer_head *page_bufs = NULL;
1598 handle_t *handle = NULL;
1599 int ret = 0, err = 0;
1600 int inline_data = ext4_has_inline_data(inode);
1601 struct buffer_head *inode_bh = NULL;
1603 ClearPageChecked(page);
1606 BUG_ON(page->index != 0);
1607 BUG_ON(len > ext4_get_max_inline_size(inode));
1608 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1609 if (inode_bh == NULL)
1612 page_bufs = page_buffers(page);
1617 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1620 /* As soon as we unlock the page, it can go away, but we have
1621 * references to buffers so we are safe */
1624 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1625 ext4_writepage_trans_blocks(inode));
1626 if (IS_ERR(handle)) {
1627 ret = PTR_ERR(handle);
1631 BUG_ON(!ext4_handle_valid(handle));
1634 BUFFER_TRACE(inode_bh, "get write access");
1635 ret = ext4_journal_get_write_access(handle, inode_bh);
1637 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1640 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1641 do_journal_get_write_access);
1643 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1648 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1649 err = ext4_journal_stop(handle);
1653 if (!ext4_has_inline_data(inode))
1654 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1656 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1663 * Note that we don't need to start a transaction unless we're journaling data
1664 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1665 * need to file the inode to the transaction's list in ordered mode because if
1666 * we are writing back data added by write(), the inode is already there and if
1667 * we are writing back data modified via mmap(), no one guarantees in which
1668 * transaction the data will hit the disk. In case we are journaling data, we
1669 * cannot start transaction directly because transaction start ranks above page
1670 * lock so we have to do some magic.
1672 * This function can get called via...
1673 * - ext4_writepages after taking page lock (have journal handle)
1674 * - journal_submit_inode_data_buffers (no journal handle)
1675 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1676 * - grab_page_cache when doing write_begin (have journal handle)
1678 * We don't do any block allocation in this function. If we have page with
1679 * multiple blocks we need to write those buffer_heads that are mapped. This
1680 * is important for mmaped based write. So if we do with blocksize 1K
1681 * truncate(f, 1024);
1682 * a = mmap(f, 0, 4096);
1684 * truncate(f, 4096);
1685 * we have in the page first buffer_head mapped via page_mkwrite call back
1686 * but other buffer_heads would be unmapped but dirty (dirty done via the
1687 * do_wp_page). So writepage should write the first block. If we modify
1688 * the mmap area beyond 1024 we will again get a page_fault and the
1689 * page_mkwrite callback will do the block allocation and mark the
1690 * buffer_heads mapped.
1692 * We redirty the page if we have any buffer_heads that is either delay or
1693 * unwritten in the page.
1695 * We can get recursively called as show below.
1697 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1700 * But since we don't do any block allocation we should not deadlock.
1701 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1703 static int ext4_writepage(struct page *page,
1704 struct writeback_control *wbc)
1709 struct buffer_head *page_bufs = NULL;
1710 struct inode *inode = page->mapping->host;
1711 struct ext4_io_submit io_submit;
1712 bool keep_towrite = false;
1714 trace_ext4_writepage(page);
1715 size = i_size_read(inode);
1716 if (page->index == size >> PAGE_CACHE_SHIFT)
1717 len = size & ~PAGE_CACHE_MASK;
1719 len = PAGE_CACHE_SIZE;
1721 page_bufs = page_buffers(page);
1723 * We cannot do block allocation or other extent handling in this
1724 * function. If there are buffers needing that, we have to redirty
1725 * the page. But we may reach here when we do a journal commit via
1726 * journal_submit_inode_data_buffers() and in that case we must write
1727 * allocated buffers to achieve data=ordered mode guarantees.
1729 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1730 ext4_bh_delay_or_unwritten)) {
1731 redirty_page_for_writepage(wbc, page);
1732 if (current->flags & PF_MEMALLOC) {
1734 * For memory cleaning there's no point in writing only
1735 * some buffers. So just bail out. Warn if we came here
1736 * from direct reclaim.
1738 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1743 keep_towrite = true;
1746 if (PageChecked(page) && ext4_should_journal_data(inode))
1748 * It's mmapped pagecache. Add buffers and journal it. There
1749 * doesn't seem much point in redirtying the page here.
1751 return __ext4_journalled_writepage(page, len);
1753 ext4_io_submit_init(&io_submit, wbc);
1754 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1755 if (!io_submit.io_end) {
1756 redirty_page_for_writepage(wbc, page);
1760 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1761 ext4_io_submit(&io_submit);
1762 /* Drop io_end reference we got from init */
1763 ext4_put_io_end_defer(io_submit.io_end);
1767 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1770 loff_t size = i_size_read(mpd->inode);
1773 BUG_ON(page->index != mpd->first_page);
1774 if (page->index == size >> PAGE_CACHE_SHIFT)
1775 len = size & ~PAGE_CACHE_MASK;
1777 len = PAGE_CACHE_SIZE;
1778 clear_page_dirty_for_io(page);
1779 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1781 mpd->wbc->nr_to_write--;
1787 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1790 * mballoc gives us at most this number of blocks...
1791 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1792 * The rest of mballoc seems to handle chunks up to full group size.
1794 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1797 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1799 * @mpd - extent of blocks
1800 * @lblk - logical number of the block in the file
1801 * @bh - buffer head we want to add to the extent
1803 * The function is used to collect contig. blocks in the same state. If the
1804 * buffer doesn't require mapping for writeback and we haven't started the
1805 * extent of buffers to map yet, the function returns 'true' immediately - the
1806 * caller can write the buffer right away. Otherwise the function returns true
1807 * if the block has been added to the extent, false if the block couldn't be
1810 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1811 struct buffer_head *bh)
1813 struct ext4_map_blocks *map = &mpd->map;
1815 /* Buffer that doesn't need mapping for writeback? */
1816 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1817 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1818 /* So far no extent to map => we write the buffer right away */
1819 if (map->m_len == 0)
1824 /* First block in the extent? */
1825 if (map->m_len == 0) {
1828 map->m_flags = bh->b_state & BH_FLAGS;
1832 /* Don't go larger than mballoc is willing to allocate */
1833 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1836 /* Can we merge the block to our big extent? */
1837 if (lblk == map->m_lblk + map->m_len &&
1838 (bh->b_state & BH_FLAGS) == map->m_flags) {
1846 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1848 * @mpd - extent of blocks for mapping
1849 * @head - the first buffer in the page
1850 * @bh - buffer we should start processing from
1851 * @lblk - logical number of the block in the file corresponding to @bh
1853 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1854 * the page for IO if all buffers in this page were mapped and there's no
1855 * accumulated extent of buffers to map or add buffers in the page to the
1856 * extent of buffers to map. The function returns 1 if the caller can continue
1857 * by processing the next page, 0 if it should stop adding buffers to the
1858 * extent to map because we cannot extend it anymore. It can also return value
1859 * < 0 in case of error during IO submission.
1861 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1862 struct buffer_head *head,
1863 struct buffer_head *bh,
1866 struct inode *inode = mpd->inode;
1868 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1869 >> inode->i_blkbits;
1872 BUG_ON(buffer_locked(bh));
1874 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1875 /* Found extent to map? */
1878 /* Everything mapped so far and we hit EOF */
1881 } while (lblk++, (bh = bh->b_this_page) != head);
1882 /* So far everything mapped? Submit the page for IO. */
1883 if (mpd->map.m_len == 0) {
1884 err = mpage_submit_page(mpd, head->b_page);
1888 return lblk < blocks;
1892 * mpage_map_buffers - update buffers corresponding to changed extent and
1893 * submit fully mapped pages for IO
1895 * @mpd - description of extent to map, on return next extent to map
1897 * Scan buffers corresponding to changed extent (we expect corresponding pages
1898 * to be already locked) and update buffer state according to new extent state.
1899 * We map delalloc buffers to their physical location, clear unwritten bits,
1900 * and mark buffers as uninit when we perform writes to unwritten extents
1901 * and do extent conversion after IO is finished. If the last page is not fully
1902 * mapped, we update @map to the next extent in the last page that needs
1903 * mapping. Otherwise we submit the page for IO.
1905 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1907 struct pagevec pvec;
1909 struct inode *inode = mpd->inode;
1910 struct buffer_head *head, *bh;
1911 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
1917 start = mpd->map.m_lblk >> bpp_bits;
1918 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
1919 lblk = start << bpp_bits;
1920 pblock = mpd->map.m_pblk;
1922 pagevec_init(&pvec, 0);
1923 while (start <= end) {
1924 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
1928 for (i = 0; i < nr_pages; i++) {
1929 struct page *page = pvec.pages[i];
1931 if (page->index > end)
1933 /* Up to 'end' pages must be contiguous */
1934 BUG_ON(page->index != start);
1935 bh = head = page_buffers(page);
1937 if (lblk < mpd->map.m_lblk)
1939 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
1941 * Buffer after end of mapped extent.
1942 * Find next buffer in the page to map.
1945 mpd->map.m_flags = 0;
1947 * FIXME: If dioread_nolock supports
1948 * blocksize < pagesize, we need to make
1949 * sure we add size mapped so far to
1950 * io_end->size as the following call
1951 * can submit the page for IO.
1953 err = mpage_process_page_bufs(mpd, head,
1955 pagevec_release(&pvec);
1960 if (buffer_delay(bh)) {
1961 clear_buffer_delay(bh);
1962 bh->b_blocknr = pblock++;
1964 clear_buffer_unwritten(bh);
1965 } while (lblk++, (bh = bh->b_this_page) != head);
1968 * FIXME: This is going to break if dioread_nolock
1969 * supports blocksize < pagesize as we will try to
1970 * convert potentially unmapped parts of inode.
1972 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
1973 /* Page fully mapped - let IO run! */
1974 err = mpage_submit_page(mpd, page);
1976 pagevec_release(&pvec);
1981 pagevec_release(&pvec);
1983 /* Extent fully mapped and matches with page boundary. We are done. */
1985 mpd->map.m_flags = 0;
1989 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
1991 struct inode *inode = mpd->inode;
1992 struct ext4_map_blocks *map = &mpd->map;
1993 int get_blocks_flags;
1994 int err, dioread_nolock;
1996 trace_ext4_da_write_pages_extent(inode, map);
1998 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
1999 * to convert an unwritten extent to be initialized (in the case
2000 * where we have written into one or more preallocated blocks). It is
2001 * possible that we're going to need more metadata blocks than
2002 * previously reserved. However we must not fail because we're in
2003 * writeback and there is nothing we can do about it so it might result
2004 * in data loss. So use reserved blocks to allocate metadata if
2007 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2008 * in question are delalloc blocks. This affects functions in many
2009 * different parts of the allocation call path. This flag exists
2010 * primarily because we don't want to change *many* call functions, so
2011 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2012 * once the inode's allocation semaphore is taken.
2014 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2015 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2016 dioread_nolock = ext4_should_dioread_nolock(inode);
2018 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2019 if (map->m_flags & (1 << BH_Delay))
2020 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2022 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2025 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2026 if (!mpd->io_submit.io_end->handle &&
2027 ext4_handle_valid(handle)) {
2028 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2029 handle->h_rsv_handle = NULL;
2031 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2034 BUG_ON(map->m_len == 0);
2035 if (map->m_flags & EXT4_MAP_NEW) {
2036 struct block_device *bdev = inode->i_sb->s_bdev;
2039 for (i = 0; i < map->m_len; i++)
2040 unmap_underlying_metadata(bdev, map->m_pblk + i);
2046 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2047 * mpd->len and submit pages underlying it for IO
2049 * @handle - handle for journal operations
2050 * @mpd - extent to map
2051 * @give_up_on_write - we set this to true iff there is a fatal error and there
2052 * is no hope of writing the data. The caller should discard
2053 * dirty pages to avoid infinite loops.
2055 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2056 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2057 * them to initialized or split the described range from larger unwritten
2058 * extent. Note that we need not map all the described range since allocation
2059 * can return less blocks or the range is covered by more unwritten extents. We
2060 * cannot map more because we are limited by reserved transaction credits. On
2061 * the other hand we always make sure that the last touched page is fully
2062 * mapped so that it can be written out (and thus forward progress is
2063 * guaranteed). After mapping we submit all mapped pages for IO.
2065 static int mpage_map_and_submit_extent(handle_t *handle,
2066 struct mpage_da_data *mpd,
2067 bool *give_up_on_write)
2069 struct inode *inode = mpd->inode;
2070 struct ext4_map_blocks *map = &mpd->map;
2075 mpd->io_submit.io_end->offset =
2076 ((loff_t)map->m_lblk) << inode->i_blkbits;
2078 err = mpage_map_one_extent(handle, mpd);
2080 struct super_block *sb = inode->i_sb;
2082 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2083 goto invalidate_dirty_pages;
2085 * Let the uper layers retry transient errors.
2086 * In the case of ENOSPC, if ext4_count_free_blocks()
2087 * is non-zero, a commit should free up blocks.
2089 if ((err == -ENOMEM) ||
2090 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2092 goto update_disksize;
2095 ext4_msg(sb, KERN_CRIT,
2096 "Delayed block allocation failed for "
2097 "inode %lu at logical offset %llu with"
2098 " max blocks %u with error %d",
2100 (unsigned long long)map->m_lblk,
2101 (unsigned)map->m_len, -err);
2102 ext4_msg(sb, KERN_CRIT,
2103 "This should not happen!! Data will "
2106 ext4_print_free_blocks(inode);
2107 invalidate_dirty_pages:
2108 *give_up_on_write = true;
2113 * Update buffer state, submit mapped pages, and get us new
2116 err = mpage_map_and_submit_buffers(mpd);
2118 goto update_disksize;
2119 } while (map->m_len);
2123 * Update on-disk size after IO is submitted. Races with
2124 * truncate are avoided by checking i_size under i_data_sem.
2126 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2127 if (disksize > EXT4_I(inode)->i_disksize) {
2131 down_write(&EXT4_I(inode)->i_data_sem);
2132 i_size = i_size_read(inode);
2133 if (disksize > i_size)
2135 if (disksize > EXT4_I(inode)->i_disksize)
2136 EXT4_I(inode)->i_disksize = disksize;
2137 err2 = ext4_mark_inode_dirty(handle, inode);
2138 up_write(&EXT4_I(inode)->i_data_sem);
2140 ext4_error(inode->i_sb,
2141 "Failed to mark inode %lu dirty",
2150 * Calculate the total number of credits to reserve for one writepages
2151 * iteration. This is called from ext4_writepages(). We map an extent of
2152 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2153 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2154 * bpp - 1 blocks in bpp different extents.
2156 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2158 int bpp = ext4_journal_blocks_per_page(inode);
2160 return ext4_meta_trans_blocks(inode,
2161 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2165 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2166 * and underlying extent to map
2168 * @mpd - where to look for pages
2170 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2171 * IO immediately. When we find a page which isn't mapped we start accumulating
2172 * extent of buffers underlying these pages that needs mapping (formed by
2173 * either delayed or unwritten buffers). We also lock the pages containing
2174 * these buffers. The extent found is returned in @mpd structure (starting at
2175 * mpd->lblk with length mpd->len blocks).
2177 * Note that this function can attach bios to one io_end structure which are
2178 * neither logically nor physically contiguous. Although it may seem as an
2179 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2180 * case as we need to track IO to all buffers underlying a page in one io_end.
2182 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2184 struct address_space *mapping = mpd->inode->i_mapping;
2185 struct pagevec pvec;
2186 unsigned int nr_pages;
2187 long left = mpd->wbc->nr_to_write;
2188 pgoff_t index = mpd->first_page;
2189 pgoff_t end = mpd->last_page;
2192 int blkbits = mpd->inode->i_blkbits;
2194 struct buffer_head *head;
2196 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2197 tag = PAGECACHE_TAG_TOWRITE;
2199 tag = PAGECACHE_TAG_DIRTY;
2201 pagevec_init(&pvec, 0);
2203 mpd->next_page = index;
2204 while (index <= end) {
2205 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2206 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2210 for (i = 0; i < nr_pages; i++) {
2211 struct page *page = pvec.pages[i];
2214 * At this point, the page may be truncated or
2215 * invalidated (changing page->mapping to NULL), or
2216 * even swizzled back from swapper_space to tmpfs file
2217 * mapping. However, page->index will not change
2218 * because we have a reference on the page.
2220 if (page->index > end)
2224 * Accumulated enough dirty pages? This doesn't apply
2225 * to WB_SYNC_ALL mode. For integrity sync we have to
2226 * keep going because someone may be concurrently
2227 * dirtying pages, and we might have synced a lot of
2228 * newly appeared dirty pages, but have not synced all
2229 * of the old dirty pages.
2231 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2234 /* If we can't merge this page, we are done. */
2235 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2240 * If the page is no longer dirty, or its mapping no
2241 * longer corresponds to inode we are writing (which
2242 * means it has been truncated or invalidated), or the
2243 * page is already under writeback and we are not doing
2244 * a data integrity writeback, skip the page
2246 if (!PageDirty(page) ||
2247 (PageWriteback(page) &&
2248 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2249 unlikely(page->mapping != mapping)) {
2254 wait_on_page_writeback(page);
2255 BUG_ON(PageWriteback(page));
2257 if (mpd->map.m_len == 0)
2258 mpd->first_page = page->index;
2259 mpd->next_page = page->index + 1;
2260 /* Add all dirty buffers to mpd */
2261 lblk = ((ext4_lblk_t)page->index) <<
2262 (PAGE_CACHE_SHIFT - blkbits);
2263 head = page_buffers(page);
2264 err = mpage_process_page_bufs(mpd, head, head, lblk);
2270 pagevec_release(&pvec);
2275 pagevec_release(&pvec);
2279 static int __writepage(struct page *page, struct writeback_control *wbc,
2282 struct address_space *mapping = data;
2283 int ret = ext4_writepage(page, wbc);
2284 mapping_set_error(mapping, ret);
2288 static int ext4_writepages(struct address_space *mapping,
2289 struct writeback_control *wbc)
2291 pgoff_t writeback_index = 0;
2292 long nr_to_write = wbc->nr_to_write;
2293 int range_whole = 0;
2295 handle_t *handle = NULL;
2296 struct mpage_da_data mpd;
2297 struct inode *inode = mapping->host;
2298 int needed_blocks, rsv_blocks = 0, ret = 0;
2299 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2301 struct blk_plug plug;
2302 bool give_up_on_write = false;
2304 trace_ext4_writepages(inode, wbc);
2307 * No pages to write? This is mainly a kludge to avoid starting
2308 * a transaction for special inodes like journal inode on last iput()
2309 * because that could violate lock ordering on umount
2311 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2312 goto out_writepages;
2314 if (ext4_should_journal_data(inode)) {
2315 struct blk_plug plug;
2317 blk_start_plug(&plug);
2318 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2319 blk_finish_plug(&plug);
2320 goto out_writepages;
2324 * If the filesystem has aborted, it is read-only, so return
2325 * right away instead of dumping stack traces later on that
2326 * will obscure the real source of the problem. We test
2327 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2328 * the latter could be true if the filesystem is mounted
2329 * read-only, and in that case, ext4_writepages should
2330 * *never* be called, so if that ever happens, we would want
2333 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2335 goto out_writepages;
2338 if (ext4_should_dioread_nolock(inode)) {
2340 * We may need to convert up to one extent per block in
2341 * the page and we may dirty the inode.
2343 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2347 * If we have inline data and arrive here, it means that
2348 * we will soon create the block for the 1st page, so
2349 * we'd better clear the inline data here.
2351 if (ext4_has_inline_data(inode)) {
2352 /* Just inode will be modified... */
2353 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2354 if (IS_ERR(handle)) {
2355 ret = PTR_ERR(handle);
2356 goto out_writepages;
2358 BUG_ON(ext4_test_inode_state(inode,
2359 EXT4_STATE_MAY_INLINE_DATA));
2360 ext4_destroy_inline_data(handle, inode);
2361 ext4_journal_stop(handle);
2364 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2367 if (wbc->range_cyclic) {
2368 writeback_index = mapping->writeback_index;
2369 if (writeback_index)
2371 mpd.first_page = writeback_index;
2374 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2375 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2380 ext4_io_submit_init(&mpd.io_submit, wbc);
2382 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2383 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2385 blk_start_plug(&plug);
2386 while (!done && mpd.first_page <= mpd.last_page) {
2387 /* For each extent of pages we use new io_end */
2388 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2389 if (!mpd.io_submit.io_end) {
2395 * We have two constraints: We find one extent to map and we
2396 * must always write out whole page (makes a difference when
2397 * blocksize < pagesize) so that we don't block on IO when we
2398 * try to write out the rest of the page. Journalled mode is
2399 * not supported by delalloc.
2401 BUG_ON(ext4_should_journal_data(inode));
2402 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2404 /* start a new transaction */
2405 handle = ext4_journal_start_with_reserve(inode,
2406 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2407 if (IS_ERR(handle)) {
2408 ret = PTR_ERR(handle);
2409 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2410 "%ld pages, ino %lu; err %d", __func__,
2411 wbc->nr_to_write, inode->i_ino, ret);
2412 /* Release allocated io_end */
2413 ext4_put_io_end(mpd.io_submit.io_end);
2417 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2418 ret = mpage_prepare_extent_to_map(&mpd);
2421 ret = mpage_map_and_submit_extent(handle, &mpd,
2425 * We scanned the whole range (or exhausted
2426 * nr_to_write), submitted what was mapped and
2427 * didn't find anything needing mapping. We are
2433 ext4_journal_stop(handle);
2434 /* Submit prepared bio */
2435 ext4_io_submit(&mpd.io_submit);
2436 /* Unlock pages we didn't use */
2437 mpage_release_unused_pages(&mpd, give_up_on_write);
2438 /* Drop our io_end reference we got from init */
2439 ext4_put_io_end(mpd.io_submit.io_end);
2441 if (ret == -ENOSPC && sbi->s_journal) {
2443 * Commit the transaction which would
2444 * free blocks released in the transaction
2447 jbd2_journal_force_commit_nested(sbi->s_journal);
2451 /* Fatal error - ENOMEM, EIO... */
2455 blk_finish_plug(&plug);
2456 if (!ret && !cycled && wbc->nr_to_write > 0) {
2458 mpd.last_page = writeback_index - 1;
2464 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2466 * Set the writeback_index so that range_cyclic
2467 * mode will write it back later
2469 mapping->writeback_index = mpd.first_page;
2472 trace_ext4_writepages_result(inode, wbc, ret,
2473 nr_to_write - wbc->nr_to_write);
2477 static int ext4_nonda_switch(struct super_block *sb)
2479 s64 free_clusters, dirty_clusters;
2480 struct ext4_sb_info *sbi = EXT4_SB(sb);
2483 * switch to non delalloc mode if we are running low
2484 * on free block. The free block accounting via percpu
2485 * counters can get slightly wrong with percpu_counter_batch getting
2486 * accumulated on each CPU without updating global counters
2487 * Delalloc need an accurate free block accounting. So switch
2488 * to non delalloc when we are near to error range.
2491 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2493 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2495 * Start pushing delalloc when 1/2 of free blocks are dirty.
2497 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2498 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2500 if (2 * free_clusters < 3 * dirty_clusters ||
2501 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2503 * free block count is less than 150% of dirty blocks
2504 * or free blocks is less than watermark
2511 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2512 loff_t pos, unsigned len, unsigned flags,
2513 struct page **pagep, void **fsdata)
2515 int ret, retries = 0;
2518 struct inode *inode = mapping->host;
2521 index = pos >> PAGE_CACHE_SHIFT;
2523 if (ext4_nonda_switch(inode->i_sb)) {
2524 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2525 return ext4_write_begin(file, mapping, pos,
2526 len, flags, pagep, fsdata);
2528 *fsdata = (void *)0;
2529 trace_ext4_da_write_begin(inode, pos, len, flags);
2531 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2532 ret = ext4_da_write_inline_data_begin(mapping, inode,
2542 * grab_cache_page_write_begin() can take a long time if the
2543 * system is thrashing due to memory pressure, or if the page
2544 * is being written back. So grab it first before we start
2545 * the transaction handle. This also allows us to allocate
2546 * the page (if needed) without using GFP_NOFS.
2549 page = grab_cache_page_write_begin(mapping, index, flags);
2555 * With delayed allocation, we don't log the i_disksize update
2556 * if there is delayed block allocation. But we still need
2557 * to journalling the i_disksize update if writes to the end
2558 * of file which has an already mapped buffer.
2561 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2562 if (IS_ERR(handle)) {
2563 page_cache_release(page);
2564 return PTR_ERR(handle);
2568 if (page->mapping != mapping) {
2569 /* The page got truncated from under us */
2571 page_cache_release(page);
2572 ext4_journal_stop(handle);
2575 /* In case writeback began while the page was unlocked */
2576 wait_for_stable_page(page);
2578 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2581 ext4_journal_stop(handle);
2583 * block_write_begin may have instantiated a few blocks
2584 * outside i_size. Trim these off again. Don't need
2585 * i_size_read because we hold i_mutex.
2587 if (pos + len > inode->i_size)
2588 ext4_truncate_failed_write(inode);
2590 if (ret == -ENOSPC &&
2591 ext4_should_retry_alloc(inode->i_sb, &retries))
2594 page_cache_release(page);
2603 * Check if we should update i_disksize
2604 * when write to the end of file but not require block allocation
2606 static int ext4_da_should_update_i_disksize(struct page *page,
2607 unsigned long offset)
2609 struct buffer_head *bh;
2610 struct inode *inode = page->mapping->host;
2614 bh = page_buffers(page);
2615 idx = offset >> inode->i_blkbits;
2617 for (i = 0; i < idx; i++)
2618 bh = bh->b_this_page;
2620 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2625 static int ext4_da_write_end(struct file *file,
2626 struct address_space *mapping,
2627 loff_t pos, unsigned len, unsigned copied,
2628 struct page *page, void *fsdata)
2630 struct inode *inode = mapping->host;
2632 handle_t *handle = ext4_journal_current_handle();
2634 unsigned long start, end;
2635 int write_mode = (int)(unsigned long)fsdata;
2637 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2638 return ext4_write_end(file, mapping, pos,
2639 len, copied, page, fsdata);
2641 trace_ext4_da_write_end(inode, pos, len, copied);
2642 start = pos & (PAGE_CACHE_SIZE - 1);
2643 end = start + copied - 1;
2646 * generic_write_end() will run mark_inode_dirty() if i_size
2647 * changes. So let's piggyback the i_disksize mark_inode_dirty
2650 new_i_size = pos + copied;
2651 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2652 if (ext4_has_inline_data(inode) ||
2653 ext4_da_should_update_i_disksize(page, end)) {
2654 ext4_update_i_disksize(inode, new_i_size);
2655 /* We need to mark inode dirty even if
2656 * new_i_size is less that inode->i_size
2657 * bu greater than i_disksize.(hint delalloc)
2659 ext4_mark_inode_dirty(handle, inode);
2663 if (write_mode != CONVERT_INLINE_DATA &&
2664 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2665 ext4_has_inline_data(inode))
2666 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2669 ret2 = generic_write_end(file, mapping, pos, len, copied,
2675 ret2 = ext4_journal_stop(handle);
2679 return ret ? ret : copied;
2682 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2683 unsigned int length)
2686 * Drop reserved blocks
2688 BUG_ON(!PageLocked(page));
2689 if (!page_has_buffers(page))
2692 ext4_da_page_release_reservation(page, offset, length);
2695 ext4_invalidatepage(page, offset, length);
2701 * Force all delayed allocation blocks to be allocated for a given inode.
2703 int ext4_alloc_da_blocks(struct inode *inode)
2705 trace_ext4_alloc_da_blocks(inode);
2707 if (!EXT4_I(inode)->i_reserved_data_blocks)
2711 * We do something simple for now. The filemap_flush() will
2712 * also start triggering a write of the data blocks, which is
2713 * not strictly speaking necessary (and for users of
2714 * laptop_mode, not even desirable). However, to do otherwise
2715 * would require replicating code paths in:
2717 * ext4_writepages() ->
2718 * write_cache_pages() ---> (via passed in callback function)
2719 * __mpage_da_writepage() -->
2720 * mpage_add_bh_to_extent()
2721 * mpage_da_map_blocks()
2723 * The problem is that write_cache_pages(), located in
2724 * mm/page-writeback.c, marks pages clean in preparation for
2725 * doing I/O, which is not desirable if we're not planning on
2728 * We could call write_cache_pages(), and then redirty all of
2729 * the pages by calling redirty_page_for_writepage() but that
2730 * would be ugly in the extreme. So instead we would need to
2731 * replicate parts of the code in the above functions,
2732 * simplifying them because we wouldn't actually intend to
2733 * write out the pages, but rather only collect contiguous
2734 * logical block extents, call the multi-block allocator, and
2735 * then update the buffer heads with the block allocations.
2737 * For now, though, we'll cheat by calling filemap_flush(),
2738 * which will map the blocks, and start the I/O, but not
2739 * actually wait for the I/O to complete.
2741 return filemap_flush(inode->i_mapping);
2745 * bmap() is special. It gets used by applications such as lilo and by
2746 * the swapper to find the on-disk block of a specific piece of data.
2748 * Naturally, this is dangerous if the block concerned is still in the
2749 * journal. If somebody makes a swapfile on an ext4 data-journaling
2750 * filesystem and enables swap, then they may get a nasty shock when the
2751 * data getting swapped to that swapfile suddenly gets overwritten by
2752 * the original zero's written out previously to the journal and
2753 * awaiting writeback in the kernel's buffer cache.
2755 * So, if we see any bmap calls here on a modified, data-journaled file,
2756 * take extra steps to flush any blocks which might be in the cache.
2758 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2760 struct inode *inode = mapping->host;
2765 * We can get here for an inline file via the FIBMAP ioctl
2767 if (ext4_has_inline_data(inode))
2770 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2771 test_opt(inode->i_sb, DELALLOC)) {
2773 * With delalloc we want to sync the file
2774 * so that we can make sure we allocate
2777 filemap_write_and_wait(mapping);
2780 if (EXT4_JOURNAL(inode) &&
2781 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2783 * This is a REALLY heavyweight approach, but the use of
2784 * bmap on dirty files is expected to be extremely rare:
2785 * only if we run lilo or swapon on a freshly made file
2786 * do we expect this to happen.
2788 * (bmap requires CAP_SYS_RAWIO so this does not
2789 * represent an unprivileged user DOS attack --- we'd be
2790 * in trouble if mortal users could trigger this path at
2793 * NB. EXT4_STATE_JDATA is not set on files other than
2794 * regular files. If somebody wants to bmap a directory
2795 * or symlink and gets confused because the buffer
2796 * hasn't yet been flushed to disk, they deserve
2797 * everything they get.
2800 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2801 journal = EXT4_JOURNAL(inode);
2802 jbd2_journal_lock_updates(journal);
2803 err = jbd2_journal_flush(journal);
2804 jbd2_journal_unlock_updates(journal);
2810 return generic_block_bmap(mapping, block, ext4_get_block);
2813 static int ext4_readpage(struct file *file, struct page *page)
2816 struct inode *inode = page->mapping->host;
2818 trace_ext4_readpage(page);
2820 if (ext4_has_inline_data(inode))
2821 ret = ext4_readpage_inline(inode, page);
2824 return mpage_readpage(page, ext4_get_block);
2830 ext4_readpages(struct file *file, struct address_space *mapping,
2831 struct list_head *pages, unsigned nr_pages)
2833 struct inode *inode = mapping->host;
2835 /* If the file has inline data, no need to do readpages. */
2836 if (ext4_has_inline_data(inode))
2839 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2842 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2843 unsigned int length)
2845 trace_ext4_invalidatepage(page, offset, length);
2847 /* No journalling happens on data buffers when this function is used */
2848 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2850 block_invalidatepage(page, offset, length);
2853 static int __ext4_journalled_invalidatepage(struct page *page,
2854 unsigned int offset,
2855 unsigned int length)
2857 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2859 trace_ext4_journalled_invalidatepage(page, offset, length);
2862 * If it's a full truncate we just forget about the pending dirtying
2864 if (offset == 0 && length == PAGE_CACHE_SIZE)
2865 ClearPageChecked(page);
2867 return jbd2_journal_invalidatepage(journal, page, offset, length);
2870 /* Wrapper for aops... */
2871 static void ext4_journalled_invalidatepage(struct page *page,
2872 unsigned int offset,
2873 unsigned int length)
2875 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2878 static int ext4_releasepage(struct page *page, gfp_t wait)
2880 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2882 trace_ext4_releasepage(page);
2884 /* Page has dirty journalled data -> cannot release */
2885 if (PageChecked(page))
2888 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2890 return try_to_free_buffers(page);
2894 * ext4_get_block used when preparing for a DIO write or buffer write.
2895 * We allocate an uinitialized extent if blocks haven't been allocated.
2896 * The extent will be converted to initialized after the IO is complete.
2898 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2899 struct buffer_head *bh_result, int create)
2901 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2902 inode->i_ino, create);
2903 return _ext4_get_block(inode, iblock, bh_result,
2904 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2907 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2908 struct buffer_head *bh_result, int create)
2910 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2911 inode->i_ino, create);
2912 return _ext4_get_block(inode, iblock, bh_result,
2913 EXT4_GET_BLOCKS_NO_LOCK);
2916 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2917 ssize_t size, void *private)
2919 ext4_io_end_t *io_end = iocb->private;
2921 /* if not async direct IO just return */
2925 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2926 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2927 iocb->private, io_end->inode->i_ino, iocb, offset,
2930 iocb->private = NULL;
2931 io_end->offset = offset;
2932 io_end->size = size;
2933 ext4_put_io_end(io_end);
2937 * For ext4 extent files, ext4 will do direct-io write to holes,
2938 * preallocated extents, and those write extend the file, no need to
2939 * fall back to buffered IO.
2941 * For holes, we fallocate those blocks, mark them as unwritten
2942 * If those blocks were preallocated, we mark sure they are split, but
2943 * still keep the range to write as unwritten.
2945 * The unwritten extents will be converted to written when DIO is completed.
2946 * For async direct IO, since the IO may still pending when return, we
2947 * set up an end_io call back function, which will do the conversion
2948 * when async direct IO completed.
2950 * If the O_DIRECT write will extend the file then add this inode to the
2951 * orphan list. So recovery will truncate it back to the original size
2952 * if the machine crashes during the write.
2955 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2956 struct iov_iter *iter, loff_t offset)
2958 struct file *file = iocb->ki_filp;
2959 struct inode *inode = file->f_mapping->host;
2961 size_t count = iov_iter_count(iter);
2963 get_block_t *get_block_func = NULL;
2965 loff_t final_size = offset + count;
2966 ext4_io_end_t *io_end = NULL;
2968 /* Use the old path for reads and writes beyond i_size. */
2969 if (rw != WRITE || final_size > inode->i_size)
2970 return ext4_ind_direct_IO(rw, iocb, iter, offset);
2972 BUG_ON(iocb->private == NULL);
2975 * Make all waiters for direct IO properly wait also for extent
2976 * conversion. This also disallows race between truncate() and
2977 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
2980 atomic_inc(&inode->i_dio_count);
2982 /* If we do a overwrite dio, i_mutex locking can be released */
2983 overwrite = *((int *)iocb->private);
2986 down_read(&EXT4_I(inode)->i_data_sem);
2987 mutex_unlock(&inode->i_mutex);
2991 * We could direct write to holes and fallocate.
2993 * Allocated blocks to fill the hole are marked as
2994 * unwritten to prevent parallel buffered read to expose
2995 * the stale data before DIO complete the data IO.
2997 * As to previously fallocated extents, ext4 get_block will
2998 * just simply mark the buffer mapped but still keep the
2999 * extents unwritten.
3001 * For non AIO case, we will convert those unwritten extents
3002 * to written after return back from blockdev_direct_IO.
3004 * For async DIO, the conversion needs to be deferred when the
3005 * IO is completed. The ext4 end_io callback function will be
3006 * called to take care of the conversion work. Here for async
3007 * case, we allocate an io_end structure to hook to the iocb.
3009 iocb->private = NULL;
3010 ext4_inode_aio_set(inode, NULL);
3011 if (!is_sync_kiocb(iocb)) {
3012 io_end = ext4_init_io_end(inode, GFP_NOFS);
3018 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3020 iocb->private = ext4_get_io_end(io_end);
3022 * we save the io structure for current async direct
3023 * IO, so that later ext4_map_blocks() could flag the
3024 * io structure whether there is a unwritten extents
3025 * needs to be converted when IO is completed.
3027 ext4_inode_aio_set(inode, io_end);
3031 get_block_func = ext4_get_block_write_nolock;
3033 get_block_func = ext4_get_block_write;
3034 dio_flags = DIO_LOCKING;
3036 ret = __blockdev_direct_IO(rw, iocb, inode,
3037 inode->i_sb->s_bdev, iter,
3045 * Put our reference to io_end. This can free the io_end structure e.g.
3046 * in sync IO case or in case of error. It can even perform extent
3047 * conversion if all bios we submitted finished before we got here.
3048 * Note that in that case iocb->private can be already set to NULL
3052 ext4_inode_aio_set(inode, NULL);
3053 ext4_put_io_end(io_end);
3055 * When no IO was submitted ext4_end_io_dio() was not
3056 * called so we have to put iocb's reference.
3058 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3059 WARN_ON(iocb->private != io_end);
3060 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3061 ext4_put_io_end(io_end);
3062 iocb->private = NULL;
3065 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3066 EXT4_STATE_DIO_UNWRITTEN)) {
3069 * for non AIO case, since the IO is already
3070 * completed, we could do the conversion right here
3072 err = ext4_convert_unwritten_extents(NULL, inode,
3076 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3081 inode_dio_done(inode);
3082 /* take i_mutex locking again if we do a ovewrite dio */
3084 up_read(&EXT4_I(inode)->i_data_sem);
3085 mutex_lock(&inode->i_mutex);
3091 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3092 struct iov_iter *iter, loff_t offset)
3094 struct file *file = iocb->ki_filp;
3095 struct inode *inode = file->f_mapping->host;
3096 size_t count = iov_iter_count(iter);
3100 * If we are doing data journalling we don't support O_DIRECT
3102 if (ext4_should_journal_data(inode))
3105 /* Let buffer I/O handle the inline data case. */
3106 if (ext4_has_inline_data(inode))
3109 trace_ext4_direct_IO_enter(inode, offset, count, rw);
3110 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3111 ret = ext4_ext_direct_IO(rw, iocb, iter, offset);
3113 ret = ext4_ind_direct_IO(rw, iocb, iter, offset);
3114 trace_ext4_direct_IO_exit(inode, offset, count, rw, ret);
3119 * Pages can be marked dirty completely asynchronously from ext4's journalling
3120 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3121 * much here because ->set_page_dirty is called under VFS locks. The page is
3122 * not necessarily locked.
3124 * We cannot just dirty the page and leave attached buffers clean, because the
3125 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3126 * or jbddirty because all the journalling code will explode.
3128 * So what we do is to mark the page "pending dirty" and next time writepage
3129 * is called, propagate that into the buffers appropriately.
3131 static int ext4_journalled_set_page_dirty(struct page *page)
3133 SetPageChecked(page);
3134 return __set_page_dirty_nobuffers(page);
3137 static const struct address_space_operations ext4_aops = {
3138 .readpage = ext4_readpage,
3139 .readpages = ext4_readpages,
3140 .writepage = ext4_writepage,
3141 .writepages = ext4_writepages,
3142 .write_begin = ext4_write_begin,
3143 .write_end = ext4_write_end,
3145 .invalidatepage = ext4_invalidatepage,
3146 .releasepage = ext4_releasepage,
3147 .direct_IO = ext4_direct_IO,
3148 .migratepage = buffer_migrate_page,
3149 .is_partially_uptodate = block_is_partially_uptodate,
3150 .error_remove_page = generic_error_remove_page,
3153 static const struct address_space_operations ext4_journalled_aops = {
3154 .readpage = ext4_readpage,
3155 .readpages = ext4_readpages,
3156 .writepage = ext4_writepage,
3157 .writepages = ext4_writepages,
3158 .write_begin = ext4_write_begin,
3159 .write_end = ext4_journalled_write_end,
3160 .set_page_dirty = ext4_journalled_set_page_dirty,
3162 .invalidatepage = ext4_journalled_invalidatepage,
3163 .releasepage = ext4_releasepage,
3164 .direct_IO = ext4_direct_IO,
3165 .is_partially_uptodate = block_is_partially_uptodate,
3166 .error_remove_page = generic_error_remove_page,
3169 static const struct address_space_operations ext4_da_aops = {
3170 .readpage = ext4_readpage,
3171 .readpages = ext4_readpages,
3172 .writepage = ext4_writepage,
3173 .writepages = ext4_writepages,
3174 .write_begin = ext4_da_write_begin,
3175 .write_end = ext4_da_write_end,
3177 .invalidatepage = ext4_da_invalidatepage,
3178 .releasepage = ext4_releasepage,
3179 .direct_IO = ext4_direct_IO,
3180 .migratepage = buffer_migrate_page,
3181 .is_partially_uptodate = block_is_partially_uptodate,
3182 .error_remove_page = generic_error_remove_page,
3185 void ext4_set_aops(struct inode *inode)
3187 switch (ext4_inode_journal_mode(inode)) {
3188 case EXT4_INODE_ORDERED_DATA_MODE:
3189 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3191 case EXT4_INODE_WRITEBACK_DATA_MODE:
3192 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3194 case EXT4_INODE_JOURNAL_DATA_MODE:
3195 inode->i_mapping->a_ops = &ext4_journalled_aops;
3200 if (test_opt(inode->i_sb, DELALLOC))
3201 inode->i_mapping->a_ops = &ext4_da_aops;
3203 inode->i_mapping->a_ops = &ext4_aops;
3207 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3208 * starting from file offset 'from'. The range to be zero'd must
3209 * be contained with in one block. If the specified range exceeds
3210 * the end of the block it will be shortened to end of the block
3211 * that cooresponds to 'from'
3213 static int ext4_block_zero_page_range(handle_t *handle,
3214 struct address_space *mapping, loff_t from, loff_t length)
3216 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3217 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3218 unsigned blocksize, max, pos;
3220 struct inode *inode = mapping->host;
3221 struct buffer_head *bh;
3225 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3226 mapping_gfp_mask(mapping) & ~__GFP_FS);
3230 blocksize = inode->i_sb->s_blocksize;
3231 max = blocksize - (offset & (blocksize - 1));
3234 * correct length if it does not fall between
3235 * 'from' and the end of the block
3237 if (length > max || length < 0)
3240 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3242 if (!page_has_buffers(page))
3243 create_empty_buffers(page, blocksize, 0);
3245 /* Find the buffer that contains "offset" */
3246 bh = page_buffers(page);
3248 while (offset >= pos) {
3249 bh = bh->b_this_page;
3253 if (buffer_freed(bh)) {
3254 BUFFER_TRACE(bh, "freed: skip");
3257 if (!buffer_mapped(bh)) {
3258 BUFFER_TRACE(bh, "unmapped");
3259 ext4_get_block(inode, iblock, bh, 0);
3260 /* unmapped? It's a hole - nothing to do */
3261 if (!buffer_mapped(bh)) {
3262 BUFFER_TRACE(bh, "still unmapped");
3267 /* Ok, it's mapped. Make sure it's up-to-date */
3268 if (PageUptodate(page))
3269 set_buffer_uptodate(bh);
3271 if (!buffer_uptodate(bh)) {
3273 ll_rw_block(READ, 1, &bh);
3275 /* Uhhuh. Read error. Complain and punt. */
3276 if (!buffer_uptodate(bh))
3279 if (ext4_should_journal_data(inode)) {
3280 BUFFER_TRACE(bh, "get write access");
3281 err = ext4_journal_get_write_access(handle, bh);
3285 zero_user(page, offset, length);
3286 BUFFER_TRACE(bh, "zeroed end of block");
3288 if (ext4_should_journal_data(inode)) {
3289 err = ext4_handle_dirty_metadata(handle, inode, bh);
3292 mark_buffer_dirty(bh);
3293 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3294 err = ext4_jbd2_file_inode(handle, inode);
3299 page_cache_release(page);
3304 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3305 * up to the end of the block which corresponds to `from'.
3306 * This required during truncate. We need to physically zero the tail end
3307 * of that block so it doesn't yield old data if the file is later grown.
3309 static int ext4_block_truncate_page(handle_t *handle,
3310 struct address_space *mapping, loff_t from)
3312 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3315 struct inode *inode = mapping->host;
3317 blocksize = inode->i_sb->s_blocksize;
3318 length = blocksize - (offset & (blocksize - 1));
3320 return ext4_block_zero_page_range(handle, mapping, from, length);
3323 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3324 loff_t lstart, loff_t length)
3326 struct super_block *sb = inode->i_sb;
3327 struct address_space *mapping = inode->i_mapping;
3328 unsigned partial_start, partial_end;
3329 ext4_fsblk_t start, end;
3330 loff_t byte_end = (lstart + length - 1);
3333 partial_start = lstart & (sb->s_blocksize - 1);
3334 partial_end = byte_end & (sb->s_blocksize - 1);
3336 start = lstart >> sb->s_blocksize_bits;
3337 end = byte_end >> sb->s_blocksize_bits;
3339 /* Handle partial zero within the single block */
3341 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3342 err = ext4_block_zero_page_range(handle, mapping,
3346 /* Handle partial zero out on the start of the range */
3347 if (partial_start) {
3348 err = ext4_block_zero_page_range(handle, mapping,
3349 lstart, sb->s_blocksize);
3353 /* Handle partial zero out on the end of the range */
3354 if (partial_end != sb->s_blocksize - 1)
3355 err = ext4_block_zero_page_range(handle, mapping,
3356 byte_end - partial_end,
3361 int ext4_can_truncate(struct inode *inode)
3363 if (S_ISREG(inode->i_mode))
3365 if (S_ISDIR(inode->i_mode))
3367 if (S_ISLNK(inode->i_mode))
3368 return !ext4_inode_is_fast_symlink(inode);
3373 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3374 * associated with the given offset and length
3376 * @inode: File inode
3377 * @offset: The offset where the hole will begin
3378 * @len: The length of the hole
3380 * Returns: 0 on success or negative on failure
3383 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3385 struct super_block *sb = inode->i_sb;
3386 ext4_lblk_t first_block, stop_block;
3387 struct address_space *mapping = inode->i_mapping;
3388 loff_t first_block_offset, last_block_offset;
3390 unsigned int credits;
3393 if (!S_ISREG(inode->i_mode))
3396 trace_ext4_punch_hole(inode, offset, length, 0);
3399 * Write out all dirty pages to avoid race conditions
3400 * Then release them.
3402 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3403 ret = filemap_write_and_wait_range(mapping, offset,
3404 offset + length - 1);
3409 mutex_lock(&inode->i_mutex);
3411 /* No need to punch hole beyond i_size */
3412 if (offset >= inode->i_size)
3416 * If the hole extends beyond i_size, set the hole
3417 * to end after the page that contains i_size
3419 if (offset + length > inode->i_size) {
3420 length = inode->i_size +
3421 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3425 if (offset & (sb->s_blocksize - 1) ||
3426 (offset + length) & (sb->s_blocksize - 1)) {
3428 * Attach jinode to inode for jbd2 if we do any zeroing of
3431 ret = ext4_inode_attach_jinode(inode);
3437 first_block_offset = round_up(offset, sb->s_blocksize);
3438 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3440 /* Now release the pages and zero block aligned part of pages*/
3441 if (last_block_offset > first_block_offset)
3442 truncate_pagecache_range(inode, first_block_offset,
3445 /* Wait all existing dio workers, newcomers will block on i_mutex */
3446 ext4_inode_block_unlocked_dio(inode);
3447 inode_dio_wait(inode);
3449 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3450 credits = ext4_writepage_trans_blocks(inode);
3452 credits = ext4_blocks_for_truncate(inode);
3453 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3454 if (IS_ERR(handle)) {
3455 ret = PTR_ERR(handle);
3456 ext4_std_error(sb, ret);
3460 ret = ext4_zero_partial_blocks(handle, inode, offset,
3465 first_block = (offset + sb->s_blocksize - 1) >>
3466 EXT4_BLOCK_SIZE_BITS(sb);
3467 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3469 /* If there are no blocks to remove, return now */
3470 if (first_block >= stop_block)
3473 down_write(&EXT4_I(inode)->i_data_sem);
3474 ext4_discard_preallocations(inode);
3476 ret = ext4_es_remove_extent(inode, first_block,
3477 stop_block - first_block);
3479 up_write(&EXT4_I(inode)->i_data_sem);
3483 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3484 ret = ext4_ext_remove_space(inode, first_block,
3487 ret = ext4_ind_remove_space(handle, inode, first_block,
3490 up_write(&EXT4_I(inode)->i_data_sem);
3492 ext4_handle_sync(handle);
3494 /* Now release the pages again to reduce race window */
3495 if (last_block_offset > first_block_offset)
3496 truncate_pagecache_range(inode, first_block_offset,
3499 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3500 ext4_mark_inode_dirty(handle, inode);
3502 ext4_journal_stop(handle);
3504 ext4_inode_resume_unlocked_dio(inode);
3506 mutex_unlock(&inode->i_mutex);
3510 int ext4_inode_attach_jinode(struct inode *inode)
3512 struct ext4_inode_info *ei = EXT4_I(inode);
3513 struct jbd2_inode *jinode;
3515 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3518 jinode = jbd2_alloc_inode(GFP_KERNEL);
3519 spin_lock(&inode->i_lock);
3522 spin_unlock(&inode->i_lock);
3525 ei->jinode = jinode;
3526 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3529 spin_unlock(&inode->i_lock);
3530 if (unlikely(jinode != NULL))
3531 jbd2_free_inode(jinode);
3538 * We block out ext4_get_block() block instantiations across the entire
3539 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3540 * simultaneously on behalf of the same inode.
3542 * As we work through the truncate and commit bits of it to the journal there
3543 * is one core, guiding principle: the file's tree must always be consistent on
3544 * disk. We must be able to restart the truncate after a crash.
3546 * The file's tree may be transiently inconsistent in memory (although it
3547 * probably isn't), but whenever we close off and commit a journal transaction,
3548 * the contents of (the filesystem + the journal) must be consistent and
3549 * restartable. It's pretty simple, really: bottom up, right to left (although
3550 * left-to-right works OK too).
3552 * Note that at recovery time, journal replay occurs *before* the restart of
3553 * truncate against the orphan inode list.
3555 * The committed inode has the new, desired i_size (which is the same as
3556 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3557 * that this inode's truncate did not complete and it will again call
3558 * ext4_truncate() to have another go. So there will be instantiated blocks
3559 * to the right of the truncation point in a crashed ext4 filesystem. But
3560 * that's fine - as long as they are linked from the inode, the post-crash
3561 * ext4_truncate() run will find them and release them.
3563 void ext4_truncate(struct inode *inode)
3565 struct ext4_inode_info *ei = EXT4_I(inode);
3566 unsigned int credits;
3568 struct address_space *mapping = inode->i_mapping;
3571 * There is a possibility that we're either freeing the inode
3572 * or it's a completely new inode. In those cases we might not
3573 * have i_mutex locked because it's not necessary.
3575 if (!(inode->i_state & (I_NEW|I_FREEING)))
3576 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3577 trace_ext4_truncate_enter(inode);
3579 if (!ext4_can_truncate(inode))
3582 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3584 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3585 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3587 if (ext4_has_inline_data(inode)) {
3590 ext4_inline_data_truncate(inode, &has_inline);
3595 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3596 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3597 if (ext4_inode_attach_jinode(inode) < 0)
3601 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3602 credits = ext4_writepage_trans_blocks(inode);
3604 credits = ext4_blocks_for_truncate(inode);
3606 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3607 if (IS_ERR(handle)) {
3608 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3612 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3613 ext4_block_truncate_page(handle, mapping, inode->i_size);
3616 * We add the inode to the orphan list, so that if this
3617 * truncate spans multiple transactions, and we crash, we will
3618 * resume the truncate when the filesystem recovers. It also
3619 * marks the inode dirty, to catch the new size.
3621 * Implication: the file must always be in a sane, consistent
3622 * truncatable state while each transaction commits.
3624 if (ext4_orphan_add(handle, inode))
3627 down_write(&EXT4_I(inode)->i_data_sem);
3629 ext4_discard_preallocations(inode);
3631 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3632 ext4_ext_truncate(handle, inode);
3634 ext4_ind_truncate(handle, inode);
3636 up_write(&ei->i_data_sem);
3639 ext4_handle_sync(handle);
3643 * If this was a simple ftruncate() and the file will remain alive,
3644 * then we need to clear up the orphan record which we created above.
3645 * However, if this was a real unlink then we were called by
3646 * ext4_delete_inode(), and we allow that function to clean up the
3647 * orphan info for us.
3650 ext4_orphan_del(handle, inode);
3652 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3653 ext4_mark_inode_dirty(handle, inode);
3654 ext4_journal_stop(handle);
3656 trace_ext4_truncate_exit(inode);
3660 * ext4_get_inode_loc returns with an extra refcount against the inode's
3661 * underlying buffer_head on success. If 'in_mem' is true, we have all
3662 * data in memory that is needed to recreate the on-disk version of this
3665 static int __ext4_get_inode_loc(struct inode *inode,
3666 struct ext4_iloc *iloc, int in_mem)
3668 struct ext4_group_desc *gdp;
3669 struct buffer_head *bh;
3670 struct super_block *sb = inode->i_sb;
3672 int inodes_per_block, inode_offset;
3675 if (!ext4_valid_inum(sb, inode->i_ino))
3678 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3679 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3684 * Figure out the offset within the block group inode table
3686 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3687 inode_offset = ((inode->i_ino - 1) %
3688 EXT4_INODES_PER_GROUP(sb));
3689 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3690 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3692 bh = sb_getblk(sb, block);
3695 if (!buffer_uptodate(bh)) {
3699 * If the buffer has the write error flag, we have failed
3700 * to write out another inode in the same block. In this
3701 * case, we don't have to read the block because we may
3702 * read the old inode data successfully.
3704 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3705 set_buffer_uptodate(bh);
3707 if (buffer_uptodate(bh)) {
3708 /* someone brought it uptodate while we waited */
3714 * If we have all information of the inode in memory and this
3715 * is the only valid inode in the block, we need not read the
3719 struct buffer_head *bitmap_bh;
3722 start = inode_offset & ~(inodes_per_block - 1);
3724 /* Is the inode bitmap in cache? */
3725 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3726 if (unlikely(!bitmap_bh))
3730 * If the inode bitmap isn't in cache then the
3731 * optimisation may end up performing two reads instead
3732 * of one, so skip it.
3734 if (!buffer_uptodate(bitmap_bh)) {
3738 for (i = start; i < start + inodes_per_block; i++) {
3739 if (i == inode_offset)
3741 if (ext4_test_bit(i, bitmap_bh->b_data))
3745 if (i == start + inodes_per_block) {
3746 /* all other inodes are free, so skip I/O */
3747 memset(bh->b_data, 0, bh->b_size);
3748 set_buffer_uptodate(bh);
3756 * If we need to do any I/O, try to pre-readahead extra
3757 * blocks from the inode table.
3759 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3760 ext4_fsblk_t b, end, table;
3762 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3764 table = ext4_inode_table(sb, gdp);
3765 /* s_inode_readahead_blks is always a power of 2 */
3766 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3770 num = EXT4_INODES_PER_GROUP(sb);
3771 if (ext4_has_group_desc_csum(sb))
3772 num -= ext4_itable_unused_count(sb, gdp);
3773 table += num / inodes_per_block;
3777 sb_breadahead(sb, b++);
3781 * There are other valid inodes in the buffer, this inode
3782 * has in-inode xattrs, or we don't have this inode in memory.
3783 * Read the block from disk.
3785 trace_ext4_load_inode(inode);
3787 bh->b_end_io = end_buffer_read_sync;
3788 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3790 if (!buffer_uptodate(bh)) {
3791 EXT4_ERROR_INODE_BLOCK(inode, block,
3792 "unable to read itable block");
3802 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3804 /* We have all inode data except xattrs in memory here. */
3805 return __ext4_get_inode_loc(inode, iloc,
3806 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3809 void ext4_set_inode_flags(struct inode *inode)
3811 unsigned int flags = EXT4_I(inode)->i_flags;
3812 unsigned int new_fl = 0;
3814 if (flags & EXT4_SYNC_FL)
3816 if (flags & EXT4_APPEND_FL)
3818 if (flags & EXT4_IMMUTABLE_FL)
3819 new_fl |= S_IMMUTABLE;
3820 if (flags & EXT4_NOATIME_FL)
3821 new_fl |= S_NOATIME;
3822 if (flags & EXT4_DIRSYNC_FL)
3823 new_fl |= S_DIRSYNC;
3824 inode_set_flags(inode, new_fl,
3825 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3828 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3829 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3831 unsigned int vfs_fl;
3832 unsigned long old_fl, new_fl;
3835 vfs_fl = ei->vfs_inode.i_flags;
3836 old_fl = ei->i_flags;
3837 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3838 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3840 if (vfs_fl & S_SYNC)
3841 new_fl |= EXT4_SYNC_FL;
3842 if (vfs_fl & S_APPEND)
3843 new_fl |= EXT4_APPEND_FL;
3844 if (vfs_fl & S_IMMUTABLE)
3845 new_fl |= EXT4_IMMUTABLE_FL;
3846 if (vfs_fl & S_NOATIME)
3847 new_fl |= EXT4_NOATIME_FL;
3848 if (vfs_fl & S_DIRSYNC)
3849 new_fl |= EXT4_DIRSYNC_FL;
3850 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3853 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3854 struct ext4_inode_info *ei)
3857 struct inode *inode = &(ei->vfs_inode);
3858 struct super_block *sb = inode->i_sb;
3860 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3861 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3862 /* we are using combined 48 bit field */
3863 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3864 le32_to_cpu(raw_inode->i_blocks_lo);
3865 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3866 /* i_blocks represent file system block size */
3867 return i_blocks << (inode->i_blkbits - 9);
3872 return le32_to_cpu(raw_inode->i_blocks_lo);
3876 static inline void ext4_iget_extra_inode(struct inode *inode,
3877 struct ext4_inode *raw_inode,
3878 struct ext4_inode_info *ei)
3880 __le32 *magic = (void *)raw_inode +
3881 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3882 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3883 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3884 ext4_find_inline_data_nolock(inode);
3886 EXT4_I(inode)->i_inline_off = 0;
3889 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3891 struct ext4_iloc iloc;
3892 struct ext4_inode *raw_inode;
3893 struct ext4_inode_info *ei;
3894 struct inode *inode;
3895 journal_t *journal = EXT4_SB(sb)->s_journal;
3901 inode = iget_locked(sb, ino);
3903 return ERR_PTR(-ENOMEM);
3904 if (!(inode->i_state & I_NEW))
3910 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3913 raw_inode = ext4_raw_inode(&iloc);
3915 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3916 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3917 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3918 EXT4_INODE_SIZE(inode->i_sb)) {
3919 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3920 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3921 EXT4_INODE_SIZE(inode->i_sb));
3926 ei->i_extra_isize = 0;
3928 /* Precompute checksum seed for inode metadata */
3929 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3930 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3931 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3933 __le32 inum = cpu_to_le32(inode->i_ino);
3934 __le32 gen = raw_inode->i_generation;
3935 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3937 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3941 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3942 EXT4_ERROR_INODE(inode, "checksum invalid");
3947 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3948 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3949 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3950 if (!(test_opt(inode->i_sb, NO_UID32))) {
3951 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3952 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3954 i_uid_write(inode, i_uid);
3955 i_gid_write(inode, i_gid);
3956 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3958 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3959 ei->i_inline_off = 0;
3960 ei->i_dir_start_lookup = 0;
3961 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3962 /* We now have enough fields to check if the inode was active or not.
3963 * This is needed because nfsd might try to access dead inodes
3964 * the test is that same one that e2fsck uses
3965 * NeilBrown 1999oct15
3967 if (inode->i_nlink == 0) {
3968 if ((inode->i_mode == 0 ||
3969 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
3970 ino != EXT4_BOOT_LOADER_INO) {
3971 /* this inode is deleted */
3975 /* The only unlinked inodes we let through here have
3976 * valid i_mode and are being read by the orphan
3977 * recovery code: that's fine, we're about to complete
3978 * the process of deleting those.
3979 * OR it is the EXT4_BOOT_LOADER_INO which is
3980 * not initialized on a new filesystem. */
3982 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3983 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3984 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3985 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3987 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3988 inode->i_size = ext4_isize(raw_inode);
3989 ei->i_disksize = inode->i_size;
3991 ei->i_reserved_quota = 0;
3993 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3994 ei->i_block_group = iloc.block_group;
3995 ei->i_last_alloc_group = ~0;
3997 * NOTE! The in-memory inode i_data array is in little-endian order
3998 * even on big-endian machines: we do NOT byteswap the block numbers!
4000 for (block = 0; block < EXT4_N_BLOCKS; block++)
4001 ei->i_data[block] = raw_inode->i_block[block];
4002 INIT_LIST_HEAD(&ei->i_orphan);
4005 * Set transaction id's of transactions that have to be committed
4006 * to finish f[data]sync. We set them to currently running transaction
4007 * as we cannot be sure that the inode or some of its metadata isn't
4008 * part of the transaction - the inode could have been reclaimed and
4009 * now it is reread from disk.
4012 transaction_t *transaction;
4015 read_lock(&journal->j_state_lock);
4016 if (journal->j_running_transaction)
4017 transaction = journal->j_running_transaction;
4019 transaction = journal->j_committing_transaction;
4021 tid = transaction->t_tid;
4023 tid = journal->j_commit_sequence;
4024 read_unlock(&journal->j_state_lock);
4025 ei->i_sync_tid = tid;
4026 ei->i_datasync_tid = tid;
4029 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4030 if (ei->i_extra_isize == 0) {
4031 /* The extra space is currently unused. Use it. */
4032 ei->i_extra_isize = sizeof(struct ext4_inode) -
4033 EXT4_GOOD_OLD_INODE_SIZE;
4035 ext4_iget_extra_inode(inode, raw_inode, ei);
4039 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4040 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4041 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4042 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4044 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4045 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4046 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4047 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4049 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4054 if (ei->i_file_acl &&
4055 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4056 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4060 } else if (!ext4_has_inline_data(inode)) {
4061 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4062 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4063 (S_ISLNK(inode->i_mode) &&
4064 !ext4_inode_is_fast_symlink(inode))))
4065 /* Validate extent which is part of inode */
4066 ret = ext4_ext_check_inode(inode);
4067 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4068 (S_ISLNK(inode->i_mode) &&
4069 !ext4_inode_is_fast_symlink(inode))) {
4070 /* Validate block references which are part of inode */
4071 ret = ext4_ind_check_inode(inode);
4077 if (S_ISREG(inode->i_mode)) {
4078 inode->i_op = &ext4_file_inode_operations;
4079 inode->i_fop = &ext4_file_operations;
4080 ext4_set_aops(inode);
4081 } else if (S_ISDIR(inode->i_mode)) {
4082 inode->i_op = &ext4_dir_inode_operations;
4083 inode->i_fop = &ext4_dir_operations;
4084 } else if (S_ISLNK(inode->i_mode)) {
4085 if (ext4_inode_is_fast_symlink(inode)) {
4086 inode->i_op = &ext4_fast_symlink_inode_operations;
4087 nd_terminate_link(ei->i_data, inode->i_size,
4088 sizeof(ei->i_data) - 1);
4090 inode->i_op = &ext4_symlink_inode_operations;
4091 ext4_set_aops(inode);
4093 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4094 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4095 inode->i_op = &ext4_special_inode_operations;
4096 if (raw_inode->i_block[0])
4097 init_special_inode(inode, inode->i_mode,
4098 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4100 init_special_inode(inode, inode->i_mode,
4101 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4102 } else if (ino == EXT4_BOOT_LOADER_INO) {
4103 make_bad_inode(inode);
4106 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4110 ext4_set_inode_flags(inode);
4111 unlock_new_inode(inode);
4117 return ERR_PTR(ret);
4120 static int ext4_inode_blocks_set(handle_t *handle,
4121 struct ext4_inode *raw_inode,
4122 struct ext4_inode_info *ei)
4124 struct inode *inode = &(ei->vfs_inode);
4125 u64 i_blocks = inode->i_blocks;
4126 struct super_block *sb = inode->i_sb;
4128 if (i_blocks <= ~0U) {
4130 * i_blocks can be represented in a 32 bit variable
4131 * as multiple of 512 bytes
4133 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4134 raw_inode->i_blocks_high = 0;
4135 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4138 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4141 if (i_blocks <= 0xffffffffffffULL) {
4143 * i_blocks can be represented in a 48 bit variable
4144 * as multiple of 512 bytes
4146 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4147 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4148 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4150 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4151 /* i_block is stored in file system block size */
4152 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4153 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4154 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4160 * Post the struct inode info into an on-disk inode location in the
4161 * buffer-cache. This gobbles the caller's reference to the
4162 * buffer_head in the inode location struct.
4164 * The caller must have write access to iloc->bh.
4166 static int ext4_do_update_inode(handle_t *handle,
4167 struct inode *inode,
4168 struct ext4_iloc *iloc)
4170 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4171 struct ext4_inode_info *ei = EXT4_I(inode);
4172 struct buffer_head *bh = iloc->bh;
4173 struct super_block *sb = inode->i_sb;
4174 int err = 0, rc, block;
4175 int need_datasync = 0, set_large_file = 0;
4179 spin_lock(&ei->i_raw_lock);
4181 /* For fields not tracked in the in-memory inode,
4182 * initialise them to zero for new inodes. */
4183 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4184 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4186 ext4_get_inode_flags(ei);
4187 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4188 i_uid = i_uid_read(inode);
4189 i_gid = i_gid_read(inode);
4190 if (!(test_opt(inode->i_sb, NO_UID32))) {
4191 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4192 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4194 * Fix up interoperability with old kernels. Otherwise, old inodes get
4195 * re-used with the upper 16 bits of the uid/gid intact
4198 raw_inode->i_uid_high =
4199 cpu_to_le16(high_16_bits(i_uid));
4200 raw_inode->i_gid_high =
4201 cpu_to_le16(high_16_bits(i_gid));
4203 raw_inode->i_uid_high = 0;
4204 raw_inode->i_gid_high = 0;
4207 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4208 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4209 raw_inode->i_uid_high = 0;
4210 raw_inode->i_gid_high = 0;
4212 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4214 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4215 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4216 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4217 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4219 if (ext4_inode_blocks_set(handle, raw_inode, ei)) {
4220 spin_unlock(&ei->i_raw_lock);
4223 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4224 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4225 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4226 raw_inode->i_file_acl_high =
4227 cpu_to_le16(ei->i_file_acl >> 32);
4228 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4229 if (ei->i_disksize != ext4_isize(raw_inode)) {
4230 ext4_isize_set(raw_inode, ei->i_disksize);
4233 if (ei->i_disksize > 0x7fffffffULL) {
4234 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4235 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4236 EXT4_SB(sb)->s_es->s_rev_level ==
4237 cpu_to_le32(EXT4_GOOD_OLD_REV))
4240 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4241 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4242 if (old_valid_dev(inode->i_rdev)) {
4243 raw_inode->i_block[0] =
4244 cpu_to_le32(old_encode_dev(inode->i_rdev));
4245 raw_inode->i_block[1] = 0;
4247 raw_inode->i_block[0] = 0;
4248 raw_inode->i_block[1] =
4249 cpu_to_le32(new_encode_dev(inode->i_rdev));
4250 raw_inode->i_block[2] = 0;
4252 } else if (!ext4_has_inline_data(inode)) {
4253 for (block = 0; block < EXT4_N_BLOCKS; block++)
4254 raw_inode->i_block[block] = ei->i_data[block];
4257 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4258 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4259 if (ei->i_extra_isize) {
4260 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4261 raw_inode->i_version_hi =
4262 cpu_to_le32(inode->i_version >> 32);
4263 raw_inode->i_extra_isize =
4264 cpu_to_le16(ei->i_extra_isize);
4268 ext4_inode_csum_set(inode, raw_inode, ei);
4270 spin_unlock(&ei->i_raw_lock);
4272 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4273 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4276 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4277 if (set_large_file) {
4278 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4279 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4282 ext4_update_dynamic_rev(sb);
4283 EXT4_SET_RO_COMPAT_FEATURE(sb,
4284 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4285 ext4_handle_sync(handle);
4286 err = ext4_handle_dirty_super(handle, sb);
4288 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4291 ext4_std_error(inode->i_sb, err);
4296 * ext4_write_inode()
4298 * We are called from a few places:
4300 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4301 * Here, there will be no transaction running. We wait for any running
4302 * transaction to commit.
4304 * - Within flush work (sys_sync(), kupdate and such).
4305 * We wait on commit, if told to.
4307 * - Within iput_final() -> write_inode_now()
4308 * We wait on commit, if told to.
4310 * In all cases it is actually safe for us to return without doing anything,
4311 * because the inode has been copied into a raw inode buffer in
4312 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4315 * Note that we are absolutely dependent upon all inode dirtiers doing the
4316 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4317 * which we are interested.
4319 * It would be a bug for them to not do this. The code:
4321 * mark_inode_dirty(inode)
4323 * inode->i_size = expr;
4325 * is in error because write_inode() could occur while `stuff()' is running,
4326 * and the new i_size will be lost. Plus the inode will no longer be on the
4327 * superblock's dirty inode list.
4329 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4333 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4336 if (EXT4_SB(inode->i_sb)->s_journal) {
4337 if (ext4_journal_current_handle()) {
4338 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4344 * No need to force transaction in WB_SYNC_NONE mode. Also
4345 * ext4_sync_fs() will force the commit after everything is
4348 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4351 err = ext4_force_commit(inode->i_sb);
4353 struct ext4_iloc iloc;
4355 err = __ext4_get_inode_loc(inode, &iloc, 0);
4359 * sync(2) will flush the whole buffer cache. No need to do
4360 * it here separately for each inode.
4362 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4363 sync_dirty_buffer(iloc.bh);
4364 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4365 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4366 "IO error syncing inode");
4375 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4376 * buffers that are attached to a page stradding i_size and are undergoing
4377 * commit. In that case we have to wait for commit to finish and try again.
4379 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4383 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4384 tid_t commit_tid = 0;
4387 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4389 * All buffers in the last page remain valid? Then there's nothing to
4390 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4393 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4396 page = find_lock_page(inode->i_mapping,
4397 inode->i_size >> PAGE_CACHE_SHIFT);
4400 ret = __ext4_journalled_invalidatepage(page, offset,
4401 PAGE_CACHE_SIZE - offset);
4403 page_cache_release(page);
4407 read_lock(&journal->j_state_lock);
4408 if (journal->j_committing_transaction)
4409 commit_tid = journal->j_committing_transaction->t_tid;
4410 read_unlock(&journal->j_state_lock);
4412 jbd2_log_wait_commit(journal, commit_tid);
4419 * Called from notify_change.
4421 * We want to trap VFS attempts to truncate the file as soon as
4422 * possible. In particular, we want to make sure that when the VFS
4423 * shrinks i_size, we put the inode on the orphan list and modify
4424 * i_disksize immediately, so that during the subsequent flushing of
4425 * dirty pages and freeing of disk blocks, we can guarantee that any
4426 * commit will leave the blocks being flushed in an unused state on
4427 * disk. (On recovery, the inode will get truncated and the blocks will
4428 * be freed, so we have a strong guarantee that no future commit will
4429 * leave these blocks visible to the user.)
4431 * Another thing we have to assure is that if we are in ordered mode
4432 * and inode is still attached to the committing transaction, we must
4433 * we start writeout of all the dirty pages which are being truncated.
4434 * This way we are sure that all the data written in the previous
4435 * transaction are already on disk (truncate waits for pages under
4438 * Called with inode->i_mutex down.
4440 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4442 struct inode *inode = dentry->d_inode;
4445 const unsigned int ia_valid = attr->ia_valid;
4447 error = inode_change_ok(inode, attr);
4451 if (is_quota_modification(inode, attr))
4452 dquot_initialize(inode);
4453 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4454 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4457 /* (user+group)*(old+new) structure, inode write (sb,
4458 * inode block, ? - but truncate inode update has it) */
4459 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4460 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4461 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4462 if (IS_ERR(handle)) {
4463 error = PTR_ERR(handle);
4466 error = dquot_transfer(inode, attr);
4468 ext4_journal_stop(handle);
4471 /* Update corresponding info in inode so that everything is in
4472 * one transaction */
4473 if (attr->ia_valid & ATTR_UID)
4474 inode->i_uid = attr->ia_uid;
4475 if (attr->ia_valid & ATTR_GID)
4476 inode->i_gid = attr->ia_gid;
4477 error = ext4_mark_inode_dirty(handle, inode);
4478 ext4_journal_stop(handle);
4481 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4484 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4485 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4487 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4491 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4492 inode_inc_iversion(inode);
4494 if (S_ISREG(inode->i_mode) &&
4495 (attr->ia_size < inode->i_size)) {
4496 if (ext4_should_order_data(inode)) {
4497 error = ext4_begin_ordered_truncate(inode,
4502 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4503 if (IS_ERR(handle)) {
4504 error = PTR_ERR(handle);
4507 if (ext4_handle_valid(handle)) {
4508 error = ext4_orphan_add(handle, inode);
4511 down_write(&EXT4_I(inode)->i_data_sem);
4512 EXT4_I(inode)->i_disksize = attr->ia_size;
4513 rc = ext4_mark_inode_dirty(handle, inode);
4517 * We have to update i_size under i_data_sem together
4518 * with i_disksize to avoid races with writeback code
4519 * running ext4_wb_update_i_disksize().
4522 i_size_write(inode, attr->ia_size);
4523 up_write(&EXT4_I(inode)->i_data_sem);
4524 ext4_journal_stop(handle);
4526 ext4_orphan_del(NULL, inode);
4530 i_size_write(inode, attr->ia_size);
4533 * Blocks are going to be removed from the inode. Wait
4534 * for dio in flight. Temporarily disable
4535 * dioread_nolock to prevent livelock.
4538 if (!ext4_should_journal_data(inode)) {
4539 ext4_inode_block_unlocked_dio(inode);
4540 inode_dio_wait(inode);
4541 ext4_inode_resume_unlocked_dio(inode);
4543 ext4_wait_for_tail_page_commit(inode);
4546 * Truncate pagecache after we've waited for commit
4547 * in data=journal mode to make pages freeable.
4549 truncate_pagecache(inode, inode->i_size);
4552 * We want to call ext4_truncate() even if attr->ia_size ==
4553 * inode->i_size for cases like truncation of fallocated space
4555 if (attr->ia_valid & ATTR_SIZE)
4556 ext4_truncate(inode);
4559 setattr_copy(inode, attr);
4560 mark_inode_dirty(inode);
4564 * If the call to ext4_truncate failed to get a transaction handle at
4565 * all, we need to clean up the in-core orphan list manually.
4567 if (orphan && inode->i_nlink)
4568 ext4_orphan_del(NULL, inode);
4570 if (!rc && (ia_valid & ATTR_MODE))
4571 rc = posix_acl_chmod(inode, inode->i_mode);
4574 ext4_std_error(inode->i_sb, error);
4580 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4583 struct inode *inode;
4584 unsigned long long delalloc_blocks;
4586 inode = dentry->d_inode;
4587 generic_fillattr(inode, stat);
4590 * If there is inline data in the inode, the inode will normally not
4591 * have data blocks allocated (it may have an external xattr block).
4592 * Report at least one sector for such files, so tools like tar, rsync,
4593 * others doen't incorrectly think the file is completely sparse.
4595 if (unlikely(ext4_has_inline_data(inode)))
4596 stat->blocks += (stat->size + 511) >> 9;
4599 * We can't update i_blocks if the block allocation is delayed
4600 * otherwise in the case of system crash before the real block
4601 * allocation is done, we will have i_blocks inconsistent with
4602 * on-disk file blocks.
4603 * We always keep i_blocks updated together with real
4604 * allocation. But to not confuse with user, stat
4605 * will return the blocks that include the delayed allocation
4606 * blocks for this file.
4608 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4609 EXT4_I(inode)->i_reserved_data_blocks);
4610 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4614 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4617 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4618 return ext4_ind_trans_blocks(inode, lblocks);
4619 return ext4_ext_index_trans_blocks(inode, pextents);
4623 * Account for index blocks, block groups bitmaps and block group
4624 * descriptor blocks if modify datablocks and index blocks
4625 * worse case, the indexs blocks spread over different block groups
4627 * If datablocks are discontiguous, they are possible to spread over
4628 * different block groups too. If they are contiguous, with flexbg,
4629 * they could still across block group boundary.
4631 * Also account for superblock, inode, quota and xattr blocks
4633 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4636 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4642 * How many index blocks need to touch to map @lblocks logical blocks
4643 * to @pextents physical extents?
4645 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4650 * Now let's see how many group bitmaps and group descriptors need
4653 groups = idxblocks + pextents;
4655 if (groups > ngroups)
4657 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4658 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4660 /* bitmaps and block group descriptor blocks */
4661 ret += groups + gdpblocks;
4663 /* Blocks for super block, inode, quota and xattr blocks */
4664 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4670 * Calculate the total number of credits to reserve to fit
4671 * the modification of a single pages into a single transaction,
4672 * which may include multiple chunks of block allocations.
4674 * This could be called via ext4_write_begin()
4676 * We need to consider the worse case, when
4677 * one new block per extent.
4679 int ext4_writepage_trans_blocks(struct inode *inode)
4681 int bpp = ext4_journal_blocks_per_page(inode);
4684 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4686 /* Account for data blocks for journalled mode */
4687 if (ext4_should_journal_data(inode))
4693 * Calculate the journal credits for a chunk of data modification.
4695 * This is called from DIO, fallocate or whoever calling
4696 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4698 * journal buffers for data blocks are not included here, as DIO
4699 * and fallocate do no need to journal data buffers.
4701 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4703 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4707 * The caller must have previously called ext4_reserve_inode_write().
4708 * Give this, we know that the caller already has write access to iloc->bh.
4710 int ext4_mark_iloc_dirty(handle_t *handle,
4711 struct inode *inode, struct ext4_iloc *iloc)
4715 if (IS_I_VERSION(inode))
4716 inode_inc_iversion(inode);
4718 /* the do_update_inode consumes one bh->b_count */
4721 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4722 err = ext4_do_update_inode(handle, inode, iloc);
4728 * On success, We end up with an outstanding reference count against
4729 * iloc->bh. This _must_ be cleaned up later.
4733 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4734 struct ext4_iloc *iloc)
4738 err = ext4_get_inode_loc(inode, iloc);
4740 BUFFER_TRACE(iloc->bh, "get_write_access");
4741 err = ext4_journal_get_write_access(handle, iloc->bh);
4747 ext4_std_error(inode->i_sb, err);
4752 * Expand an inode by new_extra_isize bytes.
4753 * Returns 0 on success or negative error number on failure.
4755 static int ext4_expand_extra_isize(struct inode *inode,
4756 unsigned int new_extra_isize,
4757 struct ext4_iloc iloc,
4760 struct ext4_inode *raw_inode;
4761 struct ext4_xattr_ibody_header *header;
4763 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4766 raw_inode = ext4_raw_inode(&iloc);
4768 header = IHDR(inode, raw_inode);
4770 /* No extended attributes present */
4771 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4772 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4773 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4775 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4779 /* try to expand with EAs present */
4780 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4785 * What we do here is to mark the in-core inode as clean with respect to inode
4786 * dirtiness (it may still be data-dirty).
4787 * This means that the in-core inode may be reaped by prune_icache
4788 * without having to perform any I/O. This is a very good thing,
4789 * because *any* task may call prune_icache - even ones which
4790 * have a transaction open against a different journal.
4792 * Is this cheating? Not really. Sure, we haven't written the
4793 * inode out, but prune_icache isn't a user-visible syncing function.
4794 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4795 * we start and wait on commits.
4797 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4799 struct ext4_iloc iloc;
4800 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4801 static unsigned int mnt_count;
4805 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4806 err = ext4_reserve_inode_write(handle, inode, &iloc);
4807 if (ext4_handle_valid(handle) &&
4808 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4809 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4811 * We need extra buffer credits since we may write into EA block
4812 * with this same handle. If journal_extend fails, then it will
4813 * only result in a minor loss of functionality for that inode.
4814 * If this is felt to be critical, then e2fsck should be run to
4815 * force a large enough s_min_extra_isize.
4817 if ((jbd2_journal_extend(handle,
4818 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4819 ret = ext4_expand_extra_isize(inode,
4820 sbi->s_want_extra_isize,
4823 ext4_set_inode_state(inode,
4824 EXT4_STATE_NO_EXPAND);
4826 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4827 ext4_warning(inode->i_sb,
4828 "Unable to expand inode %lu. Delete"
4829 " some EAs or run e2fsck.",
4832 le16_to_cpu(sbi->s_es->s_mnt_count);
4838 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4843 * ext4_dirty_inode() is called from __mark_inode_dirty()
4845 * We're really interested in the case where a file is being extended.
4846 * i_size has been changed by generic_commit_write() and we thus need
4847 * to include the updated inode in the current transaction.
4849 * Also, dquot_alloc_block() will always dirty the inode when blocks
4850 * are allocated to the file.
4852 * If the inode is marked synchronous, we don't honour that here - doing
4853 * so would cause a commit on atime updates, which we don't bother doing.
4854 * We handle synchronous inodes at the highest possible level.
4856 void ext4_dirty_inode(struct inode *inode, int flags)
4860 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4864 ext4_mark_inode_dirty(handle, inode);
4866 ext4_journal_stop(handle);
4873 * Bind an inode's backing buffer_head into this transaction, to prevent
4874 * it from being flushed to disk early. Unlike
4875 * ext4_reserve_inode_write, this leaves behind no bh reference and
4876 * returns no iloc structure, so the caller needs to repeat the iloc
4877 * lookup to mark the inode dirty later.
4879 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4881 struct ext4_iloc iloc;
4885 err = ext4_get_inode_loc(inode, &iloc);
4887 BUFFER_TRACE(iloc.bh, "get_write_access");
4888 err = jbd2_journal_get_write_access(handle, iloc.bh);
4890 err = ext4_handle_dirty_metadata(handle,
4896 ext4_std_error(inode->i_sb, err);
4901 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4908 * We have to be very careful here: changing a data block's
4909 * journaling status dynamically is dangerous. If we write a
4910 * data block to the journal, change the status and then delete
4911 * that block, we risk forgetting to revoke the old log record
4912 * from the journal and so a subsequent replay can corrupt data.
4913 * So, first we make sure that the journal is empty and that
4914 * nobody is changing anything.
4917 journal = EXT4_JOURNAL(inode);
4920 if (is_journal_aborted(journal))
4922 /* We have to allocate physical blocks for delalloc blocks
4923 * before flushing journal. otherwise delalloc blocks can not
4924 * be allocated any more. even more truncate on delalloc blocks
4925 * could trigger BUG by flushing delalloc blocks in journal.
4926 * There is no delalloc block in non-journal data mode.
4928 if (val && test_opt(inode->i_sb, DELALLOC)) {
4929 err = ext4_alloc_da_blocks(inode);
4934 /* Wait for all existing dio workers */
4935 ext4_inode_block_unlocked_dio(inode);
4936 inode_dio_wait(inode);
4938 jbd2_journal_lock_updates(journal);
4941 * OK, there are no updates running now, and all cached data is
4942 * synced to disk. We are now in a completely consistent state
4943 * which doesn't have anything in the journal, and we know that
4944 * no filesystem updates are running, so it is safe to modify
4945 * the inode's in-core data-journaling state flag now.
4949 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4951 jbd2_journal_flush(journal);
4952 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4954 ext4_set_aops(inode);
4956 jbd2_journal_unlock_updates(journal);
4957 ext4_inode_resume_unlocked_dio(inode);
4959 /* Finally we can mark the inode as dirty. */
4961 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
4963 return PTR_ERR(handle);
4965 err = ext4_mark_inode_dirty(handle, inode);
4966 ext4_handle_sync(handle);
4967 ext4_journal_stop(handle);
4968 ext4_std_error(inode->i_sb, err);
4973 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4975 return !buffer_mapped(bh);
4978 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4980 struct page *page = vmf->page;
4984 struct file *file = vma->vm_file;
4985 struct inode *inode = file_inode(file);
4986 struct address_space *mapping = inode->i_mapping;
4988 get_block_t *get_block;
4991 sb_start_pagefault(inode->i_sb);
4992 file_update_time(vma->vm_file);
4993 /* Delalloc case is easy... */
4994 if (test_opt(inode->i_sb, DELALLOC) &&
4995 !ext4_should_journal_data(inode) &&
4996 !ext4_nonda_switch(inode->i_sb)) {
4998 ret = __block_page_mkwrite(vma, vmf,
4999 ext4_da_get_block_prep);
5000 } while (ret == -ENOSPC &&
5001 ext4_should_retry_alloc(inode->i_sb, &retries));
5006 size = i_size_read(inode);
5007 /* Page got truncated from under us? */
5008 if (page->mapping != mapping || page_offset(page) > size) {
5010 ret = VM_FAULT_NOPAGE;
5014 if (page->index == size >> PAGE_CACHE_SHIFT)
5015 len = size & ~PAGE_CACHE_MASK;
5017 len = PAGE_CACHE_SIZE;
5019 * Return if we have all the buffers mapped. This avoids the need to do
5020 * journal_start/journal_stop which can block and take a long time
5022 if (page_has_buffers(page)) {
5023 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5025 ext4_bh_unmapped)) {
5026 /* Wait so that we don't change page under IO */
5027 wait_for_stable_page(page);
5028 ret = VM_FAULT_LOCKED;
5033 /* OK, we need to fill the hole... */
5034 if (ext4_should_dioread_nolock(inode))
5035 get_block = ext4_get_block_write;
5037 get_block = ext4_get_block;
5039 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5040 ext4_writepage_trans_blocks(inode));
5041 if (IS_ERR(handle)) {
5042 ret = VM_FAULT_SIGBUS;
5045 ret = __block_page_mkwrite(vma, vmf, get_block);
5046 if (!ret && ext4_should_journal_data(inode)) {
5047 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5048 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5050 ret = VM_FAULT_SIGBUS;
5051 ext4_journal_stop(handle);
5054 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5056 ext4_journal_stop(handle);
5057 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5060 ret = block_page_mkwrite_return(ret);
5062 sb_end_pagefault(inode->i_sb);