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 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
155 * Restart the transaction associated with *handle. This does a commit,
156 * so before we call here everything must be consistently dirtied against
159 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
165 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
166 * moment, get_block can be called only for blocks inside i_size since
167 * page cache has been already dropped and writes are blocked by
168 * i_mutex. So we can safely drop the i_data_sem here.
170 BUG_ON(EXT4_JOURNAL(inode) == NULL);
171 jbd_debug(2, "restarting handle %p\n", handle);
172 up_write(&EXT4_I(inode)->i_data_sem);
173 ret = ext4_journal_restart(handle, nblocks);
174 down_write(&EXT4_I(inode)->i_data_sem);
175 ext4_discard_preallocations(inode);
181 * Called at the last iput() if i_nlink is zero.
183 void ext4_evict_inode(struct inode *inode)
188 trace_ext4_evict_inode(inode);
190 if (inode->i_nlink) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode) &&
210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
211 inode->i_ino != EXT4_JOURNAL_INO) {
212 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
213 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
215 jbd2_complete_transaction(journal, commit_tid);
216 filemap_write_and_wait(&inode->i_data);
218 truncate_inode_pages(&inode->i_data, 0);
220 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
224 if (!is_bad_inode(inode))
225 dquot_initialize(inode);
227 if (ext4_should_order_data(inode))
228 ext4_begin_ordered_truncate(inode, 0);
229 truncate_inode_pages(&inode->i_data, 0);
231 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
232 if (is_bad_inode(inode))
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it
239 sb_start_intwrite(inode->i_sb);
240 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
241 ext4_blocks_for_truncate(inode)+3);
242 if (IS_ERR(handle)) {
243 ext4_std_error(inode->i_sb, PTR_ERR(handle));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL, inode);
250 sb_end_intwrite(inode->i_sb);
255 ext4_handle_sync(handle);
257 err = ext4_mark_inode_dirty(handle, inode);
259 ext4_warning(inode->i_sb,
260 "couldn't mark inode dirty (err %d)", err);
264 ext4_truncate(inode);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle, 3)) {
273 err = ext4_journal_extend(handle, 3);
275 err = ext4_journal_restart(handle, 3);
277 ext4_warning(inode->i_sb,
278 "couldn't extend journal (err %d)", err);
280 ext4_journal_stop(handle);
281 ext4_orphan_del(NULL, inode);
282 sb_end_intwrite(inode->i_sb);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle, inode);
296 EXT4_I(inode)->i_dtime = get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle, inode))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode);
309 ext4_free_inode(handle, inode);
310 ext4_journal_stop(handle);
311 sb_end_intwrite(inode->i_sb);
314 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
318 qsize_t *ext4_get_reserved_space(struct inode *inode)
320 return &EXT4_I(inode)->i_reserved_quota;
325 * Calculate the number of metadata blocks need to reserve
326 * to allocate a block located at @lblock
328 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
330 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
331 return ext4_ext_calc_metadata_amount(inode, lblock);
333 return ext4_ind_calc_metadata_amount(inode, lblock);
337 * Called with i_data_sem down, which is important since we can call
338 * ext4_discard_preallocations() from here.
340 void ext4_da_update_reserve_space(struct inode *inode,
341 int used, int quota_claim)
343 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
344 struct ext4_inode_info *ei = EXT4_I(inode);
346 spin_lock(&ei->i_block_reservation_lock);
347 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
348 if (unlikely(used > ei->i_reserved_data_blocks)) {
349 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
350 "with only %d reserved data blocks",
351 __func__, inode->i_ino, used,
352 ei->i_reserved_data_blocks);
354 used = ei->i_reserved_data_blocks;
357 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
358 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
359 "with only %d reserved metadata blocks "
360 "(releasing %d blocks with reserved %d data blocks)",
361 inode->i_ino, ei->i_allocated_meta_blocks,
362 ei->i_reserved_meta_blocks, used,
363 ei->i_reserved_data_blocks);
365 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
368 /* Update per-inode reservations */
369 ei->i_reserved_data_blocks -= used;
370 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
371 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
372 used + ei->i_allocated_meta_blocks);
373 ei->i_allocated_meta_blocks = 0;
375 if (ei->i_reserved_data_blocks == 0) {
377 * We can release all of the reserved metadata blocks
378 * only when we have written all of the delayed
381 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
382 ei->i_reserved_meta_blocks);
383 ei->i_reserved_meta_blocks = 0;
384 ei->i_da_metadata_calc_len = 0;
386 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
388 /* Update quota subsystem for data blocks */
390 dquot_claim_block(inode, EXT4_C2B(sbi, used));
393 * We did fallocate with an offset that is already delayed
394 * allocated. So on delayed allocated writeback we should
395 * not re-claim the quota for fallocated blocks.
397 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
401 * If we have done all the pending block allocations and if
402 * there aren't any writers on the inode, we can discard the
403 * inode's preallocations.
405 if ((ei->i_reserved_data_blocks == 0) &&
406 (atomic_read(&inode->i_writecount) == 0))
407 ext4_discard_preallocations(inode);
410 static int __check_block_validity(struct inode *inode, const char *func,
412 struct ext4_map_blocks *map)
414 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
416 ext4_error_inode(inode, func, line, map->m_pblk,
417 "lblock %lu mapped to illegal pblock "
418 "(length %d)", (unsigned long) map->m_lblk,
425 #define check_block_validity(inode, map) \
426 __check_block_validity((inode), __func__, __LINE__, (map))
428 #ifdef ES_AGGRESSIVE_TEST
429 static void ext4_map_blocks_es_recheck(handle_t *handle,
431 struct ext4_map_blocks *es_map,
432 struct ext4_map_blocks *map,
439 * There is a race window that the result is not the same.
440 * e.g. xfstests #223 when dioread_nolock enables. The reason
441 * is that we lookup a block mapping in extent status tree with
442 * out taking i_data_sem. So at the time the unwritten extent
443 * could be converted.
445 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
446 down_read((&EXT4_I(inode)->i_data_sem));
447 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
448 retval = ext4_ext_map_blocks(handle, inode, map, flags &
449 EXT4_GET_BLOCKS_KEEP_SIZE);
451 retval = ext4_ind_map_blocks(handle, inode, map, flags &
452 EXT4_GET_BLOCKS_KEEP_SIZE);
454 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
455 up_read((&EXT4_I(inode)->i_data_sem));
457 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
458 * because it shouldn't be marked in es_map->m_flags.
460 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
463 * We don't check m_len because extent will be collpased in status
464 * tree. So the m_len might not equal.
466 if (es_map->m_lblk != map->m_lblk ||
467 es_map->m_flags != map->m_flags ||
468 es_map->m_pblk != map->m_pblk) {
469 printk("ES cache assertion failed for inode: %lu "
470 "es_cached ex [%d/%d/%llu/%x] != "
471 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472 inode->i_ino, es_map->m_lblk, es_map->m_len,
473 es_map->m_pblk, es_map->m_flags, map->m_lblk,
474 map->m_len, map->m_pblk, map->m_flags,
478 #endif /* ES_AGGRESSIVE_TEST */
481 * The ext4_map_blocks() function tries to look up the requested blocks,
482 * and returns if the blocks are already mapped.
484 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485 * and store the allocated blocks in the result buffer head and mark it
488 * If file type is extents based, it will call ext4_ext_map_blocks(),
489 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
492 * On success, it returns the number of blocks being mapped or allocate.
493 * if create==0 and the blocks are pre-allocated and uninitialized block,
494 * the result buffer head is unmapped. If the create ==1, it will make sure
495 * the buffer head is mapped.
497 * It returns 0 if plain look up failed (blocks have not been allocated), in
498 * that case, buffer head is unmapped
500 * It returns the error in case of allocation failure.
502 int ext4_map_blocks(handle_t *handle, struct inode *inode,
503 struct ext4_map_blocks *map, int flags)
505 struct extent_status es;
507 #ifdef ES_AGGRESSIVE_TEST
508 struct ext4_map_blocks orig_map;
510 memcpy(&orig_map, map, sizeof(*map));
514 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
515 "logical block %lu\n", inode->i_ino, flags, map->m_len,
516 (unsigned long) map->m_lblk);
518 /* We can handle the block number less than EXT_MAX_BLOCKS */
519 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
522 /* Lookup extent status tree firstly */
523 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
524 ext4_es_lru_add(inode);
525 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
526 map->m_pblk = ext4_es_pblock(&es) +
527 map->m_lblk - es.es_lblk;
528 map->m_flags |= ext4_es_is_written(&es) ?
529 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
530 retval = es.es_len - (map->m_lblk - es.es_lblk);
531 if (retval > map->m_len)
534 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
539 #ifdef ES_AGGRESSIVE_TEST
540 ext4_map_blocks_es_recheck(handle, inode, map,
547 * Try to see if we can get the block without requesting a new
550 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
551 down_read((&EXT4_I(inode)->i_data_sem));
552 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
553 retval = ext4_ext_map_blocks(handle, inode, map, flags &
554 EXT4_GET_BLOCKS_KEEP_SIZE);
556 retval = ext4_ind_map_blocks(handle, inode, map, flags &
557 EXT4_GET_BLOCKS_KEEP_SIZE);
563 if (unlikely(retval != map->m_len)) {
564 ext4_warning(inode->i_sb,
565 "ES len assertion failed for inode "
566 "%lu: retval %d != map->m_len %d",
567 inode->i_ino, retval, map->m_len);
571 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
572 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
573 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
574 ext4_find_delalloc_range(inode, map->m_lblk,
575 map->m_lblk + map->m_len - 1))
576 status |= EXTENT_STATUS_DELAYED;
577 ret = ext4_es_insert_extent(inode, map->m_lblk,
578 map->m_len, map->m_pblk, status);
582 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
583 up_read((&EXT4_I(inode)->i_data_sem));
586 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
587 int ret = check_block_validity(inode, map);
592 /* If it is only a block(s) look up */
593 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
597 * Returns if the blocks have already allocated
599 * Note that if blocks have been preallocated
600 * ext4_ext_get_block() returns the create = 0
601 * with buffer head unmapped.
603 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
607 * Here we clear m_flags because after allocating an new extent,
608 * it will be set again.
610 map->m_flags &= ~EXT4_MAP_FLAGS;
613 * New blocks allocate and/or writing to uninitialized extent
614 * will possibly result in updating i_data, so we take
615 * the write lock of i_data_sem, and call get_blocks()
616 * with create == 1 flag.
618 down_write((&EXT4_I(inode)->i_data_sem));
621 * if the caller is from delayed allocation writeout path
622 * we have already reserved fs blocks for allocation
623 * let the underlying get_block() function know to
624 * avoid double accounting
626 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
627 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
629 * We need to check for EXT4 here because migrate
630 * could have changed the inode type in between
632 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
633 retval = ext4_ext_map_blocks(handle, inode, map, flags);
635 retval = ext4_ind_map_blocks(handle, inode, map, flags);
637 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
639 * We allocated new blocks which will result in
640 * i_data's format changing. Force the migrate
641 * to fail by clearing migrate flags
643 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
647 * Update reserved blocks/metadata blocks after successful
648 * block allocation which had been deferred till now. We don't
649 * support fallocate for non extent files. So we can update
650 * reserve space here.
653 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
654 ext4_da_update_reserve_space(inode, retval, 1);
656 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
657 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
663 if (unlikely(retval != map->m_len)) {
664 ext4_warning(inode->i_sb,
665 "ES len assertion failed for inode "
666 "%lu: retval %d != map->m_len %d",
667 inode->i_ino, retval, map->m_len);
672 * If the extent has been zeroed out, we don't need to update
673 * extent status tree.
675 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
676 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
677 if (ext4_es_is_written(&es))
680 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
681 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
682 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
683 ext4_find_delalloc_range(inode, map->m_lblk,
684 map->m_lblk + map->m_len - 1))
685 status |= EXTENT_STATUS_DELAYED;
686 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
687 map->m_pblk, status);
693 up_write((&EXT4_I(inode)->i_data_sem));
694 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
695 int ret = check_block_validity(inode, map);
702 /* Maximum number of blocks we map for direct IO at once. */
703 #define DIO_MAX_BLOCKS 4096
705 static int _ext4_get_block(struct inode *inode, sector_t iblock,
706 struct buffer_head *bh, int flags)
708 handle_t *handle = ext4_journal_current_handle();
709 struct ext4_map_blocks map;
710 int ret = 0, started = 0;
713 if (ext4_has_inline_data(inode))
717 map.m_len = bh->b_size >> inode->i_blkbits;
719 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
720 /* Direct IO write... */
721 if (map.m_len > DIO_MAX_BLOCKS)
722 map.m_len = DIO_MAX_BLOCKS;
723 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
724 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
726 if (IS_ERR(handle)) {
727 ret = PTR_ERR(handle);
733 ret = ext4_map_blocks(handle, inode, &map, flags);
735 ext4_io_end_t *io_end = ext4_inode_aio(inode);
737 map_bh(bh, inode->i_sb, map.m_pblk);
738 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
739 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
740 set_buffer_defer_completion(bh);
741 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
745 ext4_journal_stop(handle);
749 int ext4_get_block(struct inode *inode, sector_t iblock,
750 struct buffer_head *bh, int create)
752 return _ext4_get_block(inode, iblock, bh,
753 create ? EXT4_GET_BLOCKS_CREATE : 0);
757 * `handle' can be NULL if create is zero
759 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
760 ext4_lblk_t block, int create, int *errp)
762 struct ext4_map_blocks map;
763 struct buffer_head *bh;
766 J_ASSERT(handle != NULL || create == 0);
770 err = ext4_map_blocks(handle, inode, &map,
771 create ? EXT4_GET_BLOCKS_CREATE : 0);
773 /* ensure we send some value back into *errp */
776 if (create && err == 0)
777 err = -ENOSPC; /* should never happen */
783 bh = sb_getblk(inode->i_sb, map.m_pblk);
788 if (map.m_flags & EXT4_MAP_NEW) {
789 J_ASSERT(create != 0);
790 J_ASSERT(handle != NULL);
793 * Now that we do not always journal data, we should
794 * keep in mind whether this should always journal the
795 * new buffer as metadata. For now, regular file
796 * writes use ext4_get_block instead, so it's not a
800 BUFFER_TRACE(bh, "call get_create_access");
801 fatal = ext4_journal_get_create_access(handle, bh);
802 if (!fatal && !buffer_uptodate(bh)) {
803 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
804 set_buffer_uptodate(bh);
807 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
808 err = ext4_handle_dirty_metadata(handle, inode, bh);
812 BUFFER_TRACE(bh, "not a new buffer");
822 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
823 ext4_lblk_t block, int create, int *err)
825 struct buffer_head *bh;
827 bh = ext4_getblk(handle, inode, block, create, err);
830 if (buffer_uptodate(bh))
832 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
834 if (buffer_uptodate(bh))
841 int ext4_walk_page_buffers(handle_t *handle,
842 struct buffer_head *head,
846 int (*fn)(handle_t *handle,
847 struct buffer_head *bh))
849 struct buffer_head *bh;
850 unsigned block_start, block_end;
851 unsigned blocksize = head->b_size;
853 struct buffer_head *next;
855 for (bh = head, block_start = 0;
856 ret == 0 && (bh != head || !block_start);
857 block_start = block_end, bh = next) {
858 next = bh->b_this_page;
859 block_end = block_start + blocksize;
860 if (block_end <= from || block_start >= to) {
861 if (partial && !buffer_uptodate(bh))
865 err = (*fn)(handle, bh);
873 * To preserve ordering, it is essential that the hole instantiation and
874 * the data write be encapsulated in a single transaction. We cannot
875 * close off a transaction and start a new one between the ext4_get_block()
876 * and the commit_write(). So doing the jbd2_journal_start at the start of
877 * prepare_write() is the right place.
879 * Also, this function can nest inside ext4_writepage(). In that case, we
880 * *know* that ext4_writepage() has generated enough buffer credits to do the
881 * whole page. So we won't block on the journal in that case, which is good,
882 * because the caller may be PF_MEMALLOC.
884 * By accident, ext4 can be reentered when a transaction is open via
885 * quota file writes. If we were to commit the transaction while thus
886 * reentered, there can be a deadlock - we would be holding a quota
887 * lock, and the commit would never complete if another thread had a
888 * transaction open and was blocking on the quota lock - a ranking
891 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
892 * will _not_ run commit under these circumstances because handle->h_ref
893 * is elevated. We'll still have enough credits for the tiny quotafile
896 int do_journal_get_write_access(handle_t *handle,
897 struct buffer_head *bh)
899 int dirty = buffer_dirty(bh);
902 if (!buffer_mapped(bh) || buffer_freed(bh))
905 * __block_write_begin() could have dirtied some buffers. Clean
906 * the dirty bit as jbd2_journal_get_write_access() could complain
907 * otherwise about fs integrity issues. Setting of the dirty bit
908 * by __block_write_begin() isn't a real problem here as we clear
909 * the bit before releasing a page lock and thus writeback cannot
910 * ever write the buffer.
913 clear_buffer_dirty(bh);
914 ret = ext4_journal_get_write_access(handle, bh);
916 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
920 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
921 struct buffer_head *bh_result, int create);
922 static int ext4_write_begin(struct file *file, struct address_space *mapping,
923 loff_t pos, unsigned len, unsigned flags,
924 struct page **pagep, void **fsdata)
926 struct inode *inode = mapping->host;
927 int ret, needed_blocks;
934 trace_ext4_write_begin(inode, pos, len, flags);
936 * Reserve one block more for addition to orphan list in case
937 * we allocate blocks but write fails for some reason
939 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
940 index = pos >> PAGE_CACHE_SHIFT;
941 from = pos & (PAGE_CACHE_SIZE - 1);
944 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
945 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
954 * grab_cache_page_write_begin() can take a long time if the
955 * system is thrashing due to memory pressure, or if the page
956 * is being written back. So grab it first before we start
957 * the transaction handle. This also allows us to allocate
958 * the page (if needed) without using GFP_NOFS.
961 page = grab_cache_page_write_begin(mapping, index, flags);
967 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
968 if (IS_ERR(handle)) {
969 page_cache_release(page);
970 return PTR_ERR(handle);
974 if (page->mapping != mapping) {
975 /* The page got truncated from under us */
977 page_cache_release(page);
978 ext4_journal_stop(handle);
981 /* In case writeback began while the page was unlocked */
982 wait_for_stable_page(page);
984 if (ext4_should_dioread_nolock(inode))
985 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
987 ret = __block_write_begin(page, pos, len, ext4_get_block);
989 if (!ret && ext4_should_journal_data(inode)) {
990 ret = ext4_walk_page_buffers(handle, page_buffers(page),
992 do_journal_get_write_access);
998 * __block_write_begin may have instantiated a few blocks
999 * outside i_size. Trim these off again. Don't need
1000 * i_size_read because we hold i_mutex.
1002 * Add inode to orphan list in case we crash before
1005 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1006 ext4_orphan_add(handle, inode);
1008 ext4_journal_stop(handle);
1009 if (pos + len > inode->i_size) {
1010 ext4_truncate_failed_write(inode);
1012 * If truncate failed early the inode might
1013 * still be on the orphan list; we need to
1014 * make sure the inode is removed from the
1015 * orphan list in that case.
1018 ext4_orphan_del(NULL, inode);
1021 if (ret == -ENOSPC &&
1022 ext4_should_retry_alloc(inode->i_sb, &retries))
1024 page_cache_release(page);
1031 /* For write_end() in data=journal mode */
1032 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1035 if (!buffer_mapped(bh) || buffer_freed(bh))
1037 set_buffer_uptodate(bh);
1038 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1039 clear_buffer_meta(bh);
1040 clear_buffer_prio(bh);
1045 * We need to pick up the new inode size which generic_commit_write gave us
1046 * `file' can be NULL - eg, when called from page_symlink().
1048 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1049 * buffers are managed internally.
1051 static int ext4_write_end(struct file *file,
1052 struct address_space *mapping,
1053 loff_t pos, unsigned len, unsigned copied,
1054 struct page *page, void *fsdata)
1056 handle_t *handle = ext4_journal_current_handle();
1057 struct inode *inode = mapping->host;
1059 int i_size_changed = 0;
1061 trace_ext4_write_end(inode, pos, len, copied);
1062 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1063 ret = ext4_jbd2_file_inode(handle, inode);
1066 page_cache_release(page);
1071 if (ext4_has_inline_data(inode)) {
1072 ret = ext4_write_inline_data_end(inode, pos, len,
1078 copied = block_write_end(file, mapping, pos,
1079 len, copied, page, fsdata);
1082 * No need to use i_size_read() here, the i_size
1083 * cannot change under us because we hole i_mutex.
1085 * But it's important to update i_size while still holding page lock:
1086 * page writeout could otherwise come in and zero beyond i_size.
1088 if (pos + copied > inode->i_size) {
1089 i_size_write(inode, pos + copied);
1093 if (pos + copied > EXT4_I(inode)->i_disksize) {
1094 /* We need to mark inode dirty even if
1095 * new_i_size is less that inode->i_size
1096 * but greater than i_disksize. (hint delalloc)
1098 ext4_update_i_disksize(inode, (pos + copied));
1102 page_cache_release(page);
1105 * Don't mark the inode dirty under page lock. First, it unnecessarily
1106 * makes the holding time of page lock longer. Second, it forces lock
1107 * ordering of page lock and transaction start for journaling
1111 ext4_mark_inode_dirty(handle, inode);
1113 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1114 /* if we have allocated more blocks and copied
1115 * less. We will have blocks allocated outside
1116 * inode->i_size. So truncate them
1118 ext4_orphan_add(handle, inode);
1120 ret2 = ext4_journal_stop(handle);
1124 if (pos + len > inode->i_size) {
1125 ext4_truncate_failed_write(inode);
1127 * If truncate failed early the inode might still be
1128 * on the orphan list; we need to make sure the inode
1129 * is removed from the orphan list in that case.
1132 ext4_orphan_del(NULL, inode);
1135 return ret ? ret : copied;
1138 static int ext4_journalled_write_end(struct file *file,
1139 struct address_space *mapping,
1140 loff_t pos, unsigned len, unsigned copied,
1141 struct page *page, void *fsdata)
1143 handle_t *handle = ext4_journal_current_handle();
1144 struct inode *inode = mapping->host;
1150 trace_ext4_journalled_write_end(inode, pos, len, copied);
1151 from = pos & (PAGE_CACHE_SIZE - 1);
1154 BUG_ON(!ext4_handle_valid(handle));
1156 if (ext4_has_inline_data(inode))
1157 copied = ext4_write_inline_data_end(inode, pos, len,
1161 if (!PageUptodate(page))
1163 page_zero_new_buffers(page, from+copied, to);
1166 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1167 to, &partial, write_end_fn);
1169 SetPageUptodate(page);
1171 new_i_size = pos + copied;
1172 if (new_i_size > inode->i_size)
1173 i_size_write(inode, pos+copied);
1174 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1175 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1176 if (new_i_size > EXT4_I(inode)->i_disksize) {
1177 ext4_update_i_disksize(inode, new_i_size);
1178 ret2 = ext4_mark_inode_dirty(handle, inode);
1184 page_cache_release(page);
1185 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1186 /* if we have allocated more blocks and copied
1187 * less. We will have blocks allocated outside
1188 * inode->i_size. So truncate them
1190 ext4_orphan_add(handle, inode);
1192 ret2 = ext4_journal_stop(handle);
1195 if (pos + len > inode->i_size) {
1196 ext4_truncate_failed_write(inode);
1198 * If truncate failed early the inode might still be
1199 * on the orphan list; we need to make sure the inode
1200 * is removed from the orphan list in that case.
1203 ext4_orphan_del(NULL, inode);
1206 return ret ? ret : copied;
1210 * Reserve a metadata for a single block located at lblock
1212 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1214 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1215 struct ext4_inode_info *ei = EXT4_I(inode);
1216 unsigned int md_needed;
1217 ext4_lblk_t save_last_lblock;
1221 * recalculate the amount of metadata blocks to reserve
1222 * in order to allocate nrblocks
1223 * worse case is one extent per block
1225 spin_lock(&ei->i_block_reservation_lock);
1227 * ext4_calc_metadata_amount() has side effects, which we have
1228 * to be prepared undo if we fail to claim space.
1230 save_len = ei->i_da_metadata_calc_len;
1231 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1232 md_needed = EXT4_NUM_B2C(sbi,
1233 ext4_calc_metadata_amount(inode, lblock));
1234 trace_ext4_da_reserve_space(inode, md_needed);
1237 * We do still charge estimated metadata to the sb though;
1238 * we cannot afford to run out of free blocks.
1240 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1241 ei->i_da_metadata_calc_len = save_len;
1242 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1243 spin_unlock(&ei->i_block_reservation_lock);
1246 ei->i_reserved_meta_blocks += md_needed;
1247 spin_unlock(&ei->i_block_reservation_lock);
1249 return 0; /* success */
1253 * Reserve a single cluster located at lblock
1255 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1257 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1258 struct ext4_inode_info *ei = EXT4_I(inode);
1259 unsigned int md_needed;
1261 ext4_lblk_t save_last_lblock;
1265 * We will charge metadata quota at writeout time; this saves
1266 * us from metadata over-estimation, though we may go over by
1267 * a small amount in the end. Here we just reserve for data.
1269 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1274 * recalculate the amount of metadata blocks to reserve
1275 * in order to allocate nrblocks
1276 * worse case is one extent per block
1278 spin_lock(&ei->i_block_reservation_lock);
1280 * ext4_calc_metadata_amount() has side effects, which we have
1281 * to be prepared undo if we fail to claim space.
1283 save_len = ei->i_da_metadata_calc_len;
1284 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1285 md_needed = EXT4_NUM_B2C(sbi,
1286 ext4_calc_metadata_amount(inode, lblock));
1287 trace_ext4_da_reserve_space(inode, md_needed);
1290 * We do still charge estimated metadata to the sb though;
1291 * we cannot afford to run out of free blocks.
1293 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1294 ei->i_da_metadata_calc_len = save_len;
1295 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1296 spin_unlock(&ei->i_block_reservation_lock);
1297 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1300 ei->i_reserved_data_blocks++;
1301 ei->i_reserved_meta_blocks += md_needed;
1302 spin_unlock(&ei->i_block_reservation_lock);
1304 return 0; /* success */
1307 static void ext4_da_release_space(struct inode *inode, int to_free)
1309 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1310 struct ext4_inode_info *ei = EXT4_I(inode);
1313 return; /* Nothing to release, exit */
1315 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1317 trace_ext4_da_release_space(inode, to_free);
1318 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1320 * if there aren't enough reserved blocks, then the
1321 * counter is messed up somewhere. Since this
1322 * function is called from invalidate page, it's
1323 * harmless to return without any action.
1325 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1326 "ino %lu, to_free %d with only %d reserved "
1327 "data blocks", inode->i_ino, to_free,
1328 ei->i_reserved_data_blocks);
1330 to_free = ei->i_reserved_data_blocks;
1332 ei->i_reserved_data_blocks -= to_free;
1334 if (ei->i_reserved_data_blocks == 0) {
1336 * We can release all of the reserved metadata blocks
1337 * only when we have written all of the delayed
1338 * allocation blocks.
1339 * Note that in case of bigalloc, i_reserved_meta_blocks,
1340 * i_reserved_data_blocks, etc. refer to number of clusters.
1342 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1343 ei->i_reserved_meta_blocks);
1344 ei->i_reserved_meta_blocks = 0;
1345 ei->i_da_metadata_calc_len = 0;
1348 /* update fs dirty data blocks counter */
1349 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1351 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1353 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1356 static void ext4_da_page_release_reservation(struct page *page,
1357 unsigned int offset,
1358 unsigned int length)
1361 struct buffer_head *head, *bh;
1362 unsigned int curr_off = 0;
1363 struct inode *inode = page->mapping->host;
1364 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1365 unsigned int stop = offset + length;
1369 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1371 head = page_buffers(page);
1374 unsigned int next_off = curr_off + bh->b_size;
1376 if (next_off > stop)
1379 if ((offset <= curr_off) && (buffer_delay(bh))) {
1381 clear_buffer_delay(bh);
1383 curr_off = next_off;
1384 } while ((bh = bh->b_this_page) != head);
1387 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1388 ext4_es_remove_extent(inode, lblk, to_release);
1391 /* If we have released all the blocks belonging to a cluster, then we
1392 * need to release the reserved space for that cluster. */
1393 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1394 while (num_clusters > 0) {
1395 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1396 ((num_clusters - 1) << sbi->s_cluster_bits);
1397 if (sbi->s_cluster_ratio == 1 ||
1398 !ext4_find_delalloc_cluster(inode, lblk))
1399 ext4_da_release_space(inode, 1);
1406 * Delayed allocation stuff
1409 struct mpage_da_data {
1410 struct inode *inode;
1411 struct writeback_control *wbc;
1413 pgoff_t first_page; /* The first page to write */
1414 pgoff_t next_page; /* Current page to examine */
1415 pgoff_t last_page; /* Last page to examine */
1417 * Extent to map - this can be after first_page because that can be
1418 * fully mapped. We somewhat abuse m_flags to store whether the extent
1419 * is delalloc or unwritten.
1421 struct ext4_map_blocks map;
1422 struct ext4_io_submit io_submit; /* IO submission data */
1425 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1430 struct pagevec pvec;
1431 struct inode *inode = mpd->inode;
1432 struct address_space *mapping = inode->i_mapping;
1434 /* This is necessary when next_page == 0. */
1435 if (mpd->first_page >= mpd->next_page)
1438 index = mpd->first_page;
1439 end = mpd->next_page - 1;
1441 ext4_lblk_t start, last;
1442 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1443 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1444 ext4_es_remove_extent(inode, start, last - start + 1);
1447 pagevec_init(&pvec, 0);
1448 while (index <= end) {
1449 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1452 for (i = 0; i < nr_pages; i++) {
1453 struct page *page = pvec.pages[i];
1454 if (page->index > end)
1456 BUG_ON(!PageLocked(page));
1457 BUG_ON(PageWriteback(page));
1459 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1460 ClearPageUptodate(page);
1464 index = pvec.pages[nr_pages - 1]->index + 1;
1465 pagevec_release(&pvec);
1469 static void ext4_print_free_blocks(struct inode *inode)
1471 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1472 struct super_block *sb = inode->i_sb;
1473 struct ext4_inode_info *ei = EXT4_I(inode);
1475 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1476 EXT4_C2B(EXT4_SB(inode->i_sb),
1477 ext4_count_free_clusters(sb)));
1478 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1479 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1480 (long long) EXT4_C2B(EXT4_SB(sb),
1481 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1482 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1483 (long long) EXT4_C2B(EXT4_SB(sb),
1484 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1485 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1486 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1487 ei->i_reserved_data_blocks);
1488 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1489 ei->i_reserved_meta_blocks);
1490 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1491 ei->i_allocated_meta_blocks);
1495 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1497 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1501 * This function is grabs code from the very beginning of
1502 * ext4_map_blocks, but assumes that the caller is from delayed write
1503 * time. This function looks up the requested blocks and sets the
1504 * buffer delay bit under the protection of i_data_sem.
1506 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1507 struct ext4_map_blocks *map,
1508 struct buffer_head *bh)
1510 struct extent_status es;
1512 sector_t invalid_block = ~((sector_t) 0xffff);
1513 #ifdef ES_AGGRESSIVE_TEST
1514 struct ext4_map_blocks orig_map;
1516 memcpy(&orig_map, map, sizeof(*map));
1519 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1523 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1524 "logical block %lu\n", inode->i_ino, map->m_len,
1525 (unsigned long) map->m_lblk);
1527 /* Lookup extent status tree firstly */
1528 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1529 ext4_es_lru_add(inode);
1530 if (ext4_es_is_hole(&es)) {
1532 down_read((&EXT4_I(inode)->i_data_sem));
1537 * Delayed extent could be allocated by fallocate.
1538 * So we need to check it.
1540 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1541 map_bh(bh, inode->i_sb, invalid_block);
1543 set_buffer_delay(bh);
1547 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1548 retval = es.es_len - (iblock - es.es_lblk);
1549 if (retval > map->m_len)
1550 retval = map->m_len;
1551 map->m_len = retval;
1552 if (ext4_es_is_written(&es))
1553 map->m_flags |= EXT4_MAP_MAPPED;
1554 else if (ext4_es_is_unwritten(&es))
1555 map->m_flags |= EXT4_MAP_UNWRITTEN;
1559 #ifdef ES_AGGRESSIVE_TEST
1560 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1566 * Try to see if we can get the block without requesting a new
1567 * file system block.
1569 down_read((&EXT4_I(inode)->i_data_sem));
1570 if (ext4_has_inline_data(inode)) {
1572 * We will soon create blocks for this page, and let
1573 * us pretend as if the blocks aren't allocated yet.
1574 * In case of clusters, we have to handle the work
1575 * of mapping from cluster so that the reserved space
1576 * is calculated properly.
1578 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1579 ext4_find_delalloc_cluster(inode, map->m_lblk))
1580 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1582 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1583 retval = ext4_ext_map_blocks(NULL, inode, map,
1584 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1586 retval = ext4_ind_map_blocks(NULL, inode, map,
1587 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1593 * XXX: __block_prepare_write() unmaps passed block,
1597 * If the block was allocated from previously allocated cluster,
1598 * then we don't need to reserve it again. However we still need
1599 * to reserve metadata for every block we're going to write.
1601 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1602 ret = ext4_da_reserve_space(inode, iblock);
1604 /* not enough space to reserve */
1609 ret = ext4_da_reserve_metadata(inode, iblock);
1611 /* not enough space to reserve */
1617 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1618 ~0, EXTENT_STATUS_DELAYED);
1624 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1625 * and it should not appear on the bh->b_state.
1627 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1629 map_bh(bh, inode->i_sb, invalid_block);
1631 set_buffer_delay(bh);
1632 } else if (retval > 0) {
1634 unsigned int status;
1636 if (unlikely(retval != map->m_len)) {
1637 ext4_warning(inode->i_sb,
1638 "ES len assertion failed for inode "
1639 "%lu: retval %d != map->m_len %d",
1640 inode->i_ino, retval, map->m_len);
1644 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1645 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1646 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1647 map->m_pblk, status);
1653 up_read((&EXT4_I(inode)->i_data_sem));
1659 * This is a special get_blocks_t callback which is used by
1660 * ext4_da_write_begin(). It will either return mapped block or
1661 * reserve space for a single block.
1663 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1664 * We also have b_blocknr = -1 and b_bdev initialized properly
1666 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1667 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1668 * initialized properly.
1670 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1671 struct buffer_head *bh, int create)
1673 struct ext4_map_blocks map;
1676 BUG_ON(create == 0);
1677 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1679 map.m_lblk = iblock;
1683 * first, we need to know whether the block is allocated already
1684 * preallocated blocks are unmapped but should treated
1685 * the same as allocated blocks.
1687 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1691 map_bh(bh, inode->i_sb, map.m_pblk);
1692 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1694 if (buffer_unwritten(bh)) {
1695 /* A delayed write to unwritten bh should be marked
1696 * new and mapped. Mapped ensures that we don't do
1697 * get_block multiple times when we write to the same
1698 * offset and new ensures that we do proper zero out
1699 * for partial write.
1702 set_buffer_mapped(bh);
1707 static int bget_one(handle_t *handle, struct buffer_head *bh)
1713 static int bput_one(handle_t *handle, struct buffer_head *bh)
1719 static int __ext4_journalled_writepage(struct page *page,
1722 struct address_space *mapping = page->mapping;
1723 struct inode *inode = mapping->host;
1724 struct buffer_head *page_bufs = NULL;
1725 handle_t *handle = NULL;
1726 int ret = 0, err = 0;
1727 int inline_data = ext4_has_inline_data(inode);
1728 struct buffer_head *inode_bh = NULL;
1730 ClearPageChecked(page);
1733 BUG_ON(page->index != 0);
1734 BUG_ON(len > ext4_get_max_inline_size(inode));
1735 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1736 if (inode_bh == NULL)
1739 page_bufs = page_buffers(page);
1744 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1747 /* As soon as we unlock the page, it can go away, but we have
1748 * references to buffers so we are safe */
1751 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1752 ext4_writepage_trans_blocks(inode));
1753 if (IS_ERR(handle)) {
1754 ret = PTR_ERR(handle);
1758 BUG_ON(!ext4_handle_valid(handle));
1761 ret = ext4_journal_get_write_access(handle, inode_bh);
1763 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1766 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1767 do_journal_get_write_access);
1769 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1774 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1775 err = ext4_journal_stop(handle);
1779 if (!ext4_has_inline_data(inode))
1780 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1782 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1789 * Note that we don't need to start a transaction unless we're journaling data
1790 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1791 * need to file the inode to the transaction's list in ordered mode because if
1792 * we are writing back data added by write(), the inode is already there and if
1793 * we are writing back data modified via mmap(), no one guarantees in which
1794 * transaction the data will hit the disk. In case we are journaling data, we
1795 * cannot start transaction directly because transaction start ranks above page
1796 * lock so we have to do some magic.
1798 * This function can get called via...
1799 * - ext4_writepages after taking page lock (have journal handle)
1800 * - journal_submit_inode_data_buffers (no journal handle)
1801 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1802 * - grab_page_cache when doing write_begin (have journal handle)
1804 * We don't do any block allocation in this function. If we have page with
1805 * multiple blocks we need to write those buffer_heads that are mapped. This
1806 * is important for mmaped based write. So if we do with blocksize 1K
1807 * truncate(f, 1024);
1808 * a = mmap(f, 0, 4096);
1810 * truncate(f, 4096);
1811 * we have in the page first buffer_head mapped via page_mkwrite call back
1812 * but other buffer_heads would be unmapped but dirty (dirty done via the
1813 * do_wp_page). So writepage should write the first block. If we modify
1814 * the mmap area beyond 1024 we will again get a page_fault and the
1815 * page_mkwrite callback will do the block allocation and mark the
1816 * buffer_heads mapped.
1818 * We redirty the page if we have any buffer_heads that is either delay or
1819 * unwritten in the page.
1821 * We can get recursively called as show below.
1823 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1826 * But since we don't do any block allocation we should not deadlock.
1827 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1829 static int ext4_writepage(struct page *page,
1830 struct writeback_control *wbc)
1835 struct buffer_head *page_bufs = NULL;
1836 struct inode *inode = page->mapping->host;
1837 struct ext4_io_submit io_submit;
1838 bool keep_towrite = false;
1840 trace_ext4_writepage(page);
1841 size = i_size_read(inode);
1842 if (page->index == size >> PAGE_CACHE_SHIFT)
1843 len = size & ~PAGE_CACHE_MASK;
1845 len = PAGE_CACHE_SIZE;
1847 page_bufs = page_buffers(page);
1849 * We cannot do block allocation or other extent handling in this
1850 * function. If there are buffers needing that, we have to redirty
1851 * the page. But we may reach here when we do a journal commit via
1852 * journal_submit_inode_data_buffers() and in that case we must write
1853 * allocated buffers to achieve data=ordered mode guarantees.
1855 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1856 ext4_bh_delay_or_unwritten)) {
1857 redirty_page_for_writepage(wbc, page);
1858 if (current->flags & PF_MEMALLOC) {
1860 * For memory cleaning there's no point in writing only
1861 * some buffers. So just bail out. Warn if we came here
1862 * from direct reclaim.
1864 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1869 keep_towrite = true;
1872 if (PageChecked(page) && ext4_should_journal_data(inode))
1874 * It's mmapped pagecache. Add buffers and journal it. There
1875 * doesn't seem much point in redirtying the page here.
1877 return __ext4_journalled_writepage(page, len);
1879 ext4_io_submit_init(&io_submit, wbc);
1880 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1881 if (!io_submit.io_end) {
1882 redirty_page_for_writepage(wbc, page);
1886 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1887 ext4_io_submit(&io_submit);
1888 /* Drop io_end reference we got from init */
1889 ext4_put_io_end_defer(io_submit.io_end);
1893 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1896 loff_t size = i_size_read(mpd->inode);
1899 BUG_ON(page->index != mpd->first_page);
1900 if (page->index == size >> PAGE_CACHE_SHIFT)
1901 len = size & ~PAGE_CACHE_MASK;
1903 len = PAGE_CACHE_SIZE;
1904 clear_page_dirty_for_io(page);
1905 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1907 mpd->wbc->nr_to_write--;
1913 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1916 * mballoc gives us at most this number of blocks...
1917 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1918 * The rest of mballoc seems to handle chunks up to full group size.
1920 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1923 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1925 * @mpd - extent of blocks
1926 * @lblk - logical number of the block in the file
1927 * @bh - buffer head we want to add to the extent
1929 * The function is used to collect contig. blocks in the same state. If the
1930 * buffer doesn't require mapping for writeback and we haven't started the
1931 * extent of buffers to map yet, the function returns 'true' immediately - the
1932 * caller can write the buffer right away. Otherwise the function returns true
1933 * if the block has been added to the extent, false if the block couldn't be
1936 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1937 struct buffer_head *bh)
1939 struct ext4_map_blocks *map = &mpd->map;
1941 /* Buffer that doesn't need mapping for writeback? */
1942 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1943 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1944 /* So far no extent to map => we write the buffer right away */
1945 if (map->m_len == 0)
1950 /* First block in the extent? */
1951 if (map->m_len == 0) {
1954 map->m_flags = bh->b_state & BH_FLAGS;
1958 /* Don't go larger than mballoc is willing to allocate */
1959 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1962 /* Can we merge the block to our big extent? */
1963 if (lblk == map->m_lblk + map->m_len &&
1964 (bh->b_state & BH_FLAGS) == map->m_flags) {
1972 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1974 * @mpd - extent of blocks for mapping
1975 * @head - the first buffer in the page
1976 * @bh - buffer we should start processing from
1977 * @lblk - logical number of the block in the file corresponding to @bh
1979 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1980 * the page for IO if all buffers in this page were mapped and there's no
1981 * accumulated extent of buffers to map or add buffers in the page to the
1982 * extent of buffers to map. The function returns 1 if the caller can continue
1983 * by processing the next page, 0 if it should stop adding buffers to the
1984 * extent to map because we cannot extend it anymore. It can also return value
1985 * < 0 in case of error during IO submission.
1987 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1988 struct buffer_head *head,
1989 struct buffer_head *bh,
1992 struct inode *inode = mpd->inode;
1994 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1995 >> inode->i_blkbits;
1998 BUG_ON(buffer_locked(bh));
2000 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2001 /* Found extent to map? */
2004 /* Everything mapped so far and we hit EOF */
2007 } while (lblk++, (bh = bh->b_this_page) != head);
2008 /* So far everything mapped? Submit the page for IO. */
2009 if (mpd->map.m_len == 0) {
2010 err = mpage_submit_page(mpd, head->b_page);
2014 return lblk < blocks;
2018 * mpage_map_buffers - update buffers corresponding to changed extent and
2019 * submit fully mapped pages for IO
2021 * @mpd - description of extent to map, on return next extent to map
2023 * Scan buffers corresponding to changed extent (we expect corresponding pages
2024 * to be already locked) and update buffer state according to new extent state.
2025 * We map delalloc buffers to their physical location, clear unwritten bits,
2026 * and mark buffers as uninit when we perform writes to uninitialized extents
2027 * and do extent conversion after IO is finished. If the last page is not fully
2028 * mapped, we update @map to the next extent in the last page that needs
2029 * mapping. Otherwise we submit the page for IO.
2031 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2033 struct pagevec pvec;
2035 struct inode *inode = mpd->inode;
2036 struct buffer_head *head, *bh;
2037 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2043 start = mpd->map.m_lblk >> bpp_bits;
2044 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2045 lblk = start << bpp_bits;
2046 pblock = mpd->map.m_pblk;
2048 pagevec_init(&pvec, 0);
2049 while (start <= end) {
2050 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2054 for (i = 0; i < nr_pages; i++) {
2055 struct page *page = pvec.pages[i];
2057 if (page->index > end)
2059 /* Up to 'end' pages must be contiguous */
2060 BUG_ON(page->index != start);
2061 bh = head = page_buffers(page);
2063 if (lblk < mpd->map.m_lblk)
2065 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2067 * Buffer after end of mapped extent.
2068 * Find next buffer in the page to map.
2071 mpd->map.m_flags = 0;
2073 * FIXME: If dioread_nolock supports
2074 * blocksize < pagesize, we need to make
2075 * sure we add size mapped so far to
2076 * io_end->size as the following call
2077 * can submit the page for IO.
2079 err = mpage_process_page_bufs(mpd, head,
2081 pagevec_release(&pvec);
2086 if (buffer_delay(bh)) {
2087 clear_buffer_delay(bh);
2088 bh->b_blocknr = pblock++;
2090 clear_buffer_unwritten(bh);
2091 } while (lblk++, (bh = bh->b_this_page) != head);
2094 * FIXME: This is going to break if dioread_nolock
2095 * supports blocksize < pagesize as we will try to
2096 * convert potentially unmapped parts of inode.
2098 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2099 /* Page fully mapped - let IO run! */
2100 err = mpage_submit_page(mpd, page);
2102 pagevec_release(&pvec);
2107 pagevec_release(&pvec);
2109 /* Extent fully mapped and matches with page boundary. We are done. */
2111 mpd->map.m_flags = 0;
2115 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2117 struct inode *inode = mpd->inode;
2118 struct ext4_map_blocks *map = &mpd->map;
2119 int get_blocks_flags;
2122 trace_ext4_da_write_pages_extent(inode, map);
2124 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2125 * to convert an uninitialized extent to be initialized (in the case
2126 * where we have written into one or more preallocated blocks). It is
2127 * possible that we're going to need more metadata blocks than
2128 * previously reserved. However we must not fail because we're in
2129 * writeback and there is nothing we can do about it so it might result
2130 * in data loss. So use reserved blocks to allocate metadata if
2133 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2134 * in question are delalloc blocks. This affects functions in many
2135 * different parts of the allocation call path. This flag exists
2136 * primarily because we don't want to change *many* call functions, so
2137 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2138 * once the inode's allocation semaphore is taken.
2140 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2141 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2142 if (ext4_should_dioread_nolock(inode))
2143 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2144 if (map->m_flags & (1 << BH_Delay))
2145 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2147 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2150 if (map->m_flags & EXT4_MAP_UNINIT) {
2151 if (!mpd->io_submit.io_end->handle &&
2152 ext4_handle_valid(handle)) {
2153 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2154 handle->h_rsv_handle = NULL;
2156 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2159 BUG_ON(map->m_len == 0);
2160 if (map->m_flags & EXT4_MAP_NEW) {
2161 struct block_device *bdev = inode->i_sb->s_bdev;
2164 for (i = 0; i < map->m_len; i++)
2165 unmap_underlying_metadata(bdev, map->m_pblk + i);
2171 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2172 * mpd->len and submit pages underlying it for IO
2174 * @handle - handle for journal operations
2175 * @mpd - extent to map
2176 * @give_up_on_write - we set this to true iff there is a fatal error and there
2177 * is no hope of writing the data. The caller should discard
2178 * dirty pages to avoid infinite loops.
2180 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2181 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2182 * them to initialized or split the described range from larger unwritten
2183 * extent. Note that we need not map all the described range since allocation
2184 * can return less blocks or the range is covered by more unwritten extents. We
2185 * cannot map more because we are limited by reserved transaction credits. On
2186 * the other hand we always make sure that the last touched page is fully
2187 * mapped so that it can be written out (and thus forward progress is
2188 * guaranteed). After mapping we submit all mapped pages for IO.
2190 static int mpage_map_and_submit_extent(handle_t *handle,
2191 struct mpage_da_data *mpd,
2192 bool *give_up_on_write)
2194 struct inode *inode = mpd->inode;
2195 struct ext4_map_blocks *map = &mpd->map;
2200 mpd->io_submit.io_end->offset =
2201 ((loff_t)map->m_lblk) << inode->i_blkbits;
2203 err = mpage_map_one_extent(handle, mpd);
2205 struct super_block *sb = inode->i_sb;
2207 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2208 goto invalidate_dirty_pages;
2210 * Let the uper layers retry transient errors.
2211 * In the case of ENOSPC, if ext4_count_free_blocks()
2212 * is non-zero, a commit should free up blocks.
2214 if ((err == -ENOMEM) ||
2215 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2217 goto update_disksize;
2220 ext4_msg(sb, KERN_CRIT,
2221 "Delayed block allocation failed for "
2222 "inode %lu at logical offset %llu with"
2223 " max blocks %u with error %d",
2225 (unsigned long long)map->m_lblk,
2226 (unsigned)map->m_len, -err);
2227 ext4_msg(sb, KERN_CRIT,
2228 "This should not happen!! Data will "
2231 ext4_print_free_blocks(inode);
2232 invalidate_dirty_pages:
2233 *give_up_on_write = true;
2238 * Update buffer state, submit mapped pages, and get us new
2241 err = mpage_map_and_submit_buffers(mpd);
2243 goto update_disksize;
2244 } while (map->m_len);
2248 * Update on-disk size after IO is submitted. Races with
2249 * truncate are avoided by checking i_size under i_data_sem.
2251 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2252 if (disksize > EXT4_I(inode)->i_disksize) {
2256 down_write(&EXT4_I(inode)->i_data_sem);
2257 i_size = i_size_read(inode);
2258 if (disksize > i_size)
2260 if (disksize > EXT4_I(inode)->i_disksize)
2261 EXT4_I(inode)->i_disksize = disksize;
2262 err2 = ext4_mark_inode_dirty(handle, inode);
2263 up_write(&EXT4_I(inode)->i_data_sem);
2265 ext4_error(inode->i_sb,
2266 "Failed to mark inode %lu dirty",
2275 * Calculate the total number of credits to reserve for one writepages
2276 * iteration. This is called from ext4_writepages(). We map an extent of
2277 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2278 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2279 * bpp - 1 blocks in bpp different extents.
2281 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2283 int bpp = ext4_journal_blocks_per_page(inode);
2285 return ext4_meta_trans_blocks(inode,
2286 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2290 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2291 * and underlying extent to map
2293 * @mpd - where to look for pages
2295 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2296 * IO immediately. When we find a page which isn't mapped we start accumulating
2297 * extent of buffers underlying these pages that needs mapping (formed by
2298 * either delayed or unwritten buffers). We also lock the pages containing
2299 * these buffers. The extent found is returned in @mpd structure (starting at
2300 * mpd->lblk with length mpd->len blocks).
2302 * Note that this function can attach bios to one io_end structure which are
2303 * neither logically nor physically contiguous. Although it may seem as an
2304 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2305 * case as we need to track IO to all buffers underlying a page in one io_end.
2307 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2309 struct address_space *mapping = mpd->inode->i_mapping;
2310 struct pagevec pvec;
2311 unsigned int nr_pages;
2312 long left = mpd->wbc->nr_to_write;
2313 pgoff_t index = mpd->first_page;
2314 pgoff_t end = mpd->last_page;
2317 int blkbits = mpd->inode->i_blkbits;
2319 struct buffer_head *head;
2321 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2322 tag = PAGECACHE_TAG_TOWRITE;
2324 tag = PAGECACHE_TAG_DIRTY;
2326 pagevec_init(&pvec, 0);
2328 mpd->next_page = index;
2329 while (index <= end) {
2330 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2331 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2335 for (i = 0; i < nr_pages; i++) {
2336 struct page *page = pvec.pages[i];
2339 * At this point, the page may be truncated or
2340 * invalidated (changing page->mapping to NULL), or
2341 * even swizzled back from swapper_space to tmpfs file
2342 * mapping. However, page->index will not change
2343 * because we have a reference on the page.
2345 if (page->index > end)
2349 * Accumulated enough dirty pages? This doesn't apply
2350 * to WB_SYNC_ALL mode. For integrity sync we have to
2351 * keep going because someone may be concurrently
2352 * dirtying pages, and we might have synced a lot of
2353 * newly appeared dirty pages, but have not synced all
2354 * of the old dirty pages.
2356 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2359 /* If we can't merge this page, we are done. */
2360 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2365 * If the page is no longer dirty, or its mapping no
2366 * longer corresponds to inode we are writing (which
2367 * means it has been truncated or invalidated), or the
2368 * page is already under writeback and we are not doing
2369 * a data integrity writeback, skip the page
2371 if (!PageDirty(page) ||
2372 (PageWriteback(page) &&
2373 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2374 unlikely(page->mapping != mapping)) {
2379 wait_on_page_writeback(page);
2380 BUG_ON(PageWriteback(page));
2382 if (mpd->map.m_len == 0)
2383 mpd->first_page = page->index;
2384 mpd->next_page = page->index + 1;
2385 /* Add all dirty buffers to mpd */
2386 lblk = ((ext4_lblk_t)page->index) <<
2387 (PAGE_CACHE_SHIFT - blkbits);
2388 head = page_buffers(page);
2389 err = mpage_process_page_bufs(mpd, head, head, lblk);
2395 pagevec_release(&pvec);
2400 pagevec_release(&pvec);
2404 static int __writepage(struct page *page, struct writeback_control *wbc,
2407 struct address_space *mapping = data;
2408 int ret = ext4_writepage(page, wbc);
2409 mapping_set_error(mapping, ret);
2413 static int ext4_writepages(struct address_space *mapping,
2414 struct writeback_control *wbc)
2416 pgoff_t writeback_index = 0;
2417 long nr_to_write = wbc->nr_to_write;
2418 int range_whole = 0;
2420 handle_t *handle = NULL;
2421 struct mpage_da_data mpd;
2422 struct inode *inode = mapping->host;
2423 int needed_blocks, rsv_blocks = 0, ret = 0;
2424 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2426 struct blk_plug plug;
2427 bool give_up_on_write = false;
2429 trace_ext4_writepages(inode, wbc);
2432 * No pages to write? This is mainly a kludge to avoid starting
2433 * a transaction for special inodes like journal inode on last iput()
2434 * because that could violate lock ordering on umount
2436 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2437 goto out_writepages;
2439 if (ext4_should_journal_data(inode)) {
2440 struct blk_plug plug;
2442 blk_start_plug(&plug);
2443 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2444 blk_finish_plug(&plug);
2445 goto out_writepages;
2449 * If the filesystem has aborted, it is read-only, so return
2450 * right away instead of dumping stack traces later on that
2451 * will obscure the real source of the problem. We test
2452 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2453 * the latter could be true if the filesystem is mounted
2454 * read-only, and in that case, ext4_writepages should
2455 * *never* be called, so if that ever happens, we would want
2458 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2460 goto out_writepages;
2463 if (ext4_should_dioread_nolock(inode)) {
2465 * We may need to convert up to one extent per block in
2466 * the page and we may dirty the inode.
2468 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2472 * If we have inline data and arrive here, it means that
2473 * we will soon create the block for the 1st page, so
2474 * we'd better clear the inline data here.
2476 if (ext4_has_inline_data(inode)) {
2477 /* Just inode will be modified... */
2478 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2479 if (IS_ERR(handle)) {
2480 ret = PTR_ERR(handle);
2481 goto out_writepages;
2483 BUG_ON(ext4_test_inode_state(inode,
2484 EXT4_STATE_MAY_INLINE_DATA));
2485 ext4_destroy_inline_data(handle, inode);
2486 ext4_journal_stop(handle);
2489 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2492 if (wbc->range_cyclic) {
2493 writeback_index = mapping->writeback_index;
2494 if (writeback_index)
2496 mpd.first_page = writeback_index;
2499 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2500 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2505 ext4_io_submit_init(&mpd.io_submit, wbc);
2507 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2508 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2510 blk_start_plug(&plug);
2511 while (!done && mpd.first_page <= mpd.last_page) {
2512 /* For each extent of pages we use new io_end */
2513 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2514 if (!mpd.io_submit.io_end) {
2520 * We have two constraints: We find one extent to map and we
2521 * must always write out whole page (makes a difference when
2522 * blocksize < pagesize) so that we don't block on IO when we
2523 * try to write out the rest of the page. Journalled mode is
2524 * not supported by delalloc.
2526 BUG_ON(ext4_should_journal_data(inode));
2527 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2529 /* start a new transaction */
2530 handle = ext4_journal_start_with_reserve(inode,
2531 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2532 if (IS_ERR(handle)) {
2533 ret = PTR_ERR(handle);
2534 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2535 "%ld pages, ino %lu; err %d", __func__,
2536 wbc->nr_to_write, inode->i_ino, ret);
2537 /* Release allocated io_end */
2538 ext4_put_io_end(mpd.io_submit.io_end);
2542 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2543 ret = mpage_prepare_extent_to_map(&mpd);
2546 ret = mpage_map_and_submit_extent(handle, &mpd,
2550 * We scanned the whole range (or exhausted
2551 * nr_to_write), submitted what was mapped and
2552 * didn't find anything needing mapping. We are
2558 ext4_journal_stop(handle);
2559 /* Submit prepared bio */
2560 ext4_io_submit(&mpd.io_submit);
2561 /* Unlock pages we didn't use */
2562 mpage_release_unused_pages(&mpd, give_up_on_write);
2563 /* Drop our io_end reference we got from init */
2564 ext4_put_io_end(mpd.io_submit.io_end);
2566 if (ret == -ENOSPC && sbi->s_journal) {
2568 * Commit the transaction which would
2569 * free blocks released in the transaction
2572 jbd2_journal_force_commit_nested(sbi->s_journal);
2576 /* Fatal error - ENOMEM, EIO... */
2580 blk_finish_plug(&plug);
2581 if (!ret && !cycled && wbc->nr_to_write > 0) {
2583 mpd.last_page = writeback_index - 1;
2589 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2591 * Set the writeback_index so that range_cyclic
2592 * mode will write it back later
2594 mapping->writeback_index = mpd.first_page;
2597 trace_ext4_writepages_result(inode, wbc, ret,
2598 nr_to_write - wbc->nr_to_write);
2602 static int ext4_nonda_switch(struct super_block *sb)
2604 s64 free_clusters, dirty_clusters;
2605 struct ext4_sb_info *sbi = EXT4_SB(sb);
2608 * switch to non delalloc mode if we are running low
2609 * on free block. The free block accounting via percpu
2610 * counters can get slightly wrong with percpu_counter_batch getting
2611 * accumulated on each CPU without updating global counters
2612 * Delalloc need an accurate free block accounting. So switch
2613 * to non delalloc when we are near to error range.
2616 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2618 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2620 * Start pushing delalloc when 1/2 of free blocks are dirty.
2622 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2623 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2625 if (2 * free_clusters < 3 * dirty_clusters ||
2626 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2628 * free block count is less than 150% of dirty blocks
2629 * or free blocks is less than watermark
2636 /* We always reserve for an inode update; the superblock could be there too */
2637 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2639 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2640 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2643 if (pos + len <= 0x7fffffffULL)
2646 /* We might need to update the superblock to set LARGE_FILE */
2650 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2651 loff_t pos, unsigned len, unsigned flags,
2652 struct page **pagep, void **fsdata)
2654 int ret, retries = 0;
2657 struct inode *inode = mapping->host;
2660 index = pos >> PAGE_CACHE_SHIFT;
2662 if (ext4_nonda_switch(inode->i_sb)) {
2663 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2664 return ext4_write_begin(file, mapping, pos,
2665 len, flags, pagep, fsdata);
2667 *fsdata = (void *)0;
2668 trace_ext4_da_write_begin(inode, pos, len, flags);
2670 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2671 ret = ext4_da_write_inline_data_begin(mapping, inode,
2681 * grab_cache_page_write_begin() can take a long time if the
2682 * system is thrashing due to memory pressure, or if the page
2683 * is being written back. So grab it first before we start
2684 * the transaction handle. This also allows us to allocate
2685 * the page (if needed) without using GFP_NOFS.
2688 page = grab_cache_page_write_begin(mapping, index, flags);
2694 * With delayed allocation, we don't log the i_disksize update
2695 * if there is delayed block allocation. But we still need
2696 * to journalling the i_disksize update if writes to the end
2697 * of file which has an already mapped buffer.
2700 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2701 ext4_da_write_credits(inode, pos, len));
2702 if (IS_ERR(handle)) {
2703 page_cache_release(page);
2704 return PTR_ERR(handle);
2708 if (page->mapping != mapping) {
2709 /* The page got truncated from under us */
2711 page_cache_release(page);
2712 ext4_journal_stop(handle);
2715 /* In case writeback began while the page was unlocked */
2716 wait_for_stable_page(page);
2718 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2721 ext4_journal_stop(handle);
2723 * block_write_begin may have instantiated a few blocks
2724 * outside i_size. Trim these off again. Don't need
2725 * i_size_read because we hold i_mutex.
2727 if (pos + len > inode->i_size)
2728 ext4_truncate_failed_write(inode);
2730 if (ret == -ENOSPC &&
2731 ext4_should_retry_alloc(inode->i_sb, &retries))
2734 page_cache_release(page);
2743 * Check if we should update i_disksize
2744 * when write to the end of file but not require block allocation
2746 static int ext4_da_should_update_i_disksize(struct page *page,
2747 unsigned long offset)
2749 struct buffer_head *bh;
2750 struct inode *inode = page->mapping->host;
2754 bh = page_buffers(page);
2755 idx = offset >> inode->i_blkbits;
2757 for (i = 0; i < idx; i++)
2758 bh = bh->b_this_page;
2760 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2765 static int ext4_da_write_end(struct file *file,
2766 struct address_space *mapping,
2767 loff_t pos, unsigned len, unsigned copied,
2768 struct page *page, void *fsdata)
2770 struct inode *inode = mapping->host;
2772 handle_t *handle = ext4_journal_current_handle();
2774 unsigned long start, end;
2775 int write_mode = (int)(unsigned long)fsdata;
2777 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2778 return ext4_write_end(file, mapping, pos,
2779 len, copied, page, fsdata);
2781 trace_ext4_da_write_end(inode, pos, len, copied);
2782 start = pos & (PAGE_CACHE_SIZE - 1);
2783 end = start + copied - 1;
2786 * generic_write_end() will run mark_inode_dirty() if i_size
2787 * changes. So let's piggyback the i_disksize mark_inode_dirty
2790 new_i_size = pos + copied;
2791 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2792 if (ext4_has_inline_data(inode) ||
2793 ext4_da_should_update_i_disksize(page, end)) {
2794 down_write(&EXT4_I(inode)->i_data_sem);
2795 if (new_i_size > EXT4_I(inode)->i_disksize)
2796 EXT4_I(inode)->i_disksize = new_i_size;
2797 up_write(&EXT4_I(inode)->i_data_sem);
2798 /* We need to mark inode dirty even if
2799 * new_i_size is less that inode->i_size
2800 * bu greater than i_disksize.(hint delalloc)
2802 ext4_mark_inode_dirty(handle, inode);
2806 if (write_mode != CONVERT_INLINE_DATA &&
2807 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2808 ext4_has_inline_data(inode))
2809 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2812 ret2 = generic_write_end(file, mapping, pos, len, copied,
2818 ret2 = ext4_journal_stop(handle);
2822 return ret ? ret : copied;
2825 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2826 unsigned int length)
2829 * Drop reserved blocks
2831 BUG_ON(!PageLocked(page));
2832 if (!page_has_buffers(page))
2835 ext4_da_page_release_reservation(page, offset, length);
2838 ext4_invalidatepage(page, offset, length);
2844 * Force all delayed allocation blocks to be allocated for a given inode.
2846 int ext4_alloc_da_blocks(struct inode *inode)
2848 trace_ext4_alloc_da_blocks(inode);
2850 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2851 !EXT4_I(inode)->i_reserved_meta_blocks)
2855 * We do something simple for now. The filemap_flush() will
2856 * also start triggering a write of the data blocks, which is
2857 * not strictly speaking necessary (and for users of
2858 * laptop_mode, not even desirable). However, to do otherwise
2859 * would require replicating code paths in:
2861 * ext4_writepages() ->
2862 * write_cache_pages() ---> (via passed in callback function)
2863 * __mpage_da_writepage() -->
2864 * mpage_add_bh_to_extent()
2865 * mpage_da_map_blocks()
2867 * The problem is that write_cache_pages(), located in
2868 * mm/page-writeback.c, marks pages clean in preparation for
2869 * doing I/O, which is not desirable if we're not planning on
2872 * We could call write_cache_pages(), and then redirty all of
2873 * the pages by calling redirty_page_for_writepage() but that
2874 * would be ugly in the extreme. So instead we would need to
2875 * replicate parts of the code in the above functions,
2876 * simplifying them because we wouldn't actually intend to
2877 * write out the pages, but rather only collect contiguous
2878 * logical block extents, call the multi-block allocator, and
2879 * then update the buffer heads with the block allocations.
2881 * For now, though, we'll cheat by calling filemap_flush(),
2882 * which will map the blocks, and start the I/O, but not
2883 * actually wait for the I/O to complete.
2885 return filemap_flush(inode->i_mapping);
2889 * bmap() is special. It gets used by applications such as lilo and by
2890 * the swapper to find the on-disk block of a specific piece of data.
2892 * Naturally, this is dangerous if the block concerned is still in the
2893 * journal. If somebody makes a swapfile on an ext4 data-journaling
2894 * filesystem and enables swap, then they may get a nasty shock when the
2895 * data getting swapped to that swapfile suddenly gets overwritten by
2896 * the original zero's written out previously to the journal and
2897 * awaiting writeback in the kernel's buffer cache.
2899 * So, if we see any bmap calls here on a modified, data-journaled file,
2900 * take extra steps to flush any blocks which might be in the cache.
2902 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2904 struct inode *inode = mapping->host;
2909 * We can get here for an inline file via the FIBMAP ioctl
2911 if (ext4_has_inline_data(inode))
2914 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2915 test_opt(inode->i_sb, DELALLOC)) {
2917 * With delalloc we want to sync the file
2918 * so that we can make sure we allocate
2921 filemap_write_and_wait(mapping);
2924 if (EXT4_JOURNAL(inode) &&
2925 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2927 * This is a REALLY heavyweight approach, but the use of
2928 * bmap on dirty files is expected to be extremely rare:
2929 * only if we run lilo or swapon on a freshly made file
2930 * do we expect this to happen.
2932 * (bmap requires CAP_SYS_RAWIO so this does not
2933 * represent an unprivileged user DOS attack --- we'd be
2934 * in trouble if mortal users could trigger this path at
2937 * NB. EXT4_STATE_JDATA is not set on files other than
2938 * regular files. If somebody wants to bmap a directory
2939 * or symlink and gets confused because the buffer
2940 * hasn't yet been flushed to disk, they deserve
2941 * everything they get.
2944 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2945 journal = EXT4_JOURNAL(inode);
2946 jbd2_journal_lock_updates(journal);
2947 err = jbd2_journal_flush(journal);
2948 jbd2_journal_unlock_updates(journal);
2954 return generic_block_bmap(mapping, block, ext4_get_block);
2957 static int ext4_readpage(struct file *file, struct page *page)
2960 struct inode *inode = page->mapping->host;
2962 trace_ext4_readpage(page);
2964 if (ext4_has_inline_data(inode))
2965 ret = ext4_readpage_inline(inode, page);
2968 return mpage_readpage(page, ext4_get_block);
2974 ext4_readpages(struct file *file, struct address_space *mapping,
2975 struct list_head *pages, unsigned nr_pages)
2977 struct inode *inode = mapping->host;
2979 /* If the file has inline data, no need to do readpages. */
2980 if (ext4_has_inline_data(inode))
2983 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2986 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2987 unsigned int length)
2989 trace_ext4_invalidatepage(page, offset, length);
2991 /* No journalling happens on data buffers when this function is used */
2992 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2994 block_invalidatepage(page, offset, length);
2997 static int __ext4_journalled_invalidatepage(struct page *page,
2998 unsigned int offset,
2999 unsigned int length)
3001 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3003 trace_ext4_journalled_invalidatepage(page, offset, length);
3006 * If it's a full truncate we just forget about the pending dirtying
3008 if (offset == 0 && length == PAGE_CACHE_SIZE)
3009 ClearPageChecked(page);
3011 return jbd2_journal_invalidatepage(journal, page, offset, length);
3014 /* Wrapper for aops... */
3015 static void ext4_journalled_invalidatepage(struct page *page,
3016 unsigned int offset,
3017 unsigned int length)
3019 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3022 static int ext4_releasepage(struct page *page, gfp_t wait)
3024 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3026 trace_ext4_releasepage(page);
3028 /* Page has dirty journalled data -> cannot release */
3029 if (PageChecked(page))
3032 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3034 return try_to_free_buffers(page);
3038 * ext4_get_block used when preparing for a DIO write or buffer write.
3039 * We allocate an uinitialized extent if blocks haven't been allocated.
3040 * The extent will be converted to initialized after the IO is complete.
3042 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3043 struct buffer_head *bh_result, int create)
3045 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3046 inode->i_ino, create);
3047 return _ext4_get_block(inode, iblock, bh_result,
3048 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3051 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3052 struct buffer_head *bh_result, int create)
3054 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3055 inode->i_ino, create);
3056 return _ext4_get_block(inode, iblock, bh_result,
3057 EXT4_GET_BLOCKS_NO_LOCK);
3060 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3061 ssize_t size, void *private)
3063 ext4_io_end_t *io_end = iocb->private;
3065 /* if not async direct IO just return */
3069 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3070 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3071 iocb->private, io_end->inode->i_ino, iocb, offset,
3074 iocb->private = NULL;
3075 io_end->offset = offset;
3076 io_end->size = size;
3077 ext4_put_io_end(io_end);
3081 * For ext4 extent files, ext4 will do direct-io write to holes,
3082 * preallocated extents, and those write extend the file, no need to
3083 * fall back to buffered IO.
3085 * For holes, we fallocate those blocks, mark them as uninitialized
3086 * If those blocks were preallocated, we mark sure they are split, but
3087 * still keep the range to write as uninitialized.
3089 * The unwritten extents will be converted to written when DIO is completed.
3090 * For async direct IO, since the IO may still pending when return, we
3091 * set up an end_io call back function, which will do the conversion
3092 * when async direct IO completed.
3094 * If the O_DIRECT write will extend the file then add this inode to the
3095 * orphan list. So recovery will truncate it back to the original size
3096 * if the machine crashes during the write.
3099 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3100 const struct iovec *iov, loff_t offset,
3101 unsigned long nr_segs)
3103 struct file *file = iocb->ki_filp;
3104 struct inode *inode = file->f_mapping->host;
3106 size_t count = iov_length(iov, nr_segs);
3108 get_block_t *get_block_func = NULL;
3110 loff_t final_size = offset + count;
3111 ext4_io_end_t *io_end = NULL;
3113 /* Use the old path for reads and writes beyond i_size. */
3114 if (rw != WRITE || final_size > inode->i_size)
3115 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3117 BUG_ON(iocb->private == NULL);
3120 * Make all waiters for direct IO properly wait also for extent
3121 * conversion. This also disallows race between truncate() and
3122 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3125 atomic_inc(&inode->i_dio_count);
3127 /* If we do a overwrite dio, i_mutex locking can be released */
3128 overwrite = *((int *)iocb->private);
3131 down_read(&EXT4_I(inode)->i_data_sem);
3132 mutex_unlock(&inode->i_mutex);
3136 * We could direct write to holes and fallocate.
3138 * Allocated blocks to fill the hole are marked as
3139 * uninitialized to prevent parallel buffered read to expose
3140 * the stale data before DIO complete the data IO.
3142 * As to previously fallocated extents, ext4 get_block will
3143 * just simply mark the buffer mapped but still keep the
3144 * extents uninitialized.
3146 * For non AIO case, we will convert those unwritten extents
3147 * to written after return back from blockdev_direct_IO.
3149 * For async DIO, the conversion needs to be deferred when the
3150 * IO is completed. The ext4 end_io callback function will be
3151 * called to take care of the conversion work. Here for async
3152 * case, we allocate an io_end structure to hook to the iocb.
3154 iocb->private = NULL;
3155 ext4_inode_aio_set(inode, NULL);
3156 if (!is_sync_kiocb(iocb)) {
3157 io_end = ext4_init_io_end(inode, GFP_NOFS);
3163 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3165 iocb->private = ext4_get_io_end(io_end);
3167 * we save the io structure for current async direct
3168 * IO, so that later ext4_map_blocks() could flag the
3169 * io structure whether there is a unwritten extents
3170 * needs to be converted when IO is completed.
3172 ext4_inode_aio_set(inode, io_end);
3176 get_block_func = ext4_get_block_write_nolock;
3178 get_block_func = ext4_get_block_write;
3179 dio_flags = DIO_LOCKING;
3181 ret = __blockdev_direct_IO(rw, iocb, inode,
3182 inode->i_sb->s_bdev, iov,
3190 * Put our reference to io_end. This can free the io_end structure e.g.
3191 * in sync IO case or in case of error. It can even perform extent
3192 * conversion if all bios we submitted finished before we got here.
3193 * Note that in that case iocb->private can be already set to NULL
3197 ext4_inode_aio_set(inode, NULL);
3198 ext4_put_io_end(io_end);
3200 * When no IO was submitted ext4_end_io_dio() was not
3201 * called so we have to put iocb's reference.
3203 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3204 WARN_ON(iocb->private != io_end);
3205 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3206 ext4_put_io_end(io_end);
3207 iocb->private = NULL;
3210 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3211 EXT4_STATE_DIO_UNWRITTEN)) {
3214 * for non AIO case, since the IO is already
3215 * completed, we could do the conversion right here
3217 err = ext4_convert_unwritten_extents(NULL, inode,
3221 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3226 inode_dio_done(inode);
3227 /* take i_mutex locking again if we do a ovewrite dio */
3229 up_read(&EXT4_I(inode)->i_data_sem);
3230 mutex_lock(&inode->i_mutex);
3236 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3237 const struct iovec *iov, loff_t offset,
3238 unsigned long nr_segs)
3240 struct file *file = iocb->ki_filp;
3241 struct inode *inode = file->f_mapping->host;
3245 * If we are doing data journalling we don't support O_DIRECT
3247 if (ext4_should_journal_data(inode))
3250 /* Let buffer I/O handle the inline data case. */
3251 if (ext4_has_inline_data(inode))
3254 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3255 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3256 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3258 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3259 trace_ext4_direct_IO_exit(inode, offset,
3260 iov_length(iov, nr_segs), rw, ret);
3265 * Pages can be marked dirty completely asynchronously from ext4's journalling
3266 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3267 * much here because ->set_page_dirty is called under VFS locks. The page is
3268 * not necessarily locked.
3270 * We cannot just dirty the page and leave attached buffers clean, because the
3271 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3272 * or jbddirty because all the journalling code will explode.
3274 * So what we do is to mark the page "pending dirty" and next time writepage
3275 * is called, propagate that into the buffers appropriately.
3277 static int ext4_journalled_set_page_dirty(struct page *page)
3279 SetPageChecked(page);
3280 return __set_page_dirty_nobuffers(page);
3283 static const struct address_space_operations ext4_aops = {
3284 .readpage = ext4_readpage,
3285 .readpages = ext4_readpages,
3286 .writepage = ext4_writepage,
3287 .writepages = ext4_writepages,
3288 .write_begin = ext4_write_begin,
3289 .write_end = ext4_write_end,
3291 .invalidatepage = ext4_invalidatepage,
3292 .releasepage = ext4_releasepage,
3293 .direct_IO = ext4_direct_IO,
3294 .migratepage = buffer_migrate_page,
3295 .is_partially_uptodate = block_is_partially_uptodate,
3296 .error_remove_page = generic_error_remove_page,
3299 static const struct address_space_operations ext4_journalled_aops = {
3300 .readpage = ext4_readpage,
3301 .readpages = ext4_readpages,
3302 .writepage = ext4_writepage,
3303 .writepages = ext4_writepages,
3304 .write_begin = ext4_write_begin,
3305 .write_end = ext4_journalled_write_end,
3306 .set_page_dirty = ext4_journalled_set_page_dirty,
3308 .invalidatepage = ext4_journalled_invalidatepage,
3309 .releasepage = ext4_releasepage,
3310 .direct_IO = ext4_direct_IO,
3311 .is_partially_uptodate = block_is_partially_uptodate,
3312 .error_remove_page = generic_error_remove_page,
3315 static const struct address_space_operations ext4_da_aops = {
3316 .readpage = ext4_readpage,
3317 .readpages = ext4_readpages,
3318 .writepage = ext4_writepage,
3319 .writepages = ext4_writepages,
3320 .write_begin = ext4_da_write_begin,
3321 .write_end = ext4_da_write_end,
3323 .invalidatepage = ext4_da_invalidatepage,
3324 .releasepage = ext4_releasepage,
3325 .direct_IO = ext4_direct_IO,
3326 .migratepage = buffer_migrate_page,
3327 .is_partially_uptodate = block_is_partially_uptodate,
3328 .error_remove_page = generic_error_remove_page,
3331 void ext4_set_aops(struct inode *inode)
3333 switch (ext4_inode_journal_mode(inode)) {
3334 case EXT4_INODE_ORDERED_DATA_MODE:
3335 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3337 case EXT4_INODE_WRITEBACK_DATA_MODE:
3338 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3340 case EXT4_INODE_JOURNAL_DATA_MODE:
3341 inode->i_mapping->a_ops = &ext4_journalled_aops;
3346 if (test_opt(inode->i_sb, DELALLOC))
3347 inode->i_mapping->a_ops = &ext4_da_aops;
3349 inode->i_mapping->a_ops = &ext4_aops;
3353 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3354 * up to the end of the block which corresponds to `from'.
3355 * This required during truncate. We need to physically zero the tail end
3356 * of that block so it doesn't yield old data if the file is later grown.
3358 int ext4_block_truncate_page(handle_t *handle,
3359 struct address_space *mapping, loff_t from)
3361 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3364 struct inode *inode = mapping->host;
3366 blocksize = inode->i_sb->s_blocksize;
3367 length = blocksize - (offset & (blocksize - 1));
3369 return ext4_block_zero_page_range(handle, mapping, from, length);
3373 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3374 * starting from file offset 'from'. The range to be zero'd must
3375 * be contained with in one block. If the specified range exceeds
3376 * the end of the block it will be shortened to end of the block
3377 * that cooresponds to 'from'
3379 int ext4_block_zero_page_range(handle_t *handle,
3380 struct address_space *mapping, loff_t from, loff_t length)
3382 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3383 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3384 unsigned blocksize, max, pos;
3386 struct inode *inode = mapping->host;
3387 struct buffer_head *bh;
3391 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3392 mapping_gfp_mask(mapping) & ~__GFP_FS);
3396 blocksize = inode->i_sb->s_blocksize;
3397 max = blocksize - (offset & (blocksize - 1));
3400 * correct length if it does not fall between
3401 * 'from' and the end of the block
3403 if (length > max || length < 0)
3406 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3408 if (!page_has_buffers(page))
3409 create_empty_buffers(page, blocksize, 0);
3411 /* Find the buffer that contains "offset" */
3412 bh = page_buffers(page);
3414 while (offset >= pos) {
3415 bh = bh->b_this_page;
3419 if (buffer_freed(bh)) {
3420 BUFFER_TRACE(bh, "freed: skip");
3423 if (!buffer_mapped(bh)) {
3424 BUFFER_TRACE(bh, "unmapped");
3425 ext4_get_block(inode, iblock, bh, 0);
3426 /* unmapped? It's a hole - nothing to do */
3427 if (!buffer_mapped(bh)) {
3428 BUFFER_TRACE(bh, "still unmapped");
3433 /* Ok, it's mapped. Make sure it's up-to-date */
3434 if (PageUptodate(page))
3435 set_buffer_uptodate(bh);
3437 if (!buffer_uptodate(bh)) {
3439 ll_rw_block(READ, 1, &bh);
3441 /* Uhhuh. Read error. Complain and punt. */
3442 if (!buffer_uptodate(bh))
3445 if (ext4_should_journal_data(inode)) {
3446 BUFFER_TRACE(bh, "get write access");
3447 err = ext4_journal_get_write_access(handle, bh);
3451 zero_user(page, offset, length);
3452 BUFFER_TRACE(bh, "zeroed end of block");
3454 if (ext4_should_journal_data(inode)) {
3455 err = ext4_handle_dirty_metadata(handle, inode, bh);
3458 mark_buffer_dirty(bh);
3459 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3460 err = ext4_jbd2_file_inode(handle, inode);
3465 page_cache_release(page);
3469 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3470 loff_t lstart, loff_t length)
3472 struct super_block *sb = inode->i_sb;
3473 struct address_space *mapping = inode->i_mapping;
3474 unsigned partial_start, partial_end;
3475 ext4_fsblk_t start, end;
3476 loff_t byte_end = (lstart + length - 1);
3479 partial_start = lstart & (sb->s_blocksize - 1);
3480 partial_end = byte_end & (sb->s_blocksize - 1);
3482 start = lstart >> sb->s_blocksize_bits;
3483 end = byte_end >> sb->s_blocksize_bits;
3485 /* Handle partial zero within the single block */
3487 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3488 err = ext4_block_zero_page_range(handle, mapping,
3492 /* Handle partial zero out on the start of the range */
3493 if (partial_start) {
3494 err = ext4_block_zero_page_range(handle, mapping,
3495 lstart, sb->s_blocksize);
3499 /* Handle partial zero out on the end of the range */
3500 if (partial_end != sb->s_blocksize - 1)
3501 err = ext4_block_zero_page_range(handle, mapping,
3502 byte_end - partial_end,
3507 int ext4_can_truncate(struct inode *inode)
3509 if (S_ISREG(inode->i_mode))
3511 if (S_ISDIR(inode->i_mode))
3513 if (S_ISLNK(inode->i_mode))
3514 return !ext4_inode_is_fast_symlink(inode);
3519 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3520 * associated with the given offset and length
3522 * @inode: File inode
3523 * @offset: The offset where the hole will begin
3524 * @len: The length of the hole
3526 * Returns: 0 on success or negative on failure
3529 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3531 struct super_block *sb = inode->i_sb;
3532 ext4_lblk_t first_block, stop_block;
3533 struct address_space *mapping = inode->i_mapping;
3534 loff_t first_block_offset, last_block_offset;
3536 unsigned int credits;
3539 if (!S_ISREG(inode->i_mode))
3542 trace_ext4_punch_hole(inode, offset, length);
3545 * Write out all dirty pages to avoid race conditions
3546 * Then release them.
3548 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3549 ret = filemap_write_and_wait_range(mapping, offset,
3550 offset + length - 1);
3555 mutex_lock(&inode->i_mutex);
3556 /* It's not possible punch hole on append only file */
3557 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3561 if (IS_SWAPFILE(inode)) {
3566 /* No need to punch hole beyond i_size */
3567 if (offset >= inode->i_size)
3571 * If the hole extends beyond i_size, set the hole
3572 * to end after the page that contains i_size
3574 if (offset + length > inode->i_size) {
3575 length = inode->i_size +
3576 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3580 if (offset & (sb->s_blocksize - 1) ||
3581 (offset + length) & (sb->s_blocksize - 1)) {
3583 * Attach jinode to inode for jbd2 if we do any zeroing of
3586 ret = ext4_inode_attach_jinode(inode);
3592 first_block_offset = round_up(offset, sb->s_blocksize);
3593 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3595 /* Now release the pages and zero block aligned part of pages*/
3596 if (last_block_offset > first_block_offset)
3597 truncate_pagecache_range(inode, first_block_offset,
3600 /* Wait all existing dio workers, newcomers will block on i_mutex */
3601 ext4_inode_block_unlocked_dio(inode);
3602 inode_dio_wait(inode);
3604 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3605 credits = ext4_writepage_trans_blocks(inode);
3607 credits = ext4_blocks_for_truncate(inode);
3608 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3609 if (IS_ERR(handle)) {
3610 ret = PTR_ERR(handle);
3611 ext4_std_error(sb, ret);
3615 ret = ext4_zero_partial_blocks(handle, inode, offset,
3620 first_block = (offset + sb->s_blocksize - 1) >>
3621 EXT4_BLOCK_SIZE_BITS(sb);
3622 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3624 /* If there are no blocks to remove, return now */
3625 if (first_block >= stop_block)
3628 down_write(&EXT4_I(inode)->i_data_sem);
3629 ext4_discard_preallocations(inode);
3631 ret = ext4_es_remove_extent(inode, first_block,
3632 stop_block - first_block);
3634 up_write(&EXT4_I(inode)->i_data_sem);
3638 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3639 ret = ext4_ext_remove_space(inode, first_block,
3642 ret = ext4_free_hole_blocks(handle, inode, first_block,
3645 ext4_discard_preallocations(inode);
3646 up_write(&EXT4_I(inode)->i_data_sem);
3648 ext4_handle_sync(handle);
3649 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3650 ext4_mark_inode_dirty(handle, inode);
3652 ext4_journal_stop(handle);
3654 ext4_inode_resume_unlocked_dio(inode);
3656 mutex_unlock(&inode->i_mutex);
3660 int ext4_inode_attach_jinode(struct inode *inode)
3662 struct ext4_inode_info *ei = EXT4_I(inode);
3663 struct jbd2_inode *jinode;
3665 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3668 jinode = jbd2_alloc_inode(GFP_KERNEL);
3669 spin_lock(&inode->i_lock);
3672 spin_unlock(&inode->i_lock);
3675 ei->jinode = jinode;
3676 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3679 spin_unlock(&inode->i_lock);
3680 if (unlikely(jinode != NULL))
3681 jbd2_free_inode(jinode);
3688 * We block out ext4_get_block() block instantiations across the entire
3689 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3690 * simultaneously on behalf of the same inode.
3692 * As we work through the truncate and commit bits of it to the journal there
3693 * is one core, guiding principle: the file's tree must always be consistent on
3694 * disk. We must be able to restart the truncate after a crash.
3696 * The file's tree may be transiently inconsistent in memory (although it
3697 * probably isn't), but whenever we close off and commit a journal transaction,
3698 * the contents of (the filesystem + the journal) must be consistent and
3699 * restartable. It's pretty simple, really: bottom up, right to left (although
3700 * left-to-right works OK too).
3702 * Note that at recovery time, journal replay occurs *before* the restart of
3703 * truncate against the orphan inode list.
3705 * The committed inode has the new, desired i_size (which is the same as
3706 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3707 * that this inode's truncate did not complete and it will again call
3708 * ext4_truncate() to have another go. So there will be instantiated blocks
3709 * to the right of the truncation point in a crashed ext4 filesystem. But
3710 * that's fine - as long as they are linked from the inode, the post-crash
3711 * ext4_truncate() run will find them and release them.
3713 void ext4_truncate(struct inode *inode)
3715 struct ext4_inode_info *ei = EXT4_I(inode);
3716 unsigned int credits;
3718 struct address_space *mapping = inode->i_mapping;
3721 * There is a possibility that we're either freeing the inode
3722 * or it completely new indode. In those cases we might not
3723 * have i_mutex locked because it's not necessary.
3725 if (!(inode->i_state & (I_NEW|I_FREEING)))
3726 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3727 trace_ext4_truncate_enter(inode);
3729 if (!ext4_can_truncate(inode))
3732 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3734 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3735 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3737 if (ext4_has_inline_data(inode)) {
3740 ext4_inline_data_truncate(inode, &has_inline);
3745 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3746 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3747 if (ext4_inode_attach_jinode(inode) < 0)
3751 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3752 credits = ext4_writepage_trans_blocks(inode);
3754 credits = ext4_blocks_for_truncate(inode);
3756 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3757 if (IS_ERR(handle)) {
3758 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3762 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3763 ext4_block_truncate_page(handle, mapping, inode->i_size);
3766 * We add the inode to the orphan list, so that if this
3767 * truncate spans multiple transactions, and we crash, we will
3768 * resume the truncate when the filesystem recovers. It also
3769 * marks the inode dirty, to catch the new size.
3771 * Implication: the file must always be in a sane, consistent
3772 * truncatable state while each transaction commits.
3774 if (ext4_orphan_add(handle, inode))
3777 down_write(&EXT4_I(inode)->i_data_sem);
3779 ext4_discard_preallocations(inode);
3781 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3782 ext4_ext_truncate(handle, inode);
3784 ext4_ind_truncate(handle, inode);
3786 up_write(&ei->i_data_sem);
3789 ext4_handle_sync(handle);
3793 * If this was a simple ftruncate() and the file will remain alive,
3794 * then we need to clear up the orphan record which we created above.
3795 * However, if this was a real unlink then we were called by
3796 * ext4_delete_inode(), and we allow that function to clean up the
3797 * orphan info for us.
3800 ext4_orphan_del(handle, inode);
3802 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3803 ext4_mark_inode_dirty(handle, inode);
3804 ext4_journal_stop(handle);
3806 trace_ext4_truncate_exit(inode);
3810 * ext4_get_inode_loc returns with an extra refcount against the inode's
3811 * underlying buffer_head on success. If 'in_mem' is true, we have all
3812 * data in memory that is needed to recreate the on-disk version of this
3815 static int __ext4_get_inode_loc(struct inode *inode,
3816 struct ext4_iloc *iloc, int in_mem)
3818 struct ext4_group_desc *gdp;
3819 struct buffer_head *bh;
3820 struct super_block *sb = inode->i_sb;
3822 int inodes_per_block, inode_offset;
3825 if (!ext4_valid_inum(sb, inode->i_ino))
3828 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3829 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3834 * Figure out the offset within the block group inode table
3836 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3837 inode_offset = ((inode->i_ino - 1) %
3838 EXT4_INODES_PER_GROUP(sb));
3839 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3840 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3842 bh = sb_getblk(sb, block);
3845 if (!buffer_uptodate(bh)) {
3849 * If the buffer has the write error flag, we have failed
3850 * to write out another inode in the same block. In this
3851 * case, we don't have to read the block because we may
3852 * read the old inode data successfully.
3854 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3855 set_buffer_uptodate(bh);
3857 if (buffer_uptodate(bh)) {
3858 /* someone brought it uptodate while we waited */
3864 * If we have all information of the inode in memory and this
3865 * is the only valid inode in the block, we need not read the
3869 struct buffer_head *bitmap_bh;
3872 start = inode_offset & ~(inodes_per_block - 1);
3874 /* Is the inode bitmap in cache? */
3875 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3876 if (unlikely(!bitmap_bh))
3880 * If the inode bitmap isn't in cache then the
3881 * optimisation may end up performing two reads instead
3882 * of one, so skip it.
3884 if (!buffer_uptodate(bitmap_bh)) {
3888 for (i = start; i < start + inodes_per_block; i++) {
3889 if (i == inode_offset)
3891 if (ext4_test_bit(i, bitmap_bh->b_data))
3895 if (i == start + inodes_per_block) {
3896 /* all other inodes are free, so skip I/O */
3897 memset(bh->b_data, 0, bh->b_size);
3898 set_buffer_uptodate(bh);
3906 * If we need to do any I/O, try to pre-readahead extra
3907 * blocks from the inode table.
3909 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3910 ext4_fsblk_t b, end, table;
3912 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3914 table = ext4_inode_table(sb, gdp);
3915 /* s_inode_readahead_blks is always a power of 2 */
3916 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3920 num = EXT4_INODES_PER_GROUP(sb);
3921 if (ext4_has_group_desc_csum(sb))
3922 num -= ext4_itable_unused_count(sb, gdp);
3923 table += num / inodes_per_block;
3927 sb_breadahead(sb, b++);
3931 * There are other valid inodes in the buffer, this inode
3932 * has in-inode xattrs, or we don't have this inode in memory.
3933 * Read the block from disk.
3935 trace_ext4_load_inode(inode);
3937 bh->b_end_io = end_buffer_read_sync;
3938 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3940 if (!buffer_uptodate(bh)) {
3941 EXT4_ERROR_INODE_BLOCK(inode, block,
3942 "unable to read itable block");
3952 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3954 /* We have all inode data except xattrs in memory here. */
3955 return __ext4_get_inode_loc(inode, iloc,
3956 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3959 void ext4_set_inode_flags(struct inode *inode)
3961 unsigned int flags = EXT4_I(inode)->i_flags;
3962 unsigned int new_fl = 0;
3964 if (flags & EXT4_SYNC_FL)
3966 if (flags & EXT4_APPEND_FL)
3968 if (flags & EXT4_IMMUTABLE_FL)
3969 new_fl |= S_IMMUTABLE;
3970 if (flags & EXT4_NOATIME_FL)
3971 new_fl |= S_NOATIME;
3972 if (flags & EXT4_DIRSYNC_FL)
3973 new_fl |= S_DIRSYNC;
3974 set_mask_bits(&inode->i_flags,
3975 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC, new_fl);
3978 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3979 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3981 unsigned int vfs_fl;
3982 unsigned long old_fl, new_fl;
3985 vfs_fl = ei->vfs_inode.i_flags;
3986 old_fl = ei->i_flags;
3987 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3988 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3990 if (vfs_fl & S_SYNC)
3991 new_fl |= EXT4_SYNC_FL;
3992 if (vfs_fl & S_APPEND)
3993 new_fl |= EXT4_APPEND_FL;
3994 if (vfs_fl & S_IMMUTABLE)
3995 new_fl |= EXT4_IMMUTABLE_FL;
3996 if (vfs_fl & S_NOATIME)
3997 new_fl |= EXT4_NOATIME_FL;
3998 if (vfs_fl & S_DIRSYNC)
3999 new_fl |= EXT4_DIRSYNC_FL;
4000 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4003 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4004 struct ext4_inode_info *ei)
4007 struct inode *inode = &(ei->vfs_inode);
4008 struct super_block *sb = inode->i_sb;
4010 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4011 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4012 /* we are using combined 48 bit field */
4013 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4014 le32_to_cpu(raw_inode->i_blocks_lo);
4015 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4016 /* i_blocks represent file system block size */
4017 return i_blocks << (inode->i_blkbits - 9);
4022 return le32_to_cpu(raw_inode->i_blocks_lo);
4026 static inline void ext4_iget_extra_inode(struct inode *inode,
4027 struct ext4_inode *raw_inode,
4028 struct ext4_inode_info *ei)
4030 __le32 *magic = (void *)raw_inode +
4031 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4032 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4033 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4034 ext4_find_inline_data_nolock(inode);
4036 EXT4_I(inode)->i_inline_off = 0;
4039 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4041 struct ext4_iloc iloc;
4042 struct ext4_inode *raw_inode;
4043 struct ext4_inode_info *ei;
4044 struct inode *inode;
4045 journal_t *journal = EXT4_SB(sb)->s_journal;
4051 inode = iget_locked(sb, ino);
4053 return ERR_PTR(-ENOMEM);
4054 if (!(inode->i_state & I_NEW))
4060 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4063 raw_inode = ext4_raw_inode(&iloc);
4065 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4066 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4067 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4068 EXT4_INODE_SIZE(inode->i_sb)) {
4069 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4070 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4071 EXT4_INODE_SIZE(inode->i_sb));
4076 ei->i_extra_isize = 0;
4078 /* Precompute checksum seed for inode metadata */
4079 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4080 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4081 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4083 __le32 inum = cpu_to_le32(inode->i_ino);
4084 __le32 gen = raw_inode->i_generation;
4085 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4087 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4091 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4092 EXT4_ERROR_INODE(inode, "checksum invalid");
4097 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4098 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4099 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4100 if (!(test_opt(inode->i_sb, NO_UID32))) {
4101 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4102 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4104 i_uid_write(inode, i_uid);
4105 i_gid_write(inode, i_gid);
4106 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4108 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4109 ei->i_inline_off = 0;
4110 ei->i_dir_start_lookup = 0;
4111 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4112 /* We now have enough fields to check if the inode was active or not.
4113 * This is needed because nfsd might try to access dead inodes
4114 * the test is that same one that e2fsck uses
4115 * NeilBrown 1999oct15
4117 if (inode->i_nlink == 0) {
4118 if ((inode->i_mode == 0 ||
4119 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4120 ino != EXT4_BOOT_LOADER_INO) {
4121 /* this inode is deleted */
4125 /* The only unlinked inodes we let through here have
4126 * valid i_mode and are being read by the orphan
4127 * recovery code: that's fine, we're about to complete
4128 * the process of deleting those.
4129 * OR it is the EXT4_BOOT_LOADER_INO which is
4130 * not initialized on a new filesystem. */
4132 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4133 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4134 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4135 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4137 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4138 inode->i_size = ext4_isize(raw_inode);
4139 ei->i_disksize = inode->i_size;
4141 ei->i_reserved_quota = 0;
4143 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4144 ei->i_block_group = iloc.block_group;
4145 ei->i_last_alloc_group = ~0;
4147 * NOTE! The in-memory inode i_data array is in little-endian order
4148 * even on big-endian machines: we do NOT byteswap the block numbers!
4150 for (block = 0; block < EXT4_N_BLOCKS; block++)
4151 ei->i_data[block] = raw_inode->i_block[block];
4152 INIT_LIST_HEAD(&ei->i_orphan);
4155 * Set transaction id's of transactions that have to be committed
4156 * to finish f[data]sync. We set them to currently running transaction
4157 * as we cannot be sure that the inode or some of its metadata isn't
4158 * part of the transaction - the inode could have been reclaimed and
4159 * now it is reread from disk.
4162 transaction_t *transaction;
4165 read_lock(&journal->j_state_lock);
4166 if (journal->j_running_transaction)
4167 transaction = journal->j_running_transaction;
4169 transaction = journal->j_committing_transaction;
4171 tid = transaction->t_tid;
4173 tid = journal->j_commit_sequence;
4174 read_unlock(&journal->j_state_lock);
4175 ei->i_sync_tid = tid;
4176 ei->i_datasync_tid = tid;
4179 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4180 if (ei->i_extra_isize == 0) {
4181 /* The extra space is currently unused. Use it. */
4182 ei->i_extra_isize = sizeof(struct ext4_inode) -
4183 EXT4_GOOD_OLD_INODE_SIZE;
4185 ext4_iget_extra_inode(inode, raw_inode, ei);
4189 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4190 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4191 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4192 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4194 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4195 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4196 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4198 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4202 if (ei->i_file_acl &&
4203 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4204 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4208 } else if (!ext4_has_inline_data(inode)) {
4209 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4210 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4211 (S_ISLNK(inode->i_mode) &&
4212 !ext4_inode_is_fast_symlink(inode))))
4213 /* Validate extent which is part of inode */
4214 ret = ext4_ext_check_inode(inode);
4215 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4216 (S_ISLNK(inode->i_mode) &&
4217 !ext4_inode_is_fast_symlink(inode))) {
4218 /* Validate block references which are part of inode */
4219 ret = ext4_ind_check_inode(inode);
4225 if (S_ISREG(inode->i_mode)) {
4226 inode->i_op = &ext4_file_inode_operations;
4227 inode->i_fop = &ext4_file_operations;
4228 ext4_set_aops(inode);
4229 } else if (S_ISDIR(inode->i_mode)) {
4230 inode->i_op = &ext4_dir_inode_operations;
4231 inode->i_fop = &ext4_dir_operations;
4232 } else if (S_ISLNK(inode->i_mode)) {
4233 if (ext4_inode_is_fast_symlink(inode)) {
4234 inode->i_op = &ext4_fast_symlink_inode_operations;
4235 nd_terminate_link(ei->i_data, inode->i_size,
4236 sizeof(ei->i_data) - 1);
4238 inode->i_op = &ext4_symlink_inode_operations;
4239 ext4_set_aops(inode);
4241 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4242 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4243 inode->i_op = &ext4_special_inode_operations;
4244 if (raw_inode->i_block[0])
4245 init_special_inode(inode, inode->i_mode,
4246 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4248 init_special_inode(inode, inode->i_mode,
4249 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4250 } else if (ino == EXT4_BOOT_LOADER_INO) {
4251 make_bad_inode(inode);
4254 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4258 ext4_set_inode_flags(inode);
4259 unlock_new_inode(inode);
4265 return ERR_PTR(ret);
4268 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4270 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4271 return ERR_PTR(-EIO);
4272 return ext4_iget(sb, ino);
4275 static int ext4_inode_blocks_set(handle_t *handle,
4276 struct ext4_inode *raw_inode,
4277 struct ext4_inode_info *ei)
4279 struct inode *inode = &(ei->vfs_inode);
4280 u64 i_blocks = inode->i_blocks;
4281 struct super_block *sb = inode->i_sb;
4283 if (i_blocks <= ~0U) {
4285 * i_blocks can be represented in a 32 bit variable
4286 * as multiple of 512 bytes
4288 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4289 raw_inode->i_blocks_high = 0;
4290 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4293 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4296 if (i_blocks <= 0xffffffffffffULL) {
4298 * i_blocks can be represented in a 48 bit variable
4299 * as multiple of 512 bytes
4301 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4302 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4303 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4305 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4306 /* i_block is stored in file system block size */
4307 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4308 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4309 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4315 * Post the struct inode info into an on-disk inode location in the
4316 * buffer-cache. This gobbles the caller's reference to the
4317 * buffer_head in the inode location struct.
4319 * The caller must have write access to iloc->bh.
4321 static int ext4_do_update_inode(handle_t *handle,
4322 struct inode *inode,
4323 struct ext4_iloc *iloc)
4325 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4326 struct ext4_inode_info *ei = EXT4_I(inode);
4327 struct buffer_head *bh = iloc->bh;
4328 int err = 0, rc, block;
4329 int need_datasync = 0;
4333 /* For fields not not tracking in the in-memory inode,
4334 * initialise them to zero for new inodes. */
4335 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4336 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4338 ext4_get_inode_flags(ei);
4339 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4340 i_uid = i_uid_read(inode);
4341 i_gid = i_gid_read(inode);
4342 if (!(test_opt(inode->i_sb, NO_UID32))) {
4343 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4344 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4346 * Fix up interoperability with old kernels. Otherwise, old inodes get
4347 * re-used with the upper 16 bits of the uid/gid intact
4350 raw_inode->i_uid_high =
4351 cpu_to_le16(high_16_bits(i_uid));
4352 raw_inode->i_gid_high =
4353 cpu_to_le16(high_16_bits(i_gid));
4355 raw_inode->i_uid_high = 0;
4356 raw_inode->i_gid_high = 0;
4359 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4360 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4361 raw_inode->i_uid_high = 0;
4362 raw_inode->i_gid_high = 0;
4364 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4366 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4367 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4368 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4369 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4371 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4373 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4374 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4375 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4376 cpu_to_le32(EXT4_OS_HURD))
4377 raw_inode->i_file_acl_high =
4378 cpu_to_le16(ei->i_file_acl >> 32);
4379 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4380 if (ei->i_disksize != ext4_isize(raw_inode)) {
4381 ext4_isize_set(raw_inode, ei->i_disksize);
4384 if (ei->i_disksize > 0x7fffffffULL) {
4385 struct super_block *sb = inode->i_sb;
4386 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4387 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4388 EXT4_SB(sb)->s_es->s_rev_level ==
4389 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4390 /* If this is the first large file
4391 * created, add a flag to the superblock.
4393 err = ext4_journal_get_write_access(handle,
4394 EXT4_SB(sb)->s_sbh);
4397 ext4_update_dynamic_rev(sb);
4398 EXT4_SET_RO_COMPAT_FEATURE(sb,
4399 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4400 ext4_handle_sync(handle);
4401 err = ext4_handle_dirty_super(handle, sb);
4404 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4405 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4406 if (old_valid_dev(inode->i_rdev)) {
4407 raw_inode->i_block[0] =
4408 cpu_to_le32(old_encode_dev(inode->i_rdev));
4409 raw_inode->i_block[1] = 0;
4411 raw_inode->i_block[0] = 0;
4412 raw_inode->i_block[1] =
4413 cpu_to_le32(new_encode_dev(inode->i_rdev));
4414 raw_inode->i_block[2] = 0;
4416 } else if (!ext4_has_inline_data(inode)) {
4417 for (block = 0; block < EXT4_N_BLOCKS; block++)
4418 raw_inode->i_block[block] = ei->i_data[block];
4421 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4422 if (ei->i_extra_isize) {
4423 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4424 raw_inode->i_version_hi =
4425 cpu_to_le32(inode->i_version >> 32);
4426 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4429 ext4_inode_csum_set(inode, raw_inode, ei);
4431 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4432 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4435 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4437 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4440 ext4_std_error(inode->i_sb, err);
4445 * ext4_write_inode()
4447 * We are called from a few places:
4449 * - Within generic_file_write() for O_SYNC files.
4450 * Here, there will be no transaction running. We wait for any running
4451 * transaction to commit.
4453 * - Within sys_sync(), kupdate and such.
4454 * We wait on commit, if tol to.
4456 * - Within prune_icache() (PF_MEMALLOC == true)
4457 * Here we simply return. We can't afford to block kswapd on the
4460 * In all cases it is actually safe for us to return without doing anything,
4461 * because the inode has been copied into a raw inode buffer in
4462 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4465 * Note that we are absolutely dependent upon all inode dirtiers doing the
4466 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4467 * which we are interested.
4469 * It would be a bug for them to not do this. The code:
4471 * mark_inode_dirty(inode)
4473 * inode->i_size = expr;
4475 * is in error because a kswapd-driven write_inode() could occur while
4476 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4477 * will no longer be on the superblock's dirty inode list.
4479 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4483 if (current->flags & PF_MEMALLOC)
4486 if (EXT4_SB(inode->i_sb)->s_journal) {
4487 if (ext4_journal_current_handle()) {
4488 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4493 if (wbc->sync_mode != WB_SYNC_ALL)
4496 err = ext4_force_commit(inode->i_sb);
4498 struct ext4_iloc iloc;
4500 err = __ext4_get_inode_loc(inode, &iloc, 0);
4503 if (wbc->sync_mode == WB_SYNC_ALL)
4504 sync_dirty_buffer(iloc.bh);
4505 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4506 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4507 "IO error syncing inode");
4516 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4517 * buffers that are attached to a page stradding i_size and are undergoing
4518 * commit. In that case we have to wait for commit to finish and try again.
4520 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4524 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4525 tid_t commit_tid = 0;
4528 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4530 * All buffers in the last page remain valid? Then there's nothing to
4531 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4534 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4537 page = find_lock_page(inode->i_mapping,
4538 inode->i_size >> PAGE_CACHE_SHIFT);
4541 ret = __ext4_journalled_invalidatepage(page, offset,
4542 PAGE_CACHE_SIZE - offset);
4544 page_cache_release(page);
4548 read_lock(&journal->j_state_lock);
4549 if (journal->j_committing_transaction)
4550 commit_tid = journal->j_committing_transaction->t_tid;
4551 read_unlock(&journal->j_state_lock);
4553 jbd2_log_wait_commit(journal, commit_tid);
4560 * Called from notify_change.
4562 * We want to trap VFS attempts to truncate the file as soon as
4563 * possible. In particular, we want to make sure that when the VFS
4564 * shrinks i_size, we put the inode on the orphan list and modify
4565 * i_disksize immediately, so that during the subsequent flushing of
4566 * dirty pages and freeing of disk blocks, we can guarantee that any
4567 * commit will leave the blocks being flushed in an unused state on
4568 * disk. (On recovery, the inode will get truncated and the blocks will
4569 * be freed, so we have a strong guarantee that no future commit will
4570 * leave these blocks visible to the user.)
4572 * Another thing we have to assure is that if we are in ordered mode
4573 * and inode is still attached to the committing transaction, we must
4574 * we start writeout of all the dirty pages which are being truncated.
4575 * This way we are sure that all the data written in the previous
4576 * transaction are already on disk (truncate waits for pages under
4579 * Called with inode->i_mutex down.
4581 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4583 struct inode *inode = dentry->d_inode;
4586 const unsigned int ia_valid = attr->ia_valid;
4588 error = inode_change_ok(inode, attr);
4592 if (is_quota_modification(inode, attr))
4593 dquot_initialize(inode);
4594 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4595 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4598 /* (user+group)*(old+new) structure, inode write (sb,
4599 * inode block, ? - but truncate inode update has it) */
4600 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4601 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4602 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4603 if (IS_ERR(handle)) {
4604 error = PTR_ERR(handle);
4607 error = dquot_transfer(inode, attr);
4609 ext4_journal_stop(handle);
4612 /* Update corresponding info in inode so that everything is in
4613 * one transaction */
4614 if (attr->ia_valid & ATTR_UID)
4615 inode->i_uid = attr->ia_uid;
4616 if (attr->ia_valid & ATTR_GID)
4617 inode->i_gid = attr->ia_gid;
4618 error = ext4_mark_inode_dirty(handle, inode);
4619 ext4_journal_stop(handle);
4622 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4625 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4626 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4628 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4632 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4633 inode_inc_iversion(inode);
4635 if (S_ISREG(inode->i_mode) &&
4636 (attr->ia_size < inode->i_size)) {
4637 if (ext4_should_order_data(inode)) {
4638 error = ext4_begin_ordered_truncate(inode,
4643 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4644 if (IS_ERR(handle)) {
4645 error = PTR_ERR(handle);
4648 if (ext4_handle_valid(handle)) {
4649 error = ext4_orphan_add(handle, inode);
4652 down_write(&EXT4_I(inode)->i_data_sem);
4653 EXT4_I(inode)->i_disksize = attr->ia_size;
4654 rc = ext4_mark_inode_dirty(handle, inode);
4658 * We have to update i_size under i_data_sem together
4659 * with i_disksize to avoid races with writeback code
4660 * running ext4_wb_update_i_disksize().
4663 i_size_write(inode, attr->ia_size);
4664 up_write(&EXT4_I(inode)->i_data_sem);
4665 ext4_journal_stop(handle);
4667 ext4_orphan_del(NULL, inode);
4671 loff_t oldsize = inode->i_size;
4673 i_size_write(inode, attr->ia_size);
4674 pagecache_isize_extended(inode, oldsize, inode->i_size);
4678 * Blocks are going to be removed from the inode. Wait
4679 * for dio in flight. Temporarily disable
4680 * dioread_nolock to prevent livelock.
4683 if (!ext4_should_journal_data(inode)) {
4684 ext4_inode_block_unlocked_dio(inode);
4685 inode_dio_wait(inode);
4686 ext4_inode_resume_unlocked_dio(inode);
4688 ext4_wait_for_tail_page_commit(inode);
4691 * Truncate pagecache after we've waited for commit
4692 * in data=journal mode to make pages freeable.
4694 truncate_pagecache(inode, inode->i_size);
4697 * We want to call ext4_truncate() even if attr->ia_size ==
4698 * inode->i_size for cases like truncation of fallocated space
4700 if (attr->ia_valid & ATTR_SIZE)
4701 ext4_truncate(inode);
4704 setattr_copy(inode, attr);
4705 mark_inode_dirty(inode);
4709 * If the call to ext4_truncate failed to get a transaction handle at
4710 * all, we need to clean up the in-core orphan list manually.
4712 if (orphan && inode->i_nlink)
4713 ext4_orphan_del(NULL, inode);
4715 if (!rc && (ia_valid & ATTR_MODE))
4716 rc = posix_acl_chmod(inode, inode->i_mode);
4719 ext4_std_error(inode->i_sb, error);
4725 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4728 struct inode *inode;
4729 unsigned long long delalloc_blocks;
4731 inode = dentry->d_inode;
4732 generic_fillattr(inode, stat);
4735 * If there is inline data in the inode, the inode will normally not
4736 * have data blocks allocated (it may have an external xattr block).
4737 * Report at least one sector for such files, so tools like tar, rsync,
4738 * others doen't incorrectly think the file is completely sparse.
4740 if (unlikely(ext4_has_inline_data(inode)))
4741 stat->blocks += (stat->size + 511) >> 9;
4744 * We can't update i_blocks if the block allocation is delayed
4745 * otherwise in the case of system crash before the real block
4746 * allocation is done, we will have i_blocks inconsistent with
4747 * on-disk file blocks.
4748 * We always keep i_blocks updated together with real
4749 * allocation. But to not confuse with user, stat
4750 * will return the blocks that include the delayed allocation
4751 * blocks for this file.
4753 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4754 EXT4_I(inode)->i_reserved_data_blocks);
4755 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4759 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4762 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4763 return ext4_ind_trans_blocks(inode, lblocks);
4764 return ext4_ext_index_trans_blocks(inode, pextents);
4768 * Account for index blocks, block groups bitmaps and block group
4769 * descriptor blocks if modify datablocks and index blocks
4770 * worse case, the indexs blocks spread over different block groups
4772 * If datablocks are discontiguous, they are possible to spread over
4773 * different block groups too. If they are contiguous, with flexbg,
4774 * they could still across block group boundary.
4776 * Also account for superblock, inode, quota and xattr blocks
4778 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4781 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4787 * How many index blocks need to touch to map @lblocks logical blocks
4788 * to @pextents physical extents?
4790 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4795 * Now let's see how many group bitmaps and group descriptors need
4798 groups = idxblocks + pextents;
4800 if (groups > ngroups)
4802 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4803 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4805 /* bitmaps and block group descriptor blocks */
4806 ret += groups + gdpblocks;
4808 /* Blocks for super block, inode, quota and xattr blocks */
4809 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4815 * Calculate the total number of credits to reserve to fit
4816 * the modification of a single pages into a single transaction,
4817 * which may include multiple chunks of block allocations.
4819 * This could be called via ext4_write_begin()
4821 * We need to consider the worse case, when
4822 * one new block per extent.
4824 int ext4_writepage_trans_blocks(struct inode *inode)
4826 int bpp = ext4_journal_blocks_per_page(inode);
4829 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4831 /* Account for data blocks for journalled mode */
4832 if (ext4_should_journal_data(inode))
4838 * Calculate the journal credits for a chunk of data modification.
4840 * This is called from DIO, fallocate or whoever calling
4841 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4843 * journal buffers for data blocks are not included here, as DIO
4844 * and fallocate do no need to journal data buffers.
4846 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4848 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4852 * The caller must have previously called ext4_reserve_inode_write().
4853 * Give this, we know that the caller already has write access to iloc->bh.
4855 int ext4_mark_iloc_dirty(handle_t *handle,
4856 struct inode *inode, struct ext4_iloc *iloc)
4860 if (IS_I_VERSION(inode))
4861 inode_inc_iversion(inode);
4863 /* the do_update_inode consumes one bh->b_count */
4866 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4867 err = ext4_do_update_inode(handle, inode, iloc);
4873 * On success, We end up with an outstanding reference count against
4874 * iloc->bh. This _must_ be cleaned up later.
4878 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4879 struct ext4_iloc *iloc)
4883 err = ext4_get_inode_loc(inode, iloc);
4885 BUFFER_TRACE(iloc->bh, "get_write_access");
4886 err = ext4_journal_get_write_access(handle, iloc->bh);
4892 ext4_std_error(inode->i_sb, err);
4897 * Expand an inode by new_extra_isize bytes.
4898 * Returns 0 on success or negative error number on failure.
4900 static int ext4_expand_extra_isize(struct inode *inode,
4901 unsigned int new_extra_isize,
4902 struct ext4_iloc iloc,
4905 struct ext4_inode *raw_inode;
4906 struct ext4_xattr_ibody_header *header;
4908 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4911 raw_inode = ext4_raw_inode(&iloc);
4913 header = IHDR(inode, raw_inode);
4915 /* No extended attributes present */
4916 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4917 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4918 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4920 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4924 /* try to expand with EAs present */
4925 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4930 * What we do here is to mark the in-core inode as clean with respect to inode
4931 * dirtiness (it may still be data-dirty).
4932 * This means that the in-core inode may be reaped by prune_icache
4933 * without having to perform any I/O. This is a very good thing,
4934 * because *any* task may call prune_icache - even ones which
4935 * have a transaction open against a different journal.
4937 * Is this cheating? Not really. Sure, we haven't written the
4938 * inode out, but prune_icache isn't a user-visible syncing function.
4939 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4940 * we start and wait on commits.
4942 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4944 struct ext4_iloc iloc;
4945 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4946 static unsigned int mnt_count;
4950 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4951 err = ext4_reserve_inode_write(handle, inode, &iloc);
4952 if (ext4_handle_valid(handle) &&
4953 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4954 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4956 * We need extra buffer credits since we may write into EA block
4957 * with this same handle. If journal_extend fails, then it will
4958 * only result in a minor loss of functionality for that inode.
4959 * If this is felt to be critical, then e2fsck should be run to
4960 * force a large enough s_min_extra_isize.
4962 if ((jbd2_journal_extend(handle,
4963 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4964 ret = ext4_expand_extra_isize(inode,
4965 sbi->s_want_extra_isize,
4968 ext4_set_inode_state(inode,
4969 EXT4_STATE_NO_EXPAND);
4971 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4972 ext4_warning(inode->i_sb,
4973 "Unable to expand inode %lu. Delete"
4974 " some EAs or run e2fsck.",
4977 le16_to_cpu(sbi->s_es->s_mnt_count);
4983 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4988 * ext4_dirty_inode() is called from __mark_inode_dirty()
4990 * We're really interested in the case where a file is being extended.
4991 * i_size has been changed by generic_commit_write() and we thus need
4992 * to include the updated inode in the current transaction.
4994 * Also, dquot_alloc_block() will always dirty the inode when blocks
4995 * are allocated to the file.
4997 * If the inode is marked synchronous, we don't honour that here - doing
4998 * so would cause a commit on atime updates, which we don't bother doing.
4999 * We handle synchronous inodes at the highest possible level.
5001 void ext4_dirty_inode(struct inode *inode, int flags)
5005 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5009 ext4_mark_inode_dirty(handle, inode);
5011 ext4_journal_stop(handle);
5018 * Bind an inode's backing buffer_head into this transaction, to prevent
5019 * it from being flushed to disk early. Unlike
5020 * ext4_reserve_inode_write, this leaves behind no bh reference and
5021 * returns no iloc structure, so the caller needs to repeat the iloc
5022 * lookup to mark the inode dirty later.
5024 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5026 struct ext4_iloc iloc;
5030 err = ext4_get_inode_loc(inode, &iloc);
5032 BUFFER_TRACE(iloc.bh, "get_write_access");
5033 err = jbd2_journal_get_write_access(handle, iloc.bh);
5035 err = ext4_handle_dirty_metadata(handle,
5041 ext4_std_error(inode->i_sb, err);
5046 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5053 * We have to be very careful here: changing a data block's
5054 * journaling status dynamically is dangerous. If we write a
5055 * data block to the journal, change the status and then delete
5056 * that block, we risk forgetting to revoke the old log record
5057 * from the journal and so a subsequent replay can corrupt data.
5058 * So, first we make sure that the journal is empty and that
5059 * nobody is changing anything.
5062 journal = EXT4_JOURNAL(inode);
5065 if (is_journal_aborted(journal))
5067 /* We have to allocate physical blocks for delalloc blocks
5068 * before flushing journal. otherwise delalloc blocks can not
5069 * be allocated any more. even more truncate on delalloc blocks
5070 * could trigger BUG by flushing delalloc blocks in journal.
5071 * There is no delalloc block in non-journal data mode.
5073 if (val && test_opt(inode->i_sb, DELALLOC)) {
5074 err = ext4_alloc_da_blocks(inode);
5079 /* Wait for all existing dio workers */
5080 ext4_inode_block_unlocked_dio(inode);
5081 inode_dio_wait(inode);
5083 jbd2_journal_lock_updates(journal);
5086 * OK, there are no updates running now, and all cached data is
5087 * synced to disk. We are now in a completely consistent state
5088 * which doesn't have anything in the journal, and we know that
5089 * no filesystem updates are running, so it is safe to modify
5090 * the inode's in-core data-journaling state flag now.
5094 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5096 jbd2_journal_flush(journal);
5097 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5099 ext4_set_aops(inode);
5101 jbd2_journal_unlock_updates(journal);
5102 ext4_inode_resume_unlocked_dio(inode);
5104 /* Finally we can mark the inode as dirty. */
5106 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5108 return PTR_ERR(handle);
5110 err = ext4_mark_inode_dirty(handle, inode);
5111 ext4_handle_sync(handle);
5112 ext4_journal_stop(handle);
5113 ext4_std_error(inode->i_sb, err);
5118 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5120 return !buffer_mapped(bh);
5123 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5125 struct page *page = vmf->page;
5129 struct file *file = vma->vm_file;
5130 struct inode *inode = file_inode(file);
5131 struct address_space *mapping = inode->i_mapping;
5133 get_block_t *get_block;
5136 sb_start_pagefault(inode->i_sb);
5137 file_update_time(vma->vm_file);
5138 /* Delalloc case is easy... */
5139 if (test_opt(inode->i_sb, DELALLOC) &&
5140 !ext4_should_journal_data(inode) &&
5141 !ext4_nonda_switch(inode->i_sb)) {
5143 ret = __block_page_mkwrite(vma, vmf,
5144 ext4_da_get_block_prep);
5145 } while (ret == -ENOSPC &&
5146 ext4_should_retry_alloc(inode->i_sb, &retries));
5151 size = i_size_read(inode);
5152 /* Page got truncated from under us? */
5153 if (page->mapping != mapping || page_offset(page) > size) {
5155 ret = VM_FAULT_NOPAGE;
5159 if (page->index == size >> PAGE_CACHE_SHIFT)
5160 len = size & ~PAGE_CACHE_MASK;
5162 len = PAGE_CACHE_SIZE;
5164 * Return if we have all the buffers mapped. This avoids the need to do
5165 * journal_start/journal_stop which can block and take a long time
5167 if (page_has_buffers(page)) {
5168 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5170 ext4_bh_unmapped)) {
5171 /* Wait so that we don't change page under IO */
5172 wait_for_stable_page(page);
5173 ret = VM_FAULT_LOCKED;
5178 /* OK, we need to fill the hole... */
5179 if (ext4_should_dioread_nolock(inode))
5180 get_block = ext4_get_block_write;
5182 get_block = ext4_get_block;
5184 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5185 ext4_writepage_trans_blocks(inode));
5186 if (IS_ERR(handle)) {
5187 ret = VM_FAULT_SIGBUS;
5190 ret = __block_page_mkwrite(vma, vmf, get_block);
5191 if (!ret && ext4_should_journal_data(inode)) {
5192 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5193 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5195 ret = VM_FAULT_SIGBUS;
5196 ext4_journal_stop(handle);
5199 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5201 ext4_journal_stop(handle);
5202 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5205 ret = block_page_mkwrite_return(ret);
5207 sb_end_pagefault(inode->i_sb);