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_final(&inode->i_data);
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_final(&inode->i_data);
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 allocated.
493 * if create==0 and the blocks are pre-allocated and unwritten 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;
508 #ifdef ES_AGGRESSIVE_TEST
509 struct ext4_map_blocks orig_map;
511 memcpy(&orig_map, map, sizeof(*map));
515 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
516 "logical block %lu\n", inode->i_ino, flags, map->m_len,
517 (unsigned long) map->m_lblk);
520 * ext4_map_blocks returns an int, and m_len is an unsigned int
522 if (unlikely(map->m_len > INT_MAX))
523 map->m_len = INT_MAX;
525 /* We can handle the block number less than EXT_MAX_BLOCKS */
526 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
529 /* Lookup extent status tree firstly */
530 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
531 ext4_es_lru_add(inode);
532 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
533 map->m_pblk = ext4_es_pblock(&es) +
534 map->m_lblk - es.es_lblk;
535 map->m_flags |= ext4_es_is_written(&es) ?
536 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
537 retval = es.es_len - (map->m_lblk - es.es_lblk);
538 if (retval > map->m_len)
541 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
546 #ifdef ES_AGGRESSIVE_TEST
547 ext4_map_blocks_es_recheck(handle, inode, map,
554 * Try to see if we can get the block without requesting a new
557 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
558 down_read((&EXT4_I(inode)->i_data_sem));
559 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
560 retval = ext4_ext_map_blocks(handle, inode, map, flags &
561 EXT4_GET_BLOCKS_KEEP_SIZE);
563 retval = ext4_ind_map_blocks(handle, inode, map, flags &
564 EXT4_GET_BLOCKS_KEEP_SIZE);
569 if (unlikely(retval != map->m_len)) {
570 ext4_warning(inode->i_sb,
571 "ES len assertion failed for inode "
572 "%lu: retval %d != map->m_len %d",
573 inode->i_ino, retval, map->m_len);
577 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
578 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
579 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
580 ext4_find_delalloc_range(inode, map->m_lblk,
581 map->m_lblk + map->m_len - 1))
582 status |= EXTENT_STATUS_DELAYED;
583 ret = ext4_es_insert_extent(inode, map->m_lblk,
584 map->m_len, map->m_pblk, status);
588 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
589 up_read((&EXT4_I(inode)->i_data_sem));
592 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
593 ret = check_block_validity(inode, map);
598 /* If it is only a block(s) look up */
599 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
603 * Returns if the blocks have already allocated
605 * Note that if blocks have been preallocated
606 * ext4_ext_get_block() returns the create = 0
607 * with buffer head unmapped.
609 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
611 * If we need to convert extent to unwritten
612 * we continue and do the actual work in
613 * ext4_ext_map_blocks()
615 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
619 * Here we clear m_flags because after allocating an new extent,
620 * it will be set again.
622 map->m_flags &= ~EXT4_MAP_FLAGS;
625 * New blocks allocate and/or writing to unwritten extent
626 * will possibly result in updating i_data, so we take
627 * the write lock of i_data_sem, and call get_blocks()
628 * with create == 1 flag.
630 down_write((&EXT4_I(inode)->i_data_sem));
633 * if the caller is from delayed allocation writeout path
634 * we have already reserved fs blocks for allocation
635 * let the underlying get_block() function know to
636 * avoid double accounting
638 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
639 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
641 * We need to check for EXT4 here because migrate
642 * could have changed the inode type in between
644 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
645 retval = ext4_ext_map_blocks(handle, inode, map, flags);
647 retval = ext4_ind_map_blocks(handle, inode, map, flags);
649 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
651 * We allocated new blocks which will result in
652 * i_data's format changing. Force the migrate
653 * to fail by clearing migrate flags
655 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
659 * Update reserved blocks/metadata blocks after successful
660 * block allocation which had been deferred till now. We don't
661 * support fallocate for non extent files. So we can update
662 * reserve space here.
665 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
666 ext4_da_update_reserve_space(inode, retval, 1);
668 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
669 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
674 if (unlikely(retval != map->m_len)) {
675 ext4_warning(inode->i_sb,
676 "ES len assertion failed for inode "
677 "%lu: retval %d != map->m_len %d",
678 inode->i_ino, retval, map->m_len);
683 * If the extent has been zeroed out, we don't need to update
684 * extent status tree.
686 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
687 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
688 if (ext4_es_is_written(&es))
691 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
692 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
693 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
694 ext4_find_delalloc_range(inode, map->m_lblk,
695 map->m_lblk + map->m_len - 1))
696 status |= EXTENT_STATUS_DELAYED;
697 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
698 map->m_pblk, status);
704 up_write((&EXT4_I(inode)->i_data_sem));
705 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
706 ret = check_block_validity(inode, map);
713 /* Maximum number of blocks we map for direct IO at once. */
714 #define DIO_MAX_BLOCKS 4096
716 static int _ext4_get_block(struct inode *inode, sector_t iblock,
717 struct buffer_head *bh, int flags)
719 handle_t *handle = ext4_journal_current_handle();
720 struct ext4_map_blocks map;
721 int ret = 0, started = 0;
724 if (ext4_has_inline_data(inode))
728 map.m_len = bh->b_size >> inode->i_blkbits;
730 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
731 /* Direct IO write... */
732 if (map.m_len > DIO_MAX_BLOCKS)
733 map.m_len = DIO_MAX_BLOCKS;
734 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
735 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
737 if (IS_ERR(handle)) {
738 ret = PTR_ERR(handle);
744 ret = ext4_map_blocks(handle, inode, &map, flags);
746 ext4_io_end_t *io_end = ext4_inode_aio(inode);
748 map_bh(bh, inode->i_sb, map.m_pblk);
749 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
750 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
751 set_buffer_defer_completion(bh);
752 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
756 ext4_journal_stop(handle);
760 int ext4_get_block(struct inode *inode, sector_t iblock,
761 struct buffer_head *bh, int create)
763 return _ext4_get_block(inode, iblock, bh,
764 create ? EXT4_GET_BLOCKS_CREATE : 0);
768 * `handle' can be NULL if create is zero
770 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
771 ext4_lblk_t block, int create, int *errp)
773 struct ext4_map_blocks map;
774 struct buffer_head *bh;
777 J_ASSERT(handle != NULL || create == 0);
781 err = ext4_map_blocks(handle, inode, &map,
782 create ? EXT4_GET_BLOCKS_CREATE : 0);
784 /* ensure we send some value back into *errp */
787 if (create && err == 0)
788 err = -ENOSPC; /* should never happen */
794 bh = sb_getblk(inode->i_sb, map.m_pblk);
799 if (map.m_flags & EXT4_MAP_NEW) {
800 J_ASSERT(create != 0);
801 J_ASSERT(handle != NULL);
804 * Now that we do not always journal data, we should
805 * keep in mind whether this should always journal the
806 * new buffer as metadata. For now, regular file
807 * writes use ext4_get_block instead, so it's not a
811 BUFFER_TRACE(bh, "call get_create_access");
812 fatal = ext4_journal_get_create_access(handle, bh);
813 if (!fatal && !buffer_uptodate(bh)) {
814 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
815 set_buffer_uptodate(bh);
818 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
819 err = ext4_handle_dirty_metadata(handle, inode, bh);
823 BUFFER_TRACE(bh, "not a new buffer");
833 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
834 ext4_lblk_t block, int create, int *err)
836 struct buffer_head *bh;
838 bh = ext4_getblk(handle, inode, block, create, err);
841 if (buffer_uptodate(bh))
843 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
845 if (buffer_uptodate(bh))
852 int ext4_walk_page_buffers(handle_t *handle,
853 struct buffer_head *head,
857 int (*fn)(handle_t *handle,
858 struct buffer_head *bh))
860 struct buffer_head *bh;
861 unsigned block_start, block_end;
862 unsigned blocksize = head->b_size;
864 struct buffer_head *next;
866 for (bh = head, block_start = 0;
867 ret == 0 && (bh != head || !block_start);
868 block_start = block_end, bh = next) {
869 next = bh->b_this_page;
870 block_end = block_start + blocksize;
871 if (block_end <= from || block_start >= to) {
872 if (partial && !buffer_uptodate(bh))
876 err = (*fn)(handle, bh);
884 * To preserve ordering, it is essential that the hole instantiation and
885 * the data write be encapsulated in a single transaction. We cannot
886 * close off a transaction and start a new one between the ext4_get_block()
887 * and the commit_write(). So doing the jbd2_journal_start at the start of
888 * prepare_write() is the right place.
890 * Also, this function can nest inside ext4_writepage(). In that case, we
891 * *know* that ext4_writepage() has generated enough buffer credits to do the
892 * whole page. So we won't block on the journal in that case, which is good,
893 * because the caller may be PF_MEMALLOC.
895 * By accident, ext4 can be reentered when a transaction is open via
896 * quota file writes. If we were to commit the transaction while thus
897 * reentered, there can be a deadlock - we would be holding a quota
898 * lock, and the commit would never complete if another thread had a
899 * transaction open and was blocking on the quota lock - a ranking
902 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
903 * will _not_ run commit under these circumstances because handle->h_ref
904 * is elevated. We'll still have enough credits for the tiny quotafile
907 int do_journal_get_write_access(handle_t *handle,
908 struct buffer_head *bh)
910 int dirty = buffer_dirty(bh);
913 if (!buffer_mapped(bh) || buffer_freed(bh))
916 * __block_write_begin() could have dirtied some buffers. Clean
917 * the dirty bit as jbd2_journal_get_write_access() could complain
918 * otherwise about fs integrity issues. Setting of the dirty bit
919 * by __block_write_begin() isn't a real problem here as we clear
920 * the bit before releasing a page lock and thus writeback cannot
921 * ever write the buffer.
924 clear_buffer_dirty(bh);
925 ret = ext4_journal_get_write_access(handle, bh);
927 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
931 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
932 struct buffer_head *bh_result, int create);
933 static int ext4_write_begin(struct file *file, struct address_space *mapping,
934 loff_t pos, unsigned len, unsigned flags,
935 struct page **pagep, void **fsdata)
937 struct inode *inode = mapping->host;
938 int ret, needed_blocks;
945 trace_ext4_write_begin(inode, pos, len, flags);
947 * Reserve one block more for addition to orphan list in case
948 * we allocate blocks but write fails for some reason
950 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
951 index = pos >> PAGE_CACHE_SHIFT;
952 from = pos & (PAGE_CACHE_SIZE - 1);
955 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
956 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
965 * grab_cache_page_write_begin() can take a long time if the
966 * system is thrashing due to memory pressure, or if the page
967 * is being written back. So grab it first before we start
968 * the transaction handle. This also allows us to allocate
969 * the page (if needed) without using GFP_NOFS.
972 page = grab_cache_page_write_begin(mapping, index, flags);
978 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
979 if (IS_ERR(handle)) {
980 page_cache_release(page);
981 return PTR_ERR(handle);
985 if (page->mapping != mapping) {
986 /* The page got truncated from under us */
988 page_cache_release(page);
989 ext4_journal_stop(handle);
992 /* In case writeback began while the page was unlocked */
993 wait_for_stable_page(page);
995 if (ext4_should_dioread_nolock(inode))
996 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
998 ret = __block_write_begin(page, pos, len, ext4_get_block);
1000 if (!ret && ext4_should_journal_data(inode)) {
1001 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1003 do_journal_get_write_access);
1009 * __block_write_begin may have instantiated a few blocks
1010 * outside i_size. Trim these off again. Don't need
1011 * i_size_read because we hold i_mutex.
1013 * Add inode to orphan list in case we crash before
1016 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1017 ext4_orphan_add(handle, inode);
1019 ext4_journal_stop(handle);
1020 if (pos + len > inode->i_size) {
1021 ext4_truncate_failed_write(inode);
1023 * If truncate failed early the inode might
1024 * still be on the orphan list; we need to
1025 * make sure the inode is removed from the
1026 * orphan list in that case.
1029 ext4_orphan_del(NULL, inode);
1032 if (ret == -ENOSPC &&
1033 ext4_should_retry_alloc(inode->i_sb, &retries))
1035 page_cache_release(page);
1042 /* For write_end() in data=journal mode */
1043 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1046 if (!buffer_mapped(bh) || buffer_freed(bh))
1048 set_buffer_uptodate(bh);
1049 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1050 clear_buffer_meta(bh);
1051 clear_buffer_prio(bh);
1056 * We need to pick up the new inode size which generic_commit_write gave us
1057 * `file' can be NULL - eg, when called from page_symlink().
1059 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1060 * buffers are managed internally.
1062 static int ext4_write_end(struct file *file,
1063 struct address_space *mapping,
1064 loff_t pos, unsigned len, unsigned copied,
1065 struct page *page, void *fsdata)
1067 handle_t *handle = ext4_journal_current_handle();
1068 struct inode *inode = mapping->host;
1070 int i_size_changed = 0;
1072 trace_ext4_write_end(inode, pos, len, copied);
1073 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1074 ret = ext4_jbd2_file_inode(handle, inode);
1077 page_cache_release(page);
1082 if (ext4_has_inline_data(inode)) {
1083 ret = ext4_write_inline_data_end(inode, pos, len,
1089 copied = block_write_end(file, mapping, pos,
1090 len, copied, page, fsdata);
1093 * No need to use i_size_read() here, the i_size
1094 * cannot change under us because we hole i_mutex.
1096 * But it's important to update i_size while still holding page lock:
1097 * page writeout could otherwise come in and zero beyond i_size.
1099 if (pos + copied > inode->i_size) {
1100 i_size_write(inode, pos + copied);
1104 if (pos + copied > EXT4_I(inode)->i_disksize) {
1105 /* We need to mark inode dirty even if
1106 * new_i_size is less that inode->i_size
1107 * but greater than i_disksize. (hint delalloc)
1109 ext4_update_i_disksize(inode, (pos + copied));
1113 page_cache_release(page);
1116 * Don't mark the inode dirty under page lock. First, it unnecessarily
1117 * makes the holding time of page lock longer. Second, it forces lock
1118 * ordering of page lock and transaction start for journaling
1122 ext4_mark_inode_dirty(handle, inode);
1124 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1125 /* if we have allocated more blocks and copied
1126 * less. We will have blocks allocated outside
1127 * inode->i_size. So truncate them
1129 ext4_orphan_add(handle, inode);
1131 ret2 = ext4_journal_stop(handle);
1135 if (pos + len > inode->i_size) {
1136 ext4_truncate_failed_write(inode);
1138 * If truncate failed early the inode might still be
1139 * on the orphan list; we need to make sure the inode
1140 * is removed from the orphan list in that case.
1143 ext4_orphan_del(NULL, inode);
1146 return ret ? ret : copied;
1149 static int ext4_journalled_write_end(struct file *file,
1150 struct address_space *mapping,
1151 loff_t pos, unsigned len, unsigned copied,
1152 struct page *page, void *fsdata)
1154 handle_t *handle = ext4_journal_current_handle();
1155 struct inode *inode = mapping->host;
1161 trace_ext4_journalled_write_end(inode, pos, len, copied);
1162 from = pos & (PAGE_CACHE_SIZE - 1);
1165 BUG_ON(!ext4_handle_valid(handle));
1167 if (ext4_has_inline_data(inode))
1168 copied = ext4_write_inline_data_end(inode, pos, len,
1172 if (!PageUptodate(page))
1174 page_zero_new_buffers(page, from+copied, to);
1177 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1178 to, &partial, write_end_fn);
1180 SetPageUptodate(page);
1182 new_i_size = pos + copied;
1183 if (new_i_size > inode->i_size)
1184 i_size_write(inode, pos+copied);
1185 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1186 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1187 if (new_i_size > EXT4_I(inode)->i_disksize) {
1188 ext4_update_i_disksize(inode, new_i_size);
1189 ret2 = ext4_mark_inode_dirty(handle, inode);
1195 page_cache_release(page);
1196 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1197 /* if we have allocated more blocks and copied
1198 * less. We will have blocks allocated outside
1199 * inode->i_size. So truncate them
1201 ext4_orphan_add(handle, inode);
1203 ret2 = ext4_journal_stop(handle);
1206 if (pos + len > inode->i_size) {
1207 ext4_truncate_failed_write(inode);
1209 * If truncate failed early the inode might still be
1210 * on the orphan list; we need to make sure the inode
1211 * is removed from the orphan list in that case.
1214 ext4_orphan_del(NULL, inode);
1217 return ret ? ret : copied;
1221 * Reserve a metadata for a single block located at lblock
1223 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1225 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1226 struct ext4_inode_info *ei = EXT4_I(inode);
1227 unsigned int md_needed;
1228 ext4_lblk_t save_last_lblock;
1232 * recalculate the amount of metadata blocks to reserve
1233 * in order to allocate nrblocks
1234 * worse case is one extent per block
1236 spin_lock(&ei->i_block_reservation_lock);
1238 * ext4_calc_metadata_amount() has side effects, which we have
1239 * to be prepared undo if we fail to claim space.
1241 save_len = ei->i_da_metadata_calc_len;
1242 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1243 md_needed = EXT4_NUM_B2C(sbi,
1244 ext4_calc_metadata_amount(inode, lblock));
1245 trace_ext4_da_reserve_space(inode, md_needed);
1248 * We do still charge estimated metadata to the sb though;
1249 * we cannot afford to run out of free blocks.
1251 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1252 ei->i_da_metadata_calc_len = save_len;
1253 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1254 spin_unlock(&ei->i_block_reservation_lock);
1257 ei->i_reserved_meta_blocks += md_needed;
1258 spin_unlock(&ei->i_block_reservation_lock);
1260 return 0; /* success */
1264 * Reserve a single cluster located at lblock
1266 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1268 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1269 struct ext4_inode_info *ei = EXT4_I(inode);
1270 unsigned int md_needed;
1272 ext4_lblk_t save_last_lblock;
1276 * We will charge metadata quota at writeout time; this saves
1277 * us from metadata over-estimation, though we may go over by
1278 * a small amount in the end. Here we just reserve for data.
1280 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1285 * recalculate the amount of metadata blocks to reserve
1286 * in order to allocate nrblocks
1287 * worse case is one extent per block
1289 spin_lock(&ei->i_block_reservation_lock);
1291 * ext4_calc_metadata_amount() has side effects, which we have
1292 * to be prepared undo if we fail to claim space.
1294 save_len = ei->i_da_metadata_calc_len;
1295 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1296 md_needed = EXT4_NUM_B2C(sbi,
1297 ext4_calc_metadata_amount(inode, lblock));
1298 trace_ext4_da_reserve_space(inode, md_needed);
1301 * We do still charge estimated metadata to the sb though;
1302 * we cannot afford to run out of free blocks.
1304 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1305 ei->i_da_metadata_calc_len = save_len;
1306 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1307 spin_unlock(&ei->i_block_reservation_lock);
1308 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1311 ei->i_reserved_data_blocks++;
1312 ei->i_reserved_meta_blocks += md_needed;
1313 spin_unlock(&ei->i_block_reservation_lock);
1315 return 0; /* success */
1318 static void ext4_da_release_space(struct inode *inode, int to_free)
1320 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1321 struct ext4_inode_info *ei = EXT4_I(inode);
1324 return; /* Nothing to release, exit */
1326 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1328 trace_ext4_da_release_space(inode, to_free);
1329 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1331 * if there aren't enough reserved blocks, then the
1332 * counter is messed up somewhere. Since this
1333 * function is called from invalidate page, it's
1334 * harmless to return without any action.
1336 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1337 "ino %lu, to_free %d with only %d reserved "
1338 "data blocks", inode->i_ino, to_free,
1339 ei->i_reserved_data_blocks);
1341 to_free = ei->i_reserved_data_blocks;
1343 ei->i_reserved_data_blocks -= to_free;
1345 if (ei->i_reserved_data_blocks == 0) {
1347 * We can release all of the reserved metadata blocks
1348 * only when we have written all of the delayed
1349 * allocation blocks.
1350 * Note that in case of bigalloc, i_reserved_meta_blocks,
1351 * i_reserved_data_blocks, etc. refer to number of clusters.
1353 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1354 ei->i_reserved_meta_blocks);
1355 ei->i_reserved_meta_blocks = 0;
1356 ei->i_da_metadata_calc_len = 0;
1359 /* update fs dirty data blocks counter */
1360 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1362 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1364 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1367 static void ext4_da_page_release_reservation(struct page *page,
1368 unsigned int offset,
1369 unsigned int length)
1372 struct buffer_head *head, *bh;
1373 unsigned int curr_off = 0;
1374 struct inode *inode = page->mapping->host;
1375 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1376 unsigned int stop = offset + length;
1380 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1382 head = page_buffers(page);
1385 unsigned int next_off = curr_off + bh->b_size;
1387 if (next_off > stop)
1390 if ((offset <= curr_off) && (buffer_delay(bh))) {
1392 clear_buffer_delay(bh);
1394 curr_off = next_off;
1395 } while ((bh = bh->b_this_page) != head);
1398 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1399 ext4_es_remove_extent(inode, lblk, to_release);
1402 /* If we have released all the blocks belonging to a cluster, then we
1403 * need to release the reserved space for that cluster. */
1404 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1405 while (num_clusters > 0) {
1406 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1407 ((num_clusters - 1) << sbi->s_cluster_bits);
1408 if (sbi->s_cluster_ratio == 1 ||
1409 !ext4_find_delalloc_cluster(inode, lblk))
1410 ext4_da_release_space(inode, 1);
1417 * Delayed allocation stuff
1420 struct mpage_da_data {
1421 struct inode *inode;
1422 struct writeback_control *wbc;
1424 pgoff_t first_page; /* The first page to write */
1425 pgoff_t next_page; /* Current page to examine */
1426 pgoff_t last_page; /* Last page to examine */
1428 * Extent to map - this can be after first_page because that can be
1429 * fully mapped. We somewhat abuse m_flags to store whether the extent
1430 * is delalloc or unwritten.
1432 struct ext4_map_blocks map;
1433 struct ext4_io_submit io_submit; /* IO submission data */
1436 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1441 struct pagevec pvec;
1442 struct inode *inode = mpd->inode;
1443 struct address_space *mapping = inode->i_mapping;
1445 /* This is necessary when next_page == 0. */
1446 if (mpd->first_page >= mpd->next_page)
1449 index = mpd->first_page;
1450 end = mpd->next_page - 1;
1452 ext4_lblk_t start, last;
1453 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1454 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1455 ext4_es_remove_extent(inode, start, last - start + 1);
1458 pagevec_init(&pvec, 0);
1459 while (index <= end) {
1460 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1463 for (i = 0; i < nr_pages; i++) {
1464 struct page *page = pvec.pages[i];
1465 if (page->index > end)
1467 BUG_ON(!PageLocked(page));
1468 BUG_ON(PageWriteback(page));
1470 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1471 ClearPageUptodate(page);
1475 index = pvec.pages[nr_pages - 1]->index + 1;
1476 pagevec_release(&pvec);
1480 static void ext4_print_free_blocks(struct inode *inode)
1482 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1483 struct super_block *sb = inode->i_sb;
1484 struct ext4_inode_info *ei = EXT4_I(inode);
1486 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1487 EXT4_C2B(EXT4_SB(inode->i_sb),
1488 ext4_count_free_clusters(sb)));
1489 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1490 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1491 (long long) EXT4_C2B(EXT4_SB(sb),
1492 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1493 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1494 (long long) EXT4_C2B(EXT4_SB(sb),
1495 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1496 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1497 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1498 ei->i_reserved_data_blocks);
1499 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1500 ei->i_reserved_meta_blocks);
1501 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1502 ei->i_allocated_meta_blocks);
1506 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1508 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1512 * This function is grabs code from the very beginning of
1513 * ext4_map_blocks, but assumes that the caller is from delayed write
1514 * time. This function looks up the requested blocks and sets the
1515 * buffer delay bit under the protection of i_data_sem.
1517 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1518 struct ext4_map_blocks *map,
1519 struct buffer_head *bh)
1521 struct extent_status es;
1523 sector_t invalid_block = ~((sector_t) 0xffff);
1524 #ifdef ES_AGGRESSIVE_TEST
1525 struct ext4_map_blocks orig_map;
1527 memcpy(&orig_map, map, sizeof(*map));
1530 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1534 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1535 "logical block %lu\n", inode->i_ino, map->m_len,
1536 (unsigned long) map->m_lblk);
1538 /* Lookup extent status tree firstly */
1539 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1540 ext4_es_lru_add(inode);
1541 if (ext4_es_is_hole(&es)) {
1543 down_read((&EXT4_I(inode)->i_data_sem));
1548 * Delayed extent could be allocated by fallocate.
1549 * So we need to check it.
1551 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1552 map_bh(bh, inode->i_sb, invalid_block);
1554 set_buffer_delay(bh);
1558 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1559 retval = es.es_len - (iblock - es.es_lblk);
1560 if (retval > map->m_len)
1561 retval = map->m_len;
1562 map->m_len = retval;
1563 if (ext4_es_is_written(&es))
1564 map->m_flags |= EXT4_MAP_MAPPED;
1565 else if (ext4_es_is_unwritten(&es))
1566 map->m_flags |= EXT4_MAP_UNWRITTEN;
1570 #ifdef ES_AGGRESSIVE_TEST
1571 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1577 * Try to see if we can get the block without requesting a new
1578 * file system block.
1580 down_read((&EXT4_I(inode)->i_data_sem));
1581 if (ext4_has_inline_data(inode)) {
1583 * We will soon create blocks for this page, and let
1584 * us pretend as if the blocks aren't allocated yet.
1585 * In case of clusters, we have to handle the work
1586 * of mapping from cluster so that the reserved space
1587 * is calculated properly.
1589 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1590 ext4_find_delalloc_cluster(inode, map->m_lblk))
1591 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1593 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1594 retval = ext4_ext_map_blocks(NULL, inode, map,
1595 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1597 retval = ext4_ind_map_blocks(NULL, inode, map,
1598 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1604 * XXX: __block_prepare_write() unmaps passed block,
1608 * If the block was allocated from previously allocated cluster,
1609 * then we don't need to reserve it again. However we still need
1610 * to reserve metadata for every block we're going to write.
1612 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1613 ret = ext4_da_reserve_space(inode, iblock);
1615 /* not enough space to reserve */
1620 ret = ext4_da_reserve_metadata(inode, iblock);
1622 /* not enough space to reserve */
1628 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1629 ~0, EXTENT_STATUS_DELAYED);
1635 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1636 * and it should not appear on the bh->b_state.
1638 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1640 map_bh(bh, inode->i_sb, invalid_block);
1642 set_buffer_delay(bh);
1643 } else if (retval > 0) {
1645 unsigned int status;
1647 if (unlikely(retval != map->m_len)) {
1648 ext4_warning(inode->i_sb,
1649 "ES len assertion failed for inode "
1650 "%lu: retval %d != map->m_len %d",
1651 inode->i_ino, retval, map->m_len);
1655 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1656 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1657 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1658 map->m_pblk, status);
1664 up_read((&EXT4_I(inode)->i_data_sem));
1670 * This is a special get_blocks_t callback which is used by
1671 * ext4_da_write_begin(). It will either return mapped block or
1672 * reserve space for a single block.
1674 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1675 * We also have b_blocknr = -1 and b_bdev initialized properly
1677 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1678 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1679 * initialized properly.
1681 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1682 struct buffer_head *bh, int create)
1684 struct ext4_map_blocks map;
1687 BUG_ON(create == 0);
1688 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1690 map.m_lblk = iblock;
1694 * first, we need to know whether the block is allocated already
1695 * preallocated blocks are unmapped but should treated
1696 * the same as allocated blocks.
1698 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1702 map_bh(bh, inode->i_sb, map.m_pblk);
1703 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1705 if (buffer_unwritten(bh)) {
1706 /* A delayed write to unwritten bh should be marked
1707 * new and mapped. Mapped ensures that we don't do
1708 * get_block multiple times when we write to the same
1709 * offset and new ensures that we do proper zero out
1710 * for partial write.
1713 set_buffer_mapped(bh);
1718 static int bget_one(handle_t *handle, struct buffer_head *bh)
1724 static int bput_one(handle_t *handle, struct buffer_head *bh)
1730 static int __ext4_journalled_writepage(struct page *page,
1733 struct address_space *mapping = page->mapping;
1734 struct inode *inode = mapping->host;
1735 struct buffer_head *page_bufs = NULL;
1736 handle_t *handle = NULL;
1737 int ret = 0, err = 0;
1738 int inline_data = ext4_has_inline_data(inode);
1739 struct buffer_head *inode_bh = NULL;
1741 ClearPageChecked(page);
1744 BUG_ON(page->index != 0);
1745 BUG_ON(len > ext4_get_max_inline_size(inode));
1746 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1747 if (inode_bh == NULL)
1750 page_bufs = page_buffers(page);
1755 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1758 /* As soon as we unlock the page, it can go away, but we have
1759 * references to buffers so we are safe */
1762 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1763 ext4_writepage_trans_blocks(inode));
1764 if (IS_ERR(handle)) {
1765 ret = PTR_ERR(handle);
1769 BUG_ON(!ext4_handle_valid(handle));
1772 ret = ext4_journal_get_write_access(handle, inode_bh);
1774 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1777 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1778 do_journal_get_write_access);
1780 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1785 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1786 err = ext4_journal_stop(handle);
1790 if (!ext4_has_inline_data(inode))
1791 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1793 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1800 * Note that we don't need to start a transaction unless we're journaling data
1801 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1802 * need to file the inode to the transaction's list in ordered mode because if
1803 * we are writing back data added by write(), the inode is already there and if
1804 * we are writing back data modified via mmap(), no one guarantees in which
1805 * transaction the data will hit the disk. In case we are journaling data, we
1806 * cannot start transaction directly because transaction start ranks above page
1807 * lock so we have to do some magic.
1809 * This function can get called via...
1810 * - ext4_writepages after taking page lock (have journal handle)
1811 * - journal_submit_inode_data_buffers (no journal handle)
1812 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1813 * - grab_page_cache when doing write_begin (have journal handle)
1815 * We don't do any block allocation in this function. If we have page with
1816 * multiple blocks we need to write those buffer_heads that are mapped. This
1817 * is important for mmaped based write. So if we do with blocksize 1K
1818 * truncate(f, 1024);
1819 * a = mmap(f, 0, 4096);
1821 * truncate(f, 4096);
1822 * we have in the page first buffer_head mapped via page_mkwrite call back
1823 * but other buffer_heads would be unmapped but dirty (dirty done via the
1824 * do_wp_page). So writepage should write the first block. If we modify
1825 * the mmap area beyond 1024 we will again get a page_fault and the
1826 * page_mkwrite callback will do the block allocation and mark the
1827 * buffer_heads mapped.
1829 * We redirty the page if we have any buffer_heads that is either delay or
1830 * unwritten in the page.
1832 * We can get recursively called as show below.
1834 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1837 * But since we don't do any block allocation we should not deadlock.
1838 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1840 static int ext4_writepage(struct page *page,
1841 struct writeback_control *wbc)
1846 struct buffer_head *page_bufs = NULL;
1847 struct inode *inode = page->mapping->host;
1848 struct ext4_io_submit io_submit;
1849 bool keep_towrite = false;
1851 trace_ext4_writepage(page);
1852 size = i_size_read(inode);
1853 if (page->index == size >> PAGE_CACHE_SHIFT)
1854 len = size & ~PAGE_CACHE_MASK;
1856 len = PAGE_CACHE_SIZE;
1858 page_bufs = page_buffers(page);
1860 * We cannot do block allocation or other extent handling in this
1861 * function. If there are buffers needing that, we have to redirty
1862 * the page. But we may reach here when we do a journal commit via
1863 * journal_submit_inode_data_buffers() and in that case we must write
1864 * allocated buffers to achieve data=ordered mode guarantees.
1866 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1867 ext4_bh_delay_or_unwritten)) {
1868 redirty_page_for_writepage(wbc, page);
1869 if (current->flags & PF_MEMALLOC) {
1871 * For memory cleaning there's no point in writing only
1872 * some buffers. So just bail out. Warn if we came here
1873 * from direct reclaim.
1875 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1880 keep_towrite = true;
1883 if (PageChecked(page) && ext4_should_journal_data(inode))
1885 * It's mmapped pagecache. Add buffers and journal it. There
1886 * doesn't seem much point in redirtying the page here.
1888 return __ext4_journalled_writepage(page, len);
1890 ext4_io_submit_init(&io_submit, wbc);
1891 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1892 if (!io_submit.io_end) {
1893 redirty_page_for_writepage(wbc, page);
1897 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1898 ext4_io_submit(&io_submit);
1899 /* Drop io_end reference we got from init */
1900 ext4_put_io_end_defer(io_submit.io_end);
1904 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1907 loff_t size = i_size_read(mpd->inode);
1910 BUG_ON(page->index != mpd->first_page);
1911 if (page->index == size >> PAGE_CACHE_SHIFT)
1912 len = size & ~PAGE_CACHE_MASK;
1914 len = PAGE_CACHE_SIZE;
1915 clear_page_dirty_for_io(page);
1916 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1918 mpd->wbc->nr_to_write--;
1924 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1927 * mballoc gives us at most this number of blocks...
1928 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1929 * The rest of mballoc seems to handle chunks up to full group size.
1931 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1934 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1936 * @mpd - extent of blocks
1937 * @lblk - logical number of the block in the file
1938 * @bh - buffer head we want to add to the extent
1940 * The function is used to collect contig. blocks in the same state. If the
1941 * buffer doesn't require mapping for writeback and we haven't started the
1942 * extent of buffers to map yet, the function returns 'true' immediately - the
1943 * caller can write the buffer right away. Otherwise the function returns true
1944 * if the block has been added to the extent, false if the block couldn't be
1947 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1948 struct buffer_head *bh)
1950 struct ext4_map_blocks *map = &mpd->map;
1952 /* Buffer that doesn't need mapping for writeback? */
1953 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1954 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1955 /* So far no extent to map => we write the buffer right away */
1956 if (map->m_len == 0)
1961 /* First block in the extent? */
1962 if (map->m_len == 0) {
1965 map->m_flags = bh->b_state & BH_FLAGS;
1969 /* Don't go larger than mballoc is willing to allocate */
1970 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1973 /* Can we merge the block to our big extent? */
1974 if (lblk == map->m_lblk + map->m_len &&
1975 (bh->b_state & BH_FLAGS) == map->m_flags) {
1983 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1985 * @mpd - extent of blocks for mapping
1986 * @head - the first buffer in the page
1987 * @bh - buffer we should start processing from
1988 * @lblk - logical number of the block in the file corresponding to @bh
1990 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1991 * the page for IO if all buffers in this page were mapped and there's no
1992 * accumulated extent of buffers to map or add buffers in the page to the
1993 * extent of buffers to map. The function returns 1 if the caller can continue
1994 * by processing the next page, 0 if it should stop adding buffers to the
1995 * extent to map because we cannot extend it anymore. It can also return value
1996 * < 0 in case of error during IO submission.
1998 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1999 struct buffer_head *head,
2000 struct buffer_head *bh,
2003 struct inode *inode = mpd->inode;
2005 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2006 >> inode->i_blkbits;
2009 BUG_ON(buffer_locked(bh));
2011 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2012 /* Found extent to map? */
2015 /* Everything mapped so far and we hit EOF */
2018 } while (lblk++, (bh = bh->b_this_page) != head);
2019 /* So far everything mapped? Submit the page for IO. */
2020 if (mpd->map.m_len == 0) {
2021 err = mpage_submit_page(mpd, head->b_page);
2025 return lblk < blocks;
2029 * mpage_map_buffers - update buffers corresponding to changed extent and
2030 * submit fully mapped pages for IO
2032 * @mpd - description of extent to map, on return next extent to map
2034 * Scan buffers corresponding to changed extent (we expect corresponding pages
2035 * to be already locked) and update buffer state according to new extent state.
2036 * We map delalloc buffers to their physical location, clear unwritten bits,
2037 * and mark buffers as uninit when we perform writes to unwritten extents
2038 * and do extent conversion after IO is finished. If the last page is not fully
2039 * mapped, we update @map to the next extent in the last page that needs
2040 * mapping. Otherwise we submit the page for IO.
2042 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2044 struct pagevec pvec;
2046 struct inode *inode = mpd->inode;
2047 struct buffer_head *head, *bh;
2048 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2054 start = mpd->map.m_lblk >> bpp_bits;
2055 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2056 lblk = start << bpp_bits;
2057 pblock = mpd->map.m_pblk;
2059 pagevec_init(&pvec, 0);
2060 while (start <= end) {
2061 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2065 for (i = 0; i < nr_pages; i++) {
2066 struct page *page = pvec.pages[i];
2068 if (page->index > end)
2070 /* Up to 'end' pages must be contiguous */
2071 BUG_ON(page->index != start);
2072 bh = head = page_buffers(page);
2074 if (lblk < mpd->map.m_lblk)
2076 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2078 * Buffer after end of mapped extent.
2079 * Find next buffer in the page to map.
2082 mpd->map.m_flags = 0;
2084 * FIXME: If dioread_nolock supports
2085 * blocksize < pagesize, we need to make
2086 * sure we add size mapped so far to
2087 * io_end->size as the following call
2088 * can submit the page for IO.
2090 err = mpage_process_page_bufs(mpd, head,
2092 pagevec_release(&pvec);
2097 if (buffer_delay(bh)) {
2098 clear_buffer_delay(bh);
2099 bh->b_blocknr = pblock++;
2101 clear_buffer_unwritten(bh);
2102 } while (lblk++, (bh = bh->b_this_page) != head);
2105 * FIXME: This is going to break if dioread_nolock
2106 * supports blocksize < pagesize as we will try to
2107 * convert potentially unmapped parts of inode.
2109 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2110 /* Page fully mapped - let IO run! */
2111 err = mpage_submit_page(mpd, page);
2113 pagevec_release(&pvec);
2118 pagevec_release(&pvec);
2120 /* Extent fully mapped and matches with page boundary. We are done. */
2122 mpd->map.m_flags = 0;
2126 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2128 struct inode *inode = mpd->inode;
2129 struct ext4_map_blocks *map = &mpd->map;
2130 int get_blocks_flags;
2131 int err, dioread_nolock;
2133 trace_ext4_da_write_pages_extent(inode, map);
2135 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2136 * to convert an unwritten extent to be initialized (in the case
2137 * where we have written into one or more preallocated blocks). It is
2138 * possible that we're going to need more metadata blocks than
2139 * previously reserved. However we must not fail because we're in
2140 * writeback and there is nothing we can do about it so it might result
2141 * in data loss. So use reserved blocks to allocate metadata if
2144 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2145 * in question are delalloc blocks. This affects functions in many
2146 * different parts of the allocation call path. This flag exists
2147 * primarily because we don't want to change *many* call functions, so
2148 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2149 * once the inode's allocation semaphore is taken.
2151 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2152 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2153 dioread_nolock = ext4_should_dioread_nolock(inode);
2155 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2156 if (map->m_flags & (1 << BH_Delay))
2157 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2159 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2162 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2163 if (!mpd->io_submit.io_end->handle &&
2164 ext4_handle_valid(handle)) {
2165 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2166 handle->h_rsv_handle = NULL;
2168 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2171 BUG_ON(map->m_len == 0);
2172 if (map->m_flags & EXT4_MAP_NEW) {
2173 struct block_device *bdev = inode->i_sb->s_bdev;
2176 for (i = 0; i < map->m_len; i++)
2177 unmap_underlying_metadata(bdev, map->m_pblk + i);
2183 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2184 * mpd->len and submit pages underlying it for IO
2186 * @handle - handle for journal operations
2187 * @mpd - extent to map
2188 * @give_up_on_write - we set this to true iff there is a fatal error and there
2189 * is no hope of writing the data. The caller should discard
2190 * dirty pages to avoid infinite loops.
2192 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2193 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2194 * them to initialized or split the described range from larger unwritten
2195 * extent. Note that we need not map all the described range since allocation
2196 * can return less blocks or the range is covered by more unwritten extents. We
2197 * cannot map more because we are limited by reserved transaction credits. On
2198 * the other hand we always make sure that the last touched page is fully
2199 * mapped so that it can be written out (and thus forward progress is
2200 * guaranteed). After mapping we submit all mapped pages for IO.
2202 static int mpage_map_and_submit_extent(handle_t *handle,
2203 struct mpage_da_data *mpd,
2204 bool *give_up_on_write)
2206 struct inode *inode = mpd->inode;
2207 struct ext4_map_blocks *map = &mpd->map;
2211 mpd->io_submit.io_end->offset =
2212 ((loff_t)map->m_lblk) << inode->i_blkbits;
2214 err = mpage_map_one_extent(handle, mpd);
2216 struct super_block *sb = inode->i_sb;
2218 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2219 goto invalidate_dirty_pages;
2221 * Let the uper layers retry transient errors.
2222 * In the case of ENOSPC, if ext4_count_free_blocks()
2223 * is non-zero, a commit should free up blocks.
2225 if ((err == -ENOMEM) ||
2226 (err == -ENOSPC && ext4_count_free_clusters(sb)))
2228 ext4_msg(sb, KERN_CRIT,
2229 "Delayed block allocation failed for "
2230 "inode %lu at logical offset %llu with"
2231 " max blocks %u with error %d",
2233 (unsigned long long)map->m_lblk,
2234 (unsigned)map->m_len, -err);
2235 ext4_msg(sb, KERN_CRIT,
2236 "This should not happen!! Data will "
2239 ext4_print_free_blocks(inode);
2240 invalidate_dirty_pages:
2241 *give_up_on_write = true;
2245 * Update buffer state, submit mapped pages, and get us new
2248 err = mpage_map_and_submit_buffers(mpd);
2251 } while (map->m_len);
2254 * Update on-disk size after IO is submitted. Races with
2255 * truncate are avoided by checking i_size under i_data_sem.
2257 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2258 if (disksize > EXT4_I(inode)->i_disksize) {
2262 down_write(&EXT4_I(inode)->i_data_sem);
2263 i_size = i_size_read(inode);
2264 if (disksize > i_size)
2266 if (disksize > EXT4_I(inode)->i_disksize)
2267 EXT4_I(inode)->i_disksize = disksize;
2268 err2 = ext4_mark_inode_dirty(handle, inode);
2269 up_write(&EXT4_I(inode)->i_data_sem);
2271 ext4_error(inode->i_sb,
2272 "Failed to mark inode %lu dirty",
2281 * Calculate the total number of credits to reserve for one writepages
2282 * iteration. This is called from ext4_writepages(). We map an extent of
2283 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2284 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2285 * bpp - 1 blocks in bpp different extents.
2287 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2289 int bpp = ext4_journal_blocks_per_page(inode);
2291 return ext4_meta_trans_blocks(inode,
2292 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2296 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2297 * and underlying extent to map
2299 * @mpd - where to look for pages
2301 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2302 * IO immediately. When we find a page which isn't mapped we start accumulating
2303 * extent of buffers underlying these pages that needs mapping (formed by
2304 * either delayed or unwritten buffers). We also lock the pages containing
2305 * these buffers. The extent found is returned in @mpd structure (starting at
2306 * mpd->lblk with length mpd->len blocks).
2308 * Note that this function can attach bios to one io_end structure which are
2309 * neither logically nor physically contiguous. Although it may seem as an
2310 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2311 * case as we need to track IO to all buffers underlying a page in one io_end.
2313 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2315 struct address_space *mapping = mpd->inode->i_mapping;
2316 struct pagevec pvec;
2317 unsigned int nr_pages;
2318 long left = mpd->wbc->nr_to_write;
2319 pgoff_t index = mpd->first_page;
2320 pgoff_t end = mpd->last_page;
2323 int blkbits = mpd->inode->i_blkbits;
2325 struct buffer_head *head;
2327 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2328 tag = PAGECACHE_TAG_TOWRITE;
2330 tag = PAGECACHE_TAG_DIRTY;
2332 pagevec_init(&pvec, 0);
2334 mpd->next_page = index;
2335 while (index <= end) {
2336 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2337 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2341 for (i = 0; i < nr_pages; i++) {
2342 struct page *page = pvec.pages[i];
2345 * At this point, the page may be truncated or
2346 * invalidated (changing page->mapping to NULL), or
2347 * even swizzled back from swapper_space to tmpfs file
2348 * mapping. However, page->index will not change
2349 * because we have a reference on the page.
2351 if (page->index > end)
2355 * Accumulated enough dirty pages? This doesn't apply
2356 * to WB_SYNC_ALL mode. For integrity sync we have to
2357 * keep going because someone may be concurrently
2358 * dirtying pages, and we might have synced a lot of
2359 * newly appeared dirty pages, but have not synced all
2360 * of the old dirty pages.
2362 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2365 /* If we can't merge this page, we are done. */
2366 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2371 * If the page is no longer dirty, or its mapping no
2372 * longer corresponds to inode we are writing (which
2373 * means it has been truncated or invalidated), or the
2374 * page is already under writeback and we are not doing
2375 * a data integrity writeback, skip the page
2377 if (!PageDirty(page) ||
2378 (PageWriteback(page) &&
2379 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2380 unlikely(page->mapping != mapping)) {
2385 wait_on_page_writeback(page);
2386 BUG_ON(PageWriteback(page));
2388 if (mpd->map.m_len == 0)
2389 mpd->first_page = page->index;
2390 mpd->next_page = page->index + 1;
2391 /* Add all dirty buffers to mpd */
2392 lblk = ((ext4_lblk_t)page->index) <<
2393 (PAGE_CACHE_SHIFT - blkbits);
2394 head = page_buffers(page);
2395 err = mpage_process_page_bufs(mpd, head, head, lblk);
2401 pagevec_release(&pvec);
2406 pagevec_release(&pvec);
2410 static int __writepage(struct page *page, struct writeback_control *wbc,
2413 struct address_space *mapping = data;
2414 int ret = ext4_writepage(page, wbc);
2415 mapping_set_error(mapping, ret);
2419 static int ext4_writepages(struct address_space *mapping,
2420 struct writeback_control *wbc)
2422 pgoff_t writeback_index = 0;
2423 long nr_to_write = wbc->nr_to_write;
2424 int range_whole = 0;
2426 handle_t *handle = NULL;
2427 struct mpage_da_data mpd;
2428 struct inode *inode = mapping->host;
2429 int needed_blocks, rsv_blocks = 0, ret = 0;
2430 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2432 struct blk_plug plug;
2433 bool give_up_on_write = false;
2435 trace_ext4_writepages(inode, wbc);
2438 * No pages to write? This is mainly a kludge to avoid starting
2439 * a transaction for special inodes like journal inode on last iput()
2440 * because that could violate lock ordering on umount
2442 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2443 goto out_writepages;
2445 if (ext4_should_journal_data(inode)) {
2446 struct blk_plug plug;
2448 blk_start_plug(&plug);
2449 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2450 blk_finish_plug(&plug);
2451 goto out_writepages;
2455 * If the filesystem has aborted, it is read-only, so return
2456 * right away instead of dumping stack traces later on that
2457 * will obscure the real source of the problem. We test
2458 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2459 * the latter could be true if the filesystem is mounted
2460 * read-only, and in that case, ext4_writepages should
2461 * *never* be called, so if that ever happens, we would want
2464 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2466 goto out_writepages;
2469 if (ext4_should_dioread_nolock(inode)) {
2471 * We may need to convert up to one extent per block in
2472 * the page and we may dirty the inode.
2474 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2478 * If we have inline data and arrive here, it means that
2479 * we will soon create the block for the 1st page, so
2480 * we'd better clear the inline data here.
2482 if (ext4_has_inline_data(inode)) {
2483 /* Just inode will be modified... */
2484 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2485 if (IS_ERR(handle)) {
2486 ret = PTR_ERR(handle);
2487 goto out_writepages;
2489 BUG_ON(ext4_test_inode_state(inode,
2490 EXT4_STATE_MAY_INLINE_DATA));
2491 ext4_destroy_inline_data(handle, inode);
2492 ext4_journal_stop(handle);
2495 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2498 if (wbc->range_cyclic) {
2499 writeback_index = mapping->writeback_index;
2500 if (writeback_index)
2502 mpd.first_page = writeback_index;
2505 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2506 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2511 ext4_io_submit_init(&mpd.io_submit, wbc);
2513 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2514 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2516 blk_start_plug(&plug);
2517 while (!done && mpd.first_page <= mpd.last_page) {
2518 /* For each extent of pages we use new io_end */
2519 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2520 if (!mpd.io_submit.io_end) {
2526 * We have two constraints: We find one extent to map and we
2527 * must always write out whole page (makes a difference when
2528 * blocksize < pagesize) so that we don't block on IO when we
2529 * try to write out the rest of the page. Journalled mode is
2530 * not supported by delalloc.
2532 BUG_ON(ext4_should_journal_data(inode));
2533 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2535 /* start a new transaction */
2536 handle = ext4_journal_start_with_reserve(inode,
2537 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2538 if (IS_ERR(handle)) {
2539 ret = PTR_ERR(handle);
2540 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2541 "%ld pages, ino %lu; err %d", __func__,
2542 wbc->nr_to_write, inode->i_ino, ret);
2543 /* Release allocated io_end */
2544 ext4_put_io_end(mpd.io_submit.io_end);
2548 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2549 ret = mpage_prepare_extent_to_map(&mpd);
2552 ret = mpage_map_and_submit_extent(handle, &mpd,
2556 * We scanned the whole range (or exhausted
2557 * nr_to_write), submitted what was mapped and
2558 * didn't find anything needing mapping. We are
2564 ext4_journal_stop(handle);
2565 /* Submit prepared bio */
2566 ext4_io_submit(&mpd.io_submit);
2567 /* Unlock pages we didn't use */
2568 mpage_release_unused_pages(&mpd, give_up_on_write);
2569 /* Drop our io_end reference we got from init */
2570 ext4_put_io_end(mpd.io_submit.io_end);
2572 if (ret == -ENOSPC && sbi->s_journal) {
2574 * Commit the transaction which would
2575 * free blocks released in the transaction
2578 jbd2_journal_force_commit_nested(sbi->s_journal);
2582 /* Fatal error - ENOMEM, EIO... */
2586 blk_finish_plug(&plug);
2587 if (!ret && !cycled && wbc->nr_to_write > 0) {
2589 mpd.last_page = writeback_index - 1;
2595 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2597 * Set the writeback_index so that range_cyclic
2598 * mode will write it back later
2600 mapping->writeback_index = mpd.first_page;
2603 trace_ext4_writepages_result(inode, wbc, ret,
2604 nr_to_write - wbc->nr_to_write);
2608 static int ext4_nonda_switch(struct super_block *sb)
2610 s64 free_clusters, dirty_clusters;
2611 struct ext4_sb_info *sbi = EXT4_SB(sb);
2614 * switch to non delalloc mode if we are running low
2615 * on free block. The free block accounting via percpu
2616 * counters can get slightly wrong with percpu_counter_batch getting
2617 * accumulated on each CPU without updating global counters
2618 * Delalloc need an accurate free block accounting. So switch
2619 * to non delalloc when we are near to error range.
2622 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2624 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2626 * Start pushing delalloc when 1/2 of free blocks are dirty.
2628 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2629 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2631 if (2 * free_clusters < 3 * dirty_clusters ||
2632 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2634 * free block count is less than 150% of dirty blocks
2635 * or free blocks is less than watermark
2642 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2643 loff_t pos, unsigned len, unsigned flags,
2644 struct page **pagep, void **fsdata)
2646 int ret, retries = 0;
2649 struct inode *inode = mapping->host;
2652 index = pos >> PAGE_CACHE_SHIFT;
2654 if (ext4_nonda_switch(inode->i_sb)) {
2655 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2656 return ext4_write_begin(file, mapping, pos,
2657 len, flags, pagep, fsdata);
2659 *fsdata = (void *)0;
2660 trace_ext4_da_write_begin(inode, pos, len, flags);
2662 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2663 ret = ext4_da_write_inline_data_begin(mapping, inode,
2673 * grab_cache_page_write_begin() can take a long time if the
2674 * system is thrashing due to memory pressure, or if the page
2675 * is being written back. So grab it first before we start
2676 * the transaction handle. This also allows us to allocate
2677 * the page (if needed) without using GFP_NOFS.
2680 page = grab_cache_page_write_begin(mapping, index, flags);
2686 * With delayed allocation, we don't log the i_disksize update
2687 * if there is delayed block allocation. But we still need
2688 * to journalling the i_disksize update if writes to the end
2689 * of file which has an already mapped buffer.
2692 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2693 if (IS_ERR(handle)) {
2694 page_cache_release(page);
2695 return PTR_ERR(handle);
2699 if (page->mapping != mapping) {
2700 /* The page got truncated from under us */
2702 page_cache_release(page);
2703 ext4_journal_stop(handle);
2706 /* In case writeback began while the page was unlocked */
2707 wait_for_stable_page(page);
2709 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2712 ext4_journal_stop(handle);
2714 * block_write_begin may have instantiated a few blocks
2715 * outside i_size. Trim these off again. Don't need
2716 * i_size_read because we hold i_mutex.
2718 if (pos + len > inode->i_size)
2719 ext4_truncate_failed_write(inode);
2721 if (ret == -ENOSPC &&
2722 ext4_should_retry_alloc(inode->i_sb, &retries))
2725 page_cache_release(page);
2734 * Check if we should update i_disksize
2735 * when write to the end of file but not require block allocation
2737 static int ext4_da_should_update_i_disksize(struct page *page,
2738 unsigned long offset)
2740 struct buffer_head *bh;
2741 struct inode *inode = page->mapping->host;
2745 bh = page_buffers(page);
2746 idx = offset >> inode->i_blkbits;
2748 for (i = 0; i < idx; i++)
2749 bh = bh->b_this_page;
2751 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2756 static int ext4_da_write_end(struct file *file,
2757 struct address_space *mapping,
2758 loff_t pos, unsigned len, unsigned copied,
2759 struct page *page, void *fsdata)
2761 struct inode *inode = mapping->host;
2763 handle_t *handle = ext4_journal_current_handle();
2765 unsigned long start, end;
2766 int write_mode = (int)(unsigned long)fsdata;
2768 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2769 return ext4_write_end(file, mapping, pos,
2770 len, copied, page, fsdata);
2772 trace_ext4_da_write_end(inode, pos, len, copied);
2773 start = pos & (PAGE_CACHE_SIZE - 1);
2774 end = start + copied - 1;
2777 * generic_write_end() will run mark_inode_dirty() if i_size
2778 * changes. So let's piggyback the i_disksize mark_inode_dirty
2781 new_i_size = pos + copied;
2782 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2783 if (ext4_has_inline_data(inode) ||
2784 ext4_da_should_update_i_disksize(page, end)) {
2785 down_write(&EXT4_I(inode)->i_data_sem);
2786 if (new_i_size > EXT4_I(inode)->i_disksize)
2787 EXT4_I(inode)->i_disksize = new_i_size;
2788 up_write(&EXT4_I(inode)->i_data_sem);
2789 /* We need to mark inode dirty even if
2790 * new_i_size is less that inode->i_size
2791 * bu greater than i_disksize.(hint delalloc)
2793 ext4_mark_inode_dirty(handle, inode);
2797 if (write_mode != CONVERT_INLINE_DATA &&
2798 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2799 ext4_has_inline_data(inode))
2800 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2803 ret2 = generic_write_end(file, mapping, pos, len, copied,
2809 ret2 = ext4_journal_stop(handle);
2813 return ret ? ret : copied;
2816 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2817 unsigned int length)
2820 * Drop reserved blocks
2822 BUG_ON(!PageLocked(page));
2823 if (!page_has_buffers(page))
2826 ext4_da_page_release_reservation(page, offset, length);
2829 ext4_invalidatepage(page, offset, length);
2835 * Force all delayed allocation blocks to be allocated for a given inode.
2837 int ext4_alloc_da_blocks(struct inode *inode)
2839 trace_ext4_alloc_da_blocks(inode);
2841 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2842 !EXT4_I(inode)->i_reserved_meta_blocks)
2846 * We do something simple for now. The filemap_flush() will
2847 * also start triggering a write of the data blocks, which is
2848 * not strictly speaking necessary (and for users of
2849 * laptop_mode, not even desirable). However, to do otherwise
2850 * would require replicating code paths in:
2852 * ext4_writepages() ->
2853 * write_cache_pages() ---> (via passed in callback function)
2854 * __mpage_da_writepage() -->
2855 * mpage_add_bh_to_extent()
2856 * mpage_da_map_blocks()
2858 * The problem is that write_cache_pages(), located in
2859 * mm/page-writeback.c, marks pages clean in preparation for
2860 * doing I/O, which is not desirable if we're not planning on
2863 * We could call write_cache_pages(), and then redirty all of
2864 * the pages by calling redirty_page_for_writepage() but that
2865 * would be ugly in the extreme. So instead we would need to
2866 * replicate parts of the code in the above functions,
2867 * simplifying them because we wouldn't actually intend to
2868 * write out the pages, but rather only collect contiguous
2869 * logical block extents, call the multi-block allocator, and
2870 * then update the buffer heads with the block allocations.
2872 * For now, though, we'll cheat by calling filemap_flush(),
2873 * which will map the blocks, and start the I/O, but not
2874 * actually wait for the I/O to complete.
2876 return filemap_flush(inode->i_mapping);
2880 * bmap() is special. It gets used by applications such as lilo and by
2881 * the swapper to find the on-disk block of a specific piece of data.
2883 * Naturally, this is dangerous if the block concerned is still in the
2884 * journal. If somebody makes a swapfile on an ext4 data-journaling
2885 * filesystem and enables swap, then they may get a nasty shock when the
2886 * data getting swapped to that swapfile suddenly gets overwritten by
2887 * the original zero's written out previously to the journal and
2888 * awaiting writeback in the kernel's buffer cache.
2890 * So, if we see any bmap calls here on a modified, data-journaled file,
2891 * take extra steps to flush any blocks which might be in the cache.
2893 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2895 struct inode *inode = mapping->host;
2900 * We can get here for an inline file via the FIBMAP ioctl
2902 if (ext4_has_inline_data(inode))
2905 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2906 test_opt(inode->i_sb, DELALLOC)) {
2908 * With delalloc we want to sync the file
2909 * so that we can make sure we allocate
2912 filemap_write_and_wait(mapping);
2915 if (EXT4_JOURNAL(inode) &&
2916 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2918 * This is a REALLY heavyweight approach, but the use of
2919 * bmap on dirty files is expected to be extremely rare:
2920 * only if we run lilo or swapon on a freshly made file
2921 * do we expect this to happen.
2923 * (bmap requires CAP_SYS_RAWIO so this does not
2924 * represent an unprivileged user DOS attack --- we'd be
2925 * in trouble if mortal users could trigger this path at
2928 * NB. EXT4_STATE_JDATA is not set on files other than
2929 * regular files. If somebody wants to bmap a directory
2930 * or symlink and gets confused because the buffer
2931 * hasn't yet been flushed to disk, they deserve
2932 * everything they get.
2935 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2936 journal = EXT4_JOURNAL(inode);
2937 jbd2_journal_lock_updates(journal);
2938 err = jbd2_journal_flush(journal);
2939 jbd2_journal_unlock_updates(journal);
2945 return generic_block_bmap(mapping, block, ext4_get_block);
2948 static int ext4_readpage(struct file *file, struct page *page)
2951 struct inode *inode = page->mapping->host;
2953 trace_ext4_readpage(page);
2955 if (ext4_has_inline_data(inode))
2956 ret = ext4_readpage_inline(inode, page);
2959 return mpage_readpage(page, ext4_get_block);
2965 ext4_readpages(struct file *file, struct address_space *mapping,
2966 struct list_head *pages, unsigned nr_pages)
2968 struct inode *inode = mapping->host;
2970 /* If the file has inline data, no need to do readpages. */
2971 if (ext4_has_inline_data(inode))
2974 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2977 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2978 unsigned int length)
2980 trace_ext4_invalidatepage(page, offset, length);
2982 /* No journalling happens on data buffers when this function is used */
2983 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2985 block_invalidatepage(page, offset, length);
2988 static int __ext4_journalled_invalidatepage(struct page *page,
2989 unsigned int offset,
2990 unsigned int length)
2992 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2994 trace_ext4_journalled_invalidatepage(page, offset, length);
2997 * If it's a full truncate we just forget about the pending dirtying
2999 if (offset == 0 && length == PAGE_CACHE_SIZE)
3000 ClearPageChecked(page);
3002 return jbd2_journal_invalidatepage(journal, page, offset, length);
3005 /* Wrapper for aops... */
3006 static void ext4_journalled_invalidatepage(struct page *page,
3007 unsigned int offset,
3008 unsigned int length)
3010 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3013 static int ext4_releasepage(struct page *page, gfp_t wait)
3015 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3017 trace_ext4_releasepage(page);
3019 /* Page has dirty journalled data -> cannot release */
3020 if (PageChecked(page))
3023 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3025 return try_to_free_buffers(page);
3029 * ext4_get_block used when preparing for a DIO write or buffer write.
3030 * We allocate an uinitialized extent if blocks haven't been allocated.
3031 * The extent will be converted to initialized after the IO is complete.
3033 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3034 struct buffer_head *bh_result, int create)
3036 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3037 inode->i_ino, create);
3038 return _ext4_get_block(inode, iblock, bh_result,
3039 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3042 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3043 struct buffer_head *bh_result, int create)
3045 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3046 inode->i_ino, create);
3047 return _ext4_get_block(inode, iblock, bh_result,
3048 EXT4_GET_BLOCKS_NO_LOCK);
3051 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3052 ssize_t size, void *private)
3054 ext4_io_end_t *io_end = iocb->private;
3056 /* if not async direct IO just return */
3060 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3061 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3062 iocb->private, io_end->inode->i_ino, iocb, offset,
3065 iocb->private = NULL;
3066 io_end->offset = offset;
3067 io_end->size = size;
3068 ext4_put_io_end(io_end);
3072 * For ext4 extent files, ext4 will do direct-io write to holes,
3073 * preallocated extents, and those write extend the file, no need to
3074 * fall back to buffered IO.
3076 * For holes, we fallocate those blocks, mark them as unwritten
3077 * If those blocks were preallocated, we mark sure they are split, but
3078 * still keep the range to write as unwritten.
3080 * The unwritten extents will be converted to written when DIO is completed.
3081 * For async direct IO, since the IO may still pending when return, we
3082 * set up an end_io call back function, which will do the conversion
3083 * when async direct IO completed.
3085 * If the O_DIRECT write will extend the file then add this inode to the
3086 * orphan list. So recovery will truncate it back to the original size
3087 * if the machine crashes during the write.
3090 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3091 const struct iovec *iov, loff_t offset,
3092 unsigned long nr_segs)
3094 struct file *file = iocb->ki_filp;
3095 struct inode *inode = file->f_mapping->host;
3097 size_t count = iov_length(iov, nr_segs);
3099 get_block_t *get_block_func = NULL;
3101 loff_t final_size = offset + count;
3102 ext4_io_end_t *io_end = NULL;
3104 /* Use the old path for reads and writes beyond i_size. */
3105 if (rw != WRITE || final_size > inode->i_size)
3106 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3108 BUG_ON(iocb->private == NULL);
3111 * Make all waiters for direct IO properly wait also for extent
3112 * conversion. This also disallows race between truncate() and
3113 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3116 atomic_inc(&inode->i_dio_count);
3118 /* If we do a overwrite dio, i_mutex locking can be released */
3119 overwrite = *((int *)iocb->private);
3122 down_read(&EXT4_I(inode)->i_data_sem);
3123 mutex_unlock(&inode->i_mutex);
3127 * We could direct write to holes and fallocate.
3129 * Allocated blocks to fill the hole are marked as
3130 * unwritten to prevent parallel buffered read to expose
3131 * the stale data before DIO complete the data IO.
3133 * As to previously fallocated extents, ext4 get_block will
3134 * just simply mark the buffer mapped but still keep the
3135 * extents unwritten.
3137 * For non AIO case, we will convert those unwritten extents
3138 * to written after return back from blockdev_direct_IO.
3140 * For async DIO, the conversion needs to be deferred when the
3141 * IO is completed. The ext4 end_io callback function will be
3142 * called to take care of the conversion work. Here for async
3143 * case, we allocate an io_end structure to hook to the iocb.
3145 iocb->private = NULL;
3146 ext4_inode_aio_set(inode, NULL);
3147 if (!is_sync_kiocb(iocb)) {
3148 io_end = ext4_init_io_end(inode, GFP_NOFS);
3154 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3156 iocb->private = ext4_get_io_end(io_end);
3158 * we save the io structure for current async direct
3159 * IO, so that later ext4_map_blocks() could flag the
3160 * io structure whether there is a unwritten extents
3161 * needs to be converted when IO is completed.
3163 ext4_inode_aio_set(inode, io_end);
3167 get_block_func = ext4_get_block_write_nolock;
3169 get_block_func = ext4_get_block_write;
3170 dio_flags = DIO_LOCKING;
3172 ret = __blockdev_direct_IO(rw, iocb, inode,
3173 inode->i_sb->s_bdev, iov,
3181 * Put our reference to io_end. This can free the io_end structure e.g.
3182 * in sync IO case or in case of error. It can even perform extent
3183 * conversion if all bios we submitted finished before we got here.
3184 * Note that in that case iocb->private can be already set to NULL
3188 ext4_inode_aio_set(inode, NULL);
3189 ext4_put_io_end(io_end);
3191 * When no IO was submitted ext4_end_io_dio() was not
3192 * called so we have to put iocb's reference.
3194 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3195 WARN_ON(iocb->private != io_end);
3196 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3197 ext4_put_io_end(io_end);
3198 iocb->private = NULL;
3201 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3202 EXT4_STATE_DIO_UNWRITTEN)) {
3205 * for non AIO case, since the IO is already
3206 * completed, we could do the conversion right here
3208 err = ext4_convert_unwritten_extents(NULL, inode,
3212 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3217 inode_dio_done(inode);
3218 /* take i_mutex locking again if we do a ovewrite dio */
3220 up_read(&EXT4_I(inode)->i_data_sem);
3221 mutex_lock(&inode->i_mutex);
3227 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3228 const struct iovec *iov, loff_t offset,
3229 unsigned long nr_segs)
3231 struct file *file = iocb->ki_filp;
3232 struct inode *inode = file->f_mapping->host;
3236 * If we are doing data journalling we don't support O_DIRECT
3238 if (ext4_should_journal_data(inode))
3241 /* Let buffer I/O handle the inline data case. */
3242 if (ext4_has_inline_data(inode))
3245 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3246 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3247 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3249 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3250 trace_ext4_direct_IO_exit(inode, offset,
3251 iov_length(iov, nr_segs), rw, ret);
3256 * Pages can be marked dirty completely asynchronously from ext4's journalling
3257 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3258 * much here because ->set_page_dirty is called under VFS locks. The page is
3259 * not necessarily locked.
3261 * We cannot just dirty the page and leave attached buffers clean, because the
3262 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3263 * or jbddirty because all the journalling code will explode.
3265 * So what we do is to mark the page "pending dirty" and next time writepage
3266 * is called, propagate that into the buffers appropriately.
3268 static int ext4_journalled_set_page_dirty(struct page *page)
3270 SetPageChecked(page);
3271 return __set_page_dirty_nobuffers(page);
3274 static const struct address_space_operations ext4_aops = {
3275 .readpage = ext4_readpage,
3276 .readpages = ext4_readpages,
3277 .writepage = ext4_writepage,
3278 .writepages = ext4_writepages,
3279 .write_begin = ext4_write_begin,
3280 .write_end = ext4_write_end,
3282 .invalidatepage = ext4_invalidatepage,
3283 .releasepage = ext4_releasepage,
3284 .direct_IO = ext4_direct_IO,
3285 .migratepage = buffer_migrate_page,
3286 .is_partially_uptodate = block_is_partially_uptodate,
3287 .error_remove_page = generic_error_remove_page,
3290 static const struct address_space_operations ext4_journalled_aops = {
3291 .readpage = ext4_readpage,
3292 .readpages = ext4_readpages,
3293 .writepage = ext4_writepage,
3294 .writepages = ext4_writepages,
3295 .write_begin = ext4_write_begin,
3296 .write_end = ext4_journalled_write_end,
3297 .set_page_dirty = ext4_journalled_set_page_dirty,
3299 .invalidatepage = ext4_journalled_invalidatepage,
3300 .releasepage = ext4_releasepage,
3301 .direct_IO = ext4_direct_IO,
3302 .is_partially_uptodate = block_is_partially_uptodate,
3303 .error_remove_page = generic_error_remove_page,
3306 static const struct address_space_operations ext4_da_aops = {
3307 .readpage = ext4_readpage,
3308 .readpages = ext4_readpages,
3309 .writepage = ext4_writepage,
3310 .writepages = ext4_writepages,
3311 .write_begin = ext4_da_write_begin,
3312 .write_end = ext4_da_write_end,
3314 .invalidatepage = ext4_da_invalidatepage,
3315 .releasepage = ext4_releasepage,
3316 .direct_IO = ext4_direct_IO,
3317 .migratepage = buffer_migrate_page,
3318 .is_partially_uptodate = block_is_partially_uptodate,
3319 .error_remove_page = generic_error_remove_page,
3322 void ext4_set_aops(struct inode *inode)
3324 switch (ext4_inode_journal_mode(inode)) {
3325 case EXT4_INODE_ORDERED_DATA_MODE:
3326 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3328 case EXT4_INODE_WRITEBACK_DATA_MODE:
3329 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3331 case EXT4_INODE_JOURNAL_DATA_MODE:
3332 inode->i_mapping->a_ops = &ext4_journalled_aops;
3337 if (test_opt(inode->i_sb, DELALLOC))
3338 inode->i_mapping->a_ops = &ext4_da_aops;
3340 inode->i_mapping->a_ops = &ext4_aops;
3344 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3345 * starting from file offset 'from'. The range to be zero'd must
3346 * be contained with in one block. If the specified range exceeds
3347 * the end of the block it will be shortened to end of the block
3348 * that cooresponds to 'from'
3350 static int ext4_block_zero_page_range(handle_t *handle,
3351 struct address_space *mapping, loff_t from, loff_t length)
3353 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3354 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3355 unsigned blocksize, max, pos;
3357 struct inode *inode = mapping->host;
3358 struct buffer_head *bh;
3362 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3363 mapping_gfp_mask(mapping) & ~__GFP_FS);
3367 blocksize = inode->i_sb->s_blocksize;
3368 max = blocksize - (offset & (blocksize - 1));
3371 * correct length if it does not fall between
3372 * 'from' and the end of the block
3374 if (length > max || length < 0)
3377 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3379 if (!page_has_buffers(page))
3380 create_empty_buffers(page, blocksize, 0);
3382 /* Find the buffer that contains "offset" */
3383 bh = page_buffers(page);
3385 while (offset >= pos) {
3386 bh = bh->b_this_page;
3390 if (buffer_freed(bh)) {
3391 BUFFER_TRACE(bh, "freed: skip");
3394 if (!buffer_mapped(bh)) {
3395 BUFFER_TRACE(bh, "unmapped");
3396 ext4_get_block(inode, iblock, bh, 0);
3397 /* unmapped? It's a hole - nothing to do */
3398 if (!buffer_mapped(bh)) {
3399 BUFFER_TRACE(bh, "still unmapped");
3404 /* Ok, it's mapped. Make sure it's up-to-date */
3405 if (PageUptodate(page))
3406 set_buffer_uptodate(bh);
3408 if (!buffer_uptodate(bh)) {
3410 ll_rw_block(READ, 1, &bh);
3412 /* Uhhuh. Read error. Complain and punt. */
3413 if (!buffer_uptodate(bh))
3416 if (ext4_should_journal_data(inode)) {
3417 BUFFER_TRACE(bh, "get write access");
3418 err = ext4_journal_get_write_access(handle, bh);
3422 zero_user(page, offset, length);
3423 BUFFER_TRACE(bh, "zeroed end of block");
3425 if (ext4_should_journal_data(inode)) {
3426 err = ext4_handle_dirty_metadata(handle, inode, bh);
3429 mark_buffer_dirty(bh);
3430 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3431 err = ext4_jbd2_file_inode(handle, inode);
3436 page_cache_release(page);
3441 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3442 * up to the end of the block which corresponds to `from'.
3443 * This required during truncate. We need to physically zero the tail end
3444 * of that block so it doesn't yield old data if the file is later grown.
3446 static int ext4_block_truncate_page(handle_t *handle,
3447 struct address_space *mapping, loff_t from)
3449 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3452 struct inode *inode = mapping->host;
3454 blocksize = inode->i_sb->s_blocksize;
3455 length = blocksize - (offset & (blocksize - 1));
3457 return ext4_block_zero_page_range(handle, mapping, from, length);
3460 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3461 loff_t lstart, loff_t length)
3463 struct super_block *sb = inode->i_sb;
3464 struct address_space *mapping = inode->i_mapping;
3465 unsigned partial_start, partial_end;
3466 ext4_fsblk_t start, end;
3467 loff_t byte_end = (lstart + length - 1);
3470 partial_start = lstart & (sb->s_blocksize - 1);
3471 partial_end = byte_end & (sb->s_blocksize - 1);
3473 start = lstart >> sb->s_blocksize_bits;
3474 end = byte_end >> sb->s_blocksize_bits;
3476 /* Handle partial zero within the single block */
3478 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3479 err = ext4_block_zero_page_range(handle, mapping,
3483 /* Handle partial zero out on the start of the range */
3484 if (partial_start) {
3485 err = ext4_block_zero_page_range(handle, mapping,
3486 lstart, sb->s_blocksize);
3490 /* Handle partial zero out on the end of the range */
3491 if (partial_end != sb->s_blocksize - 1)
3492 err = ext4_block_zero_page_range(handle, mapping,
3493 byte_end - partial_end,
3498 int ext4_can_truncate(struct inode *inode)
3500 if (S_ISREG(inode->i_mode))
3502 if (S_ISDIR(inode->i_mode))
3504 if (S_ISLNK(inode->i_mode))
3505 return !ext4_inode_is_fast_symlink(inode);
3510 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3511 * associated with the given offset and length
3513 * @inode: File inode
3514 * @offset: The offset where the hole will begin
3515 * @len: The length of the hole
3517 * Returns: 0 on success or negative on failure
3520 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3522 struct super_block *sb = inode->i_sb;
3523 ext4_lblk_t first_block, stop_block;
3524 struct address_space *mapping = inode->i_mapping;
3525 loff_t first_block_offset, last_block_offset;
3527 unsigned int credits;
3530 if (!S_ISREG(inode->i_mode))
3533 trace_ext4_punch_hole(inode, offset, length, 0);
3536 * Write out all dirty pages to avoid race conditions
3537 * Then release them.
3539 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3540 ret = filemap_write_and_wait_range(mapping, offset,
3541 offset + length - 1);
3546 mutex_lock(&inode->i_mutex);
3548 /* No need to punch hole beyond i_size */
3549 if (offset >= inode->i_size)
3553 * If the hole extends beyond i_size, set the hole
3554 * to end after the page that contains i_size
3556 if (offset + length > inode->i_size) {
3557 length = inode->i_size +
3558 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3562 if (offset & (sb->s_blocksize - 1) ||
3563 (offset + length) & (sb->s_blocksize - 1)) {
3565 * Attach jinode to inode for jbd2 if we do any zeroing of
3568 ret = ext4_inode_attach_jinode(inode);
3574 first_block_offset = round_up(offset, sb->s_blocksize);
3575 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3577 /* Now release the pages and zero block aligned part of pages*/
3578 if (last_block_offset > first_block_offset)
3579 truncate_pagecache_range(inode, first_block_offset,
3582 /* Wait all existing dio workers, newcomers will block on i_mutex */
3583 ext4_inode_block_unlocked_dio(inode);
3584 inode_dio_wait(inode);
3586 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3587 credits = ext4_writepage_trans_blocks(inode);
3589 credits = ext4_blocks_for_truncate(inode);
3590 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3591 if (IS_ERR(handle)) {
3592 ret = PTR_ERR(handle);
3593 ext4_std_error(sb, ret);
3597 ret = ext4_zero_partial_blocks(handle, inode, offset,
3602 first_block = (offset + sb->s_blocksize - 1) >>
3603 EXT4_BLOCK_SIZE_BITS(sb);
3604 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3606 /* If there are no blocks to remove, return now */
3607 if (first_block >= stop_block)
3610 down_write(&EXT4_I(inode)->i_data_sem);
3611 ext4_discard_preallocations(inode);
3613 ret = ext4_es_remove_extent(inode, first_block,
3614 stop_block - first_block);
3616 up_write(&EXT4_I(inode)->i_data_sem);
3620 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3621 ret = ext4_ext_remove_space(inode, first_block,
3624 ret = ext4_free_hole_blocks(handle, inode, first_block,
3627 up_write(&EXT4_I(inode)->i_data_sem);
3629 ext4_handle_sync(handle);
3631 /* Now release the pages again to reduce race window */
3632 if (last_block_offset > first_block_offset)
3633 truncate_pagecache_range(inode, first_block_offset,
3636 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3637 ext4_mark_inode_dirty(handle, inode);
3639 ext4_journal_stop(handle);
3641 ext4_inode_resume_unlocked_dio(inode);
3643 mutex_unlock(&inode->i_mutex);
3647 int ext4_inode_attach_jinode(struct inode *inode)
3649 struct ext4_inode_info *ei = EXT4_I(inode);
3650 struct jbd2_inode *jinode;
3652 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3655 jinode = jbd2_alloc_inode(GFP_KERNEL);
3656 spin_lock(&inode->i_lock);
3659 spin_unlock(&inode->i_lock);
3662 ei->jinode = jinode;
3663 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3666 spin_unlock(&inode->i_lock);
3667 if (unlikely(jinode != NULL))
3668 jbd2_free_inode(jinode);
3675 * We block out ext4_get_block() block instantiations across the entire
3676 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3677 * simultaneously on behalf of the same inode.
3679 * As we work through the truncate and commit bits of it to the journal there
3680 * is one core, guiding principle: the file's tree must always be consistent on
3681 * disk. We must be able to restart the truncate after a crash.
3683 * The file's tree may be transiently inconsistent in memory (although it
3684 * probably isn't), but whenever we close off and commit a journal transaction,
3685 * the contents of (the filesystem + the journal) must be consistent and
3686 * restartable. It's pretty simple, really: bottom up, right to left (although
3687 * left-to-right works OK too).
3689 * Note that at recovery time, journal replay occurs *before* the restart of
3690 * truncate against the orphan inode list.
3692 * The committed inode has the new, desired i_size (which is the same as
3693 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3694 * that this inode's truncate did not complete and it will again call
3695 * ext4_truncate() to have another go. So there will be instantiated blocks
3696 * to the right of the truncation point in a crashed ext4 filesystem. But
3697 * that's fine - as long as they are linked from the inode, the post-crash
3698 * ext4_truncate() run will find them and release them.
3700 void ext4_truncate(struct inode *inode)
3702 struct ext4_inode_info *ei = EXT4_I(inode);
3703 unsigned int credits;
3705 struct address_space *mapping = inode->i_mapping;
3708 * There is a possibility that we're either freeing the inode
3709 * or it's a completely new inode. In those cases we might not
3710 * have i_mutex locked because it's not necessary.
3712 if (!(inode->i_state & (I_NEW|I_FREEING)))
3713 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3714 trace_ext4_truncate_enter(inode);
3716 if (!ext4_can_truncate(inode))
3719 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3721 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3722 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3724 if (ext4_has_inline_data(inode)) {
3727 ext4_inline_data_truncate(inode, &has_inline);
3732 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3733 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3734 if (ext4_inode_attach_jinode(inode) < 0)
3738 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3739 credits = ext4_writepage_trans_blocks(inode);
3741 credits = ext4_blocks_for_truncate(inode);
3743 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3744 if (IS_ERR(handle)) {
3745 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3749 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3750 ext4_block_truncate_page(handle, mapping, inode->i_size);
3753 * We add the inode to the orphan list, so that if this
3754 * truncate spans multiple transactions, and we crash, we will
3755 * resume the truncate when the filesystem recovers. It also
3756 * marks the inode dirty, to catch the new size.
3758 * Implication: the file must always be in a sane, consistent
3759 * truncatable state while each transaction commits.
3761 if (ext4_orphan_add(handle, inode))
3764 down_write(&EXT4_I(inode)->i_data_sem);
3766 ext4_discard_preallocations(inode);
3768 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3769 ext4_ext_truncate(handle, inode);
3771 ext4_ind_truncate(handle, inode);
3773 up_write(&ei->i_data_sem);
3776 ext4_handle_sync(handle);
3780 * If this was a simple ftruncate() and the file will remain alive,
3781 * then we need to clear up the orphan record which we created above.
3782 * However, if this was a real unlink then we were called by
3783 * ext4_delete_inode(), and we allow that function to clean up the
3784 * orphan info for us.
3787 ext4_orphan_del(handle, inode);
3789 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3790 ext4_mark_inode_dirty(handle, inode);
3791 ext4_journal_stop(handle);
3793 trace_ext4_truncate_exit(inode);
3797 * ext4_get_inode_loc returns with an extra refcount against the inode's
3798 * underlying buffer_head on success. If 'in_mem' is true, we have all
3799 * data in memory that is needed to recreate the on-disk version of this
3802 static int __ext4_get_inode_loc(struct inode *inode,
3803 struct ext4_iloc *iloc, int in_mem)
3805 struct ext4_group_desc *gdp;
3806 struct buffer_head *bh;
3807 struct super_block *sb = inode->i_sb;
3809 int inodes_per_block, inode_offset;
3812 if (!ext4_valid_inum(sb, inode->i_ino))
3815 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3816 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3821 * Figure out the offset within the block group inode table
3823 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3824 inode_offset = ((inode->i_ino - 1) %
3825 EXT4_INODES_PER_GROUP(sb));
3826 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3827 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3829 bh = sb_getblk(sb, block);
3832 if (!buffer_uptodate(bh)) {
3836 * If the buffer has the write error flag, we have failed
3837 * to write out another inode in the same block. In this
3838 * case, we don't have to read the block because we may
3839 * read the old inode data successfully.
3841 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3842 set_buffer_uptodate(bh);
3844 if (buffer_uptodate(bh)) {
3845 /* someone brought it uptodate while we waited */
3851 * If we have all information of the inode in memory and this
3852 * is the only valid inode in the block, we need not read the
3856 struct buffer_head *bitmap_bh;
3859 start = inode_offset & ~(inodes_per_block - 1);
3861 /* Is the inode bitmap in cache? */
3862 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3863 if (unlikely(!bitmap_bh))
3867 * If the inode bitmap isn't in cache then the
3868 * optimisation may end up performing two reads instead
3869 * of one, so skip it.
3871 if (!buffer_uptodate(bitmap_bh)) {
3875 for (i = start; i < start + inodes_per_block; i++) {
3876 if (i == inode_offset)
3878 if (ext4_test_bit(i, bitmap_bh->b_data))
3882 if (i == start + inodes_per_block) {
3883 /* all other inodes are free, so skip I/O */
3884 memset(bh->b_data, 0, bh->b_size);
3885 set_buffer_uptodate(bh);
3893 * If we need to do any I/O, try to pre-readahead extra
3894 * blocks from the inode table.
3896 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3897 ext4_fsblk_t b, end, table;
3899 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3901 table = ext4_inode_table(sb, gdp);
3902 /* s_inode_readahead_blks is always a power of 2 */
3903 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3907 num = EXT4_INODES_PER_GROUP(sb);
3908 if (ext4_has_group_desc_csum(sb))
3909 num -= ext4_itable_unused_count(sb, gdp);
3910 table += num / inodes_per_block;
3914 sb_breadahead(sb, b++);
3918 * There are other valid inodes in the buffer, this inode
3919 * has in-inode xattrs, or we don't have this inode in memory.
3920 * Read the block from disk.
3922 trace_ext4_load_inode(inode);
3924 bh->b_end_io = end_buffer_read_sync;
3925 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3927 if (!buffer_uptodate(bh)) {
3928 EXT4_ERROR_INODE_BLOCK(inode, block,
3929 "unable to read itable block");
3939 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3941 /* We have all inode data except xattrs in memory here. */
3942 return __ext4_get_inode_loc(inode, iloc,
3943 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3946 void ext4_set_inode_flags(struct inode *inode)
3948 unsigned int flags = EXT4_I(inode)->i_flags;
3949 unsigned int new_fl = 0;
3951 if (flags & EXT4_SYNC_FL)
3953 if (flags & EXT4_APPEND_FL)
3955 if (flags & EXT4_IMMUTABLE_FL)
3956 new_fl |= S_IMMUTABLE;
3957 if (flags & EXT4_NOATIME_FL)
3958 new_fl |= S_NOATIME;
3959 if (flags & EXT4_DIRSYNC_FL)
3960 new_fl |= S_DIRSYNC;
3961 inode_set_flags(inode, new_fl,
3962 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3965 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3966 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3968 unsigned int vfs_fl;
3969 unsigned long old_fl, new_fl;
3972 vfs_fl = ei->vfs_inode.i_flags;
3973 old_fl = ei->i_flags;
3974 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3975 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3977 if (vfs_fl & S_SYNC)
3978 new_fl |= EXT4_SYNC_FL;
3979 if (vfs_fl & S_APPEND)
3980 new_fl |= EXT4_APPEND_FL;
3981 if (vfs_fl & S_IMMUTABLE)
3982 new_fl |= EXT4_IMMUTABLE_FL;
3983 if (vfs_fl & S_NOATIME)
3984 new_fl |= EXT4_NOATIME_FL;
3985 if (vfs_fl & S_DIRSYNC)
3986 new_fl |= EXT4_DIRSYNC_FL;
3987 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3990 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3991 struct ext4_inode_info *ei)
3994 struct inode *inode = &(ei->vfs_inode);
3995 struct super_block *sb = inode->i_sb;
3997 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3998 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3999 /* we are using combined 48 bit field */
4000 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4001 le32_to_cpu(raw_inode->i_blocks_lo);
4002 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4003 /* i_blocks represent file system block size */
4004 return i_blocks << (inode->i_blkbits - 9);
4009 return le32_to_cpu(raw_inode->i_blocks_lo);
4013 static inline void ext4_iget_extra_inode(struct inode *inode,
4014 struct ext4_inode *raw_inode,
4015 struct ext4_inode_info *ei)
4017 __le32 *magic = (void *)raw_inode +
4018 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4019 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4020 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4021 ext4_find_inline_data_nolock(inode);
4023 EXT4_I(inode)->i_inline_off = 0;
4026 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4028 struct ext4_iloc iloc;
4029 struct ext4_inode *raw_inode;
4030 struct ext4_inode_info *ei;
4031 struct inode *inode;
4032 journal_t *journal = EXT4_SB(sb)->s_journal;
4038 inode = iget_locked(sb, ino);
4040 return ERR_PTR(-ENOMEM);
4041 if (!(inode->i_state & I_NEW))
4047 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4050 raw_inode = ext4_raw_inode(&iloc);
4052 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4053 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4054 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4055 EXT4_INODE_SIZE(inode->i_sb)) {
4056 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4057 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4058 EXT4_INODE_SIZE(inode->i_sb));
4063 ei->i_extra_isize = 0;
4065 /* Precompute checksum seed for inode metadata */
4066 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4067 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4068 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4070 __le32 inum = cpu_to_le32(inode->i_ino);
4071 __le32 gen = raw_inode->i_generation;
4072 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4074 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4078 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4079 EXT4_ERROR_INODE(inode, "checksum invalid");
4084 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4085 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4086 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4087 if (!(test_opt(inode->i_sb, NO_UID32))) {
4088 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4089 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4091 i_uid_write(inode, i_uid);
4092 i_gid_write(inode, i_gid);
4093 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4095 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4096 ei->i_inline_off = 0;
4097 ei->i_dir_start_lookup = 0;
4098 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4099 /* We now have enough fields to check if the inode was active or not.
4100 * This is needed because nfsd might try to access dead inodes
4101 * the test is that same one that e2fsck uses
4102 * NeilBrown 1999oct15
4104 if (inode->i_nlink == 0) {
4105 if ((inode->i_mode == 0 ||
4106 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4107 ino != EXT4_BOOT_LOADER_INO) {
4108 /* this inode is deleted */
4112 /* The only unlinked inodes we let through here have
4113 * valid i_mode and are being read by the orphan
4114 * recovery code: that's fine, we're about to complete
4115 * the process of deleting those.
4116 * OR it is the EXT4_BOOT_LOADER_INO which is
4117 * not initialized on a new filesystem. */
4119 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4120 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4121 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4122 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4124 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4125 inode->i_size = ext4_isize(raw_inode);
4126 ei->i_disksize = inode->i_size;
4128 ei->i_reserved_quota = 0;
4130 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4131 ei->i_block_group = iloc.block_group;
4132 ei->i_last_alloc_group = ~0;
4134 * NOTE! The in-memory inode i_data array is in little-endian order
4135 * even on big-endian machines: we do NOT byteswap the block numbers!
4137 for (block = 0; block < EXT4_N_BLOCKS; block++)
4138 ei->i_data[block] = raw_inode->i_block[block];
4139 INIT_LIST_HEAD(&ei->i_orphan);
4142 * Set transaction id's of transactions that have to be committed
4143 * to finish f[data]sync. We set them to currently running transaction
4144 * as we cannot be sure that the inode or some of its metadata isn't
4145 * part of the transaction - the inode could have been reclaimed and
4146 * now it is reread from disk.
4149 transaction_t *transaction;
4152 read_lock(&journal->j_state_lock);
4153 if (journal->j_running_transaction)
4154 transaction = journal->j_running_transaction;
4156 transaction = journal->j_committing_transaction;
4158 tid = transaction->t_tid;
4160 tid = journal->j_commit_sequence;
4161 read_unlock(&journal->j_state_lock);
4162 ei->i_sync_tid = tid;
4163 ei->i_datasync_tid = tid;
4166 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4167 if (ei->i_extra_isize == 0) {
4168 /* The extra space is currently unused. Use it. */
4169 ei->i_extra_isize = sizeof(struct ext4_inode) -
4170 EXT4_GOOD_OLD_INODE_SIZE;
4172 ext4_iget_extra_inode(inode, raw_inode, ei);
4176 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4177 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4178 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4179 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4181 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4182 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4183 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4184 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4186 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4191 if (ei->i_file_acl &&
4192 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4193 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4197 } else if (!ext4_has_inline_data(inode)) {
4198 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4199 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4200 (S_ISLNK(inode->i_mode) &&
4201 !ext4_inode_is_fast_symlink(inode))))
4202 /* Validate extent which is part of inode */
4203 ret = ext4_ext_check_inode(inode);
4204 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4205 (S_ISLNK(inode->i_mode) &&
4206 !ext4_inode_is_fast_symlink(inode))) {
4207 /* Validate block references which are part of inode */
4208 ret = ext4_ind_check_inode(inode);
4214 if (S_ISREG(inode->i_mode)) {
4215 inode->i_op = &ext4_file_inode_operations;
4216 inode->i_fop = &ext4_file_operations;
4217 ext4_set_aops(inode);
4218 } else if (S_ISDIR(inode->i_mode)) {
4219 inode->i_op = &ext4_dir_inode_operations;
4220 inode->i_fop = &ext4_dir_operations;
4221 } else if (S_ISLNK(inode->i_mode)) {
4222 if (ext4_inode_is_fast_symlink(inode)) {
4223 inode->i_op = &ext4_fast_symlink_inode_operations;
4224 nd_terminate_link(ei->i_data, inode->i_size,
4225 sizeof(ei->i_data) - 1);
4227 inode->i_op = &ext4_symlink_inode_operations;
4228 ext4_set_aops(inode);
4230 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4231 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4232 inode->i_op = &ext4_special_inode_operations;
4233 if (raw_inode->i_block[0])
4234 init_special_inode(inode, inode->i_mode,
4235 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4237 init_special_inode(inode, inode->i_mode,
4238 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4239 } else if (ino == EXT4_BOOT_LOADER_INO) {
4240 make_bad_inode(inode);
4243 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4247 ext4_set_inode_flags(inode);
4248 unlock_new_inode(inode);
4254 return ERR_PTR(ret);
4257 static int ext4_inode_blocks_set(handle_t *handle,
4258 struct ext4_inode *raw_inode,
4259 struct ext4_inode_info *ei)
4261 struct inode *inode = &(ei->vfs_inode);
4262 u64 i_blocks = inode->i_blocks;
4263 struct super_block *sb = inode->i_sb;
4265 if (i_blocks <= ~0U) {
4267 * i_blocks can be represented in a 32 bit variable
4268 * as multiple of 512 bytes
4270 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4271 raw_inode->i_blocks_high = 0;
4272 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4275 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4278 if (i_blocks <= 0xffffffffffffULL) {
4280 * i_blocks can be represented in a 48 bit variable
4281 * as multiple of 512 bytes
4283 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4284 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4285 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4287 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4288 /* i_block is stored in file system block size */
4289 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4290 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4291 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4297 * Post the struct inode info into an on-disk inode location in the
4298 * buffer-cache. This gobbles the caller's reference to the
4299 * buffer_head in the inode location struct.
4301 * The caller must have write access to iloc->bh.
4303 static int ext4_do_update_inode(handle_t *handle,
4304 struct inode *inode,
4305 struct ext4_iloc *iloc)
4307 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4308 struct ext4_inode_info *ei = EXT4_I(inode);
4309 struct buffer_head *bh = iloc->bh;
4310 struct super_block *sb = inode->i_sb;
4311 int err = 0, rc, block;
4312 int need_datasync = 0, set_large_file = 0;
4316 spin_lock(&ei->i_raw_lock);
4318 /* For fields not tracked in the in-memory inode,
4319 * initialise them to zero for new inodes. */
4320 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4321 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4323 ext4_get_inode_flags(ei);
4324 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4325 i_uid = i_uid_read(inode);
4326 i_gid = i_gid_read(inode);
4327 if (!(test_opt(inode->i_sb, NO_UID32))) {
4328 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4329 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4331 * Fix up interoperability with old kernels. Otherwise, old inodes get
4332 * re-used with the upper 16 bits of the uid/gid intact
4335 raw_inode->i_uid_high =
4336 cpu_to_le16(high_16_bits(i_uid));
4337 raw_inode->i_gid_high =
4338 cpu_to_le16(high_16_bits(i_gid));
4340 raw_inode->i_uid_high = 0;
4341 raw_inode->i_gid_high = 0;
4344 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4345 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4346 raw_inode->i_uid_high = 0;
4347 raw_inode->i_gid_high = 0;
4349 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4351 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4352 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4353 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4354 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4356 if (ext4_inode_blocks_set(handle, raw_inode, ei)) {
4357 spin_unlock(&ei->i_raw_lock);
4360 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4361 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4362 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4363 raw_inode->i_file_acl_high =
4364 cpu_to_le16(ei->i_file_acl >> 32);
4365 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4366 if (ei->i_disksize != ext4_isize(raw_inode)) {
4367 ext4_isize_set(raw_inode, ei->i_disksize);
4370 if (ei->i_disksize > 0x7fffffffULL) {
4371 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4372 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4373 EXT4_SB(sb)->s_es->s_rev_level ==
4374 cpu_to_le32(EXT4_GOOD_OLD_REV))
4377 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4378 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4379 if (old_valid_dev(inode->i_rdev)) {
4380 raw_inode->i_block[0] =
4381 cpu_to_le32(old_encode_dev(inode->i_rdev));
4382 raw_inode->i_block[1] = 0;
4384 raw_inode->i_block[0] = 0;
4385 raw_inode->i_block[1] =
4386 cpu_to_le32(new_encode_dev(inode->i_rdev));
4387 raw_inode->i_block[2] = 0;
4389 } else if (!ext4_has_inline_data(inode)) {
4390 for (block = 0; block < EXT4_N_BLOCKS; block++)
4391 raw_inode->i_block[block] = ei->i_data[block];
4394 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4395 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4396 if (ei->i_extra_isize) {
4397 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4398 raw_inode->i_version_hi =
4399 cpu_to_le32(inode->i_version >> 32);
4400 raw_inode->i_extra_isize =
4401 cpu_to_le16(ei->i_extra_isize);
4405 ext4_inode_csum_set(inode, raw_inode, ei);
4407 spin_unlock(&ei->i_raw_lock);
4409 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4410 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4413 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4414 if (set_large_file) {
4415 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4418 ext4_update_dynamic_rev(sb);
4419 EXT4_SET_RO_COMPAT_FEATURE(sb,
4420 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4421 ext4_handle_sync(handle);
4422 err = ext4_handle_dirty_super(handle, sb);
4424 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4427 ext4_std_error(inode->i_sb, err);
4432 * ext4_write_inode()
4434 * We are called from a few places:
4436 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4437 * Here, there will be no transaction running. We wait for any running
4438 * transaction to commit.
4440 * - Within flush work (sys_sync(), kupdate and such).
4441 * We wait on commit, if told to.
4443 * - Within iput_final() -> write_inode_now()
4444 * We wait on commit, if told to.
4446 * In all cases it is actually safe for us to return without doing anything,
4447 * because the inode has been copied into a raw inode buffer in
4448 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4451 * Note that we are absolutely dependent upon all inode dirtiers doing the
4452 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4453 * which we are interested.
4455 * It would be a bug for them to not do this. The code:
4457 * mark_inode_dirty(inode)
4459 * inode->i_size = expr;
4461 * is in error because write_inode() could occur while `stuff()' is running,
4462 * and the new i_size will be lost. Plus the inode will no longer be on the
4463 * superblock's dirty inode list.
4465 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4469 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4472 if (EXT4_SB(inode->i_sb)->s_journal) {
4473 if (ext4_journal_current_handle()) {
4474 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4480 * No need to force transaction in WB_SYNC_NONE mode. Also
4481 * ext4_sync_fs() will force the commit after everything is
4484 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4487 err = ext4_force_commit(inode->i_sb);
4489 struct ext4_iloc iloc;
4491 err = __ext4_get_inode_loc(inode, &iloc, 0);
4495 * sync(2) will flush the whole buffer cache. No need to do
4496 * it here separately for each inode.
4498 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4499 sync_dirty_buffer(iloc.bh);
4500 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4501 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4502 "IO error syncing inode");
4511 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4512 * buffers that are attached to a page stradding i_size and are undergoing
4513 * commit. In that case we have to wait for commit to finish and try again.
4515 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4519 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4520 tid_t commit_tid = 0;
4523 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4525 * All buffers in the last page remain valid? Then there's nothing to
4526 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4529 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4532 page = find_lock_page(inode->i_mapping,
4533 inode->i_size >> PAGE_CACHE_SHIFT);
4536 ret = __ext4_journalled_invalidatepage(page, offset,
4537 PAGE_CACHE_SIZE - offset);
4539 page_cache_release(page);
4543 read_lock(&journal->j_state_lock);
4544 if (journal->j_committing_transaction)
4545 commit_tid = journal->j_committing_transaction->t_tid;
4546 read_unlock(&journal->j_state_lock);
4548 jbd2_log_wait_commit(journal, commit_tid);
4555 * Called from notify_change.
4557 * We want to trap VFS attempts to truncate the file as soon as
4558 * possible. In particular, we want to make sure that when the VFS
4559 * shrinks i_size, we put the inode on the orphan list and modify
4560 * i_disksize immediately, so that during the subsequent flushing of
4561 * dirty pages and freeing of disk blocks, we can guarantee that any
4562 * commit will leave the blocks being flushed in an unused state on
4563 * disk. (On recovery, the inode will get truncated and the blocks will
4564 * be freed, so we have a strong guarantee that no future commit will
4565 * leave these blocks visible to the user.)
4567 * Another thing we have to assure is that if we are in ordered mode
4568 * and inode is still attached to the committing transaction, we must
4569 * we start writeout of all the dirty pages which are being truncated.
4570 * This way we are sure that all the data written in the previous
4571 * transaction are already on disk (truncate waits for pages under
4574 * Called with inode->i_mutex down.
4576 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4578 struct inode *inode = dentry->d_inode;
4581 const unsigned int ia_valid = attr->ia_valid;
4583 error = inode_change_ok(inode, attr);
4587 if (is_quota_modification(inode, attr))
4588 dquot_initialize(inode);
4589 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4590 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4593 /* (user+group)*(old+new) structure, inode write (sb,
4594 * inode block, ? - but truncate inode update has it) */
4595 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4596 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4597 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4598 if (IS_ERR(handle)) {
4599 error = PTR_ERR(handle);
4602 error = dquot_transfer(inode, attr);
4604 ext4_journal_stop(handle);
4607 /* Update corresponding info in inode so that everything is in
4608 * one transaction */
4609 if (attr->ia_valid & ATTR_UID)
4610 inode->i_uid = attr->ia_uid;
4611 if (attr->ia_valid & ATTR_GID)
4612 inode->i_gid = attr->ia_gid;
4613 error = ext4_mark_inode_dirty(handle, inode);
4614 ext4_journal_stop(handle);
4617 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4620 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4621 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4623 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4627 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4628 inode_inc_iversion(inode);
4630 if (S_ISREG(inode->i_mode) &&
4631 (attr->ia_size < inode->i_size)) {
4632 if (ext4_should_order_data(inode)) {
4633 error = ext4_begin_ordered_truncate(inode,
4638 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4639 if (IS_ERR(handle)) {
4640 error = PTR_ERR(handle);
4643 if (ext4_handle_valid(handle)) {
4644 error = ext4_orphan_add(handle, inode);
4647 down_write(&EXT4_I(inode)->i_data_sem);
4648 EXT4_I(inode)->i_disksize = attr->ia_size;
4649 rc = ext4_mark_inode_dirty(handle, inode);
4653 * We have to update i_size under i_data_sem together
4654 * with i_disksize to avoid races with writeback code
4655 * running ext4_wb_update_i_disksize().
4658 i_size_write(inode, attr->ia_size);
4659 up_write(&EXT4_I(inode)->i_data_sem);
4660 ext4_journal_stop(handle);
4662 ext4_orphan_del(NULL, inode);
4666 i_size_write(inode, attr->ia_size);
4669 * Blocks are going to be removed from the inode. Wait
4670 * for dio in flight. Temporarily disable
4671 * dioread_nolock to prevent livelock.
4674 if (!ext4_should_journal_data(inode)) {
4675 ext4_inode_block_unlocked_dio(inode);
4676 inode_dio_wait(inode);
4677 ext4_inode_resume_unlocked_dio(inode);
4679 ext4_wait_for_tail_page_commit(inode);
4682 * Truncate pagecache after we've waited for commit
4683 * in data=journal mode to make pages freeable.
4685 truncate_pagecache(inode, inode->i_size);
4688 * We want to call ext4_truncate() even if attr->ia_size ==
4689 * inode->i_size for cases like truncation of fallocated space
4691 if (attr->ia_valid & ATTR_SIZE)
4692 ext4_truncate(inode);
4695 setattr_copy(inode, attr);
4696 mark_inode_dirty(inode);
4700 * If the call to ext4_truncate failed to get a transaction handle at
4701 * all, we need to clean up the in-core orphan list manually.
4703 if (orphan && inode->i_nlink)
4704 ext4_orphan_del(NULL, inode);
4706 if (!rc && (ia_valid & ATTR_MODE))
4707 rc = posix_acl_chmod(inode, inode->i_mode);
4710 ext4_std_error(inode->i_sb, error);
4716 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4719 struct inode *inode;
4720 unsigned long long delalloc_blocks;
4722 inode = dentry->d_inode;
4723 generic_fillattr(inode, stat);
4726 * If there is inline data in the inode, the inode will normally not
4727 * have data blocks allocated (it may have an external xattr block).
4728 * Report at least one sector for such files, so tools like tar, rsync,
4729 * others doen't incorrectly think the file is completely sparse.
4731 if (unlikely(ext4_has_inline_data(inode)))
4732 stat->blocks += (stat->size + 511) >> 9;
4735 * We can't update i_blocks if the block allocation is delayed
4736 * otherwise in the case of system crash before the real block
4737 * allocation is done, we will have i_blocks inconsistent with
4738 * on-disk file blocks.
4739 * We always keep i_blocks updated together with real
4740 * allocation. But to not confuse with user, stat
4741 * will return the blocks that include the delayed allocation
4742 * blocks for this file.
4744 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4745 EXT4_I(inode)->i_reserved_data_blocks);
4746 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4750 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4753 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4754 return ext4_ind_trans_blocks(inode, lblocks);
4755 return ext4_ext_index_trans_blocks(inode, pextents);
4759 * Account for index blocks, block groups bitmaps and block group
4760 * descriptor blocks if modify datablocks and index blocks
4761 * worse case, the indexs blocks spread over different block groups
4763 * If datablocks are discontiguous, they are possible to spread over
4764 * different block groups too. If they are contiguous, with flexbg,
4765 * they could still across block group boundary.
4767 * Also account for superblock, inode, quota and xattr blocks
4769 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4772 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4778 * How many index blocks need to touch to map @lblocks logical blocks
4779 * to @pextents physical extents?
4781 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4786 * Now let's see how many group bitmaps and group descriptors need
4789 groups = idxblocks + pextents;
4791 if (groups > ngroups)
4793 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4794 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4796 /* bitmaps and block group descriptor blocks */
4797 ret += groups + gdpblocks;
4799 /* Blocks for super block, inode, quota and xattr blocks */
4800 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4806 * Calculate the total number of credits to reserve to fit
4807 * the modification of a single pages into a single transaction,
4808 * which may include multiple chunks of block allocations.
4810 * This could be called via ext4_write_begin()
4812 * We need to consider the worse case, when
4813 * one new block per extent.
4815 int ext4_writepage_trans_blocks(struct inode *inode)
4817 int bpp = ext4_journal_blocks_per_page(inode);
4820 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4822 /* Account for data blocks for journalled mode */
4823 if (ext4_should_journal_data(inode))
4829 * Calculate the journal credits for a chunk of data modification.
4831 * This is called from DIO, fallocate or whoever calling
4832 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4834 * journal buffers for data blocks are not included here, as DIO
4835 * and fallocate do no need to journal data buffers.
4837 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4839 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4843 * The caller must have previously called ext4_reserve_inode_write().
4844 * Give this, we know that the caller already has write access to iloc->bh.
4846 int ext4_mark_iloc_dirty(handle_t *handle,
4847 struct inode *inode, struct ext4_iloc *iloc)
4851 if (IS_I_VERSION(inode))
4852 inode_inc_iversion(inode);
4854 /* the do_update_inode consumes one bh->b_count */
4857 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4858 err = ext4_do_update_inode(handle, inode, iloc);
4864 * On success, We end up with an outstanding reference count against
4865 * iloc->bh. This _must_ be cleaned up later.
4869 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4870 struct ext4_iloc *iloc)
4874 err = ext4_get_inode_loc(inode, iloc);
4876 BUFFER_TRACE(iloc->bh, "get_write_access");
4877 err = ext4_journal_get_write_access(handle, iloc->bh);
4883 ext4_std_error(inode->i_sb, err);
4888 * Expand an inode by new_extra_isize bytes.
4889 * Returns 0 on success or negative error number on failure.
4891 static int ext4_expand_extra_isize(struct inode *inode,
4892 unsigned int new_extra_isize,
4893 struct ext4_iloc iloc,
4896 struct ext4_inode *raw_inode;
4897 struct ext4_xattr_ibody_header *header;
4899 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4902 raw_inode = ext4_raw_inode(&iloc);
4904 header = IHDR(inode, raw_inode);
4906 /* No extended attributes present */
4907 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4908 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4909 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4911 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4915 /* try to expand with EAs present */
4916 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4921 * What we do here is to mark the in-core inode as clean with respect to inode
4922 * dirtiness (it may still be data-dirty).
4923 * This means that the in-core inode may be reaped by prune_icache
4924 * without having to perform any I/O. This is a very good thing,
4925 * because *any* task may call prune_icache - even ones which
4926 * have a transaction open against a different journal.
4928 * Is this cheating? Not really. Sure, we haven't written the
4929 * inode out, but prune_icache isn't a user-visible syncing function.
4930 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4931 * we start and wait on commits.
4933 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4935 struct ext4_iloc iloc;
4936 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4937 static unsigned int mnt_count;
4941 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4942 err = ext4_reserve_inode_write(handle, inode, &iloc);
4943 if (ext4_handle_valid(handle) &&
4944 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4945 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4947 * We need extra buffer credits since we may write into EA block
4948 * with this same handle. If journal_extend fails, then it will
4949 * only result in a minor loss of functionality for that inode.
4950 * If this is felt to be critical, then e2fsck should be run to
4951 * force a large enough s_min_extra_isize.
4953 if ((jbd2_journal_extend(handle,
4954 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4955 ret = ext4_expand_extra_isize(inode,
4956 sbi->s_want_extra_isize,
4959 ext4_set_inode_state(inode,
4960 EXT4_STATE_NO_EXPAND);
4962 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4963 ext4_warning(inode->i_sb,
4964 "Unable to expand inode %lu. Delete"
4965 " some EAs or run e2fsck.",
4968 le16_to_cpu(sbi->s_es->s_mnt_count);
4974 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4979 * ext4_dirty_inode() is called from __mark_inode_dirty()
4981 * We're really interested in the case where a file is being extended.
4982 * i_size has been changed by generic_commit_write() and we thus need
4983 * to include the updated inode in the current transaction.
4985 * Also, dquot_alloc_block() will always dirty the inode when blocks
4986 * are allocated to the file.
4988 * If the inode is marked synchronous, we don't honour that here - doing
4989 * so would cause a commit on atime updates, which we don't bother doing.
4990 * We handle synchronous inodes at the highest possible level.
4992 void ext4_dirty_inode(struct inode *inode, int flags)
4996 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5000 ext4_mark_inode_dirty(handle, inode);
5002 ext4_journal_stop(handle);
5009 * Bind an inode's backing buffer_head into this transaction, to prevent
5010 * it from being flushed to disk early. Unlike
5011 * ext4_reserve_inode_write, this leaves behind no bh reference and
5012 * returns no iloc structure, so the caller needs to repeat the iloc
5013 * lookup to mark the inode dirty later.
5015 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5017 struct ext4_iloc iloc;
5021 err = ext4_get_inode_loc(inode, &iloc);
5023 BUFFER_TRACE(iloc.bh, "get_write_access");
5024 err = jbd2_journal_get_write_access(handle, iloc.bh);
5026 err = ext4_handle_dirty_metadata(handle,
5032 ext4_std_error(inode->i_sb, err);
5037 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5044 * We have to be very careful here: changing a data block's
5045 * journaling status dynamically is dangerous. If we write a
5046 * data block to the journal, change the status and then delete
5047 * that block, we risk forgetting to revoke the old log record
5048 * from the journal and so a subsequent replay can corrupt data.
5049 * So, first we make sure that the journal is empty and that
5050 * nobody is changing anything.
5053 journal = EXT4_JOURNAL(inode);
5056 if (is_journal_aborted(journal))
5058 /* We have to allocate physical blocks for delalloc blocks
5059 * before flushing journal. otherwise delalloc blocks can not
5060 * be allocated any more. even more truncate on delalloc blocks
5061 * could trigger BUG by flushing delalloc blocks in journal.
5062 * There is no delalloc block in non-journal data mode.
5064 if (val && test_opt(inode->i_sb, DELALLOC)) {
5065 err = ext4_alloc_da_blocks(inode);
5070 /* Wait for all existing dio workers */
5071 ext4_inode_block_unlocked_dio(inode);
5072 inode_dio_wait(inode);
5074 jbd2_journal_lock_updates(journal);
5077 * OK, there are no updates running now, and all cached data is
5078 * synced to disk. We are now in a completely consistent state
5079 * which doesn't have anything in the journal, and we know that
5080 * no filesystem updates are running, so it is safe to modify
5081 * the inode's in-core data-journaling state flag now.
5085 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5087 jbd2_journal_flush(journal);
5088 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5090 ext4_set_aops(inode);
5092 jbd2_journal_unlock_updates(journal);
5093 ext4_inode_resume_unlocked_dio(inode);
5095 /* Finally we can mark the inode as dirty. */
5097 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5099 return PTR_ERR(handle);
5101 err = ext4_mark_inode_dirty(handle, inode);
5102 ext4_handle_sync(handle);
5103 ext4_journal_stop(handle);
5104 ext4_std_error(inode->i_sb, err);
5109 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5111 return !buffer_mapped(bh);
5114 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5116 struct page *page = vmf->page;
5120 struct file *file = vma->vm_file;
5121 struct inode *inode = file_inode(file);
5122 struct address_space *mapping = inode->i_mapping;
5124 get_block_t *get_block;
5127 sb_start_pagefault(inode->i_sb);
5128 file_update_time(vma->vm_file);
5129 /* Delalloc case is easy... */
5130 if (test_opt(inode->i_sb, DELALLOC) &&
5131 !ext4_should_journal_data(inode) &&
5132 !ext4_nonda_switch(inode->i_sb)) {
5134 ret = __block_page_mkwrite(vma, vmf,
5135 ext4_da_get_block_prep);
5136 } while (ret == -ENOSPC &&
5137 ext4_should_retry_alloc(inode->i_sb, &retries));
5142 size = i_size_read(inode);
5143 /* Page got truncated from under us? */
5144 if (page->mapping != mapping || page_offset(page) > size) {
5146 ret = VM_FAULT_NOPAGE;
5150 if (page->index == size >> PAGE_CACHE_SHIFT)
5151 len = size & ~PAGE_CACHE_MASK;
5153 len = PAGE_CACHE_SIZE;
5155 * Return if we have all the buffers mapped. This avoids the need to do
5156 * journal_start/journal_stop which can block and take a long time
5158 if (page_has_buffers(page)) {
5159 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5161 ext4_bh_unmapped)) {
5162 /* Wait so that we don't change page under IO */
5163 wait_for_stable_page(page);
5164 ret = VM_FAULT_LOCKED;
5169 /* OK, we need to fill the hole... */
5170 if (ext4_should_dioread_nolock(inode))
5171 get_block = ext4_get_block_write;
5173 get_block = ext4_get_block;
5175 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5176 ext4_writepage_trans_blocks(inode));
5177 if (IS_ERR(handle)) {
5178 ret = VM_FAULT_SIGBUS;
5181 ret = __block_page_mkwrite(vma, vmf, get_block);
5182 if (!ret && ext4_should_journal_data(inode)) {
5183 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5184 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5186 ret = VM_FAULT_SIGBUS;
5187 ext4_journal_stop(handle);
5190 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5192 ext4_journal_stop(handle);
5193 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5196 ret = block_page_mkwrite_return(ret);
5198 sb_end_pagefault(inode->i_sb);