2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/aio.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
60 csum_lo = le16_to_cpu(raw->i_checksum_lo);
61 raw->i_checksum_lo = 0;
62 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
63 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
64 csum_hi = le16_to_cpu(raw->i_checksum_hi);
65 raw->i_checksum_hi = 0;
68 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
69 EXT4_INODE_SIZE(inode->i_sb));
71 raw->i_checksum_lo = cpu_to_le16(csum_lo);
72 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
73 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
74 raw->i_checksum_hi = cpu_to_le16(csum_hi);
79 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
80 struct ext4_inode_info *ei)
82 __u32 provided, calculated;
84 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
85 cpu_to_le32(EXT4_OS_LINUX) ||
86 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
87 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
90 provided = le16_to_cpu(raw->i_checksum_lo);
91 calculated = ext4_inode_csum(inode, raw, ei);
92 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
93 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
94 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
98 return provided == calculated;
101 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
102 struct ext4_inode_info *ei)
106 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
107 cpu_to_le32(EXT4_OS_LINUX) ||
108 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
109 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
112 csum = ext4_inode_csum(inode, raw, ei);
113 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
114 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
115 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
116 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
119 static inline int ext4_begin_ordered_truncate(struct inode *inode,
122 trace_ext4_begin_ordered_truncate(inode, new_size);
124 * If jinode is zero, then we never opened the file for
125 * writing, so there's no need to call
126 * jbd2_journal_begin_ordered_truncate() since there's no
127 * outstanding writes we need to flush.
129 if (!EXT4_I(inode)->jinode)
131 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
132 EXT4_I(inode)->jinode,
136 static void ext4_invalidatepage(struct page *page, unsigned int offset,
137 unsigned int length);
138 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
139 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
140 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
144 * Test whether an inode is a fast symlink.
146 static int ext4_inode_is_fast_symlink(struct inode *inode)
148 int ea_blocks = EXT4_I(inode)->i_file_acl ?
149 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
151 if (ext4_has_inline_data(inode))
154 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
158 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against
162 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
168 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
169 * moment, get_block can be called only for blocks inside i_size since
170 * page cache has been already dropped and writes are blocked by
171 * i_mutex. So we can safely drop the i_data_sem here.
173 BUG_ON(EXT4_JOURNAL(inode) == NULL);
174 jbd_debug(2, "restarting handle %p\n", handle);
175 up_write(&EXT4_I(inode)->i_data_sem);
176 ret = ext4_journal_restart(handle, nblocks);
177 down_write(&EXT4_I(inode)->i_data_sem);
178 ext4_discard_preallocations(inode);
184 * Called at the last iput() if i_nlink is zero.
186 void ext4_evict_inode(struct inode *inode)
191 trace_ext4_evict_inode(inode);
193 if (inode->i_nlink) {
195 * When journalling data dirty buffers are tracked only in the
196 * journal. So although mm thinks everything is clean and
197 * ready for reaping the inode might still have some pages to
198 * write in the running transaction or waiting to be
199 * checkpointed. Thus calling jbd2_journal_invalidatepage()
200 * (via truncate_inode_pages()) to discard these buffers can
201 * cause data loss. Also even if we did not discard these
202 * buffers, we would have no way to find them after the inode
203 * is reaped and thus user could see stale data if he tries to
204 * read them before the transaction is checkpointed. So be
205 * careful and force everything to disk here... We use
206 * ei->i_datasync_tid to store the newest transaction
207 * containing inode's data.
209 * Note that directories do not have this problem because they
210 * don't use page cache.
212 if (ext4_should_journal_data(inode) &&
213 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
214 inode->i_ino != EXT4_JOURNAL_INO) {
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
218 jbd2_complete_transaction(journal, commit_tid);
219 filemap_write_and_wait(&inode->i_data);
221 truncate_inode_pages_final(&inode->i_data);
223 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
227 if (!is_bad_inode(inode))
228 dquot_initialize(inode);
230 if (ext4_should_order_data(inode))
231 ext4_begin_ordered_truncate(inode, 0);
232 truncate_inode_pages_final(&inode->i_data);
234 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
235 if (is_bad_inode(inode))
239 * Protect us against freezing - iput() caller didn't have to have any
240 * protection against it
242 sb_start_intwrite(inode->i_sb);
243 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
244 ext4_blocks_for_truncate(inode)+3);
245 if (IS_ERR(handle)) {
246 ext4_std_error(inode->i_sb, PTR_ERR(handle));
248 * If we're going to skip the normal cleanup, we still need to
249 * make sure that the in-core orphan linked list is properly
252 ext4_orphan_del(NULL, inode);
253 sb_end_intwrite(inode->i_sb);
258 ext4_handle_sync(handle);
260 err = ext4_mark_inode_dirty(handle, inode);
262 ext4_warning(inode->i_sb,
263 "couldn't mark inode dirty (err %d)", err);
267 ext4_truncate(inode);
270 * ext4_ext_truncate() doesn't reserve any slop when it
271 * restarts journal transactions; therefore there may not be
272 * enough credits left in the handle to remove the inode from
273 * the orphan list and set the dtime field.
275 if (!ext4_handle_has_enough_credits(handle, 3)) {
276 err = ext4_journal_extend(handle, 3);
278 err = ext4_journal_restart(handle, 3);
280 ext4_warning(inode->i_sb,
281 "couldn't extend journal (err %d)", err);
283 ext4_journal_stop(handle);
284 ext4_orphan_del(NULL, inode);
285 sb_end_intwrite(inode->i_sb);
291 * Kill off the orphan record which ext4_truncate created.
292 * AKPM: I think this can be inside the above `if'.
293 * Note that ext4_orphan_del() has to be able to cope with the
294 * deletion of a non-existent orphan - this is because we don't
295 * know if ext4_truncate() actually created an orphan record.
296 * (Well, we could do this if we need to, but heck - it works)
298 ext4_orphan_del(handle, inode);
299 EXT4_I(inode)->i_dtime = get_seconds();
302 * One subtle ordering requirement: if anything has gone wrong
303 * (transaction abort, IO errors, whatever), then we can still
304 * do these next steps (the fs will already have been marked as
305 * having errors), but we can't free the inode if the mark_dirty
308 if (ext4_mark_inode_dirty(handle, inode))
309 /* If that failed, just do the required in-core inode clear. */
310 ext4_clear_inode(inode);
312 ext4_free_inode(handle, inode);
313 ext4_journal_stop(handle);
314 sb_end_intwrite(inode->i_sb);
317 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
321 qsize_t *ext4_get_reserved_space(struct inode *inode)
323 return &EXT4_I(inode)->i_reserved_quota;
328 * Calculate the number of metadata blocks need to reserve
329 * to allocate a block located at @lblock
331 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
333 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
334 return ext4_ext_calc_metadata_amount(inode, lblock);
336 return ext4_ind_calc_metadata_amount(inode, lblock);
340 * Called with i_data_sem down, which is important since we can call
341 * ext4_discard_preallocations() from here.
343 void ext4_da_update_reserve_space(struct inode *inode,
344 int used, int quota_claim)
346 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
347 struct ext4_inode_info *ei = EXT4_I(inode);
349 spin_lock(&ei->i_block_reservation_lock);
350 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
351 if (unlikely(used > ei->i_reserved_data_blocks)) {
352 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
353 "with only %d reserved data blocks",
354 __func__, inode->i_ino, used,
355 ei->i_reserved_data_blocks);
357 used = ei->i_reserved_data_blocks;
360 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
361 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
362 "with only %d reserved metadata blocks "
363 "(releasing %d blocks with reserved %d data blocks)",
364 inode->i_ino, ei->i_allocated_meta_blocks,
365 ei->i_reserved_meta_blocks, used,
366 ei->i_reserved_data_blocks);
368 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
371 /* Update per-inode reservations */
372 ei->i_reserved_data_blocks -= used;
373 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
374 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
375 used + ei->i_allocated_meta_blocks);
376 ei->i_allocated_meta_blocks = 0;
378 if (ei->i_reserved_data_blocks == 0) {
380 * We can release all of the reserved metadata blocks
381 * only when we have written all of the delayed
384 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
385 ei->i_reserved_meta_blocks);
386 ei->i_reserved_meta_blocks = 0;
387 ei->i_da_metadata_calc_len = 0;
389 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
391 /* Update quota subsystem for data blocks */
393 dquot_claim_block(inode, EXT4_C2B(sbi, used));
396 * We did fallocate with an offset that is already delayed
397 * allocated. So on delayed allocated writeback we should
398 * not re-claim the quota for fallocated blocks.
400 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
404 * If we have done all the pending block allocations and if
405 * there aren't any writers on the inode, we can discard the
406 * inode's preallocations.
408 if ((ei->i_reserved_data_blocks == 0) &&
409 (atomic_read(&inode->i_writecount) == 0))
410 ext4_discard_preallocations(inode);
413 static int __check_block_validity(struct inode *inode, const char *func,
415 struct ext4_map_blocks *map)
417 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
419 ext4_error_inode(inode, func, line, map->m_pblk,
420 "lblock %lu mapped to illegal pblock "
421 "(length %d)", (unsigned long) map->m_lblk,
428 #define check_block_validity(inode, map) \
429 __check_block_validity((inode), __func__, __LINE__, (map))
431 #ifdef ES_AGGRESSIVE_TEST
432 static void ext4_map_blocks_es_recheck(handle_t *handle,
434 struct ext4_map_blocks *es_map,
435 struct ext4_map_blocks *map,
442 * There is a race window that the result is not the same.
443 * e.g. xfstests #223 when dioread_nolock enables. The reason
444 * is that we lookup a block mapping in extent status tree with
445 * out taking i_data_sem. So at the time the unwritten extent
446 * could be converted.
448 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
449 down_read(&EXT4_I(inode)->i_data_sem);
450 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
451 retval = ext4_ext_map_blocks(handle, inode, map, flags &
452 EXT4_GET_BLOCKS_KEEP_SIZE);
454 retval = ext4_ind_map_blocks(handle, inode, map, flags &
455 EXT4_GET_BLOCKS_KEEP_SIZE);
457 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
458 up_read((&EXT4_I(inode)->i_data_sem));
460 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
461 * because it shouldn't be marked in es_map->m_flags.
463 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
466 * We don't check m_len because extent will be collpased in status
467 * tree. So the m_len might not equal.
469 if (es_map->m_lblk != map->m_lblk ||
470 es_map->m_flags != map->m_flags ||
471 es_map->m_pblk != map->m_pblk) {
472 printk("ES cache assertion failed for inode: %lu "
473 "es_cached ex [%d/%d/%llu/%x] != "
474 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
475 inode->i_ino, es_map->m_lblk, es_map->m_len,
476 es_map->m_pblk, es_map->m_flags, map->m_lblk,
477 map->m_len, map->m_pblk, map->m_flags,
481 #endif /* ES_AGGRESSIVE_TEST */
484 * The ext4_map_blocks() function tries to look up the requested blocks,
485 * and returns if the blocks are already mapped.
487 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
488 * and store the allocated blocks in the result buffer head and mark it
491 * If file type is extents based, it will call ext4_ext_map_blocks(),
492 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
495 * On success, it returns the number of blocks being mapped or allocated.
496 * if create==0 and the blocks are pre-allocated and unwritten block,
497 * the result buffer head is unmapped. If the create ==1, it will make sure
498 * the buffer head is mapped.
500 * It returns 0 if plain look up failed (blocks have not been allocated), in
501 * that case, buffer head is unmapped
503 * It returns the error in case of allocation failure.
505 int ext4_map_blocks(handle_t *handle, struct inode *inode,
506 struct ext4_map_blocks *map, int flags)
508 struct extent_status es;
511 #ifdef ES_AGGRESSIVE_TEST
512 struct ext4_map_blocks orig_map;
514 memcpy(&orig_map, map, sizeof(*map));
518 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
519 "logical block %lu\n", inode->i_ino, flags, map->m_len,
520 (unsigned long) map->m_lblk);
523 * ext4_map_blocks returns an int, and m_len is an unsigned int
525 if (unlikely(map->m_len > INT_MAX))
526 map->m_len = INT_MAX;
528 /* We can handle the block number less than EXT_MAX_BLOCKS */
529 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
532 /* Lookup extent status tree firstly */
533 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
534 ext4_es_lru_add(inode);
535 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
536 map->m_pblk = ext4_es_pblock(&es) +
537 map->m_lblk - es.es_lblk;
538 map->m_flags |= ext4_es_is_written(&es) ?
539 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
540 retval = es.es_len - (map->m_lblk - es.es_lblk);
541 if (retval > map->m_len)
544 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
549 #ifdef ES_AGGRESSIVE_TEST
550 ext4_map_blocks_es_recheck(handle, inode, map,
557 * Try to see if we can get the block without requesting a new
560 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
561 down_read(&EXT4_I(inode)->i_data_sem);
562 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
563 retval = ext4_ext_map_blocks(handle, inode, map, flags &
564 EXT4_GET_BLOCKS_KEEP_SIZE);
566 retval = ext4_ind_map_blocks(handle, inode, map, flags &
567 EXT4_GET_BLOCKS_KEEP_SIZE);
572 if (unlikely(retval != map->m_len)) {
573 ext4_warning(inode->i_sb,
574 "ES len assertion failed for inode "
575 "%lu: retval %d != map->m_len %d",
576 inode->i_ino, retval, map->m_len);
580 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
581 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
582 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
583 ext4_find_delalloc_range(inode, map->m_lblk,
584 map->m_lblk + map->m_len - 1))
585 status |= EXTENT_STATUS_DELAYED;
586 ret = ext4_es_insert_extent(inode, map->m_lblk,
587 map->m_len, map->m_pblk, status);
591 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
592 up_read((&EXT4_I(inode)->i_data_sem));
595 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
596 ret = check_block_validity(inode, map);
601 /* If it is only a block(s) look up */
602 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
606 * Returns if the blocks have already allocated
608 * Note that if blocks have been preallocated
609 * ext4_ext_get_block() returns the create = 0
610 * with buffer head unmapped.
612 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
614 * If we need to convert extent to unwritten
615 * we continue and do the actual work in
616 * ext4_ext_map_blocks()
618 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
622 * Here we clear m_flags because after allocating an new extent,
623 * it will be set again.
625 map->m_flags &= ~EXT4_MAP_FLAGS;
628 * New blocks allocate and/or writing to unwritten extent
629 * will possibly result in updating i_data, so we take
630 * the write lock of i_data_sem, and call get_blocks()
631 * with create == 1 flag.
633 down_write(&EXT4_I(inode)->i_data_sem);
636 * if the caller is from delayed allocation writeout path
637 * we have already reserved fs blocks for allocation
638 * let the underlying get_block() function know to
639 * avoid double accounting
641 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
642 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
644 * We need to check for EXT4 here because migrate
645 * could have changed the inode type in between
647 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
648 retval = ext4_ext_map_blocks(handle, inode, map, flags);
650 retval = ext4_ind_map_blocks(handle, inode, map, flags);
652 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
654 * We allocated new blocks which will result in
655 * i_data's format changing. Force the migrate
656 * to fail by clearing migrate flags
658 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
662 * Update reserved blocks/metadata blocks after successful
663 * block allocation which had been deferred till now. We don't
664 * support fallocate for non extent files. So we can update
665 * reserve space here.
668 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
669 ext4_da_update_reserve_space(inode, retval, 1);
671 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
672 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
677 if (unlikely(retval != map->m_len)) {
678 ext4_warning(inode->i_sb,
679 "ES len assertion failed for inode "
680 "%lu: retval %d != map->m_len %d",
681 inode->i_ino, retval, map->m_len);
686 * If the extent has been zeroed out, we don't need to update
687 * extent status tree.
689 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
690 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
691 if (ext4_es_is_written(&es))
694 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
695 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
696 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
697 ext4_find_delalloc_range(inode, map->m_lblk,
698 map->m_lblk + map->m_len - 1))
699 status |= EXTENT_STATUS_DELAYED;
700 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
701 map->m_pblk, status);
707 up_write((&EXT4_I(inode)->i_data_sem));
708 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
709 ret = check_block_validity(inode, map);
716 /* Maximum number of blocks we map for direct IO at once. */
717 #define DIO_MAX_BLOCKS 4096
719 static int _ext4_get_block(struct inode *inode, sector_t iblock,
720 struct buffer_head *bh, int flags)
722 handle_t *handle = ext4_journal_current_handle();
723 struct ext4_map_blocks map;
724 int ret = 0, started = 0;
727 if (ext4_has_inline_data(inode))
731 map.m_len = bh->b_size >> inode->i_blkbits;
733 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
734 /* Direct IO write... */
735 if (map.m_len > DIO_MAX_BLOCKS)
736 map.m_len = DIO_MAX_BLOCKS;
737 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
738 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
740 if (IS_ERR(handle)) {
741 ret = PTR_ERR(handle);
747 ret = ext4_map_blocks(handle, inode, &map, flags);
749 ext4_io_end_t *io_end = ext4_inode_aio(inode);
751 map_bh(bh, inode->i_sb, map.m_pblk);
752 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
753 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
754 set_buffer_defer_completion(bh);
755 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
759 ext4_journal_stop(handle);
763 int ext4_get_block(struct inode *inode, sector_t iblock,
764 struct buffer_head *bh, int create)
766 return _ext4_get_block(inode, iblock, bh,
767 create ? EXT4_GET_BLOCKS_CREATE : 0);
771 * `handle' can be NULL if create is zero
773 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
774 ext4_lblk_t block, int create, int *errp)
776 struct ext4_map_blocks map;
777 struct buffer_head *bh;
780 J_ASSERT(handle != NULL || create == 0);
784 err = ext4_map_blocks(handle, inode, &map,
785 create ? EXT4_GET_BLOCKS_CREATE : 0);
787 /* ensure we send some value back into *errp */
790 if (create && err == 0)
791 err = -ENOSPC; /* should never happen */
797 bh = sb_getblk(inode->i_sb, map.m_pblk);
802 if (map.m_flags & EXT4_MAP_NEW) {
803 J_ASSERT(create != 0);
804 J_ASSERT(handle != NULL);
807 * Now that we do not always journal data, we should
808 * keep in mind whether this should always journal the
809 * new buffer as metadata. For now, regular file
810 * writes use ext4_get_block instead, so it's not a
814 BUFFER_TRACE(bh, "call get_create_access");
815 fatal = ext4_journal_get_create_access(handle, bh);
816 if (!fatal && !buffer_uptodate(bh)) {
817 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
818 set_buffer_uptodate(bh);
821 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
822 err = ext4_handle_dirty_metadata(handle, inode, bh);
826 BUFFER_TRACE(bh, "not a new buffer");
836 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
837 ext4_lblk_t block, int create, int *err)
839 struct buffer_head *bh;
841 bh = ext4_getblk(handle, inode, block, create, err);
844 if (buffer_uptodate(bh))
846 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
848 if (buffer_uptodate(bh))
855 int ext4_walk_page_buffers(handle_t *handle,
856 struct buffer_head *head,
860 int (*fn)(handle_t *handle,
861 struct buffer_head *bh))
863 struct buffer_head *bh;
864 unsigned block_start, block_end;
865 unsigned blocksize = head->b_size;
867 struct buffer_head *next;
869 for (bh = head, block_start = 0;
870 ret == 0 && (bh != head || !block_start);
871 block_start = block_end, bh = next) {
872 next = bh->b_this_page;
873 block_end = block_start + blocksize;
874 if (block_end <= from || block_start >= to) {
875 if (partial && !buffer_uptodate(bh))
879 err = (*fn)(handle, bh);
887 * To preserve ordering, it is essential that the hole instantiation and
888 * the data write be encapsulated in a single transaction. We cannot
889 * close off a transaction and start a new one between the ext4_get_block()
890 * and the commit_write(). So doing the jbd2_journal_start at the start of
891 * prepare_write() is the right place.
893 * Also, this function can nest inside ext4_writepage(). In that case, we
894 * *know* that ext4_writepage() has generated enough buffer credits to do the
895 * whole page. So we won't block on the journal in that case, which is good,
896 * because the caller may be PF_MEMALLOC.
898 * By accident, ext4 can be reentered when a transaction is open via
899 * quota file writes. If we were to commit the transaction while thus
900 * reentered, there can be a deadlock - we would be holding a quota
901 * lock, and the commit would never complete if another thread had a
902 * transaction open and was blocking on the quota lock - a ranking
905 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
906 * will _not_ run commit under these circumstances because handle->h_ref
907 * is elevated. We'll still have enough credits for the tiny quotafile
910 int do_journal_get_write_access(handle_t *handle,
911 struct buffer_head *bh)
913 int dirty = buffer_dirty(bh);
916 if (!buffer_mapped(bh) || buffer_freed(bh))
919 * __block_write_begin() could have dirtied some buffers. Clean
920 * the dirty bit as jbd2_journal_get_write_access() could complain
921 * otherwise about fs integrity issues. Setting of the dirty bit
922 * by __block_write_begin() isn't a real problem here as we clear
923 * the bit before releasing a page lock and thus writeback cannot
924 * ever write the buffer.
927 clear_buffer_dirty(bh);
928 BUFFER_TRACE(bh, "get write access");
929 ret = ext4_journal_get_write_access(handle, bh);
931 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
935 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
936 struct buffer_head *bh_result, int create);
937 static int ext4_write_begin(struct file *file, struct address_space *mapping,
938 loff_t pos, unsigned len, unsigned flags,
939 struct page **pagep, void **fsdata)
941 struct inode *inode = mapping->host;
942 int ret, needed_blocks;
949 trace_ext4_write_begin(inode, pos, len, flags);
951 * Reserve one block more for addition to orphan list in case
952 * we allocate blocks but write fails for some reason
954 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
955 index = pos >> PAGE_CACHE_SHIFT;
956 from = pos & (PAGE_CACHE_SIZE - 1);
959 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
960 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
969 * grab_cache_page_write_begin() can take a long time if the
970 * system is thrashing due to memory pressure, or if the page
971 * is being written back. So grab it first before we start
972 * the transaction handle. This also allows us to allocate
973 * the page (if needed) without using GFP_NOFS.
976 page = grab_cache_page_write_begin(mapping, index, flags);
982 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
983 if (IS_ERR(handle)) {
984 page_cache_release(page);
985 return PTR_ERR(handle);
989 if (page->mapping != mapping) {
990 /* The page got truncated from under us */
992 page_cache_release(page);
993 ext4_journal_stop(handle);
996 /* In case writeback began while the page was unlocked */
997 wait_for_stable_page(page);
999 if (ext4_should_dioread_nolock(inode))
1000 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1002 ret = __block_write_begin(page, pos, len, ext4_get_block);
1004 if (!ret && ext4_should_journal_data(inode)) {
1005 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1007 do_journal_get_write_access);
1013 * __block_write_begin may have instantiated a few blocks
1014 * outside i_size. Trim these off again. Don't need
1015 * i_size_read because we hold i_mutex.
1017 * Add inode to orphan list in case we crash before
1020 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1021 ext4_orphan_add(handle, inode);
1023 ext4_journal_stop(handle);
1024 if (pos + len > inode->i_size) {
1025 ext4_truncate_failed_write(inode);
1027 * If truncate failed early the inode might
1028 * still be on the orphan list; we need to
1029 * make sure the inode is removed from the
1030 * orphan list in that case.
1033 ext4_orphan_del(NULL, inode);
1036 if (ret == -ENOSPC &&
1037 ext4_should_retry_alloc(inode->i_sb, &retries))
1039 page_cache_release(page);
1046 /* For write_end() in data=journal mode */
1047 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1050 if (!buffer_mapped(bh) || buffer_freed(bh))
1052 set_buffer_uptodate(bh);
1053 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1054 clear_buffer_meta(bh);
1055 clear_buffer_prio(bh);
1060 * We need to pick up the new inode size which generic_commit_write gave us
1061 * `file' can be NULL - eg, when called from page_symlink().
1063 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1064 * buffers are managed internally.
1066 static int ext4_write_end(struct file *file,
1067 struct address_space *mapping,
1068 loff_t pos, unsigned len, unsigned copied,
1069 struct page *page, void *fsdata)
1071 handle_t *handle = ext4_journal_current_handle();
1072 struct inode *inode = mapping->host;
1074 int i_size_changed = 0;
1076 trace_ext4_write_end(inode, pos, len, copied);
1077 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1078 ret = ext4_jbd2_file_inode(handle, inode);
1081 page_cache_release(page);
1086 if (ext4_has_inline_data(inode)) {
1087 ret = ext4_write_inline_data_end(inode, pos, len,
1093 copied = block_write_end(file, mapping, pos,
1094 len, copied, page, fsdata);
1097 * No need to use i_size_read() here, the i_size
1098 * cannot change under us because we hole i_mutex.
1100 * But it's important to update i_size while still holding page lock:
1101 * page writeout could otherwise come in and zero beyond i_size.
1103 if (pos + copied > inode->i_size) {
1104 i_size_write(inode, pos + copied);
1108 if (pos + copied > EXT4_I(inode)->i_disksize) {
1109 /* We need to mark inode dirty even if
1110 * new_i_size is less that inode->i_size
1111 * but greater than i_disksize. (hint delalloc)
1113 ext4_update_i_disksize(inode, (pos + copied));
1117 page_cache_release(page);
1120 * Don't mark the inode dirty under page lock. First, it unnecessarily
1121 * makes the holding time of page lock longer. Second, it forces lock
1122 * ordering of page lock and transaction start for journaling
1126 ext4_mark_inode_dirty(handle, inode);
1128 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1129 /* if we have allocated more blocks and copied
1130 * less. We will have blocks allocated outside
1131 * inode->i_size. So truncate them
1133 ext4_orphan_add(handle, inode);
1135 ret2 = ext4_journal_stop(handle);
1139 if (pos + len > inode->i_size) {
1140 ext4_truncate_failed_write(inode);
1142 * If truncate failed early the inode might still be
1143 * on the orphan list; we need to make sure the inode
1144 * is removed from the orphan list in that case.
1147 ext4_orphan_del(NULL, inode);
1150 return ret ? ret : copied;
1153 static int ext4_journalled_write_end(struct file *file,
1154 struct address_space *mapping,
1155 loff_t pos, unsigned len, unsigned copied,
1156 struct page *page, void *fsdata)
1158 handle_t *handle = ext4_journal_current_handle();
1159 struct inode *inode = mapping->host;
1165 trace_ext4_journalled_write_end(inode, pos, len, copied);
1166 from = pos & (PAGE_CACHE_SIZE - 1);
1169 BUG_ON(!ext4_handle_valid(handle));
1171 if (ext4_has_inline_data(inode))
1172 copied = ext4_write_inline_data_end(inode, pos, len,
1176 if (!PageUptodate(page))
1178 page_zero_new_buffers(page, from+copied, to);
1181 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1182 to, &partial, write_end_fn);
1184 SetPageUptodate(page);
1186 new_i_size = pos + copied;
1187 if (new_i_size > inode->i_size)
1188 i_size_write(inode, pos+copied);
1189 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1190 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1191 if (new_i_size > EXT4_I(inode)->i_disksize) {
1192 ext4_update_i_disksize(inode, new_i_size);
1193 ret2 = ext4_mark_inode_dirty(handle, inode);
1199 page_cache_release(page);
1200 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1201 /* if we have allocated more blocks and copied
1202 * less. We will have blocks allocated outside
1203 * inode->i_size. So truncate them
1205 ext4_orphan_add(handle, inode);
1207 ret2 = ext4_journal_stop(handle);
1210 if (pos + len > inode->i_size) {
1211 ext4_truncate_failed_write(inode);
1213 * If truncate failed early the inode might still be
1214 * on the orphan list; we need to make sure the inode
1215 * is removed from the orphan list in that case.
1218 ext4_orphan_del(NULL, inode);
1221 return ret ? ret : copied;
1225 * Reserve a metadata for a single block located at lblock
1227 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1229 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1230 struct ext4_inode_info *ei = EXT4_I(inode);
1231 unsigned int md_needed;
1232 ext4_lblk_t save_last_lblock;
1236 * recalculate the amount of metadata blocks to reserve
1237 * in order to allocate nrblocks
1238 * worse case is one extent per block
1240 spin_lock(&ei->i_block_reservation_lock);
1242 * ext4_calc_metadata_amount() has side effects, which we have
1243 * to be prepared undo if we fail to claim space.
1245 save_len = ei->i_da_metadata_calc_len;
1246 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1247 md_needed = EXT4_NUM_B2C(sbi,
1248 ext4_calc_metadata_amount(inode, lblock));
1249 trace_ext4_da_reserve_space(inode, md_needed);
1252 * We do still charge estimated metadata to the sb though;
1253 * we cannot afford to run out of free blocks.
1255 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1256 ei->i_da_metadata_calc_len = save_len;
1257 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1258 spin_unlock(&ei->i_block_reservation_lock);
1261 ei->i_reserved_meta_blocks += md_needed;
1262 spin_unlock(&ei->i_block_reservation_lock);
1264 return 0; /* success */
1268 * Reserve a single cluster located at lblock
1270 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1272 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1273 struct ext4_inode_info *ei = EXT4_I(inode);
1274 unsigned int md_needed;
1276 ext4_lblk_t save_last_lblock;
1280 * We will charge metadata quota at writeout time; this saves
1281 * us from metadata over-estimation, though we may go over by
1282 * a small amount in the end. Here we just reserve for data.
1284 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1289 * recalculate the amount of metadata blocks to reserve
1290 * in order to allocate nrblocks
1291 * worse case is one extent per block
1293 spin_lock(&ei->i_block_reservation_lock);
1295 * ext4_calc_metadata_amount() has side effects, which we have
1296 * to be prepared undo if we fail to claim space.
1298 save_len = ei->i_da_metadata_calc_len;
1299 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1300 md_needed = EXT4_NUM_B2C(sbi,
1301 ext4_calc_metadata_amount(inode, lblock));
1302 trace_ext4_da_reserve_space(inode, md_needed);
1305 * We do still charge estimated metadata to the sb though;
1306 * we cannot afford to run out of free blocks.
1308 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1309 ei->i_da_metadata_calc_len = save_len;
1310 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1311 spin_unlock(&ei->i_block_reservation_lock);
1312 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1315 ei->i_reserved_data_blocks++;
1316 ei->i_reserved_meta_blocks += md_needed;
1317 spin_unlock(&ei->i_block_reservation_lock);
1319 return 0; /* success */
1322 static void ext4_da_release_space(struct inode *inode, int to_free)
1324 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1325 struct ext4_inode_info *ei = EXT4_I(inode);
1328 return; /* Nothing to release, exit */
1330 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1332 trace_ext4_da_release_space(inode, to_free);
1333 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1335 * if there aren't enough reserved blocks, then the
1336 * counter is messed up somewhere. Since this
1337 * function is called from invalidate page, it's
1338 * harmless to return without any action.
1340 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1341 "ino %lu, to_free %d with only %d reserved "
1342 "data blocks", inode->i_ino, to_free,
1343 ei->i_reserved_data_blocks);
1345 to_free = ei->i_reserved_data_blocks;
1347 ei->i_reserved_data_blocks -= to_free;
1349 if (ei->i_reserved_data_blocks == 0) {
1351 * We can release all of the reserved metadata blocks
1352 * only when we have written all of the delayed
1353 * allocation blocks.
1354 * Note that in case of bigalloc, i_reserved_meta_blocks,
1355 * i_reserved_data_blocks, etc. refer to number of clusters.
1357 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1358 ei->i_reserved_meta_blocks);
1359 ei->i_reserved_meta_blocks = 0;
1360 ei->i_da_metadata_calc_len = 0;
1363 /* update fs dirty data blocks counter */
1364 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1366 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1368 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1371 static void ext4_da_page_release_reservation(struct page *page,
1372 unsigned int offset,
1373 unsigned int length)
1376 struct buffer_head *head, *bh;
1377 unsigned int curr_off = 0;
1378 struct inode *inode = page->mapping->host;
1379 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1380 unsigned int stop = offset + length;
1384 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1386 head = page_buffers(page);
1389 unsigned int next_off = curr_off + bh->b_size;
1391 if (next_off > stop)
1394 if ((offset <= curr_off) && (buffer_delay(bh))) {
1396 clear_buffer_delay(bh);
1398 curr_off = next_off;
1399 } while ((bh = bh->b_this_page) != head);
1402 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1403 ext4_es_remove_extent(inode, lblk, to_release);
1406 /* If we have released all the blocks belonging to a cluster, then we
1407 * need to release the reserved space for that cluster. */
1408 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1409 while (num_clusters > 0) {
1410 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1411 ((num_clusters - 1) << sbi->s_cluster_bits);
1412 if (sbi->s_cluster_ratio == 1 ||
1413 !ext4_find_delalloc_cluster(inode, lblk))
1414 ext4_da_release_space(inode, 1);
1421 * Delayed allocation stuff
1424 struct mpage_da_data {
1425 struct inode *inode;
1426 struct writeback_control *wbc;
1428 pgoff_t first_page; /* The first page to write */
1429 pgoff_t next_page; /* Current page to examine */
1430 pgoff_t last_page; /* Last page to examine */
1432 * Extent to map - this can be after first_page because that can be
1433 * fully mapped. We somewhat abuse m_flags to store whether the extent
1434 * is delalloc or unwritten.
1436 struct ext4_map_blocks map;
1437 struct ext4_io_submit io_submit; /* IO submission data */
1440 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1445 struct pagevec pvec;
1446 struct inode *inode = mpd->inode;
1447 struct address_space *mapping = inode->i_mapping;
1449 /* This is necessary when next_page == 0. */
1450 if (mpd->first_page >= mpd->next_page)
1453 index = mpd->first_page;
1454 end = mpd->next_page - 1;
1456 ext4_lblk_t start, last;
1457 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1458 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1459 ext4_es_remove_extent(inode, start, last - start + 1);
1462 pagevec_init(&pvec, 0);
1463 while (index <= end) {
1464 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1467 for (i = 0; i < nr_pages; i++) {
1468 struct page *page = pvec.pages[i];
1469 if (page->index > end)
1471 BUG_ON(!PageLocked(page));
1472 BUG_ON(PageWriteback(page));
1474 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1475 ClearPageUptodate(page);
1479 index = pvec.pages[nr_pages - 1]->index + 1;
1480 pagevec_release(&pvec);
1484 static void ext4_print_free_blocks(struct inode *inode)
1486 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1487 struct super_block *sb = inode->i_sb;
1488 struct ext4_inode_info *ei = EXT4_I(inode);
1490 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1491 EXT4_C2B(EXT4_SB(inode->i_sb),
1492 ext4_count_free_clusters(sb)));
1493 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1494 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1495 (long long) EXT4_C2B(EXT4_SB(sb),
1496 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1497 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1498 (long long) EXT4_C2B(EXT4_SB(sb),
1499 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1500 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1501 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1502 ei->i_reserved_data_blocks);
1503 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1504 ei->i_reserved_meta_blocks);
1505 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1506 ei->i_allocated_meta_blocks);
1510 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1512 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1516 * This function is grabs code from the very beginning of
1517 * ext4_map_blocks, but assumes that the caller is from delayed write
1518 * time. This function looks up the requested blocks and sets the
1519 * buffer delay bit under the protection of i_data_sem.
1521 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1522 struct ext4_map_blocks *map,
1523 struct buffer_head *bh)
1525 struct extent_status es;
1527 sector_t invalid_block = ~((sector_t) 0xffff);
1528 #ifdef ES_AGGRESSIVE_TEST
1529 struct ext4_map_blocks orig_map;
1531 memcpy(&orig_map, map, sizeof(*map));
1534 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1538 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1539 "logical block %lu\n", inode->i_ino, map->m_len,
1540 (unsigned long) map->m_lblk);
1542 /* Lookup extent status tree firstly */
1543 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1544 ext4_es_lru_add(inode);
1545 if (ext4_es_is_hole(&es)) {
1547 down_read(&EXT4_I(inode)->i_data_sem);
1552 * Delayed extent could be allocated by fallocate.
1553 * So we need to check it.
1555 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1556 map_bh(bh, inode->i_sb, invalid_block);
1558 set_buffer_delay(bh);
1562 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1563 retval = es.es_len - (iblock - es.es_lblk);
1564 if (retval > map->m_len)
1565 retval = map->m_len;
1566 map->m_len = retval;
1567 if (ext4_es_is_written(&es))
1568 map->m_flags |= EXT4_MAP_MAPPED;
1569 else if (ext4_es_is_unwritten(&es))
1570 map->m_flags |= EXT4_MAP_UNWRITTEN;
1574 #ifdef ES_AGGRESSIVE_TEST
1575 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1581 * Try to see if we can get the block without requesting a new
1582 * file system block.
1584 down_read(&EXT4_I(inode)->i_data_sem);
1585 if (ext4_has_inline_data(inode)) {
1587 * We will soon create blocks for this page, and let
1588 * us pretend as if the blocks aren't allocated yet.
1589 * In case of clusters, we have to handle the work
1590 * of mapping from cluster so that the reserved space
1591 * is calculated properly.
1593 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1594 ext4_find_delalloc_cluster(inode, map->m_lblk))
1595 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1597 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1598 retval = ext4_ext_map_blocks(NULL, inode, map,
1599 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1601 retval = ext4_ind_map_blocks(NULL, inode, map,
1602 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1608 * XXX: __block_prepare_write() unmaps passed block,
1612 * If the block was allocated from previously allocated cluster,
1613 * then we don't need to reserve it again. However we still need
1614 * to reserve metadata for every block we're going to write.
1616 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1617 ret = ext4_da_reserve_space(inode, iblock);
1619 /* not enough space to reserve */
1624 ret = ext4_da_reserve_metadata(inode, iblock);
1626 /* not enough space to reserve */
1632 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1633 ~0, EXTENT_STATUS_DELAYED);
1639 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1640 * and it should not appear on the bh->b_state.
1642 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1644 map_bh(bh, inode->i_sb, invalid_block);
1646 set_buffer_delay(bh);
1647 } else if (retval > 0) {
1649 unsigned int status;
1651 if (unlikely(retval != map->m_len)) {
1652 ext4_warning(inode->i_sb,
1653 "ES len assertion failed for inode "
1654 "%lu: retval %d != map->m_len %d",
1655 inode->i_ino, retval, map->m_len);
1659 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1660 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1661 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1662 map->m_pblk, status);
1668 up_read((&EXT4_I(inode)->i_data_sem));
1674 * This is a special get_blocks_t callback which is used by
1675 * ext4_da_write_begin(). It will either return mapped block or
1676 * reserve space for a single block.
1678 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1679 * We also have b_blocknr = -1 and b_bdev initialized properly
1681 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1682 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1683 * initialized properly.
1685 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1686 struct buffer_head *bh, int create)
1688 struct ext4_map_blocks map;
1691 BUG_ON(create == 0);
1692 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1694 map.m_lblk = iblock;
1698 * first, we need to know whether the block is allocated already
1699 * preallocated blocks are unmapped but should treated
1700 * the same as allocated blocks.
1702 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1706 map_bh(bh, inode->i_sb, map.m_pblk);
1707 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1709 if (buffer_unwritten(bh)) {
1710 /* A delayed write to unwritten bh should be marked
1711 * new and mapped. Mapped ensures that we don't do
1712 * get_block multiple times when we write to the same
1713 * offset and new ensures that we do proper zero out
1714 * for partial write.
1717 set_buffer_mapped(bh);
1722 static int bget_one(handle_t *handle, struct buffer_head *bh)
1728 static int bput_one(handle_t *handle, struct buffer_head *bh)
1734 static int __ext4_journalled_writepage(struct page *page,
1737 struct address_space *mapping = page->mapping;
1738 struct inode *inode = mapping->host;
1739 struct buffer_head *page_bufs = NULL;
1740 handle_t *handle = NULL;
1741 int ret = 0, err = 0;
1742 int inline_data = ext4_has_inline_data(inode);
1743 struct buffer_head *inode_bh = NULL;
1745 ClearPageChecked(page);
1748 BUG_ON(page->index != 0);
1749 BUG_ON(len > ext4_get_max_inline_size(inode));
1750 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1751 if (inode_bh == NULL)
1754 page_bufs = page_buffers(page);
1759 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1762 /* As soon as we unlock the page, it can go away, but we have
1763 * references to buffers so we are safe */
1766 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1767 ext4_writepage_trans_blocks(inode));
1768 if (IS_ERR(handle)) {
1769 ret = PTR_ERR(handle);
1773 BUG_ON(!ext4_handle_valid(handle));
1776 BUFFER_TRACE(inode_bh, "get write access");
1777 ret = ext4_journal_get_write_access(handle, inode_bh);
1779 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1782 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1783 do_journal_get_write_access);
1785 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1790 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1791 err = ext4_journal_stop(handle);
1795 if (!ext4_has_inline_data(inode))
1796 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1798 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1805 * Note that we don't need to start a transaction unless we're journaling data
1806 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1807 * need to file the inode to the transaction's list in ordered mode because if
1808 * we are writing back data added by write(), the inode is already there and if
1809 * we are writing back data modified via mmap(), no one guarantees in which
1810 * transaction the data will hit the disk. In case we are journaling data, we
1811 * cannot start transaction directly because transaction start ranks above page
1812 * lock so we have to do some magic.
1814 * This function can get called via...
1815 * - ext4_writepages after taking page lock (have journal handle)
1816 * - journal_submit_inode_data_buffers (no journal handle)
1817 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1818 * - grab_page_cache when doing write_begin (have journal handle)
1820 * We don't do any block allocation in this function. If we have page with
1821 * multiple blocks we need to write those buffer_heads that are mapped. This
1822 * is important for mmaped based write. So if we do with blocksize 1K
1823 * truncate(f, 1024);
1824 * a = mmap(f, 0, 4096);
1826 * truncate(f, 4096);
1827 * we have in the page first buffer_head mapped via page_mkwrite call back
1828 * but other buffer_heads would be unmapped but dirty (dirty done via the
1829 * do_wp_page). So writepage should write the first block. If we modify
1830 * the mmap area beyond 1024 we will again get a page_fault and the
1831 * page_mkwrite callback will do the block allocation and mark the
1832 * buffer_heads mapped.
1834 * We redirty the page if we have any buffer_heads that is either delay or
1835 * unwritten in the page.
1837 * We can get recursively called as show below.
1839 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1842 * But since we don't do any block allocation we should not deadlock.
1843 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1845 static int ext4_writepage(struct page *page,
1846 struct writeback_control *wbc)
1851 struct buffer_head *page_bufs = NULL;
1852 struct inode *inode = page->mapping->host;
1853 struct ext4_io_submit io_submit;
1854 bool keep_towrite = false;
1856 trace_ext4_writepage(page);
1857 size = i_size_read(inode);
1858 if (page->index == size >> PAGE_CACHE_SHIFT)
1859 len = size & ~PAGE_CACHE_MASK;
1861 len = PAGE_CACHE_SIZE;
1863 page_bufs = page_buffers(page);
1865 * We cannot do block allocation or other extent handling in this
1866 * function. If there are buffers needing that, we have to redirty
1867 * the page. But we may reach here when we do a journal commit via
1868 * journal_submit_inode_data_buffers() and in that case we must write
1869 * allocated buffers to achieve data=ordered mode guarantees.
1871 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1872 ext4_bh_delay_or_unwritten)) {
1873 redirty_page_for_writepage(wbc, page);
1874 if (current->flags & PF_MEMALLOC) {
1876 * For memory cleaning there's no point in writing only
1877 * some buffers. So just bail out. Warn if we came here
1878 * from direct reclaim.
1880 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1885 keep_towrite = true;
1888 if (PageChecked(page) && ext4_should_journal_data(inode))
1890 * It's mmapped pagecache. Add buffers and journal it. There
1891 * doesn't seem much point in redirtying the page here.
1893 return __ext4_journalled_writepage(page, len);
1895 ext4_io_submit_init(&io_submit, wbc);
1896 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1897 if (!io_submit.io_end) {
1898 redirty_page_for_writepage(wbc, page);
1902 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
1903 ext4_io_submit(&io_submit);
1904 /* Drop io_end reference we got from init */
1905 ext4_put_io_end_defer(io_submit.io_end);
1909 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1912 loff_t size = i_size_read(mpd->inode);
1915 BUG_ON(page->index != mpd->first_page);
1916 if (page->index == size >> PAGE_CACHE_SHIFT)
1917 len = size & ~PAGE_CACHE_MASK;
1919 len = PAGE_CACHE_SIZE;
1920 clear_page_dirty_for_io(page);
1921 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
1923 mpd->wbc->nr_to_write--;
1929 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1932 * mballoc gives us at most this number of blocks...
1933 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1934 * The rest of mballoc seems to handle chunks up to full group size.
1936 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1939 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1941 * @mpd - extent of blocks
1942 * @lblk - logical number of the block in the file
1943 * @bh - buffer head we want to add to the extent
1945 * The function is used to collect contig. blocks in the same state. If the
1946 * buffer doesn't require mapping for writeback and we haven't started the
1947 * extent of buffers to map yet, the function returns 'true' immediately - the
1948 * caller can write the buffer right away. Otherwise the function returns true
1949 * if the block has been added to the extent, false if the block couldn't be
1952 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1953 struct buffer_head *bh)
1955 struct ext4_map_blocks *map = &mpd->map;
1957 /* Buffer that doesn't need mapping for writeback? */
1958 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1959 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1960 /* So far no extent to map => we write the buffer right away */
1961 if (map->m_len == 0)
1966 /* First block in the extent? */
1967 if (map->m_len == 0) {
1970 map->m_flags = bh->b_state & BH_FLAGS;
1974 /* Don't go larger than mballoc is willing to allocate */
1975 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1978 /* Can we merge the block to our big extent? */
1979 if (lblk == map->m_lblk + map->m_len &&
1980 (bh->b_state & BH_FLAGS) == map->m_flags) {
1988 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1990 * @mpd - extent of blocks for mapping
1991 * @head - the first buffer in the page
1992 * @bh - buffer we should start processing from
1993 * @lblk - logical number of the block in the file corresponding to @bh
1995 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1996 * the page for IO if all buffers in this page were mapped and there's no
1997 * accumulated extent of buffers to map or add buffers in the page to the
1998 * extent of buffers to map. The function returns 1 if the caller can continue
1999 * by processing the next page, 0 if it should stop adding buffers to the
2000 * extent to map because we cannot extend it anymore. It can also return value
2001 * < 0 in case of error during IO submission.
2003 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2004 struct buffer_head *head,
2005 struct buffer_head *bh,
2008 struct inode *inode = mpd->inode;
2010 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2011 >> inode->i_blkbits;
2014 BUG_ON(buffer_locked(bh));
2016 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2017 /* Found extent to map? */
2020 /* Everything mapped so far and we hit EOF */
2023 } while (lblk++, (bh = bh->b_this_page) != head);
2024 /* So far everything mapped? Submit the page for IO. */
2025 if (mpd->map.m_len == 0) {
2026 err = mpage_submit_page(mpd, head->b_page);
2030 return lblk < blocks;
2034 * mpage_map_buffers - update buffers corresponding to changed extent and
2035 * submit fully mapped pages for IO
2037 * @mpd - description of extent to map, on return next extent to map
2039 * Scan buffers corresponding to changed extent (we expect corresponding pages
2040 * to be already locked) and update buffer state according to new extent state.
2041 * We map delalloc buffers to their physical location, clear unwritten bits,
2042 * and mark buffers as uninit when we perform writes to unwritten extents
2043 * and do extent conversion after IO is finished. If the last page is not fully
2044 * mapped, we update @map to the next extent in the last page that needs
2045 * mapping. Otherwise we submit the page for IO.
2047 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2049 struct pagevec pvec;
2051 struct inode *inode = mpd->inode;
2052 struct buffer_head *head, *bh;
2053 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2059 start = mpd->map.m_lblk >> bpp_bits;
2060 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2061 lblk = start << bpp_bits;
2062 pblock = mpd->map.m_pblk;
2064 pagevec_init(&pvec, 0);
2065 while (start <= end) {
2066 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2070 for (i = 0; i < nr_pages; i++) {
2071 struct page *page = pvec.pages[i];
2073 if (page->index > end)
2075 /* Up to 'end' pages must be contiguous */
2076 BUG_ON(page->index != start);
2077 bh = head = page_buffers(page);
2079 if (lblk < mpd->map.m_lblk)
2081 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2083 * Buffer after end of mapped extent.
2084 * Find next buffer in the page to map.
2087 mpd->map.m_flags = 0;
2089 * FIXME: If dioread_nolock supports
2090 * blocksize < pagesize, we need to make
2091 * sure we add size mapped so far to
2092 * io_end->size as the following call
2093 * can submit the page for IO.
2095 err = mpage_process_page_bufs(mpd, head,
2097 pagevec_release(&pvec);
2102 if (buffer_delay(bh)) {
2103 clear_buffer_delay(bh);
2104 bh->b_blocknr = pblock++;
2106 clear_buffer_unwritten(bh);
2107 } while (lblk++, (bh = bh->b_this_page) != head);
2110 * FIXME: This is going to break if dioread_nolock
2111 * supports blocksize < pagesize as we will try to
2112 * convert potentially unmapped parts of inode.
2114 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2115 /* Page fully mapped - let IO run! */
2116 err = mpage_submit_page(mpd, page);
2118 pagevec_release(&pvec);
2123 pagevec_release(&pvec);
2125 /* Extent fully mapped and matches with page boundary. We are done. */
2127 mpd->map.m_flags = 0;
2131 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2133 struct inode *inode = mpd->inode;
2134 struct ext4_map_blocks *map = &mpd->map;
2135 int get_blocks_flags;
2136 int err, dioread_nolock;
2138 trace_ext4_da_write_pages_extent(inode, map);
2140 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2141 * to convert an unwritten extent to be initialized (in the case
2142 * where we have written into one or more preallocated blocks). It is
2143 * possible that we're going to need more metadata blocks than
2144 * previously reserved. However we must not fail because we're in
2145 * writeback and there is nothing we can do about it so it might result
2146 * in data loss. So use reserved blocks to allocate metadata if
2149 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2150 * in question are delalloc blocks. This affects functions in many
2151 * different parts of the allocation call path. This flag exists
2152 * primarily because we don't want to change *many* call functions, so
2153 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2154 * once the inode's allocation semaphore is taken.
2156 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2157 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2158 dioread_nolock = ext4_should_dioread_nolock(inode);
2160 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2161 if (map->m_flags & (1 << BH_Delay))
2162 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2164 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2167 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2168 if (!mpd->io_submit.io_end->handle &&
2169 ext4_handle_valid(handle)) {
2170 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2171 handle->h_rsv_handle = NULL;
2173 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2176 BUG_ON(map->m_len == 0);
2177 if (map->m_flags & EXT4_MAP_NEW) {
2178 struct block_device *bdev = inode->i_sb->s_bdev;
2181 for (i = 0; i < map->m_len; i++)
2182 unmap_underlying_metadata(bdev, map->m_pblk + i);
2188 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2189 * mpd->len and submit pages underlying it for IO
2191 * @handle - handle for journal operations
2192 * @mpd - extent to map
2193 * @give_up_on_write - we set this to true iff there is a fatal error and there
2194 * is no hope of writing the data. The caller should discard
2195 * dirty pages to avoid infinite loops.
2197 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2198 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2199 * them to initialized or split the described range from larger unwritten
2200 * extent. Note that we need not map all the described range since allocation
2201 * can return less blocks or the range is covered by more unwritten extents. We
2202 * cannot map more because we are limited by reserved transaction credits. On
2203 * the other hand we always make sure that the last touched page is fully
2204 * mapped so that it can be written out (and thus forward progress is
2205 * guaranteed). After mapping we submit all mapped pages for IO.
2207 static int mpage_map_and_submit_extent(handle_t *handle,
2208 struct mpage_da_data *mpd,
2209 bool *give_up_on_write)
2211 struct inode *inode = mpd->inode;
2212 struct ext4_map_blocks *map = &mpd->map;
2216 mpd->io_submit.io_end->offset =
2217 ((loff_t)map->m_lblk) << inode->i_blkbits;
2219 err = mpage_map_one_extent(handle, mpd);
2221 struct super_block *sb = inode->i_sb;
2223 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2224 goto invalidate_dirty_pages;
2226 * Let the uper layers retry transient errors.
2227 * In the case of ENOSPC, if ext4_count_free_blocks()
2228 * is non-zero, a commit should free up blocks.
2230 if ((err == -ENOMEM) ||
2231 (err == -ENOSPC && ext4_count_free_clusters(sb)))
2233 ext4_msg(sb, KERN_CRIT,
2234 "Delayed block allocation failed for "
2235 "inode %lu at logical offset %llu with"
2236 " max blocks %u with error %d",
2238 (unsigned long long)map->m_lblk,
2239 (unsigned)map->m_len, -err);
2240 ext4_msg(sb, KERN_CRIT,
2241 "This should not happen!! Data will "
2244 ext4_print_free_blocks(inode);
2245 invalidate_dirty_pages:
2246 *give_up_on_write = true;
2250 * Update buffer state, submit mapped pages, and get us new
2253 err = mpage_map_and_submit_buffers(mpd);
2256 } while (map->m_len);
2259 * Update on-disk size after IO is submitted. Races with
2260 * truncate are avoided by checking i_size under i_data_sem.
2262 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2263 if (disksize > EXT4_I(inode)->i_disksize) {
2267 down_write(&EXT4_I(inode)->i_data_sem);
2268 i_size = i_size_read(inode);
2269 if (disksize > i_size)
2271 if (disksize > EXT4_I(inode)->i_disksize)
2272 EXT4_I(inode)->i_disksize = disksize;
2273 err2 = ext4_mark_inode_dirty(handle, inode);
2274 up_write(&EXT4_I(inode)->i_data_sem);
2276 ext4_error(inode->i_sb,
2277 "Failed to mark inode %lu dirty",
2286 * Calculate the total number of credits to reserve for one writepages
2287 * iteration. This is called from ext4_writepages(). We map an extent of
2288 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2289 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2290 * bpp - 1 blocks in bpp different extents.
2292 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2294 int bpp = ext4_journal_blocks_per_page(inode);
2296 return ext4_meta_trans_blocks(inode,
2297 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2301 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2302 * and underlying extent to map
2304 * @mpd - where to look for pages
2306 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2307 * IO immediately. When we find a page which isn't mapped we start accumulating
2308 * extent of buffers underlying these pages that needs mapping (formed by
2309 * either delayed or unwritten buffers). We also lock the pages containing
2310 * these buffers. The extent found is returned in @mpd structure (starting at
2311 * mpd->lblk with length mpd->len blocks).
2313 * Note that this function can attach bios to one io_end structure which are
2314 * neither logically nor physically contiguous. Although it may seem as an
2315 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2316 * case as we need to track IO to all buffers underlying a page in one io_end.
2318 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2320 struct address_space *mapping = mpd->inode->i_mapping;
2321 struct pagevec pvec;
2322 unsigned int nr_pages;
2323 long left = mpd->wbc->nr_to_write;
2324 pgoff_t index = mpd->first_page;
2325 pgoff_t end = mpd->last_page;
2328 int blkbits = mpd->inode->i_blkbits;
2330 struct buffer_head *head;
2332 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2333 tag = PAGECACHE_TAG_TOWRITE;
2335 tag = PAGECACHE_TAG_DIRTY;
2337 pagevec_init(&pvec, 0);
2339 mpd->next_page = index;
2340 while (index <= end) {
2341 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2342 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2346 for (i = 0; i < nr_pages; i++) {
2347 struct page *page = pvec.pages[i];
2350 * At this point, the page may be truncated or
2351 * invalidated (changing page->mapping to NULL), or
2352 * even swizzled back from swapper_space to tmpfs file
2353 * mapping. However, page->index will not change
2354 * because we have a reference on the page.
2356 if (page->index > end)
2360 * Accumulated enough dirty pages? This doesn't apply
2361 * to WB_SYNC_ALL mode. For integrity sync we have to
2362 * keep going because someone may be concurrently
2363 * dirtying pages, and we might have synced a lot of
2364 * newly appeared dirty pages, but have not synced all
2365 * of the old dirty pages.
2367 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2370 /* If we can't merge this page, we are done. */
2371 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2376 * If the page is no longer dirty, or its mapping no
2377 * longer corresponds to inode we are writing (which
2378 * means it has been truncated or invalidated), or the
2379 * page is already under writeback and we are not doing
2380 * a data integrity writeback, skip the page
2382 if (!PageDirty(page) ||
2383 (PageWriteback(page) &&
2384 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2385 unlikely(page->mapping != mapping)) {
2390 wait_on_page_writeback(page);
2391 BUG_ON(PageWriteback(page));
2393 if (mpd->map.m_len == 0)
2394 mpd->first_page = page->index;
2395 mpd->next_page = page->index + 1;
2396 /* Add all dirty buffers to mpd */
2397 lblk = ((ext4_lblk_t)page->index) <<
2398 (PAGE_CACHE_SHIFT - blkbits);
2399 head = page_buffers(page);
2400 err = mpage_process_page_bufs(mpd, head, head, lblk);
2406 pagevec_release(&pvec);
2411 pagevec_release(&pvec);
2415 static int __writepage(struct page *page, struct writeback_control *wbc,
2418 struct address_space *mapping = data;
2419 int ret = ext4_writepage(page, wbc);
2420 mapping_set_error(mapping, ret);
2424 static int ext4_writepages(struct address_space *mapping,
2425 struct writeback_control *wbc)
2427 pgoff_t writeback_index = 0;
2428 long nr_to_write = wbc->nr_to_write;
2429 int range_whole = 0;
2431 handle_t *handle = NULL;
2432 struct mpage_da_data mpd;
2433 struct inode *inode = mapping->host;
2434 int needed_blocks, rsv_blocks = 0, ret = 0;
2435 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2437 struct blk_plug plug;
2438 bool give_up_on_write = false;
2440 trace_ext4_writepages(inode, wbc);
2443 * No pages to write? This is mainly a kludge to avoid starting
2444 * a transaction for special inodes like journal inode on last iput()
2445 * because that could violate lock ordering on umount
2447 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2448 goto out_writepages;
2450 if (ext4_should_journal_data(inode)) {
2451 struct blk_plug plug;
2453 blk_start_plug(&plug);
2454 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2455 blk_finish_plug(&plug);
2456 goto out_writepages;
2460 * If the filesystem has aborted, it is read-only, so return
2461 * right away instead of dumping stack traces later on that
2462 * will obscure the real source of the problem. We test
2463 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2464 * the latter could be true if the filesystem is mounted
2465 * read-only, and in that case, ext4_writepages should
2466 * *never* be called, so if that ever happens, we would want
2469 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2471 goto out_writepages;
2474 if (ext4_should_dioread_nolock(inode)) {
2476 * We may need to convert up to one extent per block in
2477 * the page and we may dirty the inode.
2479 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2483 * If we have inline data and arrive here, it means that
2484 * we will soon create the block for the 1st page, so
2485 * we'd better clear the inline data here.
2487 if (ext4_has_inline_data(inode)) {
2488 /* Just inode will be modified... */
2489 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2490 if (IS_ERR(handle)) {
2491 ret = PTR_ERR(handle);
2492 goto out_writepages;
2494 BUG_ON(ext4_test_inode_state(inode,
2495 EXT4_STATE_MAY_INLINE_DATA));
2496 ext4_destroy_inline_data(handle, inode);
2497 ext4_journal_stop(handle);
2500 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2503 if (wbc->range_cyclic) {
2504 writeback_index = mapping->writeback_index;
2505 if (writeback_index)
2507 mpd.first_page = writeback_index;
2510 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2511 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2516 ext4_io_submit_init(&mpd.io_submit, wbc);
2518 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2519 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2521 blk_start_plug(&plug);
2522 while (!done && mpd.first_page <= mpd.last_page) {
2523 /* For each extent of pages we use new io_end */
2524 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2525 if (!mpd.io_submit.io_end) {
2531 * We have two constraints: We find one extent to map and we
2532 * must always write out whole page (makes a difference when
2533 * blocksize < pagesize) so that we don't block on IO when we
2534 * try to write out the rest of the page. Journalled mode is
2535 * not supported by delalloc.
2537 BUG_ON(ext4_should_journal_data(inode));
2538 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2540 /* start a new transaction */
2541 handle = ext4_journal_start_with_reserve(inode,
2542 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2543 if (IS_ERR(handle)) {
2544 ret = PTR_ERR(handle);
2545 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2546 "%ld pages, ino %lu; err %d", __func__,
2547 wbc->nr_to_write, inode->i_ino, ret);
2548 /* Release allocated io_end */
2549 ext4_put_io_end(mpd.io_submit.io_end);
2553 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2554 ret = mpage_prepare_extent_to_map(&mpd);
2557 ret = mpage_map_and_submit_extent(handle, &mpd,
2561 * We scanned the whole range (or exhausted
2562 * nr_to_write), submitted what was mapped and
2563 * didn't find anything needing mapping. We are
2569 ext4_journal_stop(handle);
2570 /* Submit prepared bio */
2571 ext4_io_submit(&mpd.io_submit);
2572 /* Unlock pages we didn't use */
2573 mpage_release_unused_pages(&mpd, give_up_on_write);
2574 /* Drop our io_end reference we got from init */
2575 ext4_put_io_end(mpd.io_submit.io_end);
2577 if (ret == -ENOSPC && sbi->s_journal) {
2579 * Commit the transaction which would
2580 * free blocks released in the transaction
2583 jbd2_journal_force_commit_nested(sbi->s_journal);
2587 /* Fatal error - ENOMEM, EIO... */
2591 blk_finish_plug(&plug);
2592 if (!ret && !cycled && wbc->nr_to_write > 0) {
2594 mpd.last_page = writeback_index - 1;
2600 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2602 * Set the writeback_index so that range_cyclic
2603 * mode will write it back later
2605 mapping->writeback_index = mpd.first_page;
2608 trace_ext4_writepages_result(inode, wbc, ret,
2609 nr_to_write - wbc->nr_to_write);
2613 static int ext4_nonda_switch(struct super_block *sb)
2615 s64 free_clusters, dirty_clusters;
2616 struct ext4_sb_info *sbi = EXT4_SB(sb);
2619 * switch to non delalloc mode if we are running low
2620 * on free block. The free block accounting via percpu
2621 * counters can get slightly wrong with percpu_counter_batch getting
2622 * accumulated on each CPU without updating global counters
2623 * Delalloc need an accurate free block accounting. So switch
2624 * to non delalloc when we are near to error range.
2627 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2629 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2631 * Start pushing delalloc when 1/2 of free blocks are dirty.
2633 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2634 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2636 if (2 * free_clusters < 3 * dirty_clusters ||
2637 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2639 * free block count is less than 150% of dirty blocks
2640 * or free blocks is less than watermark
2647 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2648 loff_t pos, unsigned len, unsigned flags,
2649 struct page **pagep, void **fsdata)
2651 int ret, retries = 0;
2654 struct inode *inode = mapping->host;
2657 index = pos >> PAGE_CACHE_SHIFT;
2659 if (ext4_nonda_switch(inode->i_sb)) {
2660 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2661 return ext4_write_begin(file, mapping, pos,
2662 len, flags, pagep, fsdata);
2664 *fsdata = (void *)0;
2665 trace_ext4_da_write_begin(inode, pos, len, flags);
2667 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2668 ret = ext4_da_write_inline_data_begin(mapping, inode,
2678 * grab_cache_page_write_begin() can take a long time if the
2679 * system is thrashing due to memory pressure, or if the page
2680 * is being written back. So grab it first before we start
2681 * the transaction handle. This also allows us to allocate
2682 * the page (if needed) without using GFP_NOFS.
2685 page = grab_cache_page_write_begin(mapping, index, flags);
2691 * With delayed allocation, we don't log the i_disksize update
2692 * if there is delayed block allocation. But we still need
2693 * to journalling the i_disksize update if writes to the end
2694 * of file which has an already mapped buffer.
2697 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2698 if (IS_ERR(handle)) {
2699 page_cache_release(page);
2700 return PTR_ERR(handle);
2704 if (page->mapping != mapping) {
2705 /* The page got truncated from under us */
2707 page_cache_release(page);
2708 ext4_journal_stop(handle);
2711 /* In case writeback began while the page was unlocked */
2712 wait_for_stable_page(page);
2714 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2717 ext4_journal_stop(handle);
2719 * block_write_begin may have instantiated a few blocks
2720 * outside i_size. Trim these off again. Don't need
2721 * i_size_read because we hold i_mutex.
2723 if (pos + len > inode->i_size)
2724 ext4_truncate_failed_write(inode);
2726 if (ret == -ENOSPC &&
2727 ext4_should_retry_alloc(inode->i_sb, &retries))
2730 page_cache_release(page);
2739 * Check if we should update i_disksize
2740 * when write to the end of file but not require block allocation
2742 static int ext4_da_should_update_i_disksize(struct page *page,
2743 unsigned long offset)
2745 struct buffer_head *bh;
2746 struct inode *inode = page->mapping->host;
2750 bh = page_buffers(page);
2751 idx = offset >> inode->i_blkbits;
2753 for (i = 0; i < idx; i++)
2754 bh = bh->b_this_page;
2756 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2761 static int ext4_da_write_end(struct file *file,
2762 struct address_space *mapping,
2763 loff_t pos, unsigned len, unsigned copied,
2764 struct page *page, void *fsdata)
2766 struct inode *inode = mapping->host;
2768 handle_t *handle = ext4_journal_current_handle();
2770 unsigned long start, end;
2771 int write_mode = (int)(unsigned long)fsdata;
2773 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2774 return ext4_write_end(file, mapping, pos,
2775 len, copied, page, fsdata);
2777 trace_ext4_da_write_end(inode, pos, len, copied);
2778 start = pos & (PAGE_CACHE_SIZE - 1);
2779 end = start + copied - 1;
2782 * generic_write_end() will run mark_inode_dirty() if i_size
2783 * changes. So let's piggyback the i_disksize mark_inode_dirty
2786 new_i_size = pos + copied;
2787 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2788 if (ext4_has_inline_data(inode) ||
2789 ext4_da_should_update_i_disksize(page, end)) {
2790 down_write(&EXT4_I(inode)->i_data_sem);
2791 if (new_i_size > EXT4_I(inode)->i_disksize)
2792 EXT4_I(inode)->i_disksize = new_i_size;
2793 up_write(&EXT4_I(inode)->i_data_sem);
2794 /* We need to mark inode dirty even if
2795 * new_i_size is less that inode->i_size
2796 * bu greater than i_disksize.(hint delalloc)
2798 ext4_mark_inode_dirty(handle, inode);
2802 if (write_mode != CONVERT_INLINE_DATA &&
2803 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2804 ext4_has_inline_data(inode))
2805 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2808 ret2 = generic_write_end(file, mapping, pos, len, copied,
2814 ret2 = ext4_journal_stop(handle);
2818 return ret ? ret : copied;
2821 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2822 unsigned int length)
2825 * Drop reserved blocks
2827 BUG_ON(!PageLocked(page));
2828 if (!page_has_buffers(page))
2831 ext4_da_page_release_reservation(page, offset, length);
2834 ext4_invalidatepage(page, offset, length);
2840 * Force all delayed allocation blocks to be allocated for a given inode.
2842 int ext4_alloc_da_blocks(struct inode *inode)
2844 trace_ext4_alloc_da_blocks(inode);
2846 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2847 !EXT4_I(inode)->i_reserved_meta_blocks)
2851 * We do something simple for now. The filemap_flush() will
2852 * also start triggering a write of the data blocks, which is
2853 * not strictly speaking necessary (and for users of
2854 * laptop_mode, not even desirable). However, to do otherwise
2855 * would require replicating code paths in:
2857 * ext4_writepages() ->
2858 * write_cache_pages() ---> (via passed in callback function)
2859 * __mpage_da_writepage() -->
2860 * mpage_add_bh_to_extent()
2861 * mpage_da_map_blocks()
2863 * The problem is that write_cache_pages(), located in
2864 * mm/page-writeback.c, marks pages clean in preparation for
2865 * doing I/O, which is not desirable if we're not planning on
2868 * We could call write_cache_pages(), and then redirty all of
2869 * the pages by calling redirty_page_for_writepage() but that
2870 * would be ugly in the extreme. So instead we would need to
2871 * replicate parts of the code in the above functions,
2872 * simplifying them because we wouldn't actually intend to
2873 * write out the pages, but rather only collect contiguous
2874 * logical block extents, call the multi-block allocator, and
2875 * then update the buffer heads with the block allocations.
2877 * For now, though, we'll cheat by calling filemap_flush(),
2878 * which will map the blocks, and start the I/O, but not
2879 * actually wait for the I/O to complete.
2881 return filemap_flush(inode->i_mapping);
2885 * bmap() is special. It gets used by applications such as lilo and by
2886 * the swapper to find the on-disk block of a specific piece of data.
2888 * Naturally, this is dangerous if the block concerned is still in the
2889 * journal. If somebody makes a swapfile on an ext4 data-journaling
2890 * filesystem and enables swap, then they may get a nasty shock when the
2891 * data getting swapped to that swapfile suddenly gets overwritten by
2892 * the original zero's written out previously to the journal and
2893 * awaiting writeback in the kernel's buffer cache.
2895 * So, if we see any bmap calls here on a modified, data-journaled file,
2896 * take extra steps to flush any blocks which might be in the cache.
2898 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2900 struct inode *inode = mapping->host;
2905 * We can get here for an inline file via the FIBMAP ioctl
2907 if (ext4_has_inline_data(inode))
2910 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2911 test_opt(inode->i_sb, DELALLOC)) {
2913 * With delalloc we want to sync the file
2914 * so that we can make sure we allocate
2917 filemap_write_and_wait(mapping);
2920 if (EXT4_JOURNAL(inode) &&
2921 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2923 * This is a REALLY heavyweight approach, but the use of
2924 * bmap on dirty files is expected to be extremely rare:
2925 * only if we run lilo or swapon on a freshly made file
2926 * do we expect this to happen.
2928 * (bmap requires CAP_SYS_RAWIO so this does not
2929 * represent an unprivileged user DOS attack --- we'd be
2930 * in trouble if mortal users could trigger this path at
2933 * NB. EXT4_STATE_JDATA is not set on files other than
2934 * regular files. If somebody wants to bmap a directory
2935 * or symlink and gets confused because the buffer
2936 * hasn't yet been flushed to disk, they deserve
2937 * everything they get.
2940 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2941 journal = EXT4_JOURNAL(inode);
2942 jbd2_journal_lock_updates(journal);
2943 err = jbd2_journal_flush(journal);
2944 jbd2_journal_unlock_updates(journal);
2950 return generic_block_bmap(mapping, block, ext4_get_block);
2953 static int ext4_readpage(struct file *file, struct page *page)
2956 struct inode *inode = page->mapping->host;
2958 trace_ext4_readpage(page);
2960 if (ext4_has_inline_data(inode))
2961 ret = ext4_readpage_inline(inode, page);
2964 return mpage_readpage(page, ext4_get_block);
2970 ext4_readpages(struct file *file, struct address_space *mapping,
2971 struct list_head *pages, unsigned nr_pages)
2973 struct inode *inode = mapping->host;
2975 /* If the file has inline data, no need to do readpages. */
2976 if (ext4_has_inline_data(inode))
2979 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2982 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2983 unsigned int length)
2985 trace_ext4_invalidatepage(page, offset, length);
2987 /* No journalling happens on data buffers when this function is used */
2988 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2990 block_invalidatepage(page, offset, length);
2993 static int __ext4_journalled_invalidatepage(struct page *page,
2994 unsigned int offset,
2995 unsigned int length)
2997 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2999 trace_ext4_journalled_invalidatepage(page, offset, length);
3002 * If it's a full truncate we just forget about the pending dirtying
3004 if (offset == 0 && length == PAGE_CACHE_SIZE)
3005 ClearPageChecked(page);
3007 return jbd2_journal_invalidatepage(journal, page, offset, length);
3010 /* Wrapper for aops... */
3011 static void ext4_journalled_invalidatepage(struct page *page,
3012 unsigned int offset,
3013 unsigned int length)
3015 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3018 static int ext4_releasepage(struct page *page, gfp_t wait)
3020 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3022 trace_ext4_releasepage(page);
3024 /* Page has dirty journalled data -> cannot release */
3025 if (PageChecked(page))
3028 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3030 return try_to_free_buffers(page);
3034 * ext4_get_block used when preparing for a DIO write or buffer write.
3035 * We allocate an uinitialized extent if blocks haven't been allocated.
3036 * The extent will be converted to initialized after the IO is complete.
3038 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3039 struct buffer_head *bh_result, int create)
3041 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3042 inode->i_ino, create);
3043 return _ext4_get_block(inode, iblock, bh_result,
3044 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3047 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3048 struct buffer_head *bh_result, int create)
3050 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3051 inode->i_ino, create);
3052 return _ext4_get_block(inode, iblock, bh_result,
3053 EXT4_GET_BLOCKS_NO_LOCK);
3056 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3057 ssize_t size, void *private)
3059 ext4_io_end_t *io_end = iocb->private;
3061 /* if not async direct IO just return */
3065 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3066 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3067 iocb->private, io_end->inode->i_ino, iocb, offset,
3070 iocb->private = NULL;
3071 io_end->offset = offset;
3072 io_end->size = size;
3073 ext4_put_io_end(io_end);
3077 * For ext4 extent files, ext4 will do direct-io write to holes,
3078 * preallocated extents, and those write extend the file, no need to
3079 * fall back to buffered IO.
3081 * For holes, we fallocate those blocks, mark them as unwritten
3082 * If those blocks were preallocated, we mark sure they are split, but
3083 * still keep the range to write as unwritten.
3085 * The unwritten extents will be converted to written when DIO is completed.
3086 * For async direct IO, since the IO may still pending when return, we
3087 * set up an end_io call back function, which will do the conversion
3088 * when async direct IO completed.
3090 * If the O_DIRECT write will extend the file then add this inode to the
3091 * orphan list. So recovery will truncate it back to the original size
3092 * if the machine crashes during the write.
3095 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3096 struct iov_iter *iter, loff_t offset)
3098 struct file *file = iocb->ki_filp;
3099 struct inode *inode = file->f_mapping->host;
3101 size_t count = iov_iter_count(iter);
3103 get_block_t *get_block_func = NULL;
3105 loff_t final_size = offset + count;
3106 ext4_io_end_t *io_end = NULL;
3108 /* Use the old path for reads and writes beyond i_size. */
3109 if (rw != WRITE || final_size > inode->i_size)
3110 return ext4_ind_direct_IO(rw, iocb, iter, offset);
3112 BUG_ON(iocb->private == NULL);
3115 * Make all waiters for direct IO properly wait also for extent
3116 * conversion. This also disallows race between truncate() and
3117 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3120 atomic_inc(&inode->i_dio_count);
3122 /* If we do a overwrite dio, i_mutex locking can be released */
3123 overwrite = *((int *)iocb->private);
3126 down_read(&EXT4_I(inode)->i_data_sem);
3127 mutex_unlock(&inode->i_mutex);
3131 * We could direct write to holes and fallocate.
3133 * Allocated blocks to fill the hole are marked as
3134 * unwritten to prevent parallel buffered read to expose
3135 * the stale data before DIO complete the data IO.
3137 * As to previously fallocated extents, ext4 get_block will
3138 * just simply mark the buffer mapped but still keep the
3139 * extents unwritten.
3141 * For non AIO case, we will convert those unwritten extents
3142 * to written after return back from blockdev_direct_IO.
3144 * For async DIO, the conversion needs to be deferred when the
3145 * IO is completed. The ext4 end_io callback function will be
3146 * called to take care of the conversion work. Here for async
3147 * case, we allocate an io_end structure to hook to the iocb.
3149 iocb->private = NULL;
3150 ext4_inode_aio_set(inode, NULL);
3151 if (!is_sync_kiocb(iocb)) {
3152 io_end = ext4_init_io_end(inode, GFP_NOFS);
3158 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3160 iocb->private = ext4_get_io_end(io_end);
3162 * we save the io structure for current async direct
3163 * IO, so that later ext4_map_blocks() could flag the
3164 * io structure whether there is a unwritten extents
3165 * needs to be converted when IO is completed.
3167 ext4_inode_aio_set(inode, io_end);
3171 get_block_func = ext4_get_block_write_nolock;
3173 get_block_func = ext4_get_block_write;
3174 dio_flags = DIO_LOCKING;
3176 ret = __blockdev_direct_IO(rw, iocb, inode,
3177 inode->i_sb->s_bdev, iter,
3185 * Put our reference to io_end. This can free the io_end structure e.g.
3186 * in sync IO case or in case of error. It can even perform extent
3187 * conversion if all bios we submitted finished before we got here.
3188 * Note that in that case iocb->private can be already set to NULL
3192 ext4_inode_aio_set(inode, NULL);
3193 ext4_put_io_end(io_end);
3195 * When no IO was submitted ext4_end_io_dio() was not
3196 * called so we have to put iocb's reference.
3198 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3199 WARN_ON(iocb->private != io_end);
3200 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3201 ext4_put_io_end(io_end);
3202 iocb->private = NULL;
3205 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3206 EXT4_STATE_DIO_UNWRITTEN)) {
3209 * for non AIO case, since the IO is already
3210 * completed, we could do the conversion right here
3212 err = ext4_convert_unwritten_extents(NULL, inode,
3216 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3221 inode_dio_done(inode);
3222 /* take i_mutex locking again if we do a ovewrite dio */
3224 up_read(&EXT4_I(inode)->i_data_sem);
3225 mutex_lock(&inode->i_mutex);
3231 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3232 struct iov_iter *iter, loff_t offset)
3234 struct file *file = iocb->ki_filp;
3235 struct inode *inode = file->f_mapping->host;
3236 size_t count = iov_iter_count(iter);
3240 * If we are doing data journalling we don't support O_DIRECT
3242 if (ext4_should_journal_data(inode))
3245 /* Let buffer I/O handle the inline data case. */
3246 if (ext4_has_inline_data(inode))
3249 trace_ext4_direct_IO_enter(inode, offset, count, rw);
3250 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3251 ret = ext4_ext_direct_IO(rw, iocb, iter, offset);
3253 ret = ext4_ind_direct_IO(rw, iocb, iter, offset);
3254 trace_ext4_direct_IO_exit(inode, offset, count, rw, ret);
3259 * Pages can be marked dirty completely asynchronously from ext4's journalling
3260 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3261 * much here because ->set_page_dirty is called under VFS locks. The page is
3262 * not necessarily locked.
3264 * We cannot just dirty the page and leave attached buffers clean, because the
3265 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3266 * or jbddirty because all the journalling code will explode.
3268 * So what we do is to mark the page "pending dirty" and next time writepage
3269 * is called, propagate that into the buffers appropriately.
3271 static int ext4_journalled_set_page_dirty(struct page *page)
3273 SetPageChecked(page);
3274 return __set_page_dirty_nobuffers(page);
3277 static const struct address_space_operations ext4_aops = {
3278 .readpage = ext4_readpage,
3279 .readpages = ext4_readpages,
3280 .writepage = ext4_writepage,
3281 .writepages = ext4_writepages,
3282 .write_begin = ext4_write_begin,
3283 .write_end = ext4_write_end,
3285 .invalidatepage = ext4_invalidatepage,
3286 .releasepage = ext4_releasepage,
3287 .direct_IO = ext4_direct_IO,
3288 .migratepage = buffer_migrate_page,
3289 .is_partially_uptodate = block_is_partially_uptodate,
3290 .error_remove_page = generic_error_remove_page,
3293 static const struct address_space_operations ext4_journalled_aops = {
3294 .readpage = ext4_readpage,
3295 .readpages = ext4_readpages,
3296 .writepage = ext4_writepage,
3297 .writepages = ext4_writepages,
3298 .write_begin = ext4_write_begin,
3299 .write_end = ext4_journalled_write_end,
3300 .set_page_dirty = ext4_journalled_set_page_dirty,
3302 .invalidatepage = ext4_journalled_invalidatepage,
3303 .releasepage = ext4_releasepage,
3304 .direct_IO = ext4_direct_IO,
3305 .is_partially_uptodate = block_is_partially_uptodate,
3306 .error_remove_page = generic_error_remove_page,
3309 static const struct address_space_operations ext4_da_aops = {
3310 .readpage = ext4_readpage,
3311 .readpages = ext4_readpages,
3312 .writepage = ext4_writepage,
3313 .writepages = ext4_writepages,
3314 .write_begin = ext4_da_write_begin,
3315 .write_end = ext4_da_write_end,
3317 .invalidatepage = ext4_da_invalidatepage,
3318 .releasepage = ext4_releasepage,
3319 .direct_IO = ext4_direct_IO,
3320 .migratepage = buffer_migrate_page,
3321 .is_partially_uptodate = block_is_partially_uptodate,
3322 .error_remove_page = generic_error_remove_page,
3325 void ext4_set_aops(struct inode *inode)
3327 switch (ext4_inode_journal_mode(inode)) {
3328 case EXT4_INODE_ORDERED_DATA_MODE:
3329 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3331 case EXT4_INODE_WRITEBACK_DATA_MODE:
3332 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3334 case EXT4_INODE_JOURNAL_DATA_MODE:
3335 inode->i_mapping->a_ops = &ext4_journalled_aops;
3340 if (test_opt(inode->i_sb, DELALLOC))
3341 inode->i_mapping->a_ops = &ext4_da_aops;
3343 inode->i_mapping->a_ops = &ext4_aops;
3347 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3348 * starting from file offset 'from'. The range to be zero'd must
3349 * be contained with in one block. If the specified range exceeds
3350 * the end of the block it will be shortened to end of the block
3351 * that cooresponds to 'from'
3353 static int ext4_block_zero_page_range(handle_t *handle,
3354 struct address_space *mapping, loff_t from, loff_t length)
3356 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3357 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3358 unsigned blocksize, max, pos;
3360 struct inode *inode = mapping->host;
3361 struct buffer_head *bh;
3365 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3366 mapping_gfp_mask(mapping) & ~__GFP_FS);
3370 blocksize = inode->i_sb->s_blocksize;
3371 max = blocksize - (offset & (blocksize - 1));
3374 * correct length if it does not fall between
3375 * 'from' and the end of the block
3377 if (length > max || length < 0)
3380 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3382 if (!page_has_buffers(page))
3383 create_empty_buffers(page, blocksize, 0);
3385 /* Find the buffer that contains "offset" */
3386 bh = page_buffers(page);
3388 while (offset >= pos) {
3389 bh = bh->b_this_page;
3393 if (buffer_freed(bh)) {
3394 BUFFER_TRACE(bh, "freed: skip");
3397 if (!buffer_mapped(bh)) {
3398 BUFFER_TRACE(bh, "unmapped");
3399 ext4_get_block(inode, iblock, bh, 0);
3400 /* unmapped? It's a hole - nothing to do */
3401 if (!buffer_mapped(bh)) {
3402 BUFFER_TRACE(bh, "still unmapped");
3407 /* Ok, it's mapped. Make sure it's up-to-date */
3408 if (PageUptodate(page))
3409 set_buffer_uptodate(bh);
3411 if (!buffer_uptodate(bh)) {
3413 ll_rw_block(READ, 1, &bh);
3415 /* Uhhuh. Read error. Complain and punt. */
3416 if (!buffer_uptodate(bh))
3419 if (ext4_should_journal_data(inode)) {
3420 BUFFER_TRACE(bh, "get write access");
3421 err = ext4_journal_get_write_access(handle, bh);
3425 zero_user(page, offset, length);
3426 BUFFER_TRACE(bh, "zeroed end of block");
3428 if (ext4_should_journal_data(inode)) {
3429 err = ext4_handle_dirty_metadata(handle, inode, bh);
3432 mark_buffer_dirty(bh);
3433 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3434 err = ext4_jbd2_file_inode(handle, inode);
3439 page_cache_release(page);
3444 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3445 * up to the end of the block which corresponds to `from'.
3446 * This required during truncate. We need to physically zero the tail end
3447 * of that block so it doesn't yield old data if the file is later grown.
3449 static int ext4_block_truncate_page(handle_t *handle,
3450 struct address_space *mapping, loff_t from)
3452 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3455 struct inode *inode = mapping->host;
3457 blocksize = inode->i_sb->s_blocksize;
3458 length = blocksize - (offset & (blocksize - 1));
3460 return ext4_block_zero_page_range(handle, mapping, from, length);
3463 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3464 loff_t lstart, loff_t length)
3466 struct super_block *sb = inode->i_sb;
3467 struct address_space *mapping = inode->i_mapping;
3468 unsigned partial_start, partial_end;
3469 ext4_fsblk_t start, end;
3470 loff_t byte_end = (lstart + length - 1);
3473 partial_start = lstart & (sb->s_blocksize - 1);
3474 partial_end = byte_end & (sb->s_blocksize - 1);
3476 start = lstart >> sb->s_blocksize_bits;
3477 end = byte_end >> sb->s_blocksize_bits;
3479 /* Handle partial zero within the single block */
3481 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3482 err = ext4_block_zero_page_range(handle, mapping,
3486 /* Handle partial zero out on the start of the range */
3487 if (partial_start) {
3488 err = ext4_block_zero_page_range(handle, mapping,
3489 lstart, sb->s_blocksize);
3493 /* Handle partial zero out on the end of the range */
3494 if (partial_end != sb->s_blocksize - 1)
3495 err = ext4_block_zero_page_range(handle, mapping,
3496 byte_end - partial_end,
3501 int ext4_can_truncate(struct inode *inode)
3503 if (S_ISREG(inode->i_mode))
3505 if (S_ISDIR(inode->i_mode))
3507 if (S_ISLNK(inode->i_mode))
3508 return !ext4_inode_is_fast_symlink(inode);
3513 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3514 * associated with the given offset and length
3516 * @inode: File inode
3517 * @offset: The offset where the hole will begin
3518 * @len: The length of the hole
3520 * Returns: 0 on success or negative on failure
3523 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3525 struct super_block *sb = inode->i_sb;
3526 ext4_lblk_t first_block, stop_block;
3527 struct address_space *mapping = inode->i_mapping;
3528 loff_t first_block_offset, last_block_offset;
3530 unsigned int credits;
3533 if (!S_ISREG(inode->i_mode))
3536 trace_ext4_punch_hole(inode, offset, length, 0);
3539 * Write out all dirty pages to avoid race conditions
3540 * Then release them.
3542 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3543 ret = filemap_write_and_wait_range(mapping, offset,
3544 offset + length - 1);
3549 mutex_lock(&inode->i_mutex);
3551 /* No need to punch hole beyond i_size */
3552 if (offset >= inode->i_size)
3556 * If the hole extends beyond i_size, set the hole
3557 * to end after the page that contains i_size
3559 if (offset + length > inode->i_size) {
3560 length = inode->i_size +
3561 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3565 if (offset & (sb->s_blocksize - 1) ||
3566 (offset + length) & (sb->s_blocksize - 1)) {
3568 * Attach jinode to inode for jbd2 if we do any zeroing of
3571 ret = ext4_inode_attach_jinode(inode);
3577 first_block_offset = round_up(offset, sb->s_blocksize);
3578 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3580 /* Now release the pages and zero block aligned part of pages*/
3581 if (last_block_offset > first_block_offset)
3582 truncate_pagecache_range(inode, first_block_offset,
3585 /* Wait all existing dio workers, newcomers will block on i_mutex */
3586 ext4_inode_block_unlocked_dio(inode);
3587 inode_dio_wait(inode);
3589 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3590 credits = ext4_writepage_trans_blocks(inode);
3592 credits = ext4_blocks_for_truncate(inode);
3593 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3594 if (IS_ERR(handle)) {
3595 ret = PTR_ERR(handle);
3596 ext4_std_error(sb, ret);
3600 ret = ext4_zero_partial_blocks(handle, inode, offset,
3605 first_block = (offset + sb->s_blocksize - 1) >>
3606 EXT4_BLOCK_SIZE_BITS(sb);
3607 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3609 /* If there are no blocks to remove, return now */
3610 if (first_block >= stop_block)
3613 down_write(&EXT4_I(inode)->i_data_sem);
3614 ext4_discard_preallocations(inode);
3616 ret = ext4_es_remove_extent(inode, first_block,
3617 stop_block - first_block);
3619 up_write(&EXT4_I(inode)->i_data_sem);
3623 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3624 ret = ext4_ext_remove_space(inode, first_block,
3627 ret = ext4_free_hole_blocks(handle, inode, first_block,
3630 up_write(&EXT4_I(inode)->i_data_sem);
3632 ext4_handle_sync(handle);
3634 /* Now release the pages again to reduce race window */
3635 if (last_block_offset > first_block_offset)
3636 truncate_pagecache_range(inode, first_block_offset,
3639 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3640 ext4_mark_inode_dirty(handle, inode);
3642 ext4_journal_stop(handle);
3644 ext4_inode_resume_unlocked_dio(inode);
3646 mutex_unlock(&inode->i_mutex);
3650 int ext4_inode_attach_jinode(struct inode *inode)
3652 struct ext4_inode_info *ei = EXT4_I(inode);
3653 struct jbd2_inode *jinode;
3655 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3658 jinode = jbd2_alloc_inode(GFP_KERNEL);
3659 spin_lock(&inode->i_lock);
3662 spin_unlock(&inode->i_lock);
3665 ei->jinode = jinode;
3666 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3669 spin_unlock(&inode->i_lock);
3670 if (unlikely(jinode != NULL))
3671 jbd2_free_inode(jinode);
3678 * We block out ext4_get_block() block instantiations across the entire
3679 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3680 * simultaneously on behalf of the same inode.
3682 * As we work through the truncate and commit bits of it to the journal there
3683 * is one core, guiding principle: the file's tree must always be consistent on
3684 * disk. We must be able to restart the truncate after a crash.
3686 * The file's tree may be transiently inconsistent in memory (although it
3687 * probably isn't), but whenever we close off and commit a journal transaction,
3688 * the contents of (the filesystem + the journal) must be consistent and
3689 * restartable. It's pretty simple, really: bottom up, right to left (although
3690 * left-to-right works OK too).
3692 * Note that at recovery time, journal replay occurs *before* the restart of
3693 * truncate against the orphan inode list.
3695 * The committed inode has the new, desired i_size (which is the same as
3696 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3697 * that this inode's truncate did not complete and it will again call
3698 * ext4_truncate() to have another go. So there will be instantiated blocks
3699 * to the right of the truncation point in a crashed ext4 filesystem. But
3700 * that's fine - as long as they are linked from the inode, the post-crash
3701 * ext4_truncate() run will find them and release them.
3703 void ext4_truncate(struct inode *inode)
3705 struct ext4_inode_info *ei = EXT4_I(inode);
3706 unsigned int credits;
3708 struct address_space *mapping = inode->i_mapping;
3711 * There is a possibility that we're either freeing the inode
3712 * or it's a completely new inode. In those cases we might not
3713 * have i_mutex locked because it's not necessary.
3715 if (!(inode->i_state & (I_NEW|I_FREEING)))
3716 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3717 trace_ext4_truncate_enter(inode);
3719 if (!ext4_can_truncate(inode))
3722 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3724 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3725 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3727 if (ext4_has_inline_data(inode)) {
3730 ext4_inline_data_truncate(inode, &has_inline);
3735 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3736 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3737 if (ext4_inode_attach_jinode(inode) < 0)
3741 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3742 credits = ext4_writepage_trans_blocks(inode);
3744 credits = ext4_blocks_for_truncate(inode);
3746 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3747 if (IS_ERR(handle)) {
3748 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3752 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3753 ext4_block_truncate_page(handle, mapping, inode->i_size);
3756 * We add the inode to the orphan list, so that if this
3757 * truncate spans multiple transactions, and we crash, we will
3758 * resume the truncate when the filesystem recovers. It also
3759 * marks the inode dirty, to catch the new size.
3761 * Implication: the file must always be in a sane, consistent
3762 * truncatable state while each transaction commits.
3764 if (ext4_orphan_add(handle, inode))
3767 down_write(&EXT4_I(inode)->i_data_sem);
3769 ext4_discard_preallocations(inode);
3771 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3772 ext4_ext_truncate(handle, inode);
3774 ext4_ind_truncate(handle, inode);
3776 up_write(&ei->i_data_sem);
3779 ext4_handle_sync(handle);
3783 * If this was a simple ftruncate() and the file will remain alive,
3784 * then we need to clear up the orphan record which we created above.
3785 * However, if this was a real unlink then we were called by
3786 * ext4_delete_inode(), and we allow that function to clean up the
3787 * orphan info for us.
3790 ext4_orphan_del(handle, inode);
3792 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3793 ext4_mark_inode_dirty(handle, inode);
3794 ext4_journal_stop(handle);
3796 trace_ext4_truncate_exit(inode);
3800 * ext4_get_inode_loc returns with an extra refcount against the inode's
3801 * underlying buffer_head on success. If 'in_mem' is true, we have all
3802 * data in memory that is needed to recreate the on-disk version of this
3805 static int __ext4_get_inode_loc(struct inode *inode,
3806 struct ext4_iloc *iloc, int in_mem)
3808 struct ext4_group_desc *gdp;
3809 struct buffer_head *bh;
3810 struct super_block *sb = inode->i_sb;
3812 int inodes_per_block, inode_offset;
3815 if (!ext4_valid_inum(sb, inode->i_ino))
3818 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3819 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3824 * Figure out the offset within the block group inode table
3826 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3827 inode_offset = ((inode->i_ino - 1) %
3828 EXT4_INODES_PER_GROUP(sb));
3829 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3830 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3832 bh = sb_getblk(sb, block);
3835 if (!buffer_uptodate(bh)) {
3839 * If the buffer has the write error flag, we have failed
3840 * to write out another inode in the same block. In this
3841 * case, we don't have to read the block because we may
3842 * read the old inode data successfully.
3844 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3845 set_buffer_uptodate(bh);
3847 if (buffer_uptodate(bh)) {
3848 /* someone brought it uptodate while we waited */
3854 * If we have all information of the inode in memory and this
3855 * is the only valid inode in the block, we need not read the
3859 struct buffer_head *bitmap_bh;
3862 start = inode_offset & ~(inodes_per_block - 1);
3864 /* Is the inode bitmap in cache? */
3865 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3866 if (unlikely(!bitmap_bh))
3870 * If the inode bitmap isn't in cache then the
3871 * optimisation may end up performing two reads instead
3872 * of one, so skip it.
3874 if (!buffer_uptodate(bitmap_bh)) {
3878 for (i = start; i < start + inodes_per_block; i++) {
3879 if (i == inode_offset)
3881 if (ext4_test_bit(i, bitmap_bh->b_data))
3885 if (i == start + inodes_per_block) {
3886 /* all other inodes are free, so skip I/O */
3887 memset(bh->b_data, 0, bh->b_size);
3888 set_buffer_uptodate(bh);
3896 * If we need to do any I/O, try to pre-readahead extra
3897 * blocks from the inode table.
3899 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3900 ext4_fsblk_t b, end, table;
3902 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3904 table = ext4_inode_table(sb, gdp);
3905 /* s_inode_readahead_blks is always a power of 2 */
3906 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3910 num = EXT4_INODES_PER_GROUP(sb);
3911 if (ext4_has_group_desc_csum(sb))
3912 num -= ext4_itable_unused_count(sb, gdp);
3913 table += num / inodes_per_block;
3917 sb_breadahead(sb, b++);
3921 * There are other valid inodes in the buffer, this inode
3922 * has in-inode xattrs, or we don't have this inode in memory.
3923 * Read the block from disk.
3925 trace_ext4_load_inode(inode);
3927 bh->b_end_io = end_buffer_read_sync;
3928 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3930 if (!buffer_uptodate(bh)) {
3931 EXT4_ERROR_INODE_BLOCK(inode, block,
3932 "unable to read itable block");
3942 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3944 /* We have all inode data except xattrs in memory here. */
3945 return __ext4_get_inode_loc(inode, iloc,
3946 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3949 void ext4_set_inode_flags(struct inode *inode)
3951 unsigned int flags = EXT4_I(inode)->i_flags;
3952 unsigned int new_fl = 0;
3954 if (flags & EXT4_SYNC_FL)
3956 if (flags & EXT4_APPEND_FL)
3958 if (flags & EXT4_IMMUTABLE_FL)
3959 new_fl |= S_IMMUTABLE;
3960 if (flags & EXT4_NOATIME_FL)
3961 new_fl |= S_NOATIME;
3962 if (flags & EXT4_DIRSYNC_FL)
3963 new_fl |= S_DIRSYNC;
3964 inode_set_flags(inode, new_fl,
3965 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3968 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3969 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3971 unsigned int vfs_fl;
3972 unsigned long old_fl, new_fl;
3975 vfs_fl = ei->vfs_inode.i_flags;
3976 old_fl = ei->i_flags;
3977 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3978 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3980 if (vfs_fl & S_SYNC)
3981 new_fl |= EXT4_SYNC_FL;
3982 if (vfs_fl & S_APPEND)
3983 new_fl |= EXT4_APPEND_FL;
3984 if (vfs_fl & S_IMMUTABLE)
3985 new_fl |= EXT4_IMMUTABLE_FL;
3986 if (vfs_fl & S_NOATIME)
3987 new_fl |= EXT4_NOATIME_FL;
3988 if (vfs_fl & S_DIRSYNC)
3989 new_fl |= EXT4_DIRSYNC_FL;
3990 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3993 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3994 struct ext4_inode_info *ei)
3997 struct inode *inode = &(ei->vfs_inode);
3998 struct super_block *sb = inode->i_sb;
4000 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4001 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4002 /* we are using combined 48 bit field */
4003 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4004 le32_to_cpu(raw_inode->i_blocks_lo);
4005 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4006 /* i_blocks represent file system block size */
4007 return i_blocks << (inode->i_blkbits - 9);
4012 return le32_to_cpu(raw_inode->i_blocks_lo);
4016 static inline void ext4_iget_extra_inode(struct inode *inode,
4017 struct ext4_inode *raw_inode,
4018 struct ext4_inode_info *ei)
4020 __le32 *magic = (void *)raw_inode +
4021 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4022 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4023 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4024 ext4_find_inline_data_nolock(inode);
4026 EXT4_I(inode)->i_inline_off = 0;
4029 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4031 struct ext4_iloc iloc;
4032 struct ext4_inode *raw_inode;
4033 struct ext4_inode_info *ei;
4034 struct inode *inode;
4035 journal_t *journal = EXT4_SB(sb)->s_journal;
4041 inode = iget_locked(sb, ino);
4043 return ERR_PTR(-ENOMEM);
4044 if (!(inode->i_state & I_NEW))
4050 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4053 raw_inode = ext4_raw_inode(&iloc);
4055 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4056 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4057 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4058 EXT4_INODE_SIZE(inode->i_sb)) {
4059 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4060 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4061 EXT4_INODE_SIZE(inode->i_sb));
4066 ei->i_extra_isize = 0;
4068 /* Precompute checksum seed for inode metadata */
4069 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4070 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4071 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4073 __le32 inum = cpu_to_le32(inode->i_ino);
4074 __le32 gen = raw_inode->i_generation;
4075 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4077 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4081 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4082 EXT4_ERROR_INODE(inode, "checksum invalid");
4087 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4088 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4089 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4090 if (!(test_opt(inode->i_sb, NO_UID32))) {
4091 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4092 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4094 i_uid_write(inode, i_uid);
4095 i_gid_write(inode, i_gid);
4096 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4098 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4099 ei->i_inline_off = 0;
4100 ei->i_dir_start_lookup = 0;
4101 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4102 /* We now have enough fields to check if the inode was active or not.
4103 * This is needed because nfsd might try to access dead inodes
4104 * the test is that same one that e2fsck uses
4105 * NeilBrown 1999oct15
4107 if (inode->i_nlink == 0) {
4108 if ((inode->i_mode == 0 ||
4109 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4110 ino != EXT4_BOOT_LOADER_INO) {
4111 /* this inode is deleted */
4115 /* The only unlinked inodes we let through here have
4116 * valid i_mode and are being read by the orphan
4117 * recovery code: that's fine, we're about to complete
4118 * the process of deleting those.
4119 * OR it is the EXT4_BOOT_LOADER_INO which is
4120 * not initialized on a new filesystem. */
4122 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4123 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4124 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4125 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4127 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4128 inode->i_size = ext4_isize(raw_inode);
4129 ei->i_disksize = inode->i_size;
4131 ei->i_reserved_quota = 0;
4133 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4134 ei->i_block_group = iloc.block_group;
4135 ei->i_last_alloc_group = ~0;
4137 * NOTE! The in-memory inode i_data array is in little-endian order
4138 * even on big-endian machines: we do NOT byteswap the block numbers!
4140 for (block = 0; block < EXT4_N_BLOCKS; block++)
4141 ei->i_data[block] = raw_inode->i_block[block];
4142 INIT_LIST_HEAD(&ei->i_orphan);
4145 * Set transaction id's of transactions that have to be committed
4146 * to finish f[data]sync. We set them to currently running transaction
4147 * as we cannot be sure that the inode or some of its metadata isn't
4148 * part of the transaction - the inode could have been reclaimed and
4149 * now it is reread from disk.
4152 transaction_t *transaction;
4155 read_lock(&journal->j_state_lock);
4156 if (journal->j_running_transaction)
4157 transaction = journal->j_running_transaction;
4159 transaction = journal->j_committing_transaction;
4161 tid = transaction->t_tid;
4163 tid = journal->j_commit_sequence;
4164 read_unlock(&journal->j_state_lock);
4165 ei->i_sync_tid = tid;
4166 ei->i_datasync_tid = tid;
4169 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4170 if (ei->i_extra_isize == 0) {
4171 /* The extra space is currently unused. Use it. */
4172 ei->i_extra_isize = sizeof(struct ext4_inode) -
4173 EXT4_GOOD_OLD_INODE_SIZE;
4175 ext4_iget_extra_inode(inode, raw_inode, ei);
4179 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4180 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4181 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4182 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4184 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4185 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4186 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4187 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4189 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4194 if (ei->i_file_acl &&
4195 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4196 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4200 } else if (!ext4_has_inline_data(inode)) {
4201 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4202 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4203 (S_ISLNK(inode->i_mode) &&
4204 !ext4_inode_is_fast_symlink(inode))))
4205 /* Validate extent which is part of inode */
4206 ret = ext4_ext_check_inode(inode);
4207 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4208 (S_ISLNK(inode->i_mode) &&
4209 !ext4_inode_is_fast_symlink(inode))) {
4210 /* Validate block references which are part of inode */
4211 ret = ext4_ind_check_inode(inode);
4217 if (S_ISREG(inode->i_mode)) {
4218 inode->i_op = &ext4_file_inode_operations;
4219 inode->i_fop = &ext4_file_operations;
4220 ext4_set_aops(inode);
4221 } else if (S_ISDIR(inode->i_mode)) {
4222 inode->i_op = &ext4_dir_inode_operations;
4223 inode->i_fop = &ext4_dir_operations;
4224 } else if (S_ISLNK(inode->i_mode)) {
4225 if (ext4_inode_is_fast_symlink(inode)) {
4226 inode->i_op = &ext4_fast_symlink_inode_operations;
4227 nd_terminate_link(ei->i_data, inode->i_size,
4228 sizeof(ei->i_data) - 1);
4230 inode->i_op = &ext4_symlink_inode_operations;
4231 ext4_set_aops(inode);
4233 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4234 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4235 inode->i_op = &ext4_special_inode_operations;
4236 if (raw_inode->i_block[0])
4237 init_special_inode(inode, inode->i_mode,
4238 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4240 init_special_inode(inode, inode->i_mode,
4241 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4242 } else if (ino == EXT4_BOOT_LOADER_INO) {
4243 make_bad_inode(inode);
4246 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4250 ext4_set_inode_flags(inode);
4251 unlock_new_inode(inode);
4257 return ERR_PTR(ret);
4260 static int ext4_inode_blocks_set(handle_t *handle,
4261 struct ext4_inode *raw_inode,
4262 struct ext4_inode_info *ei)
4264 struct inode *inode = &(ei->vfs_inode);
4265 u64 i_blocks = inode->i_blocks;
4266 struct super_block *sb = inode->i_sb;
4268 if (i_blocks <= ~0U) {
4270 * i_blocks can be represented in a 32 bit variable
4271 * as multiple of 512 bytes
4273 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4274 raw_inode->i_blocks_high = 0;
4275 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4278 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4281 if (i_blocks <= 0xffffffffffffULL) {
4283 * i_blocks can be represented in a 48 bit variable
4284 * as multiple of 512 bytes
4286 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4287 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4288 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4290 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4291 /* i_block is stored in file system block size */
4292 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4293 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4294 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4300 * Post the struct inode info into an on-disk inode location in the
4301 * buffer-cache. This gobbles the caller's reference to the
4302 * buffer_head in the inode location struct.
4304 * The caller must have write access to iloc->bh.
4306 static int ext4_do_update_inode(handle_t *handle,
4307 struct inode *inode,
4308 struct ext4_iloc *iloc)
4310 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4311 struct ext4_inode_info *ei = EXT4_I(inode);
4312 struct buffer_head *bh = iloc->bh;
4313 struct super_block *sb = inode->i_sb;
4314 int err = 0, rc, block;
4315 int need_datasync = 0, set_large_file = 0;
4319 spin_lock(&ei->i_raw_lock);
4321 /* For fields not tracked in the in-memory inode,
4322 * initialise them to zero for new inodes. */
4323 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4324 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4326 ext4_get_inode_flags(ei);
4327 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4328 i_uid = i_uid_read(inode);
4329 i_gid = i_gid_read(inode);
4330 if (!(test_opt(inode->i_sb, NO_UID32))) {
4331 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4332 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4334 * Fix up interoperability with old kernels. Otherwise, old inodes get
4335 * re-used with the upper 16 bits of the uid/gid intact
4338 raw_inode->i_uid_high =
4339 cpu_to_le16(high_16_bits(i_uid));
4340 raw_inode->i_gid_high =
4341 cpu_to_le16(high_16_bits(i_gid));
4343 raw_inode->i_uid_high = 0;
4344 raw_inode->i_gid_high = 0;
4347 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4348 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4349 raw_inode->i_uid_high = 0;
4350 raw_inode->i_gid_high = 0;
4352 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4354 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4355 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4356 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4357 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4359 if (ext4_inode_blocks_set(handle, raw_inode, ei)) {
4360 spin_unlock(&ei->i_raw_lock);
4363 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4364 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4365 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4366 raw_inode->i_file_acl_high =
4367 cpu_to_le16(ei->i_file_acl >> 32);
4368 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4369 if (ei->i_disksize != ext4_isize(raw_inode)) {
4370 ext4_isize_set(raw_inode, ei->i_disksize);
4373 if (ei->i_disksize > 0x7fffffffULL) {
4374 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4375 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4376 EXT4_SB(sb)->s_es->s_rev_level ==
4377 cpu_to_le32(EXT4_GOOD_OLD_REV))
4380 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4381 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4382 if (old_valid_dev(inode->i_rdev)) {
4383 raw_inode->i_block[0] =
4384 cpu_to_le32(old_encode_dev(inode->i_rdev));
4385 raw_inode->i_block[1] = 0;
4387 raw_inode->i_block[0] = 0;
4388 raw_inode->i_block[1] =
4389 cpu_to_le32(new_encode_dev(inode->i_rdev));
4390 raw_inode->i_block[2] = 0;
4392 } else if (!ext4_has_inline_data(inode)) {
4393 for (block = 0; block < EXT4_N_BLOCKS; block++)
4394 raw_inode->i_block[block] = ei->i_data[block];
4397 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4398 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4399 if (ei->i_extra_isize) {
4400 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4401 raw_inode->i_version_hi =
4402 cpu_to_le32(inode->i_version >> 32);
4403 raw_inode->i_extra_isize =
4404 cpu_to_le16(ei->i_extra_isize);
4408 ext4_inode_csum_set(inode, raw_inode, ei);
4410 spin_unlock(&ei->i_raw_lock);
4412 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4413 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4416 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4417 if (set_large_file) {
4418 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
4419 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4422 ext4_update_dynamic_rev(sb);
4423 EXT4_SET_RO_COMPAT_FEATURE(sb,
4424 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4425 ext4_handle_sync(handle);
4426 err = ext4_handle_dirty_super(handle, sb);
4428 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4431 ext4_std_error(inode->i_sb, err);
4436 * ext4_write_inode()
4438 * We are called from a few places:
4440 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4441 * Here, there will be no transaction running. We wait for any running
4442 * transaction to commit.
4444 * - Within flush work (sys_sync(), kupdate and such).
4445 * We wait on commit, if told to.
4447 * - Within iput_final() -> write_inode_now()
4448 * We wait on commit, if told to.
4450 * In all cases it is actually safe for us to return without doing anything,
4451 * because the inode has been copied into a raw inode buffer in
4452 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4455 * Note that we are absolutely dependent upon all inode dirtiers doing the
4456 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4457 * which we are interested.
4459 * It would be a bug for them to not do this. The code:
4461 * mark_inode_dirty(inode)
4463 * inode->i_size = expr;
4465 * is in error because write_inode() could occur while `stuff()' is running,
4466 * and the new i_size will be lost. Plus the inode will no longer be on the
4467 * superblock's dirty inode list.
4469 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4473 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
4476 if (EXT4_SB(inode->i_sb)->s_journal) {
4477 if (ext4_journal_current_handle()) {
4478 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4484 * No need to force transaction in WB_SYNC_NONE mode. Also
4485 * ext4_sync_fs() will force the commit after everything is
4488 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
4491 err = ext4_force_commit(inode->i_sb);
4493 struct ext4_iloc iloc;
4495 err = __ext4_get_inode_loc(inode, &iloc, 0);
4499 * sync(2) will flush the whole buffer cache. No need to do
4500 * it here separately for each inode.
4502 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
4503 sync_dirty_buffer(iloc.bh);
4504 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4505 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4506 "IO error syncing inode");
4515 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4516 * buffers that are attached to a page stradding i_size and are undergoing
4517 * commit. In that case we have to wait for commit to finish and try again.
4519 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4523 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4524 tid_t commit_tid = 0;
4527 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4529 * All buffers in the last page remain valid? Then there's nothing to
4530 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4533 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4536 page = find_lock_page(inode->i_mapping,
4537 inode->i_size >> PAGE_CACHE_SHIFT);
4540 ret = __ext4_journalled_invalidatepage(page, offset,
4541 PAGE_CACHE_SIZE - offset);
4543 page_cache_release(page);
4547 read_lock(&journal->j_state_lock);
4548 if (journal->j_committing_transaction)
4549 commit_tid = journal->j_committing_transaction->t_tid;
4550 read_unlock(&journal->j_state_lock);
4552 jbd2_log_wait_commit(journal, commit_tid);
4559 * Called from notify_change.
4561 * We want to trap VFS attempts to truncate the file as soon as
4562 * possible. In particular, we want to make sure that when the VFS
4563 * shrinks i_size, we put the inode on the orphan list and modify
4564 * i_disksize immediately, so that during the subsequent flushing of
4565 * dirty pages and freeing of disk blocks, we can guarantee that any
4566 * commit will leave the blocks being flushed in an unused state on
4567 * disk. (On recovery, the inode will get truncated and the blocks will
4568 * be freed, so we have a strong guarantee that no future commit will
4569 * leave these blocks visible to the user.)
4571 * Another thing we have to assure is that if we are in ordered mode
4572 * and inode is still attached to the committing transaction, we must
4573 * we start writeout of all the dirty pages which are being truncated.
4574 * This way we are sure that all the data written in the previous
4575 * transaction are already on disk (truncate waits for pages under
4578 * Called with inode->i_mutex down.
4580 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4582 struct inode *inode = dentry->d_inode;
4585 const unsigned int ia_valid = attr->ia_valid;
4587 error = inode_change_ok(inode, attr);
4591 if (is_quota_modification(inode, attr))
4592 dquot_initialize(inode);
4593 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4594 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4597 /* (user+group)*(old+new) structure, inode write (sb,
4598 * inode block, ? - but truncate inode update has it) */
4599 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4600 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4601 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4602 if (IS_ERR(handle)) {
4603 error = PTR_ERR(handle);
4606 error = dquot_transfer(inode, attr);
4608 ext4_journal_stop(handle);
4611 /* Update corresponding info in inode so that everything is in
4612 * one transaction */
4613 if (attr->ia_valid & ATTR_UID)
4614 inode->i_uid = attr->ia_uid;
4615 if (attr->ia_valid & ATTR_GID)
4616 inode->i_gid = attr->ia_gid;
4617 error = ext4_mark_inode_dirty(handle, inode);
4618 ext4_journal_stop(handle);
4621 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4624 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4625 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4627 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4631 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4632 inode_inc_iversion(inode);
4634 if (S_ISREG(inode->i_mode) &&
4635 (attr->ia_size < inode->i_size)) {
4636 if (ext4_should_order_data(inode)) {
4637 error = ext4_begin_ordered_truncate(inode,
4642 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4643 if (IS_ERR(handle)) {
4644 error = PTR_ERR(handle);
4647 if (ext4_handle_valid(handle)) {
4648 error = ext4_orphan_add(handle, inode);
4651 down_write(&EXT4_I(inode)->i_data_sem);
4652 EXT4_I(inode)->i_disksize = attr->ia_size;
4653 rc = ext4_mark_inode_dirty(handle, inode);
4657 * We have to update i_size under i_data_sem together
4658 * with i_disksize to avoid races with writeback code
4659 * running ext4_wb_update_i_disksize().
4662 i_size_write(inode, attr->ia_size);
4663 up_write(&EXT4_I(inode)->i_data_sem);
4664 ext4_journal_stop(handle);
4666 ext4_orphan_del(NULL, inode);
4670 i_size_write(inode, attr->ia_size);
4673 * Blocks are going to be removed from the inode. Wait
4674 * for dio in flight. Temporarily disable
4675 * dioread_nolock to prevent livelock.
4678 if (!ext4_should_journal_data(inode)) {
4679 ext4_inode_block_unlocked_dio(inode);
4680 inode_dio_wait(inode);
4681 ext4_inode_resume_unlocked_dio(inode);
4683 ext4_wait_for_tail_page_commit(inode);
4686 * Truncate pagecache after we've waited for commit
4687 * in data=journal mode to make pages freeable.
4689 truncate_pagecache(inode, inode->i_size);
4692 * We want to call ext4_truncate() even if attr->ia_size ==
4693 * inode->i_size for cases like truncation of fallocated space
4695 if (attr->ia_valid & ATTR_SIZE)
4696 ext4_truncate(inode);
4699 setattr_copy(inode, attr);
4700 mark_inode_dirty(inode);
4704 * If the call to ext4_truncate failed to get a transaction handle at
4705 * all, we need to clean up the in-core orphan list manually.
4707 if (orphan && inode->i_nlink)
4708 ext4_orphan_del(NULL, inode);
4710 if (!rc && (ia_valid & ATTR_MODE))
4711 rc = posix_acl_chmod(inode, inode->i_mode);
4714 ext4_std_error(inode->i_sb, error);
4720 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4723 struct inode *inode;
4724 unsigned long long delalloc_blocks;
4726 inode = dentry->d_inode;
4727 generic_fillattr(inode, stat);
4730 * If there is inline data in the inode, the inode will normally not
4731 * have data blocks allocated (it may have an external xattr block).
4732 * Report at least one sector for such files, so tools like tar, rsync,
4733 * others doen't incorrectly think the file is completely sparse.
4735 if (unlikely(ext4_has_inline_data(inode)))
4736 stat->blocks += (stat->size + 511) >> 9;
4739 * We can't update i_blocks if the block allocation is delayed
4740 * otherwise in the case of system crash before the real block
4741 * allocation is done, we will have i_blocks inconsistent with
4742 * on-disk file blocks.
4743 * We always keep i_blocks updated together with real
4744 * allocation. But to not confuse with user, stat
4745 * will return the blocks that include the delayed allocation
4746 * blocks for this file.
4748 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4749 EXT4_I(inode)->i_reserved_data_blocks);
4750 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
4754 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4757 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4758 return ext4_ind_trans_blocks(inode, lblocks);
4759 return ext4_ext_index_trans_blocks(inode, pextents);
4763 * Account for index blocks, block groups bitmaps and block group
4764 * descriptor blocks if modify datablocks and index blocks
4765 * worse case, the indexs blocks spread over different block groups
4767 * If datablocks are discontiguous, they are possible to spread over
4768 * different block groups too. If they are contiguous, with flexbg,
4769 * they could still across block group boundary.
4771 * Also account for superblock, inode, quota and xattr blocks
4773 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4776 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4782 * How many index blocks need to touch to map @lblocks logical blocks
4783 * to @pextents physical extents?
4785 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4790 * Now let's see how many group bitmaps and group descriptors need
4793 groups = idxblocks + pextents;
4795 if (groups > ngroups)
4797 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4798 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4800 /* bitmaps and block group descriptor blocks */
4801 ret += groups + gdpblocks;
4803 /* Blocks for super block, inode, quota and xattr blocks */
4804 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4810 * Calculate the total number of credits to reserve to fit
4811 * the modification of a single pages into a single transaction,
4812 * which may include multiple chunks of block allocations.
4814 * This could be called via ext4_write_begin()
4816 * We need to consider the worse case, when
4817 * one new block per extent.
4819 int ext4_writepage_trans_blocks(struct inode *inode)
4821 int bpp = ext4_journal_blocks_per_page(inode);
4824 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4826 /* Account for data blocks for journalled mode */
4827 if (ext4_should_journal_data(inode))
4833 * Calculate the journal credits for a chunk of data modification.
4835 * This is called from DIO, fallocate or whoever calling
4836 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4838 * journal buffers for data blocks are not included here, as DIO
4839 * and fallocate do no need to journal data buffers.
4841 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4843 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4847 * The caller must have previously called ext4_reserve_inode_write().
4848 * Give this, we know that the caller already has write access to iloc->bh.
4850 int ext4_mark_iloc_dirty(handle_t *handle,
4851 struct inode *inode, struct ext4_iloc *iloc)
4855 if (IS_I_VERSION(inode))
4856 inode_inc_iversion(inode);
4858 /* the do_update_inode consumes one bh->b_count */
4861 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4862 err = ext4_do_update_inode(handle, inode, iloc);
4868 * On success, We end up with an outstanding reference count against
4869 * iloc->bh. This _must_ be cleaned up later.
4873 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4874 struct ext4_iloc *iloc)
4878 err = ext4_get_inode_loc(inode, iloc);
4880 BUFFER_TRACE(iloc->bh, "get_write_access");
4881 err = ext4_journal_get_write_access(handle, iloc->bh);
4887 ext4_std_error(inode->i_sb, err);
4892 * Expand an inode by new_extra_isize bytes.
4893 * Returns 0 on success or negative error number on failure.
4895 static int ext4_expand_extra_isize(struct inode *inode,
4896 unsigned int new_extra_isize,
4897 struct ext4_iloc iloc,
4900 struct ext4_inode *raw_inode;
4901 struct ext4_xattr_ibody_header *header;
4903 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4906 raw_inode = ext4_raw_inode(&iloc);
4908 header = IHDR(inode, raw_inode);
4910 /* No extended attributes present */
4911 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4912 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4913 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4915 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4919 /* try to expand with EAs present */
4920 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4925 * What we do here is to mark the in-core inode as clean with respect to inode
4926 * dirtiness (it may still be data-dirty).
4927 * This means that the in-core inode may be reaped by prune_icache
4928 * without having to perform any I/O. This is a very good thing,
4929 * because *any* task may call prune_icache - even ones which
4930 * have a transaction open against a different journal.
4932 * Is this cheating? Not really. Sure, we haven't written the
4933 * inode out, but prune_icache isn't a user-visible syncing function.
4934 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4935 * we start and wait on commits.
4937 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4939 struct ext4_iloc iloc;
4940 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4941 static unsigned int mnt_count;
4945 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4946 err = ext4_reserve_inode_write(handle, inode, &iloc);
4947 if (ext4_handle_valid(handle) &&
4948 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4949 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4951 * We need extra buffer credits since we may write into EA block
4952 * with this same handle. If journal_extend fails, then it will
4953 * only result in a minor loss of functionality for that inode.
4954 * If this is felt to be critical, then e2fsck should be run to
4955 * force a large enough s_min_extra_isize.
4957 if ((jbd2_journal_extend(handle,
4958 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4959 ret = ext4_expand_extra_isize(inode,
4960 sbi->s_want_extra_isize,
4963 ext4_set_inode_state(inode,
4964 EXT4_STATE_NO_EXPAND);
4966 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4967 ext4_warning(inode->i_sb,
4968 "Unable to expand inode %lu. Delete"
4969 " some EAs or run e2fsck.",
4972 le16_to_cpu(sbi->s_es->s_mnt_count);
4978 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4983 * ext4_dirty_inode() is called from __mark_inode_dirty()
4985 * We're really interested in the case where a file is being extended.
4986 * i_size has been changed by generic_commit_write() and we thus need
4987 * to include the updated inode in the current transaction.
4989 * Also, dquot_alloc_block() will always dirty the inode when blocks
4990 * are allocated to the file.
4992 * If the inode is marked synchronous, we don't honour that here - doing
4993 * so would cause a commit on atime updates, which we don't bother doing.
4994 * We handle synchronous inodes at the highest possible level.
4996 void ext4_dirty_inode(struct inode *inode, int flags)
5000 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5004 ext4_mark_inode_dirty(handle, inode);
5006 ext4_journal_stop(handle);
5013 * Bind an inode's backing buffer_head into this transaction, to prevent
5014 * it from being flushed to disk early. Unlike
5015 * ext4_reserve_inode_write, this leaves behind no bh reference and
5016 * returns no iloc structure, so the caller needs to repeat the iloc
5017 * lookup to mark the inode dirty later.
5019 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5021 struct ext4_iloc iloc;
5025 err = ext4_get_inode_loc(inode, &iloc);
5027 BUFFER_TRACE(iloc.bh, "get_write_access");
5028 err = jbd2_journal_get_write_access(handle, iloc.bh);
5030 err = ext4_handle_dirty_metadata(handle,
5036 ext4_std_error(inode->i_sb, err);
5041 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5048 * We have to be very careful here: changing a data block's
5049 * journaling status dynamically is dangerous. If we write a
5050 * data block to the journal, change the status and then delete
5051 * that block, we risk forgetting to revoke the old log record
5052 * from the journal and so a subsequent replay can corrupt data.
5053 * So, first we make sure that the journal is empty and that
5054 * nobody is changing anything.
5057 journal = EXT4_JOURNAL(inode);
5060 if (is_journal_aborted(journal))
5062 /* We have to allocate physical blocks for delalloc blocks
5063 * before flushing journal. otherwise delalloc blocks can not
5064 * be allocated any more. even more truncate on delalloc blocks
5065 * could trigger BUG by flushing delalloc blocks in journal.
5066 * There is no delalloc block in non-journal data mode.
5068 if (val && test_opt(inode->i_sb, DELALLOC)) {
5069 err = ext4_alloc_da_blocks(inode);
5074 /* Wait for all existing dio workers */
5075 ext4_inode_block_unlocked_dio(inode);
5076 inode_dio_wait(inode);
5078 jbd2_journal_lock_updates(journal);
5081 * OK, there are no updates running now, and all cached data is
5082 * synced to disk. We are now in a completely consistent state
5083 * which doesn't have anything in the journal, and we know that
5084 * no filesystem updates are running, so it is safe to modify
5085 * the inode's in-core data-journaling state flag now.
5089 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5091 jbd2_journal_flush(journal);
5092 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5094 ext4_set_aops(inode);
5096 jbd2_journal_unlock_updates(journal);
5097 ext4_inode_resume_unlocked_dio(inode);
5099 /* Finally we can mark the inode as dirty. */
5101 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5103 return PTR_ERR(handle);
5105 err = ext4_mark_inode_dirty(handle, inode);
5106 ext4_handle_sync(handle);
5107 ext4_journal_stop(handle);
5108 ext4_std_error(inode->i_sb, err);
5113 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5115 return !buffer_mapped(bh);
5118 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5120 struct page *page = vmf->page;
5124 struct file *file = vma->vm_file;
5125 struct inode *inode = file_inode(file);
5126 struct address_space *mapping = inode->i_mapping;
5128 get_block_t *get_block;
5131 sb_start_pagefault(inode->i_sb);
5132 file_update_time(vma->vm_file);
5133 /* Delalloc case is easy... */
5134 if (test_opt(inode->i_sb, DELALLOC) &&
5135 !ext4_should_journal_data(inode) &&
5136 !ext4_nonda_switch(inode->i_sb)) {
5138 ret = __block_page_mkwrite(vma, vmf,
5139 ext4_da_get_block_prep);
5140 } while (ret == -ENOSPC &&
5141 ext4_should_retry_alloc(inode->i_sb, &retries));
5146 size = i_size_read(inode);
5147 /* Page got truncated from under us? */
5148 if (page->mapping != mapping || page_offset(page) > size) {
5150 ret = VM_FAULT_NOPAGE;
5154 if (page->index == size >> PAGE_CACHE_SHIFT)
5155 len = size & ~PAGE_CACHE_MASK;
5157 len = PAGE_CACHE_SIZE;
5159 * Return if we have all the buffers mapped. This avoids the need to do
5160 * journal_start/journal_stop which can block and take a long time
5162 if (page_has_buffers(page)) {
5163 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5165 ext4_bh_unmapped)) {
5166 /* Wait so that we don't change page under IO */
5167 wait_for_stable_page(page);
5168 ret = VM_FAULT_LOCKED;
5173 /* OK, we need to fill the hole... */
5174 if (ext4_should_dioread_nolock(inode))
5175 get_block = ext4_get_block_write;
5177 get_block = ext4_get_block;
5179 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5180 ext4_writepage_trans_blocks(inode));
5181 if (IS_ERR(handle)) {
5182 ret = VM_FAULT_SIGBUS;
5185 ret = __block_page_mkwrite(vma, vmf, get_block);
5186 if (!ret && ext4_should_journal_data(inode)) {
5187 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5188 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5190 ret = VM_FAULT_SIGBUS;
5191 ext4_journal_stop(handle);
5194 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5196 ext4_journal_stop(handle);
5197 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5200 ret = block_page_mkwrite_return(ret);
5202 sb_end_pagefault(inode->i_sb);