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>
42 #include "ext4_jbd2.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 struct ext4_inode_info *ei)
54 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 csum_lo = le16_to_cpu(raw->i_checksum_lo);
60 raw->i_checksum_lo = 0;
61 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
62 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
63 csum_hi = le16_to_cpu(raw->i_checksum_hi);
64 raw->i_checksum_hi = 0;
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
68 EXT4_INODE_SIZE(inode->i_sb));
70 raw->i_checksum_lo = cpu_to_le16(csum_lo);
71 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
72 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
73 raw->i_checksum_hi = cpu_to_le16(csum_hi);
78 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
79 struct ext4_inode_info *ei)
81 __u32 provided, calculated;
83 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
84 cpu_to_le32(EXT4_OS_LINUX) ||
85 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
86 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
89 provided = le16_to_cpu(raw->i_checksum_lo);
90 calculated = ext4_inode_csum(inode, raw, ei);
91 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
92 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
93 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
97 return provided == calculated;
100 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
101 struct ext4_inode_info *ei)
105 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
106 cpu_to_le32(EXT4_OS_LINUX) ||
107 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
108 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
111 csum = ext4_inode_csum(inode, raw, ei);
112 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
113 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
114 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
115 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
118 static inline int ext4_begin_ordered_truncate(struct inode *inode,
121 trace_ext4_begin_ordered_truncate(inode, new_size);
123 * If jinode is zero, then we never opened the file for
124 * writing, so there's no need to call
125 * jbd2_journal_begin_ordered_truncate() since there's no
126 * outstanding writes we need to flush.
128 if (!EXT4_I(inode)->jinode)
130 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
131 EXT4_I(inode)->jinode,
135 static void ext4_invalidatepage(struct page *page, unsigned int offset,
136 unsigned int length);
137 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
138 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
139 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
143 * Test whether an inode is a fast symlink.
145 static int ext4_inode_is_fast_symlink(struct inode *inode)
147 int ea_blocks = EXT4_I(inode)->i_file_acl ?
148 (inode->i_sb->s_blocksize >> 9) : 0;
150 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode) == NULL);
170 jbd_debug(2, "restarting handle %p\n", handle);
171 up_write(&EXT4_I(inode)->i_data_sem);
172 ret = ext4_journal_restart(handle, nblocks);
173 down_write(&EXT4_I(inode)->i_data_sem);
174 ext4_discard_preallocations(inode);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode *inode)
187 trace_ext4_evict_inode(inode);
189 if (inode->i_nlink) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode) &&
209 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
210 inode->i_ino != EXT4_JOURNAL_INO) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_complete_transaction(journal, commit_tid);
215 filemap_write_and_wait(&inode->i_data);
217 truncate_inode_pages(&inode->i_data, 0);
218 ext4_ioend_shutdown(inode);
222 if (!is_bad_inode(inode))
223 dquot_initialize(inode);
225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
227 truncate_inode_pages(&inode->i_data, 0);
228 ext4_ioend_shutdown(inode);
230 if (is_bad_inode(inode))
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode->i_sb);
238 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
239 ext4_blocks_for_truncate(inode)+3);
240 if (IS_ERR(handle)) {
241 ext4_std_error(inode->i_sb, PTR_ERR(handle));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL, inode);
248 sb_end_intwrite(inode->i_sb);
253 ext4_handle_sync(handle);
255 err = ext4_mark_inode_dirty(handle, inode);
257 ext4_warning(inode->i_sb,
258 "couldn't mark inode dirty (err %d)", err);
262 ext4_truncate(inode);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle, 3)) {
271 err = ext4_journal_extend(handle, 3);
273 err = ext4_journal_restart(handle, 3);
275 ext4_warning(inode->i_sb,
276 "couldn't extend journal (err %d)", err);
278 ext4_journal_stop(handle);
279 ext4_orphan_del(NULL, inode);
280 sb_end_intwrite(inode->i_sb);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle, inode);
294 EXT4_I(inode)->i_dtime = get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle, inode))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode);
307 ext4_free_inode(handle, inode);
308 ext4_journal_stop(handle);
309 sb_end_intwrite(inode->i_sb);
312 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
316 qsize_t *ext4_get_reserved_space(struct inode *inode)
318 return &EXT4_I(inode)->i_reserved_quota;
323 * Calculate the number of metadata blocks need to reserve
324 * to allocate a block located at @lblock
326 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
328 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
329 return ext4_ext_calc_metadata_amount(inode, lblock);
331 return ext4_ind_calc_metadata_amount(inode, lblock);
335 * Called with i_data_sem down, which is important since we can call
336 * ext4_discard_preallocations() from here.
338 void ext4_da_update_reserve_space(struct inode *inode,
339 int used, int quota_claim)
341 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
342 struct ext4_inode_info *ei = EXT4_I(inode);
344 spin_lock(&ei->i_block_reservation_lock);
345 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
346 if (unlikely(used > ei->i_reserved_data_blocks)) {
347 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
348 "with only %d reserved data blocks",
349 __func__, inode->i_ino, used,
350 ei->i_reserved_data_blocks);
352 used = ei->i_reserved_data_blocks;
355 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
356 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
357 "with only %d reserved metadata blocks "
358 "(releasing %d blocks with reserved %d data blocks)",
359 inode->i_ino, ei->i_allocated_meta_blocks,
360 ei->i_reserved_meta_blocks, used,
361 ei->i_reserved_data_blocks);
363 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
366 /* Update per-inode reservations */
367 ei->i_reserved_data_blocks -= used;
368 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
369 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
370 used + ei->i_allocated_meta_blocks);
371 ei->i_allocated_meta_blocks = 0;
373 if (ei->i_reserved_data_blocks == 0) {
375 * We can release all of the reserved metadata blocks
376 * only when we have written all of the delayed
379 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
380 ei->i_reserved_meta_blocks);
381 ei->i_reserved_meta_blocks = 0;
382 ei->i_da_metadata_calc_len = 0;
384 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
386 /* Update quota subsystem for data blocks */
388 dquot_claim_block(inode, EXT4_C2B(sbi, used));
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
393 * not re-claim the quota for fallocated blocks.
395 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
403 if ((ei->i_reserved_data_blocks == 0) &&
404 (atomic_read(&inode->i_writecount) == 0))
405 ext4_discard_preallocations(inode);
408 static int __check_block_validity(struct inode *inode, const char *func,
410 struct ext4_map_blocks *map)
412 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
414 ext4_error_inode(inode, func, line, map->m_pblk,
415 "lblock %lu mapped to illegal pblock "
416 "(length %d)", (unsigned long) map->m_lblk,
423 #define check_block_validity(inode, map) \
424 __check_block_validity((inode), __func__, __LINE__, (map))
426 #ifdef ES_AGGRESSIVE_TEST
427 static void ext4_map_blocks_es_recheck(handle_t *handle,
429 struct ext4_map_blocks *es_map,
430 struct ext4_map_blocks *map,
437 * There is a race window that the result is not the same.
438 * e.g. xfstests #223 when dioread_nolock enables. The reason
439 * is that we lookup a block mapping in extent status tree with
440 * out taking i_data_sem. So at the time the unwritten extent
441 * could be converted.
443 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
444 down_read((&EXT4_I(inode)->i_data_sem));
445 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
446 retval = ext4_ext_map_blocks(handle, inode, map, flags &
447 EXT4_GET_BLOCKS_KEEP_SIZE);
449 retval = ext4_ind_map_blocks(handle, inode, map, flags &
450 EXT4_GET_BLOCKS_KEEP_SIZE);
452 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
453 up_read((&EXT4_I(inode)->i_data_sem));
455 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
456 * because it shouldn't be marked in es_map->m_flags.
458 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
461 * We don't check m_len because extent will be collpased in status
462 * tree. So the m_len might not equal.
464 if (es_map->m_lblk != map->m_lblk ||
465 es_map->m_flags != map->m_flags ||
466 es_map->m_pblk != map->m_pblk) {
467 printk("ES cache assertation failed for inode: %lu "
468 "es_cached ex [%d/%d/%llu/%x] != "
469 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
470 inode->i_ino, es_map->m_lblk, es_map->m_len,
471 es_map->m_pblk, es_map->m_flags, map->m_lblk,
472 map->m_len, map->m_pblk, map->m_flags,
476 #endif /* ES_AGGRESSIVE_TEST */
479 * The ext4_map_blocks() function tries to look up the requested blocks,
480 * and returns if the blocks are already mapped.
482 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
483 * and store the allocated blocks in the result buffer head and mark it
486 * If file type is extents based, it will call ext4_ext_map_blocks(),
487 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
490 * On success, it returns the number of blocks being mapped or allocate.
491 * if create==0 and the blocks are pre-allocated and uninitialized block,
492 * the result buffer head is unmapped. If the create ==1, it will make sure
493 * the buffer head is mapped.
495 * It returns 0 if plain look up failed (blocks have not been allocated), in
496 * that case, buffer head is unmapped
498 * It returns the error in case of allocation failure.
500 int ext4_map_blocks(handle_t *handle, struct inode *inode,
501 struct ext4_map_blocks *map, int flags)
503 struct extent_status es;
505 #ifdef ES_AGGRESSIVE_TEST
506 struct ext4_map_blocks orig_map;
508 memcpy(&orig_map, map, sizeof(*map));
512 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
513 "logical block %lu\n", inode->i_ino, flags, map->m_len,
514 (unsigned long) map->m_lblk);
516 /* Lookup extent status tree firstly */
517 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
518 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
519 map->m_pblk = ext4_es_pblock(&es) +
520 map->m_lblk - es.es_lblk;
521 map->m_flags |= ext4_es_is_written(&es) ?
522 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
523 retval = es.es_len - (map->m_lblk - es.es_lblk);
524 if (retval > map->m_len)
527 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
532 #ifdef ES_AGGRESSIVE_TEST
533 ext4_map_blocks_es_recheck(handle, inode, map,
540 * Try to see if we can get the block without requesting a new
543 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
544 down_read((&EXT4_I(inode)->i_data_sem));
545 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
546 retval = ext4_ext_map_blocks(handle, inode, map, flags &
547 EXT4_GET_BLOCKS_KEEP_SIZE);
549 retval = ext4_ind_map_blocks(handle, inode, map, flags &
550 EXT4_GET_BLOCKS_KEEP_SIZE);
554 unsigned long long status;
556 #ifdef ES_AGGRESSIVE_TEST
557 if (retval != map->m_len) {
558 printk("ES len assertation failed for inode: %lu "
559 "retval %d != map->m_len %d "
560 "in %s (lookup)\n", inode->i_ino, retval,
561 map->m_len, __func__);
565 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
566 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
567 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
568 ext4_find_delalloc_range(inode, map->m_lblk,
569 map->m_lblk + map->m_len - 1))
570 status |= EXTENT_STATUS_DELAYED;
571 ret = ext4_es_insert_extent(inode, map->m_lblk,
572 map->m_len, map->m_pblk, status);
576 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
577 up_read((&EXT4_I(inode)->i_data_sem));
580 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
581 int ret = check_block_validity(inode, map);
586 /* If it is only a block(s) look up */
587 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
591 * Returns if the blocks have already allocated
593 * Note that if blocks have been preallocated
594 * ext4_ext_get_block() returns the create = 0
595 * with buffer head unmapped.
597 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
601 * Here we clear m_flags because after allocating an new extent,
602 * it will be set again.
604 map->m_flags &= ~EXT4_MAP_FLAGS;
607 * New blocks allocate and/or writing to uninitialized extent
608 * will possibly result in updating i_data, so we take
609 * the write lock of i_data_sem, and call get_blocks()
610 * with create == 1 flag.
612 down_write((&EXT4_I(inode)->i_data_sem));
615 * if the caller is from delayed allocation writeout path
616 * we have already reserved fs blocks for allocation
617 * let the underlying get_block() function know to
618 * avoid double accounting
620 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
621 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
623 * We need to check for EXT4 here because migrate
624 * could have changed the inode type in between
626 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
627 retval = ext4_ext_map_blocks(handle, inode, map, flags);
629 retval = ext4_ind_map_blocks(handle, inode, map, flags);
631 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
633 * We allocated new blocks which will result in
634 * i_data's format changing. Force the migrate
635 * to fail by clearing migrate flags
637 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
641 * Update reserved blocks/metadata blocks after successful
642 * block allocation which had been deferred till now. We don't
643 * support fallocate for non extent files. So we can update
644 * reserve space here.
647 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
648 ext4_da_update_reserve_space(inode, retval, 1);
650 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
651 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
655 unsigned long long status;
657 #ifdef ES_AGGRESSIVE_TEST
658 if (retval != map->m_len) {
659 printk("ES len assertation failed for inode: %lu "
660 "retval %d != map->m_len %d "
661 "in %s (allocation)\n", inode->i_ino, retval,
662 map->m_len, __func__);
667 * If the extent has been zeroed out, we don't need to update
668 * extent status tree.
670 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
671 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
672 if (ext4_es_is_written(&es))
675 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
676 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
677 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
678 ext4_find_delalloc_range(inode, map->m_lblk,
679 map->m_lblk + map->m_len - 1))
680 status |= EXTENT_STATUS_DELAYED;
681 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
682 map->m_pblk, status);
688 up_write((&EXT4_I(inode)->i_data_sem));
689 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
690 int ret = check_block_validity(inode, map);
697 /* Maximum number of blocks we map for direct IO at once. */
698 #define DIO_MAX_BLOCKS 4096
700 static int _ext4_get_block(struct inode *inode, sector_t iblock,
701 struct buffer_head *bh, int flags)
703 handle_t *handle = ext4_journal_current_handle();
704 struct ext4_map_blocks map;
705 int ret = 0, started = 0;
708 if (ext4_has_inline_data(inode))
712 map.m_len = bh->b_size >> inode->i_blkbits;
714 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
715 /* Direct IO write... */
716 if (map.m_len > DIO_MAX_BLOCKS)
717 map.m_len = DIO_MAX_BLOCKS;
718 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
719 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
721 if (IS_ERR(handle)) {
722 ret = PTR_ERR(handle);
728 ret = ext4_map_blocks(handle, inode, &map, flags);
730 map_bh(bh, inode->i_sb, map.m_pblk);
731 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
732 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
736 ext4_journal_stop(handle);
740 int ext4_get_block(struct inode *inode, sector_t iblock,
741 struct buffer_head *bh, int create)
743 return _ext4_get_block(inode, iblock, bh,
744 create ? EXT4_GET_BLOCKS_CREATE : 0);
748 * `handle' can be NULL if create is zero
750 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
751 ext4_lblk_t block, int create, int *errp)
753 struct ext4_map_blocks map;
754 struct buffer_head *bh;
757 J_ASSERT(handle != NULL || create == 0);
761 err = ext4_map_blocks(handle, inode, &map,
762 create ? EXT4_GET_BLOCKS_CREATE : 0);
764 /* ensure we send some value back into *errp */
767 if (create && err == 0)
768 err = -ENOSPC; /* should never happen */
774 bh = sb_getblk(inode->i_sb, map.m_pblk);
779 if (map.m_flags & EXT4_MAP_NEW) {
780 J_ASSERT(create != 0);
781 J_ASSERT(handle != NULL);
784 * Now that we do not always journal data, we should
785 * keep in mind whether this should always journal the
786 * new buffer as metadata. For now, regular file
787 * writes use ext4_get_block instead, so it's not a
791 BUFFER_TRACE(bh, "call get_create_access");
792 fatal = ext4_journal_get_create_access(handle, bh);
793 if (!fatal && !buffer_uptodate(bh)) {
794 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
795 set_buffer_uptodate(bh);
798 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
799 err = ext4_handle_dirty_metadata(handle, inode, bh);
803 BUFFER_TRACE(bh, "not a new buffer");
813 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
814 ext4_lblk_t block, int create, int *err)
816 struct buffer_head *bh;
818 bh = ext4_getblk(handle, inode, block, create, err);
821 if (buffer_uptodate(bh))
823 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
825 if (buffer_uptodate(bh))
832 int ext4_walk_page_buffers(handle_t *handle,
833 struct buffer_head *head,
837 int (*fn)(handle_t *handle,
838 struct buffer_head *bh))
840 struct buffer_head *bh;
841 unsigned block_start, block_end;
842 unsigned blocksize = head->b_size;
844 struct buffer_head *next;
846 for (bh = head, block_start = 0;
847 ret == 0 && (bh != head || !block_start);
848 block_start = block_end, bh = next) {
849 next = bh->b_this_page;
850 block_end = block_start + blocksize;
851 if (block_end <= from || block_start >= to) {
852 if (partial && !buffer_uptodate(bh))
856 err = (*fn)(handle, bh);
864 * To preserve ordering, it is essential that the hole instantiation and
865 * the data write be encapsulated in a single transaction. We cannot
866 * close off a transaction and start a new one between the ext4_get_block()
867 * and the commit_write(). So doing the jbd2_journal_start at the start of
868 * prepare_write() is the right place.
870 * Also, this function can nest inside ext4_writepage(). In that case, we
871 * *know* that ext4_writepage() has generated enough buffer credits to do the
872 * whole page. So we won't block on the journal in that case, which is good,
873 * because the caller may be PF_MEMALLOC.
875 * By accident, ext4 can be reentered when a transaction is open via
876 * quota file writes. If we were to commit the transaction while thus
877 * reentered, there can be a deadlock - we would be holding a quota
878 * lock, and the commit would never complete if another thread had a
879 * transaction open and was blocking on the quota lock - a ranking
882 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
883 * will _not_ run commit under these circumstances because handle->h_ref
884 * is elevated. We'll still have enough credits for the tiny quotafile
887 int do_journal_get_write_access(handle_t *handle,
888 struct buffer_head *bh)
890 int dirty = buffer_dirty(bh);
893 if (!buffer_mapped(bh) || buffer_freed(bh))
896 * __block_write_begin() could have dirtied some buffers. Clean
897 * the dirty bit as jbd2_journal_get_write_access() could complain
898 * otherwise about fs integrity issues. Setting of the dirty bit
899 * by __block_write_begin() isn't a real problem here as we clear
900 * the bit before releasing a page lock and thus writeback cannot
901 * ever write the buffer.
904 clear_buffer_dirty(bh);
905 ret = ext4_journal_get_write_access(handle, bh);
907 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
911 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
912 struct buffer_head *bh_result, int create);
913 static int ext4_write_begin(struct file *file, struct address_space *mapping,
914 loff_t pos, unsigned len, unsigned flags,
915 struct page **pagep, void **fsdata)
917 struct inode *inode = mapping->host;
918 int ret, needed_blocks;
925 trace_ext4_write_begin(inode, pos, len, flags);
927 * Reserve one block more for addition to orphan list in case
928 * we allocate blocks but write fails for some reason
930 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
931 index = pos >> PAGE_CACHE_SHIFT;
932 from = pos & (PAGE_CACHE_SIZE - 1);
935 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
936 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
945 * grab_cache_page_write_begin() can take a long time if the
946 * system is thrashing due to memory pressure, or if the page
947 * is being written back. So grab it first before we start
948 * the transaction handle. This also allows us to allocate
949 * the page (if needed) without using GFP_NOFS.
952 page = grab_cache_page_write_begin(mapping, index, flags);
958 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
959 if (IS_ERR(handle)) {
960 page_cache_release(page);
961 return PTR_ERR(handle);
965 if (page->mapping != mapping) {
966 /* The page got truncated from under us */
968 page_cache_release(page);
969 ext4_journal_stop(handle);
972 wait_on_page_writeback(page);
974 if (ext4_should_dioread_nolock(inode))
975 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
977 ret = __block_write_begin(page, pos, len, ext4_get_block);
979 if (!ret && ext4_should_journal_data(inode)) {
980 ret = ext4_walk_page_buffers(handle, page_buffers(page),
982 do_journal_get_write_access);
988 * __block_write_begin may have instantiated a few blocks
989 * outside i_size. Trim these off again. Don't need
990 * i_size_read because we hold i_mutex.
992 * Add inode to orphan list in case we crash before
995 if (pos + len > inode->i_size && ext4_can_truncate(inode))
996 ext4_orphan_add(handle, inode);
998 ext4_journal_stop(handle);
999 if (pos + len > inode->i_size) {
1000 ext4_truncate_failed_write(inode);
1002 * If truncate failed early the inode might
1003 * still be on the orphan list; we need to
1004 * make sure the inode is removed from the
1005 * orphan list in that case.
1008 ext4_orphan_del(NULL, inode);
1011 if (ret == -ENOSPC &&
1012 ext4_should_retry_alloc(inode->i_sb, &retries))
1014 page_cache_release(page);
1021 /* For write_end() in data=journal mode */
1022 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1025 if (!buffer_mapped(bh) || buffer_freed(bh))
1027 set_buffer_uptodate(bh);
1028 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1029 clear_buffer_meta(bh);
1030 clear_buffer_prio(bh);
1035 * We need to pick up the new inode size which generic_commit_write gave us
1036 * `file' can be NULL - eg, when called from page_symlink().
1038 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1039 * buffers are managed internally.
1041 static int ext4_write_end(struct file *file,
1042 struct address_space *mapping,
1043 loff_t pos, unsigned len, unsigned copied,
1044 struct page *page, void *fsdata)
1046 handle_t *handle = ext4_journal_current_handle();
1047 struct inode *inode = mapping->host;
1049 int i_size_changed = 0;
1051 trace_ext4_write_end(inode, pos, len, copied);
1052 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1053 ret = ext4_jbd2_file_inode(handle, inode);
1056 page_cache_release(page);
1061 if (ext4_has_inline_data(inode))
1062 copied = ext4_write_inline_data_end(inode, pos, len,
1065 copied = block_write_end(file, mapping, pos,
1066 len, copied, page, fsdata);
1069 * No need to use i_size_read() here, the i_size
1070 * cannot change under us because we hole i_mutex.
1072 * But it's important to update i_size while still holding page lock:
1073 * page writeout could otherwise come in and zero beyond i_size.
1075 if (pos + copied > inode->i_size) {
1076 i_size_write(inode, pos + copied);
1080 if (pos + copied > EXT4_I(inode)->i_disksize) {
1081 /* We need to mark inode dirty even if
1082 * new_i_size is less that inode->i_size
1083 * but greater than i_disksize. (hint delalloc)
1085 ext4_update_i_disksize(inode, (pos + copied));
1089 page_cache_release(page);
1092 * Don't mark the inode dirty under page lock. First, it unnecessarily
1093 * makes the holding time of page lock longer. Second, it forces lock
1094 * ordering of page lock and transaction start for journaling
1098 ext4_mark_inode_dirty(handle, inode);
1102 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1103 /* if we have allocated more blocks and copied
1104 * less. We will have blocks allocated outside
1105 * inode->i_size. So truncate them
1107 ext4_orphan_add(handle, inode);
1109 ret2 = ext4_journal_stop(handle);
1113 if (pos + len > inode->i_size) {
1114 ext4_truncate_failed_write(inode);
1116 * If truncate failed early the inode might still be
1117 * on the orphan list; we need to make sure the inode
1118 * is removed from the orphan list in that case.
1121 ext4_orphan_del(NULL, inode);
1124 return ret ? ret : copied;
1127 static int ext4_journalled_write_end(struct file *file,
1128 struct address_space *mapping,
1129 loff_t pos, unsigned len, unsigned copied,
1130 struct page *page, void *fsdata)
1132 handle_t *handle = ext4_journal_current_handle();
1133 struct inode *inode = mapping->host;
1139 trace_ext4_journalled_write_end(inode, pos, len, copied);
1140 from = pos & (PAGE_CACHE_SIZE - 1);
1143 BUG_ON(!ext4_handle_valid(handle));
1145 if (ext4_has_inline_data(inode))
1146 copied = ext4_write_inline_data_end(inode, pos, len,
1150 if (!PageUptodate(page))
1152 page_zero_new_buffers(page, from+copied, to);
1155 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1156 to, &partial, write_end_fn);
1158 SetPageUptodate(page);
1160 new_i_size = pos + copied;
1161 if (new_i_size > inode->i_size)
1162 i_size_write(inode, pos+copied);
1163 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1164 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1165 if (new_i_size > EXT4_I(inode)->i_disksize) {
1166 ext4_update_i_disksize(inode, new_i_size);
1167 ret2 = ext4_mark_inode_dirty(handle, inode);
1173 page_cache_release(page);
1174 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1175 /* if we have allocated more blocks and copied
1176 * less. We will have blocks allocated outside
1177 * inode->i_size. So truncate them
1179 ext4_orphan_add(handle, inode);
1181 ret2 = ext4_journal_stop(handle);
1184 if (pos + len > inode->i_size) {
1185 ext4_truncate_failed_write(inode);
1187 * If truncate failed early the inode might still be
1188 * on the orphan list; we need to make sure the inode
1189 * is removed from the orphan list in that case.
1192 ext4_orphan_del(NULL, inode);
1195 return ret ? ret : copied;
1199 * Reserve a metadata for a single block located at lblock
1201 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1204 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1205 struct ext4_inode_info *ei = EXT4_I(inode);
1206 unsigned int md_needed;
1207 ext4_lblk_t save_last_lblock;
1211 * recalculate the amount of metadata blocks to reserve
1212 * in order to allocate nrblocks
1213 * worse case is one extent per block
1216 spin_lock(&ei->i_block_reservation_lock);
1218 * ext4_calc_metadata_amount() has side effects, which we have
1219 * to be prepared undo if we fail to claim space.
1221 save_len = ei->i_da_metadata_calc_len;
1222 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1223 md_needed = EXT4_NUM_B2C(sbi,
1224 ext4_calc_metadata_amount(inode, lblock));
1225 trace_ext4_da_reserve_space(inode, md_needed);
1228 * We do still charge estimated metadata to the sb though;
1229 * we cannot afford to run out of free blocks.
1231 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1232 ei->i_da_metadata_calc_len = save_len;
1233 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1234 spin_unlock(&ei->i_block_reservation_lock);
1235 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1241 ei->i_reserved_meta_blocks += md_needed;
1242 spin_unlock(&ei->i_block_reservation_lock);
1244 return 0; /* success */
1248 * Reserve a single cluster located at lblock
1250 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1253 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1254 struct ext4_inode_info *ei = EXT4_I(inode);
1255 unsigned int md_needed;
1257 ext4_lblk_t save_last_lblock;
1261 * We will charge metadata quota at writeout time; this saves
1262 * us from metadata over-estimation, though we may go over by
1263 * a small amount in the end. Here we just reserve for data.
1265 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1270 * recalculate the amount of metadata blocks to reserve
1271 * in order to allocate nrblocks
1272 * worse case is one extent per block
1275 spin_lock(&ei->i_block_reservation_lock);
1277 * ext4_calc_metadata_amount() has side effects, which we have
1278 * to be prepared undo if we fail to claim space.
1280 save_len = ei->i_da_metadata_calc_len;
1281 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1282 md_needed = EXT4_NUM_B2C(sbi,
1283 ext4_calc_metadata_amount(inode, lblock));
1284 trace_ext4_da_reserve_space(inode, md_needed);
1287 * We do still charge estimated metadata to the sb though;
1288 * we cannot afford to run out of free blocks.
1290 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1291 ei->i_da_metadata_calc_len = save_len;
1292 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1293 spin_unlock(&ei->i_block_reservation_lock);
1294 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1298 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1301 ei->i_reserved_data_blocks++;
1302 ei->i_reserved_meta_blocks += md_needed;
1303 spin_unlock(&ei->i_block_reservation_lock);
1305 return 0; /* success */
1308 static void ext4_da_release_space(struct inode *inode, int to_free)
1310 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1311 struct ext4_inode_info *ei = EXT4_I(inode);
1314 return; /* Nothing to release, exit */
1316 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1318 trace_ext4_da_release_space(inode, to_free);
1319 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1321 * if there aren't enough reserved blocks, then the
1322 * counter is messed up somewhere. Since this
1323 * function is called from invalidate page, it's
1324 * harmless to return without any action.
1326 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1327 "ino %lu, to_free %d with only %d reserved "
1328 "data blocks", inode->i_ino, to_free,
1329 ei->i_reserved_data_blocks);
1331 to_free = ei->i_reserved_data_blocks;
1333 ei->i_reserved_data_blocks -= to_free;
1335 if (ei->i_reserved_data_blocks == 0) {
1337 * We can release all of the reserved metadata blocks
1338 * only when we have written all of the delayed
1339 * allocation blocks.
1340 * Note that in case of bigalloc, i_reserved_meta_blocks,
1341 * i_reserved_data_blocks, etc. refer to number of clusters.
1343 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1344 ei->i_reserved_meta_blocks);
1345 ei->i_reserved_meta_blocks = 0;
1346 ei->i_da_metadata_calc_len = 0;
1349 /* update fs dirty data blocks counter */
1350 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1352 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1354 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1357 static void ext4_da_page_release_reservation(struct page *page,
1358 unsigned int offset,
1359 unsigned int length)
1362 struct buffer_head *head, *bh;
1363 unsigned int curr_off = 0;
1364 struct inode *inode = page->mapping->host;
1365 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1366 unsigned int stop = offset + length;
1370 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1372 head = page_buffers(page);
1375 unsigned int next_off = curr_off + bh->b_size;
1377 if (next_off > stop)
1380 if ((offset <= curr_off) && (buffer_delay(bh))) {
1382 clear_buffer_delay(bh);
1384 curr_off = next_off;
1385 } while ((bh = bh->b_this_page) != head);
1388 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1389 ext4_es_remove_extent(inode, lblk, to_release);
1392 /* If we have released all the blocks belonging to a cluster, then we
1393 * need to release the reserved space for that cluster. */
1394 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1395 while (num_clusters > 0) {
1396 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1397 ((num_clusters - 1) << sbi->s_cluster_bits);
1398 if (sbi->s_cluster_ratio == 1 ||
1399 !ext4_find_delalloc_cluster(inode, lblk))
1400 ext4_da_release_space(inode, 1);
1407 * Delayed allocation stuff
1410 struct mpage_da_data {
1411 struct inode *inode;
1412 struct writeback_control *wbc;
1413 pgoff_t first_page; /* The first page to write */
1414 pgoff_t next_page; /* Current page to examine */
1415 pgoff_t last_page; /* Last page to examine */
1417 * Extent to map - this can be after first_page because that can be
1418 * fully mapped. We somewhat abuse m_flags to store whether the extent
1419 * is delalloc or unwritten.
1421 struct ext4_map_blocks map;
1422 struct ext4_io_submit io_submit; /* IO submission data */
1425 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1430 struct pagevec pvec;
1431 struct inode *inode = mpd->inode;
1432 struct address_space *mapping = inode->i_mapping;
1434 /* This is necessary when next_page == 0. */
1435 if (mpd->first_page >= mpd->next_page)
1438 index = mpd->first_page;
1439 end = mpd->next_page - 1;
1441 ext4_lblk_t start, last;
1442 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1443 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1444 ext4_es_remove_extent(inode, start, last - start + 1);
1447 pagevec_init(&pvec, 0);
1448 while (index <= end) {
1449 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1452 for (i = 0; i < nr_pages; i++) {
1453 struct page *page = pvec.pages[i];
1454 if (page->index > end)
1456 BUG_ON(!PageLocked(page));
1457 BUG_ON(PageWriteback(page));
1459 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1460 ClearPageUptodate(page);
1464 index = pvec.pages[nr_pages - 1]->index + 1;
1465 pagevec_release(&pvec);
1469 static void ext4_print_free_blocks(struct inode *inode)
1471 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1472 struct super_block *sb = inode->i_sb;
1473 struct ext4_inode_info *ei = EXT4_I(inode);
1475 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1476 EXT4_C2B(EXT4_SB(inode->i_sb),
1477 ext4_count_free_clusters(sb)));
1478 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1479 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1480 (long long) EXT4_C2B(EXT4_SB(sb),
1481 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1482 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1483 (long long) EXT4_C2B(EXT4_SB(sb),
1484 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1485 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1486 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1487 ei->i_reserved_data_blocks);
1488 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1489 ei->i_reserved_meta_blocks);
1490 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1491 ei->i_allocated_meta_blocks);
1495 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1497 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1501 * This function is grabs code from the very beginning of
1502 * ext4_map_blocks, but assumes that the caller is from delayed write
1503 * time. This function looks up the requested blocks and sets the
1504 * buffer delay bit under the protection of i_data_sem.
1506 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1507 struct ext4_map_blocks *map,
1508 struct buffer_head *bh)
1510 struct extent_status es;
1512 sector_t invalid_block = ~((sector_t) 0xffff);
1513 #ifdef ES_AGGRESSIVE_TEST
1514 struct ext4_map_blocks orig_map;
1516 memcpy(&orig_map, map, sizeof(*map));
1519 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1523 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1524 "logical block %lu\n", inode->i_ino, map->m_len,
1525 (unsigned long) map->m_lblk);
1527 /* Lookup extent status tree firstly */
1528 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1530 if (ext4_es_is_hole(&es)) {
1532 down_read((&EXT4_I(inode)->i_data_sem));
1537 * Delayed extent could be allocated by fallocate.
1538 * So we need to check it.
1540 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1541 map_bh(bh, inode->i_sb, invalid_block);
1543 set_buffer_delay(bh);
1547 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1548 retval = es.es_len - (iblock - es.es_lblk);
1549 if (retval > map->m_len)
1550 retval = map->m_len;
1551 map->m_len = retval;
1552 if (ext4_es_is_written(&es))
1553 map->m_flags |= EXT4_MAP_MAPPED;
1554 else if (ext4_es_is_unwritten(&es))
1555 map->m_flags |= EXT4_MAP_UNWRITTEN;
1559 #ifdef ES_AGGRESSIVE_TEST
1560 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1566 * Try to see if we can get the block without requesting a new
1567 * file system block.
1569 down_read((&EXT4_I(inode)->i_data_sem));
1570 if (ext4_has_inline_data(inode)) {
1572 * We will soon create blocks for this page, and let
1573 * us pretend as if the blocks aren't allocated yet.
1574 * In case of clusters, we have to handle the work
1575 * of mapping from cluster so that the reserved space
1576 * is calculated properly.
1578 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1579 ext4_find_delalloc_cluster(inode, map->m_lblk))
1580 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1582 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1583 retval = ext4_ext_map_blocks(NULL, inode, map,
1584 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1586 retval = ext4_ind_map_blocks(NULL, inode, map,
1587 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1593 * XXX: __block_prepare_write() unmaps passed block,
1597 * If the block was allocated from previously allocated cluster,
1598 * then we don't need to reserve it again. However we still need
1599 * to reserve metadata for every block we're going to write.
1601 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1602 ret = ext4_da_reserve_space(inode, iblock);
1604 /* not enough space to reserve */
1609 ret = ext4_da_reserve_metadata(inode, iblock);
1611 /* not enough space to reserve */
1617 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1618 ~0, EXTENT_STATUS_DELAYED);
1624 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1625 * and it should not appear on the bh->b_state.
1627 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1629 map_bh(bh, inode->i_sb, invalid_block);
1631 set_buffer_delay(bh);
1632 } else if (retval > 0) {
1634 unsigned long long status;
1636 #ifdef ES_AGGRESSIVE_TEST
1637 if (retval != map->m_len) {
1638 printk("ES len assertation failed for inode: %lu "
1639 "retval %d != map->m_len %d "
1640 "in %s (lookup)\n", inode->i_ino, retval,
1641 map->m_len, __func__);
1645 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1646 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1647 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1648 map->m_pblk, status);
1654 up_read((&EXT4_I(inode)->i_data_sem));
1660 * This is a special get_blocks_t callback which is used by
1661 * ext4_da_write_begin(). It will either return mapped block or
1662 * reserve space for a single block.
1664 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1665 * We also have b_blocknr = -1 and b_bdev initialized properly
1667 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1668 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1669 * initialized properly.
1671 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1672 struct buffer_head *bh, int create)
1674 struct ext4_map_blocks map;
1677 BUG_ON(create == 0);
1678 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1680 map.m_lblk = iblock;
1684 * first, we need to know whether the block is allocated already
1685 * preallocated blocks are unmapped but should treated
1686 * the same as allocated blocks.
1688 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1692 map_bh(bh, inode->i_sb, map.m_pblk);
1693 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1695 if (buffer_unwritten(bh)) {
1696 /* A delayed write to unwritten bh should be marked
1697 * new and mapped. Mapped ensures that we don't do
1698 * get_block multiple times when we write to the same
1699 * offset and new ensures that we do proper zero out
1700 * for partial write.
1703 set_buffer_mapped(bh);
1708 static int bget_one(handle_t *handle, struct buffer_head *bh)
1714 static int bput_one(handle_t *handle, struct buffer_head *bh)
1720 static int __ext4_journalled_writepage(struct page *page,
1723 struct address_space *mapping = page->mapping;
1724 struct inode *inode = mapping->host;
1725 struct buffer_head *page_bufs = NULL;
1726 handle_t *handle = NULL;
1727 int ret = 0, err = 0;
1728 int inline_data = ext4_has_inline_data(inode);
1729 struct buffer_head *inode_bh = NULL;
1731 ClearPageChecked(page);
1734 BUG_ON(page->index != 0);
1735 BUG_ON(len > ext4_get_max_inline_size(inode));
1736 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1737 if (inode_bh == NULL)
1740 page_bufs = page_buffers(page);
1745 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1748 /* As soon as we unlock the page, it can go away, but we have
1749 * references to buffers so we are safe */
1752 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1753 ext4_writepage_trans_blocks(inode));
1754 if (IS_ERR(handle)) {
1755 ret = PTR_ERR(handle);
1759 BUG_ON(!ext4_handle_valid(handle));
1762 ret = ext4_journal_get_write_access(handle, inode_bh);
1764 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1767 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1768 do_journal_get_write_access);
1770 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1775 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1776 err = ext4_journal_stop(handle);
1780 if (!ext4_has_inline_data(inode))
1781 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1783 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1790 * Note that we don't need to start a transaction unless we're journaling data
1791 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1792 * need to file the inode to the transaction's list in ordered mode because if
1793 * we are writing back data added by write(), the inode is already there and if
1794 * we are writing back data modified via mmap(), no one guarantees in which
1795 * transaction the data will hit the disk. In case we are journaling data, we
1796 * cannot start transaction directly because transaction start ranks above page
1797 * lock so we have to do some magic.
1799 * This function can get called via...
1800 * - ext4_da_writepages after taking page lock (have journal handle)
1801 * - journal_submit_inode_data_buffers (no journal handle)
1802 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1803 * - grab_page_cache when doing write_begin (have journal handle)
1805 * We don't do any block allocation in this function. If we have page with
1806 * multiple blocks we need to write those buffer_heads that are mapped. This
1807 * is important for mmaped based write. So if we do with blocksize 1K
1808 * truncate(f, 1024);
1809 * a = mmap(f, 0, 4096);
1811 * truncate(f, 4096);
1812 * we have in the page first buffer_head mapped via page_mkwrite call back
1813 * but other buffer_heads would be unmapped but dirty (dirty done via the
1814 * do_wp_page). So writepage should write the first block. If we modify
1815 * the mmap area beyond 1024 we will again get a page_fault and the
1816 * page_mkwrite callback will do the block allocation and mark the
1817 * buffer_heads mapped.
1819 * We redirty the page if we have any buffer_heads that is either delay or
1820 * unwritten in the page.
1822 * We can get recursively called as show below.
1824 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1827 * But since we don't do any block allocation we should not deadlock.
1828 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1830 static int ext4_writepage(struct page *page,
1831 struct writeback_control *wbc)
1836 struct buffer_head *page_bufs = NULL;
1837 struct inode *inode = page->mapping->host;
1838 struct ext4_io_submit io_submit;
1840 trace_ext4_writepage(page);
1841 size = i_size_read(inode);
1842 if (page->index == size >> PAGE_CACHE_SHIFT)
1843 len = size & ~PAGE_CACHE_MASK;
1845 len = PAGE_CACHE_SIZE;
1847 page_bufs = page_buffers(page);
1849 * We cannot do block allocation or other extent handling in this
1850 * function. If there are buffers needing that, we have to redirty
1851 * the page. But we may reach here when we do a journal commit via
1852 * journal_submit_inode_data_buffers() and in that case we must write
1853 * allocated buffers to achieve data=ordered mode guarantees.
1855 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1856 ext4_bh_delay_or_unwritten)) {
1857 redirty_page_for_writepage(wbc, page);
1858 if (current->flags & PF_MEMALLOC) {
1860 * For memory cleaning there's no point in writing only
1861 * some buffers. So just bail out. Warn if we came here
1862 * from direct reclaim.
1864 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1871 if (PageChecked(page) && ext4_should_journal_data(inode))
1873 * It's mmapped pagecache. Add buffers and journal it. There
1874 * doesn't seem much point in redirtying the page here.
1876 return __ext4_journalled_writepage(page, len);
1878 ext4_io_submit_init(&io_submit, wbc);
1879 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1880 if (!io_submit.io_end) {
1881 redirty_page_for_writepage(wbc, page);
1885 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
1886 ext4_io_submit(&io_submit);
1887 /* Drop io_end reference we got from init */
1888 ext4_put_io_end_defer(io_submit.io_end);
1892 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1895 * mballoc gives us at most this number of blocks...
1896 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1897 * The rest of mballoc seems to handle chunks upto full group size.
1899 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1902 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1904 * @mpd - extent of blocks
1905 * @lblk - logical number of the block in the file
1906 * @b_state - b_state of the buffer head added
1908 * the function is used to collect contig. blocks in same state
1910 static int mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1911 unsigned long b_state)
1913 struct ext4_map_blocks *map = &mpd->map;
1915 /* Don't go larger than mballoc is willing to allocate */
1916 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1919 /* First block in the extent? */
1920 if (map->m_len == 0) {
1923 map->m_flags = b_state & BH_FLAGS;
1927 /* Can we merge the block to our big extent? */
1928 if (lblk == map->m_lblk + map->m_len &&
1929 (b_state & BH_FLAGS) == map->m_flags) {
1936 static bool add_page_bufs_to_extent(struct mpage_da_data *mpd,
1937 struct buffer_head *head,
1938 struct buffer_head *bh,
1941 struct inode *inode = mpd->inode;
1942 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1943 >> inode->i_blkbits;
1946 BUG_ON(buffer_locked(bh));
1948 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1949 (!buffer_delay(bh) && !buffer_unwritten(bh)) ||
1951 /* Found extent to map? */
1958 if (!mpage_add_bh_to_extent(mpd, lblk, bh->b_state))
1960 } while (lblk++, (bh = bh->b_this_page) != head);
1964 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1967 loff_t size = i_size_read(mpd->inode);
1970 BUG_ON(page->index != mpd->first_page);
1971 if (page->index == size >> PAGE_CACHE_SHIFT)
1972 len = size & ~PAGE_CACHE_MASK;
1974 len = PAGE_CACHE_SIZE;
1975 clear_page_dirty_for_io(page);
1976 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc);
1978 mpd->wbc->nr_to_write--;
1985 * mpage_map_buffers - update buffers corresponding to changed extent and
1986 * submit fully mapped pages for IO
1988 * @mpd - description of extent to map, on return next extent to map
1990 * Scan buffers corresponding to changed extent (we expect corresponding pages
1991 * to be already locked) and update buffer state according to new extent state.
1992 * We map delalloc buffers to their physical location, clear unwritten bits,
1993 * and mark buffers as uninit when we perform writes to uninitialized extents
1994 * and do extent conversion after IO is finished. If the last page is not fully
1995 * mapped, we update @map to the next extent in the last page that needs
1996 * mapping. Otherwise we submit the page for IO.
1998 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2000 struct pagevec pvec;
2002 struct inode *inode = mpd->inode;
2003 struct buffer_head *head, *bh;
2004 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
2005 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2006 >> inode->i_blkbits;
2012 start = mpd->map.m_lblk >> bpp_bits;
2013 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2014 lblk = start << bpp_bits;
2015 pblock = mpd->map.m_pblk;
2017 pagevec_init(&pvec, 0);
2018 while (start <= end) {
2019 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2023 for (i = 0; i < nr_pages; i++) {
2024 struct page *page = pvec.pages[i];
2026 if (page->index > end)
2028 /* Upto 'end' pages must be contiguous */
2029 BUG_ON(page->index != start);
2030 bh = head = page_buffers(page);
2032 if (lblk < mpd->map.m_lblk)
2034 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2036 * Buffer after end of mapped extent.
2037 * Find next buffer in the page to map.
2040 mpd->map.m_flags = 0;
2041 add_page_bufs_to_extent(mpd, head, bh,
2043 pagevec_release(&pvec);
2046 if (buffer_delay(bh)) {
2047 clear_buffer_delay(bh);
2048 bh->b_blocknr = pblock++;
2050 clear_buffer_unwritten(bh);
2051 } while (++lblk < blocks &&
2052 (bh = bh->b_this_page) != head);
2055 * FIXME: This is going to break if dioread_nolock
2056 * supports blocksize < pagesize as we will try to
2057 * convert potentially unmapped parts of inode.
2059 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2060 /* Page fully mapped - let IO run! */
2061 err = mpage_submit_page(mpd, page);
2063 pagevec_release(&pvec);
2068 pagevec_release(&pvec);
2070 /* Extent fully mapped and matches with page boundary. We are done. */
2072 mpd->map.m_flags = 0;
2076 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2078 struct inode *inode = mpd->inode;
2079 struct ext4_map_blocks *map = &mpd->map;
2080 int get_blocks_flags;
2083 trace_ext4_da_write_pages_extent(inode, map);
2085 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2086 * to convert an uninitialized extent to be initialized (in the case
2087 * where we have written into one or more preallocated blocks). It is
2088 * possible that we're going to need more metadata blocks than
2089 * previously reserved. However we must not fail because we're in
2090 * writeback and there is nothing we can do about it so it might result
2091 * in data loss. So use reserved blocks to allocate metadata if
2094 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2095 * in question are delalloc blocks. This affects functions in many
2096 * different parts of the allocation call path. This flag exists
2097 * primarily because we don't want to change *many* call functions, so
2098 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2099 * once the inode's allocation semaphore is taken.
2101 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2102 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2103 if (ext4_should_dioread_nolock(inode))
2104 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2105 if (map->m_flags & (1 << BH_Delay))
2106 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2108 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2111 if (map->m_flags & EXT4_MAP_UNINIT)
2112 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2114 BUG_ON(map->m_len == 0);
2115 if (map->m_flags & EXT4_MAP_NEW) {
2116 struct block_device *bdev = inode->i_sb->s_bdev;
2119 for (i = 0; i < map->m_len; i++)
2120 unmap_underlying_metadata(bdev, map->m_pblk + i);
2126 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2127 * mpd->len and submit pages underlying it for IO
2129 * @handle - handle for journal operations
2130 * @mpd - extent to map
2132 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2133 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2134 * them to initialized or split the described range from larger unwritten
2135 * extent. Note that we need not map all the described range since allocation
2136 * can return less blocks or the range is covered by more unwritten extents. We
2137 * cannot map more because we are limited by reserved transaction credits. On
2138 * the other hand we always make sure that the last touched page is fully
2139 * mapped so that it can be written out (and thus forward progress is
2140 * guaranteed). After mapping we submit all mapped pages for IO.
2142 static int mpage_map_and_submit_extent(handle_t *handle,
2143 struct mpage_da_data *mpd)
2145 struct inode *inode = mpd->inode;
2146 struct ext4_map_blocks *map = &mpd->map;
2150 mpd->io_submit.io_end->offset =
2151 ((loff_t)map->m_lblk) << inode->i_blkbits;
2152 while (map->m_len) {
2153 err = mpage_map_one_extent(handle, mpd);
2155 struct super_block *sb = inode->i_sb;
2158 * Need to commit transaction to free blocks. Let upper
2159 * layers sort it out.
2161 if (err == -ENOSPC && ext4_count_free_clusters(sb))
2164 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2165 ext4_msg(sb, KERN_CRIT,
2166 "Delayed block allocation failed for "
2167 "inode %lu at logical offset %llu with"
2168 " max blocks %u with error %d",
2170 (unsigned long long)map->m_lblk,
2171 (unsigned)map->m_len, err);
2172 ext4_msg(sb, KERN_CRIT,
2173 "This should not happen!! Data will "
2176 ext4_print_free_blocks(inode);
2178 /* invalidate all the pages */
2179 mpage_release_unused_pages(mpd, true);
2183 * Update buffer state, submit mapped pages, and get us new
2186 err = mpage_map_and_submit_buffers(mpd);
2191 /* Update on-disk size after IO is submitted */
2192 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
2193 if (disksize > i_size_read(inode))
2194 disksize = i_size_read(inode);
2195 if (disksize > EXT4_I(inode)->i_disksize) {
2198 ext4_update_i_disksize(inode, disksize);
2199 err2 = ext4_mark_inode_dirty(handle, inode);
2201 ext4_error(inode->i_sb,
2202 "Failed to mark inode %lu dirty",
2211 * Calculate the total number of credits to reserve for one writepages
2212 * iteration. This is called from ext4_da_writepages(). We map an extent of
2213 * upto MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2214 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2215 * bpp - 1 blocks in bpp different extents.
2217 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2219 int bpp = ext4_journal_blocks_per_page(inode);
2221 return ext4_meta_trans_blocks(inode,
2222 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2226 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2227 * and underlying extent to map
2229 * @mpd - where to look for pages
2231 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2232 * IO immediately. When we find a page which isn't mapped we start accumulating
2233 * extent of buffers underlying these pages that needs mapping (formed by
2234 * either delayed or unwritten buffers). We also lock the pages containing
2235 * these buffers. The extent found is returned in @mpd structure (starting at
2236 * mpd->lblk with length mpd->len blocks).
2238 * Note that this function can attach bios to one io_end structure which are
2239 * neither logically nor physically contiguous. Although it may seem as an
2240 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2241 * case as we need to track IO to all buffers underlying a page in one io_end.
2243 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2245 struct address_space *mapping = mpd->inode->i_mapping;
2246 struct pagevec pvec;
2247 unsigned int nr_pages;
2248 pgoff_t index = mpd->first_page;
2249 pgoff_t end = mpd->last_page;
2252 int blkbits = mpd->inode->i_blkbits;
2254 struct buffer_head *head;
2256 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2257 tag = PAGECACHE_TAG_TOWRITE;
2259 tag = PAGECACHE_TAG_DIRTY;
2261 pagevec_init(&pvec, 0);
2263 mpd->next_page = index;
2264 while (index <= end) {
2265 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2266 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2270 for (i = 0; i < nr_pages; i++) {
2271 struct page *page = pvec.pages[i];
2274 * At this point, the page may be truncated or
2275 * invalidated (changing page->mapping to NULL), or
2276 * even swizzled back from swapper_space to tmpfs file
2277 * mapping. However, page->index will not change
2278 * because we have a reference on the page.
2280 if (page->index > end)
2283 /* If we can't merge this page, we are done. */
2284 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2289 * If the page is no longer dirty, or its mapping no
2290 * longer corresponds to inode we are writing (which
2291 * means it has been truncated or invalidated), or the
2292 * page is already under writeback and we are not doing
2293 * a data integrity writeback, skip the page
2295 if (!PageDirty(page) ||
2296 (PageWriteback(page) &&
2297 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2298 unlikely(page->mapping != mapping)) {
2303 wait_on_page_writeback(page);
2304 BUG_ON(PageWriteback(page));
2306 if (mpd->map.m_len == 0)
2307 mpd->first_page = page->index;
2308 mpd->next_page = page->index + 1;
2309 /* Add all dirty buffers to mpd */
2310 lblk = ((ext4_lblk_t)page->index) <<
2311 (PAGE_CACHE_SHIFT - blkbits);
2312 head = page_buffers(page);
2313 if (!add_page_bufs_to_extent(mpd, head, head, lblk))
2315 /* So far everything mapped? Submit the page for IO. */
2316 if (mpd->map.m_len == 0) {
2317 err = mpage_submit_page(mpd, page);
2323 * Accumulated enough dirty pages? This doesn't apply
2324 * to WB_SYNC_ALL mode. For integrity sync we have to
2325 * keep going because someone may be concurrently
2326 * dirtying pages, and we might have synced a lot of
2327 * newly appeared dirty pages, but have not synced all
2328 * of the old dirty pages.
2330 if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2331 mpd->next_page - mpd->first_page >=
2332 mpd->wbc->nr_to_write)
2335 pagevec_release(&pvec);
2340 pagevec_release(&pvec);
2344 static int ext4_da_writepages(struct address_space *mapping,
2345 struct writeback_control *wbc)
2347 pgoff_t writeback_index = 0;
2348 long nr_to_write = wbc->nr_to_write;
2349 int range_whole = 0;
2351 handle_t *handle = NULL;
2352 struct mpage_da_data mpd;
2353 struct inode *inode = mapping->host;
2354 int needed_blocks, ret = 0;
2355 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2357 struct blk_plug plug;
2359 trace_ext4_da_writepages(inode, wbc);
2362 * No pages to write? This is mainly a kludge to avoid starting
2363 * a transaction for special inodes like journal inode on last iput()
2364 * because that could violate lock ordering on umount
2366 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2370 * If the filesystem has aborted, it is read-only, so return
2371 * right away instead of dumping stack traces later on that
2372 * will obscure the real source of the problem. We test
2373 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2374 * the latter could be true if the filesystem is mounted
2375 * read-only, and in that case, ext4_da_writepages should
2376 * *never* be called, so if that ever happens, we would want
2379 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2383 * If we have inline data and arrive here, it means that
2384 * we will soon create the block for the 1st page, so
2385 * we'd better clear the inline data here.
2387 if (ext4_has_inline_data(inode)) {
2388 /* Just inode will be modified... */
2389 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2390 if (IS_ERR(handle)) {
2391 ret = PTR_ERR(handle);
2392 goto out_writepages;
2394 BUG_ON(ext4_test_inode_state(inode,
2395 EXT4_STATE_MAY_INLINE_DATA));
2396 ext4_destroy_inline_data(handle, inode);
2397 ext4_journal_stop(handle);
2400 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2403 if (wbc->range_cyclic) {
2404 writeback_index = mapping->writeback_index;
2405 if (writeback_index)
2407 mpd.first_page = writeback_index;
2410 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2411 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
2416 ext4_io_submit_init(&mpd.io_submit, wbc);
2418 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2419 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2421 blk_start_plug(&plug);
2422 while (!done && mpd.first_page <= mpd.last_page) {
2423 /* For each extent of pages we use new io_end */
2424 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2425 if (!mpd.io_submit.io_end) {
2431 * We have two constraints: We find one extent to map and we
2432 * must always write out whole page (makes a difference when
2433 * blocksize < pagesize) so that we don't block on IO when we
2434 * try to write out the rest of the page. Journalled mode is
2435 * not supported by delalloc.
2437 BUG_ON(ext4_should_journal_data(inode));
2438 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2440 /* start a new transaction */
2441 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2443 if (IS_ERR(handle)) {
2444 ret = PTR_ERR(handle);
2445 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2446 "%ld pages, ino %lu; err %d", __func__,
2447 wbc->nr_to_write, inode->i_ino, ret);
2448 /* Release allocated io_end */
2449 ext4_put_io_end(mpd.io_submit.io_end);
2453 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2454 ret = mpage_prepare_extent_to_map(&mpd);
2457 ret = mpage_map_and_submit_extent(handle, &mpd);
2460 * We scanned the whole range (or exhausted
2461 * nr_to_write), submitted what was mapped and
2462 * didn't find anything needing mapping. We are
2468 ext4_journal_stop(handle);
2469 /* Submit prepared bio */
2470 ext4_io_submit(&mpd.io_submit);
2471 /* Unlock pages we didn't use */
2472 mpage_release_unused_pages(&mpd, false);
2473 /* Drop our io_end reference we got from init */
2474 ext4_put_io_end(mpd.io_submit.io_end);
2476 if (ret == -ENOSPC && sbi->s_journal) {
2478 * Commit the transaction which would
2479 * free blocks released in the transaction
2482 jbd2_journal_force_commit_nested(sbi->s_journal);
2486 /* Fatal error - ENOMEM, EIO... */
2490 blk_finish_plug(&plug);
2491 if (!ret && !cycled) {
2493 mpd.last_page = writeback_index - 1;
2499 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2501 * Set the writeback_index so that range_cyclic
2502 * mode will write it back later
2504 mapping->writeback_index = mpd.first_page;
2507 trace_ext4_da_writepages_result(inode, wbc, ret,
2508 nr_to_write - wbc->nr_to_write);
2512 static int ext4_nonda_switch(struct super_block *sb)
2514 s64 free_clusters, dirty_clusters;
2515 struct ext4_sb_info *sbi = EXT4_SB(sb);
2518 * switch to non delalloc mode if we are running low
2519 * on free block. The free block accounting via percpu
2520 * counters can get slightly wrong with percpu_counter_batch getting
2521 * accumulated on each CPU without updating global counters
2522 * Delalloc need an accurate free block accounting. So switch
2523 * to non delalloc when we are near to error range.
2526 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2528 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2530 * Start pushing delalloc when 1/2 of free blocks are dirty.
2532 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2533 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2535 if (2 * free_clusters < 3 * dirty_clusters ||
2536 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2538 * free block count is less than 150% of dirty blocks
2539 * or free blocks is less than watermark
2546 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2547 loff_t pos, unsigned len, unsigned flags,
2548 struct page **pagep, void **fsdata)
2550 int ret, retries = 0;
2553 struct inode *inode = mapping->host;
2556 index = pos >> PAGE_CACHE_SHIFT;
2558 if (ext4_nonda_switch(inode->i_sb)) {
2559 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2560 return ext4_write_begin(file, mapping, pos,
2561 len, flags, pagep, fsdata);
2563 *fsdata = (void *)0;
2564 trace_ext4_da_write_begin(inode, pos, len, flags);
2566 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2567 ret = ext4_da_write_inline_data_begin(mapping, inode,
2577 * grab_cache_page_write_begin() can take a long time if the
2578 * system is thrashing due to memory pressure, or if the page
2579 * is being written back. So grab it first before we start
2580 * the transaction handle. This also allows us to allocate
2581 * the page (if needed) without using GFP_NOFS.
2584 page = grab_cache_page_write_begin(mapping, index, flags);
2590 * With delayed allocation, we don't log the i_disksize update
2591 * if there is delayed block allocation. But we still need
2592 * to journalling the i_disksize update if writes to the end
2593 * of file which has an already mapped buffer.
2596 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2597 if (IS_ERR(handle)) {
2598 page_cache_release(page);
2599 return PTR_ERR(handle);
2603 if (page->mapping != mapping) {
2604 /* The page got truncated from under us */
2606 page_cache_release(page);
2607 ext4_journal_stop(handle);
2610 /* In case writeback began while the page was unlocked */
2611 wait_on_page_writeback(page);
2613 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2616 ext4_journal_stop(handle);
2618 * block_write_begin may have instantiated a few blocks
2619 * outside i_size. Trim these off again. Don't need
2620 * i_size_read because we hold i_mutex.
2622 if (pos + len > inode->i_size)
2623 ext4_truncate_failed_write(inode);
2625 if (ret == -ENOSPC &&
2626 ext4_should_retry_alloc(inode->i_sb, &retries))
2629 page_cache_release(page);
2638 * Check if we should update i_disksize
2639 * when write to the end of file but not require block allocation
2641 static int ext4_da_should_update_i_disksize(struct page *page,
2642 unsigned long offset)
2644 struct buffer_head *bh;
2645 struct inode *inode = page->mapping->host;
2649 bh = page_buffers(page);
2650 idx = offset >> inode->i_blkbits;
2652 for (i = 0; i < idx; i++)
2653 bh = bh->b_this_page;
2655 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2660 static int ext4_da_write_end(struct file *file,
2661 struct address_space *mapping,
2662 loff_t pos, unsigned len, unsigned copied,
2663 struct page *page, void *fsdata)
2665 struct inode *inode = mapping->host;
2667 handle_t *handle = ext4_journal_current_handle();
2669 unsigned long start, end;
2670 int write_mode = (int)(unsigned long)fsdata;
2672 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2673 return ext4_write_end(file, mapping, pos,
2674 len, copied, page, fsdata);
2676 trace_ext4_da_write_end(inode, pos, len, copied);
2677 start = pos & (PAGE_CACHE_SIZE - 1);
2678 end = start + copied - 1;
2681 * generic_write_end() will run mark_inode_dirty() if i_size
2682 * changes. So let's piggyback the i_disksize mark_inode_dirty
2685 new_i_size = pos + copied;
2686 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2687 if (ext4_has_inline_data(inode) ||
2688 ext4_da_should_update_i_disksize(page, end)) {
2689 down_write(&EXT4_I(inode)->i_data_sem);
2690 if (new_i_size > EXT4_I(inode)->i_disksize)
2691 EXT4_I(inode)->i_disksize = new_i_size;
2692 up_write(&EXT4_I(inode)->i_data_sem);
2693 /* We need to mark inode dirty even if
2694 * new_i_size is less that inode->i_size
2695 * bu greater than i_disksize.(hint delalloc)
2697 ext4_mark_inode_dirty(handle, inode);
2701 if (write_mode != CONVERT_INLINE_DATA &&
2702 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2703 ext4_has_inline_data(inode))
2704 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2707 ret2 = generic_write_end(file, mapping, pos, len, copied,
2713 ret2 = ext4_journal_stop(handle);
2717 return ret ? ret : copied;
2720 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2721 unsigned int length)
2724 * Drop reserved blocks
2726 BUG_ON(!PageLocked(page));
2727 if (!page_has_buffers(page))
2730 ext4_da_page_release_reservation(page, offset, length);
2733 ext4_invalidatepage(page, offset, length);
2739 * Force all delayed allocation blocks to be allocated for a given inode.
2741 int ext4_alloc_da_blocks(struct inode *inode)
2743 trace_ext4_alloc_da_blocks(inode);
2745 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2746 !EXT4_I(inode)->i_reserved_meta_blocks)
2750 * We do something simple for now. The filemap_flush() will
2751 * also start triggering a write of the data blocks, which is
2752 * not strictly speaking necessary (and for users of
2753 * laptop_mode, not even desirable). However, to do otherwise
2754 * would require replicating code paths in:
2756 * ext4_da_writepages() ->
2757 * write_cache_pages() ---> (via passed in callback function)
2758 * __mpage_da_writepage() -->
2759 * mpage_add_bh_to_extent()
2760 * mpage_da_map_blocks()
2762 * The problem is that write_cache_pages(), located in
2763 * mm/page-writeback.c, marks pages clean in preparation for
2764 * doing I/O, which is not desirable if we're not planning on
2767 * We could call write_cache_pages(), and then redirty all of
2768 * the pages by calling redirty_page_for_writepage() but that
2769 * would be ugly in the extreme. So instead we would need to
2770 * replicate parts of the code in the above functions,
2771 * simplifying them because we wouldn't actually intend to
2772 * write out the pages, but rather only collect contiguous
2773 * logical block extents, call the multi-block allocator, and
2774 * then update the buffer heads with the block allocations.
2776 * For now, though, we'll cheat by calling filemap_flush(),
2777 * which will map the blocks, and start the I/O, but not
2778 * actually wait for the I/O to complete.
2780 return filemap_flush(inode->i_mapping);
2784 * bmap() is special. It gets used by applications such as lilo and by
2785 * the swapper to find the on-disk block of a specific piece of data.
2787 * Naturally, this is dangerous if the block concerned is still in the
2788 * journal. If somebody makes a swapfile on an ext4 data-journaling
2789 * filesystem and enables swap, then they may get a nasty shock when the
2790 * data getting swapped to that swapfile suddenly gets overwritten by
2791 * the original zero's written out previously to the journal and
2792 * awaiting writeback in the kernel's buffer cache.
2794 * So, if we see any bmap calls here on a modified, data-journaled file,
2795 * take extra steps to flush any blocks which might be in the cache.
2797 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2799 struct inode *inode = mapping->host;
2804 * We can get here for an inline file via the FIBMAP ioctl
2806 if (ext4_has_inline_data(inode))
2809 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2810 test_opt(inode->i_sb, DELALLOC)) {
2812 * With delalloc we want to sync the file
2813 * so that we can make sure we allocate
2816 filemap_write_and_wait(mapping);
2819 if (EXT4_JOURNAL(inode) &&
2820 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2822 * This is a REALLY heavyweight approach, but the use of
2823 * bmap on dirty files is expected to be extremely rare:
2824 * only if we run lilo or swapon on a freshly made file
2825 * do we expect this to happen.
2827 * (bmap requires CAP_SYS_RAWIO so this does not
2828 * represent an unprivileged user DOS attack --- we'd be
2829 * in trouble if mortal users could trigger this path at
2832 * NB. EXT4_STATE_JDATA is not set on files other than
2833 * regular files. If somebody wants to bmap a directory
2834 * or symlink and gets confused because the buffer
2835 * hasn't yet been flushed to disk, they deserve
2836 * everything they get.
2839 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2840 journal = EXT4_JOURNAL(inode);
2841 jbd2_journal_lock_updates(journal);
2842 err = jbd2_journal_flush(journal);
2843 jbd2_journal_unlock_updates(journal);
2849 return generic_block_bmap(mapping, block, ext4_get_block);
2852 static int ext4_readpage(struct file *file, struct page *page)
2855 struct inode *inode = page->mapping->host;
2857 trace_ext4_readpage(page);
2859 if (ext4_has_inline_data(inode))
2860 ret = ext4_readpage_inline(inode, page);
2863 return mpage_readpage(page, ext4_get_block);
2869 ext4_readpages(struct file *file, struct address_space *mapping,
2870 struct list_head *pages, unsigned nr_pages)
2872 struct inode *inode = mapping->host;
2874 /* If the file has inline data, no need to do readpages. */
2875 if (ext4_has_inline_data(inode))
2878 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2881 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2882 unsigned int length)
2884 trace_ext4_invalidatepage(page, offset, length);
2886 /* No journalling happens on data buffers when this function is used */
2887 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2889 block_invalidatepage(page, offset, length);
2892 static int __ext4_journalled_invalidatepage(struct page *page,
2893 unsigned int offset,
2894 unsigned int length)
2896 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2898 trace_ext4_journalled_invalidatepage(page, offset, length);
2901 * If it's a full truncate we just forget about the pending dirtying
2903 if (offset == 0 && length == PAGE_CACHE_SIZE)
2904 ClearPageChecked(page);
2906 return jbd2_journal_invalidatepage(journal, page, offset, length);
2909 /* Wrapper for aops... */
2910 static void ext4_journalled_invalidatepage(struct page *page,
2911 unsigned int offset,
2912 unsigned int length)
2914 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2917 static int ext4_releasepage(struct page *page, gfp_t wait)
2919 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2921 trace_ext4_releasepage(page);
2923 /* Page has dirty journalled data -> cannot release */
2924 if (PageChecked(page))
2927 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2929 return try_to_free_buffers(page);
2933 * ext4_get_block used when preparing for a DIO write or buffer write.
2934 * We allocate an uinitialized extent if blocks haven't been allocated.
2935 * The extent will be converted to initialized after the IO is complete.
2937 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2938 struct buffer_head *bh_result, int create)
2940 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2941 inode->i_ino, create);
2942 return _ext4_get_block(inode, iblock, bh_result,
2943 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2946 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2947 struct buffer_head *bh_result, int create)
2949 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2950 inode->i_ino, create);
2951 return _ext4_get_block(inode, iblock, bh_result,
2952 EXT4_GET_BLOCKS_NO_LOCK);
2955 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2956 ssize_t size, void *private, int ret,
2959 struct inode *inode = file_inode(iocb->ki_filp);
2960 ext4_io_end_t *io_end = iocb->private;
2962 /* if not async direct IO just return */
2964 inode_dio_done(inode);
2966 aio_complete(iocb, ret, 0);
2970 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2971 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2972 iocb->private, io_end->inode->i_ino, iocb, offset,
2975 iocb->private = NULL;
2976 io_end->offset = offset;
2977 io_end->size = size;
2979 io_end->iocb = iocb;
2980 io_end->result = ret;
2982 ext4_put_io_end_defer(io_end);
2986 * For ext4 extent files, ext4 will do direct-io write to holes,
2987 * preallocated extents, and those write extend the file, no need to
2988 * fall back to buffered IO.
2990 * For holes, we fallocate those blocks, mark them as uninitialized
2991 * If those blocks were preallocated, we mark sure they are split, but
2992 * still keep the range to write as uninitialized.
2994 * The unwritten extents will be converted to written when DIO is completed.
2995 * For async direct IO, since the IO may still pending when return, we
2996 * set up an end_io call back function, which will do the conversion
2997 * when async direct IO completed.
2999 * If the O_DIRECT write will extend the file then add this inode to the
3000 * orphan list. So recovery will truncate it back to the original size
3001 * if the machine crashes during the write.
3004 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3005 const struct iovec *iov, loff_t offset,
3006 unsigned long nr_segs)
3008 struct file *file = iocb->ki_filp;
3009 struct inode *inode = file->f_mapping->host;
3011 size_t count = iov_length(iov, nr_segs);
3013 get_block_t *get_block_func = NULL;
3015 loff_t final_size = offset + count;
3016 ext4_io_end_t *io_end = NULL;
3018 /* Use the old path for reads and writes beyond i_size. */
3019 if (rw != WRITE || final_size > inode->i_size)
3020 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3022 BUG_ON(iocb->private == NULL);
3024 /* If we do a overwrite dio, i_mutex locking can be released */
3025 overwrite = *((int *)iocb->private);
3028 atomic_inc(&inode->i_dio_count);
3029 down_read(&EXT4_I(inode)->i_data_sem);
3030 mutex_unlock(&inode->i_mutex);
3034 * We could direct write to holes and fallocate.
3036 * Allocated blocks to fill the hole are marked as
3037 * uninitialized to prevent parallel buffered read to expose
3038 * the stale data before DIO complete the data IO.
3040 * As to previously fallocated extents, ext4 get_block will
3041 * just simply mark the buffer mapped but still keep the
3042 * extents uninitialized.
3044 * For non AIO case, we will convert those unwritten extents
3045 * to written after return back from blockdev_direct_IO.
3047 * For async DIO, the conversion needs to be deferred when the
3048 * IO is completed. The ext4 end_io callback function will be
3049 * called to take care of the conversion work. Here for async
3050 * case, we allocate an io_end structure to hook to the iocb.
3052 iocb->private = NULL;
3053 ext4_inode_aio_set(inode, NULL);
3054 if (!is_sync_kiocb(iocb)) {
3055 io_end = ext4_init_io_end(inode, GFP_NOFS);
3060 io_end->flag |= EXT4_IO_END_DIRECT;
3062 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3064 iocb->private = ext4_get_io_end(io_end);
3066 * we save the io structure for current async direct
3067 * IO, so that later ext4_map_blocks() could flag the
3068 * io structure whether there is a unwritten extents
3069 * needs to be converted when IO is completed.
3071 ext4_inode_aio_set(inode, io_end);
3075 get_block_func = ext4_get_block_write_nolock;
3077 get_block_func = ext4_get_block_write;
3078 dio_flags = DIO_LOCKING;
3080 ret = __blockdev_direct_IO(rw, iocb, inode,
3081 inode->i_sb->s_bdev, iov,
3089 * Put our reference to io_end. This can free the io_end structure e.g.
3090 * in sync IO case or in case of error. It can even perform extent
3091 * conversion if all bios we submitted finished before we got here.
3092 * Note that in that case iocb->private can be already set to NULL
3096 ext4_inode_aio_set(inode, NULL);
3097 ext4_put_io_end(io_end);
3099 * When no IO was submitted ext4_end_io_dio() was not
3100 * called so we have to put iocb's reference.
3102 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3103 WARN_ON(iocb->private != io_end);
3104 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3105 WARN_ON(io_end->iocb);
3107 * Generic code already did inode_dio_done() so we
3108 * have to clear EXT4_IO_END_DIRECT to not do it for
3112 ext4_put_io_end(io_end);
3113 iocb->private = NULL;
3116 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3117 EXT4_STATE_DIO_UNWRITTEN)) {
3120 * for non AIO case, since the IO is already
3121 * completed, we could do the conversion right here
3123 err = ext4_convert_unwritten_extents(inode,
3127 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3131 /* take i_mutex locking again if we do a ovewrite dio */
3133 inode_dio_done(inode);
3134 up_read(&EXT4_I(inode)->i_data_sem);
3135 mutex_lock(&inode->i_mutex);
3141 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3142 const struct iovec *iov, loff_t offset,
3143 unsigned long nr_segs)
3145 struct file *file = iocb->ki_filp;
3146 struct inode *inode = file->f_mapping->host;
3150 * If we are doing data journalling we don't support O_DIRECT
3152 if (ext4_should_journal_data(inode))
3155 /* Let buffer I/O handle the inline data case. */
3156 if (ext4_has_inline_data(inode))
3159 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3160 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3161 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3163 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3164 trace_ext4_direct_IO_exit(inode, offset,
3165 iov_length(iov, nr_segs), rw, ret);
3170 * Pages can be marked dirty completely asynchronously from ext4's journalling
3171 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3172 * much here because ->set_page_dirty is called under VFS locks. The page is
3173 * not necessarily locked.
3175 * We cannot just dirty the page and leave attached buffers clean, because the
3176 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3177 * or jbddirty because all the journalling code will explode.
3179 * So what we do is to mark the page "pending dirty" and next time writepage
3180 * is called, propagate that into the buffers appropriately.
3182 static int ext4_journalled_set_page_dirty(struct page *page)
3184 SetPageChecked(page);
3185 return __set_page_dirty_nobuffers(page);
3188 static const struct address_space_operations ext4_aops = {
3189 .readpage = ext4_readpage,
3190 .readpages = ext4_readpages,
3191 .writepage = ext4_writepage,
3192 .write_begin = ext4_write_begin,
3193 .write_end = ext4_write_end,
3195 .invalidatepage = ext4_invalidatepage,
3196 .releasepage = ext4_releasepage,
3197 .direct_IO = ext4_direct_IO,
3198 .migratepage = buffer_migrate_page,
3199 .is_partially_uptodate = block_is_partially_uptodate,
3200 .error_remove_page = generic_error_remove_page,
3203 static const struct address_space_operations ext4_journalled_aops = {
3204 .readpage = ext4_readpage,
3205 .readpages = ext4_readpages,
3206 .writepage = ext4_writepage,
3207 .write_begin = ext4_write_begin,
3208 .write_end = ext4_journalled_write_end,
3209 .set_page_dirty = ext4_journalled_set_page_dirty,
3211 .invalidatepage = ext4_journalled_invalidatepage,
3212 .releasepage = ext4_releasepage,
3213 .direct_IO = ext4_direct_IO,
3214 .is_partially_uptodate = block_is_partially_uptodate,
3215 .error_remove_page = generic_error_remove_page,
3218 static const struct address_space_operations ext4_da_aops = {
3219 .readpage = ext4_readpage,
3220 .readpages = ext4_readpages,
3221 .writepage = ext4_writepage,
3222 .writepages = ext4_da_writepages,
3223 .write_begin = ext4_da_write_begin,
3224 .write_end = ext4_da_write_end,
3226 .invalidatepage = ext4_da_invalidatepage,
3227 .releasepage = ext4_releasepage,
3228 .direct_IO = ext4_direct_IO,
3229 .migratepage = buffer_migrate_page,
3230 .is_partially_uptodate = block_is_partially_uptodate,
3231 .error_remove_page = generic_error_remove_page,
3234 void ext4_set_aops(struct inode *inode)
3236 switch (ext4_inode_journal_mode(inode)) {
3237 case EXT4_INODE_ORDERED_DATA_MODE:
3238 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3240 case EXT4_INODE_WRITEBACK_DATA_MODE:
3241 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3243 case EXT4_INODE_JOURNAL_DATA_MODE:
3244 inode->i_mapping->a_ops = &ext4_journalled_aops;
3249 if (test_opt(inode->i_sb, DELALLOC))
3250 inode->i_mapping->a_ops = &ext4_da_aops;
3252 inode->i_mapping->a_ops = &ext4_aops;
3256 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3257 * up to the end of the block which corresponds to `from'.
3258 * This required during truncate. We need to physically zero the tail end
3259 * of that block so it doesn't yield old data if the file is later grown.
3261 int ext4_block_truncate_page(handle_t *handle,
3262 struct address_space *mapping, loff_t from)
3264 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3267 struct inode *inode = mapping->host;
3269 blocksize = inode->i_sb->s_blocksize;
3270 length = blocksize - (offset & (blocksize - 1));
3272 return ext4_block_zero_page_range(handle, mapping, from, length);
3276 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3277 * starting from file offset 'from'. The range to be zero'd must
3278 * be contained with in one block. If the specified range exceeds
3279 * the end of the block it will be shortened to end of the block
3280 * that cooresponds to 'from'
3282 int ext4_block_zero_page_range(handle_t *handle,
3283 struct address_space *mapping, loff_t from, loff_t length)
3285 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3286 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3287 unsigned blocksize, max, pos;
3289 struct inode *inode = mapping->host;
3290 struct buffer_head *bh;
3294 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3295 mapping_gfp_mask(mapping) & ~__GFP_FS);
3299 blocksize = inode->i_sb->s_blocksize;
3300 max = blocksize - (offset & (blocksize - 1));
3303 * correct length if it does not fall between
3304 * 'from' and the end of the block
3306 if (length > max || length < 0)
3309 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3311 if (!page_has_buffers(page))
3312 create_empty_buffers(page, blocksize, 0);
3314 /* Find the buffer that contains "offset" */
3315 bh = page_buffers(page);
3317 while (offset >= pos) {
3318 bh = bh->b_this_page;
3324 if (buffer_freed(bh)) {
3325 BUFFER_TRACE(bh, "freed: skip");
3329 if (!buffer_mapped(bh)) {
3330 BUFFER_TRACE(bh, "unmapped");
3331 ext4_get_block(inode, iblock, bh, 0);
3332 /* unmapped? It's a hole - nothing to do */
3333 if (!buffer_mapped(bh)) {
3334 BUFFER_TRACE(bh, "still unmapped");
3339 /* Ok, it's mapped. Make sure it's up-to-date */
3340 if (PageUptodate(page))
3341 set_buffer_uptodate(bh);
3343 if (!buffer_uptodate(bh)) {
3345 ll_rw_block(READ, 1, &bh);
3347 /* Uhhuh. Read error. Complain and punt. */
3348 if (!buffer_uptodate(bh))
3352 if (ext4_should_journal_data(inode)) {
3353 BUFFER_TRACE(bh, "get write access");
3354 err = ext4_journal_get_write_access(handle, bh);
3359 zero_user(page, offset, length);
3361 BUFFER_TRACE(bh, "zeroed end of block");
3364 if (ext4_should_journal_data(inode)) {
3365 err = ext4_handle_dirty_metadata(handle, inode, bh);
3367 mark_buffer_dirty(bh);
3368 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3369 err = ext4_jbd2_file_inode(handle, inode);
3374 page_cache_release(page);
3378 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3379 loff_t lstart, loff_t length)
3381 struct super_block *sb = inode->i_sb;
3382 struct address_space *mapping = inode->i_mapping;
3383 unsigned partial = lstart & (sb->s_blocksize - 1);
3384 ext4_fsblk_t start, end;
3385 loff_t byte_end = (lstart + length - 1);
3388 start = lstart >> sb->s_blocksize_bits;
3389 end = byte_end >> sb->s_blocksize_bits;
3391 /* Handle partial zero within the single block */
3393 err = ext4_block_zero_page_range(handle, mapping,
3397 /* Handle partial zero out on the start of the range */
3399 err = ext4_block_zero_page_range(handle, mapping,
3400 lstart, sb->s_blocksize);
3404 /* Handle partial zero out on the end of the range */
3405 partial = byte_end & (sb->s_blocksize - 1);
3406 if (partial != sb->s_blocksize - 1)
3407 err = ext4_block_zero_page_range(handle, mapping,
3413 int ext4_can_truncate(struct inode *inode)
3415 if (S_ISREG(inode->i_mode))
3417 if (S_ISDIR(inode->i_mode))
3419 if (S_ISLNK(inode->i_mode))
3420 return !ext4_inode_is_fast_symlink(inode);
3425 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3426 * associated with the given offset and length
3428 * @inode: File inode
3429 * @offset: The offset where the hole will begin
3430 * @len: The length of the hole
3432 * Returns: 0 on success or negative on failure
3435 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3437 struct inode *inode = file_inode(file);
3438 struct super_block *sb = inode->i_sb;
3439 ext4_lblk_t first_block, stop_block;
3440 struct address_space *mapping = inode->i_mapping;
3441 loff_t first_block_offset, last_block_offset;
3443 unsigned int credits;
3446 if (!S_ISREG(inode->i_mode))
3449 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3450 /* TODO: Add support for bigalloc file systems */
3454 trace_ext4_punch_hole(inode, offset, length);
3457 * Write out all dirty pages to avoid race conditions
3458 * Then release them.
3460 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3461 ret = filemap_write_and_wait_range(mapping, offset,
3462 offset + length - 1);
3467 mutex_lock(&inode->i_mutex);
3468 /* It's not possible punch hole on append only file */
3469 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3473 if (IS_SWAPFILE(inode)) {
3478 /* No need to punch hole beyond i_size */
3479 if (offset >= inode->i_size)
3483 * If the hole extends beyond i_size, set the hole
3484 * to end after the page that contains i_size
3486 if (offset + length > inode->i_size) {
3487 length = inode->i_size +
3488 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3492 first_block_offset = round_up(offset, sb->s_blocksize);
3493 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3495 /* Now release the pages and zero block aligned part of pages*/
3496 if (last_block_offset > first_block_offset)
3497 truncate_pagecache_range(inode, first_block_offset,
3500 /* Wait all existing dio workers, newcomers will block on i_mutex */
3501 ext4_inode_block_unlocked_dio(inode);
3502 ret = ext4_flush_unwritten_io(inode);
3505 inode_dio_wait(inode);
3507 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3508 credits = ext4_writepage_trans_blocks(inode);
3510 credits = ext4_blocks_for_truncate(inode);
3511 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3512 if (IS_ERR(handle)) {
3513 ret = PTR_ERR(handle);
3514 ext4_std_error(sb, ret);
3518 ret = ext4_zero_partial_blocks(handle, inode, offset,
3523 first_block = (offset + sb->s_blocksize - 1) >>
3524 EXT4_BLOCK_SIZE_BITS(sb);
3525 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3527 /* If there are no blocks to remove, return now */
3528 if (first_block >= stop_block)
3531 down_write(&EXT4_I(inode)->i_data_sem);
3532 ext4_discard_preallocations(inode);
3534 ret = ext4_es_remove_extent(inode, first_block,
3535 stop_block - first_block);
3537 up_write(&EXT4_I(inode)->i_data_sem);
3541 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3542 ret = ext4_ext_remove_space(inode, first_block,
3545 ret = ext4_free_hole_blocks(handle, inode, first_block,
3548 ext4_discard_preallocations(inode);
3549 up_write(&EXT4_I(inode)->i_data_sem);
3551 ext4_handle_sync(handle);
3552 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3553 ext4_mark_inode_dirty(handle, inode);
3555 ext4_journal_stop(handle);
3557 ext4_inode_resume_unlocked_dio(inode);
3559 mutex_unlock(&inode->i_mutex);
3566 * We block out ext4_get_block() block instantiations across the entire
3567 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3568 * simultaneously on behalf of the same inode.
3570 * As we work through the truncate and commit bits of it to the journal there
3571 * is one core, guiding principle: the file's tree must always be consistent on
3572 * disk. We must be able to restart the truncate after a crash.
3574 * The file's tree may be transiently inconsistent in memory (although it
3575 * probably isn't), but whenever we close off and commit a journal transaction,
3576 * the contents of (the filesystem + the journal) must be consistent and
3577 * restartable. It's pretty simple, really: bottom up, right to left (although
3578 * left-to-right works OK too).
3580 * Note that at recovery time, journal replay occurs *before* the restart of
3581 * truncate against the orphan inode list.
3583 * The committed inode has the new, desired i_size (which is the same as
3584 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3585 * that this inode's truncate did not complete and it will again call
3586 * ext4_truncate() to have another go. So there will be instantiated blocks
3587 * to the right of the truncation point in a crashed ext4 filesystem. But
3588 * that's fine - as long as they are linked from the inode, the post-crash
3589 * ext4_truncate() run will find them and release them.
3591 void ext4_truncate(struct inode *inode)
3593 struct ext4_inode_info *ei = EXT4_I(inode);
3594 unsigned int credits;
3596 struct address_space *mapping = inode->i_mapping;
3599 * There is a possibility that we're either freeing the inode
3600 * or it completely new indode. In those cases we might not
3601 * have i_mutex locked because it's not necessary.
3603 if (!(inode->i_state & (I_NEW|I_FREEING)))
3604 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3605 trace_ext4_truncate_enter(inode);
3607 if (!ext4_can_truncate(inode))
3610 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3612 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3613 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3615 if (ext4_has_inline_data(inode)) {
3618 ext4_inline_data_truncate(inode, &has_inline);
3624 * finish any pending end_io work so we won't run the risk of
3625 * converting any truncated blocks to initialized later
3627 ext4_flush_unwritten_io(inode);
3629 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3630 credits = ext4_writepage_trans_blocks(inode);
3632 credits = ext4_blocks_for_truncate(inode);
3634 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3635 if (IS_ERR(handle)) {
3636 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3640 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3641 ext4_block_truncate_page(handle, mapping, inode->i_size);
3644 * We add the inode to the orphan list, so that if this
3645 * truncate spans multiple transactions, and we crash, we will
3646 * resume the truncate when the filesystem recovers. It also
3647 * marks the inode dirty, to catch the new size.
3649 * Implication: the file must always be in a sane, consistent
3650 * truncatable state while each transaction commits.
3652 if (ext4_orphan_add(handle, inode))
3655 down_write(&EXT4_I(inode)->i_data_sem);
3657 ext4_discard_preallocations(inode);
3659 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3660 ext4_ext_truncate(handle, inode);
3662 ext4_ind_truncate(handle, inode);
3664 up_write(&ei->i_data_sem);
3667 ext4_handle_sync(handle);
3671 * If this was a simple ftruncate() and the file will remain alive,
3672 * then we need to clear up the orphan record which we created above.
3673 * However, if this was a real unlink then we were called by
3674 * ext4_delete_inode(), and we allow that function to clean up the
3675 * orphan info for us.
3678 ext4_orphan_del(handle, inode);
3680 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3681 ext4_mark_inode_dirty(handle, inode);
3682 ext4_journal_stop(handle);
3684 trace_ext4_truncate_exit(inode);
3688 * ext4_get_inode_loc returns with an extra refcount against the inode's
3689 * underlying buffer_head on success. If 'in_mem' is true, we have all
3690 * data in memory that is needed to recreate the on-disk version of this
3693 static int __ext4_get_inode_loc(struct inode *inode,
3694 struct ext4_iloc *iloc, int in_mem)
3696 struct ext4_group_desc *gdp;
3697 struct buffer_head *bh;
3698 struct super_block *sb = inode->i_sb;
3700 int inodes_per_block, inode_offset;
3703 if (!ext4_valid_inum(sb, inode->i_ino))
3706 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3707 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3712 * Figure out the offset within the block group inode table
3714 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3715 inode_offset = ((inode->i_ino - 1) %
3716 EXT4_INODES_PER_GROUP(sb));
3717 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3718 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3720 bh = sb_getblk(sb, block);
3723 if (!buffer_uptodate(bh)) {
3727 * If the buffer has the write error flag, we have failed
3728 * to write out another inode in the same block. In this
3729 * case, we don't have to read the block because we may
3730 * read the old inode data successfully.
3732 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3733 set_buffer_uptodate(bh);
3735 if (buffer_uptodate(bh)) {
3736 /* someone brought it uptodate while we waited */
3742 * If we have all information of the inode in memory and this
3743 * is the only valid inode in the block, we need not read the
3747 struct buffer_head *bitmap_bh;
3750 start = inode_offset & ~(inodes_per_block - 1);
3752 /* Is the inode bitmap in cache? */
3753 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3754 if (unlikely(!bitmap_bh))
3758 * If the inode bitmap isn't in cache then the
3759 * optimisation may end up performing two reads instead
3760 * of one, so skip it.
3762 if (!buffer_uptodate(bitmap_bh)) {
3766 for (i = start; i < start + inodes_per_block; i++) {
3767 if (i == inode_offset)
3769 if (ext4_test_bit(i, bitmap_bh->b_data))
3773 if (i == start + inodes_per_block) {
3774 /* all other inodes are free, so skip I/O */
3775 memset(bh->b_data, 0, bh->b_size);
3776 set_buffer_uptodate(bh);
3784 * If we need to do any I/O, try to pre-readahead extra
3785 * blocks from the inode table.
3787 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3788 ext4_fsblk_t b, end, table;
3790 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3792 table = ext4_inode_table(sb, gdp);
3793 /* s_inode_readahead_blks is always a power of 2 */
3794 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3798 num = EXT4_INODES_PER_GROUP(sb);
3799 if (ext4_has_group_desc_csum(sb))
3800 num -= ext4_itable_unused_count(sb, gdp);
3801 table += num / inodes_per_block;
3805 sb_breadahead(sb, b++);
3809 * There are other valid inodes in the buffer, this inode
3810 * has in-inode xattrs, or we don't have this inode in memory.
3811 * Read the block from disk.
3813 trace_ext4_load_inode(inode);
3815 bh->b_end_io = end_buffer_read_sync;
3816 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3818 if (!buffer_uptodate(bh)) {
3819 EXT4_ERROR_INODE_BLOCK(inode, block,
3820 "unable to read itable block");
3830 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3832 /* We have all inode data except xattrs in memory here. */
3833 return __ext4_get_inode_loc(inode, iloc,
3834 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3837 void ext4_set_inode_flags(struct inode *inode)
3839 unsigned int flags = EXT4_I(inode)->i_flags;
3841 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3842 if (flags & EXT4_SYNC_FL)
3843 inode->i_flags |= S_SYNC;
3844 if (flags & EXT4_APPEND_FL)
3845 inode->i_flags |= S_APPEND;
3846 if (flags & EXT4_IMMUTABLE_FL)
3847 inode->i_flags |= S_IMMUTABLE;
3848 if (flags & EXT4_NOATIME_FL)
3849 inode->i_flags |= S_NOATIME;
3850 if (flags & EXT4_DIRSYNC_FL)
3851 inode->i_flags |= S_DIRSYNC;
3854 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3855 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3857 unsigned int vfs_fl;
3858 unsigned long old_fl, new_fl;
3861 vfs_fl = ei->vfs_inode.i_flags;
3862 old_fl = ei->i_flags;
3863 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3864 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3866 if (vfs_fl & S_SYNC)
3867 new_fl |= EXT4_SYNC_FL;
3868 if (vfs_fl & S_APPEND)
3869 new_fl |= EXT4_APPEND_FL;
3870 if (vfs_fl & S_IMMUTABLE)
3871 new_fl |= EXT4_IMMUTABLE_FL;
3872 if (vfs_fl & S_NOATIME)
3873 new_fl |= EXT4_NOATIME_FL;
3874 if (vfs_fl & S_DIRSYNC)
3875 new_fl |= EXT4_DIRSYNC_FL;
3876 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3879 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3880 struct ext4_inode_info *ei)
3883 struct inode *inode = &(ei->vfs_inode);
3884 struct super_block *sb = inode->i_sb;
3886 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3887 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3888 /* we are using combined 48 bit field */
3889 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3890 le32_to_cpu(raw_inode->i_blocks_lo);
3891 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3892 /* i_blocks represent file system block size */
3893 return i_blocks << (inode->i_blkbits - 9);
3898 return le32_to_cpu(raw_inode->i_blocks_lo);
3902 static inline void ext4_iget_extra_inode(struct inode *inode,
3903 struct ext4_inode *raw_inode,
3904 struct ext4_inode_info *ei)
3906 __le32 *magic = (void *)raw_inode +
3907 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3908 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3909 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3910 ext4_find_inline_data_nolock(inode);
3912 EXT4_I(inode)->i_inline_off = 0;
3915 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3917 struct ext4_iloc iloc;
3918 struct ext4_inode *raw_inode;
3919 struct ext4_inode_info *ei;
3920 struct inode *inode;
3921 journal_t *journal = EXT4_SB(sb)->s_journal;
3927 inode = iget_locked(sb, ino);
3929 return ERR_PTR(-ENOMEM);
3930 if (!(inode->i_state & I_NEW))
3936 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3939 raw_inode = ext4_raw_inode(&iloc);
3941 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3942 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3943 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3944 EXT4_INODE_SIZE(inode->i_sb)) {
3945 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3946 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3947 EXT4_INODE_SIZE(inode->i_sb));
3952 ei->i_extra_isize = 0;
3954 /* Precompute checksum seed for inode metadata */
3955 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3956 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3957 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3959 __le32 inum = cpu_to_le32(inode->i_ino);
3960 __le32 gen = raw_inode->i_generation;
3961 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3963 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3967 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3968 EXT4_ERROR_INODE(inode, "checksum invalid");
3973 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3974 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3975 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3976 if (!(test_opt(inode->i_sb, NO_UID32))) {
3977 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3978 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3980 i_uid_write(inode, i_uid);
3981 i_gid_write(inode, i_gid);
3982 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3984 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3985 ei->i_inline_off = 0;
3986 ei->i_dir_start_lookup = 0;
3987 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3988 /* We now have enough fields to check if the inode was active or not.
3989 * This is needed because nfsd might try to access dead inodes
3990 * the test is that same one that e2fsck uses
3991 * NeilBrown 1999oct15
3993 if (inode->i_nlink == 0) {
3994 if ((inode->i_mode == 0 ||
3995 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
3996 ino != EXT4_BOOT_LOADER_INO) {
3997 /* this inode is deleted */
4001 /* The only unlinked inodes we let through here have
4002 * valid i_mode and are being read by the orphan
4003 * recovery code: that's fine, we're about to complete
4004 * the process of deleting those.
4005 * OR it is the EXT4_BOOT_LOADER_INO which is
4006 * not initialized on a new filesystem. */
4008 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4009 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4010 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4011 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4013 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4014 inode->i_size = ext4_isize(raw_inode);
4015 ei->i_disksize = inode->i_size;
4017 ei->i_reserved_quota = 0;
4019 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4020 ei->i_block_group = iloc.block_group;
4021 ei->i_last_alloc_group = ~0;
4023 * NOTE! The in-memory inode i_data array is in little-endian order
4024 * even on big-endian machines: we do NOT byteswap the block numbers!
4026 for (block = 0; block < EXT4_N_BLOCKS; block++)
4027 ei->i_data[block] = raw_inode->i_block[block];
4028 INIT_LIST_HEAD(&ei->i_orphan);
4031 * Set transaction id's of transactions that have to be committed
4032 * to finish f[data]sync. We set them to currently running transaction
4033 * as we cannot be sure that the inode or some of its metadata isn't
4034 * part of the transaction - the inode could have been reclaimed and
4035 * now it is reread from disk.
4038 transaction_t *transaction;
4041 read_lock(&journal->j_state_lock);
4042 if (journal->j_running_transaction)
4043 transaction = journal->j_running_transaction;
4045 transaction = journal->j_committing_transaction;
4047 tid = transaction->t_tid;
4049 tid = journal->j_commit_sequence;
4050 read_unlock(&journal->j_state_lock);
4051 ei->i_sync_tid = tid;
4052 ei->i_datasync_tid = tid;
4055 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4056 if (ei->i_extra_isize == 0) {
4057 /* The extra space is currently unused. Use it. */
4058 ei->i_extra_isize = sizeof(struct ext4_inode) -
4059 EXT4_GOOD_OLD_INODE_SIZE;
4061 ext4_iget_extra_inode(inode, raw_inode, ei);
4065 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4066 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4067 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4068 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4070 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4071 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4072 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4074 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4078 if (ei->i_file_acl &&
4079 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4080 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4084 } else if (!ext4_has_inline_data(inode)) {
4085 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4086 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4087 (S_ISLNK(inode->i_mode) &&
4088 !ext4_inode_is_fast_symlink(inode))))
4089 /* Validate extent which is part of inode */
4090 ret = ext4_ext_check_inode(inode);
4091 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4092 (S_ISLNK(inode->i_mode) &&
4093 !ext4_inode_is_fast_symlink(inode))) {
4094 /* Validate block references which are part of inode */
4095 ret = ext4_ind_check_inode(inode);
4101 if (S_ISREG(inode->i_mode)) {
4102 inode->i_op = &ext4_file_inode_operations;
4103 inode->i_fop = &ext4_file_operations;
4104 ext4_set_aops(inode);
4105 } else if (S_ISDIR(inode->i_mode)) {
4106 inode->i_op = &ext4_dir_inode_operations;
4107 inode->i_fop = &ext4_dir_operations;
4108 } else if (S_ISLNK(inode->i_mode)) {
4109 if (ext4_inode_is_fast_symlink(inode)) {
4110 inode->i_op = &ext4_fast_symlink_inode_operations;
4111 nd_terminate_link(ei->i_data, inode->i_size,
4112 sizeof(ei->i_data) - 1);
4114 inode->i_op = &ext4_symlink_inode_operations;
4115 ext4_set_aops(inode);
4117 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4118 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4119 inode->i_op = &ext4_special_inode_operations;
4120 if (raw_inode->i_block[0])
4121 init_special_inode(inode, inode->i_mode,
4122 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4124 init_special_inode(inode, inode->i_mode,
4125 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4126 } else if (ino == EXT4_BOOT_LOADER_INO) {
4127 make_bad_inode(inode);
4130 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4134 ext4_set_inode_flags(inode);
4135 unlock_new_inode(inode);
4141 return ERR_PTR(ret);
4144 static int ext4_inode_blocks_set(handle_t *handle,
4145 struct ext4_inode *raw_inode,
4146 struct ext4_inode_info *ei)
4148 struct inode *inode = &(ei->vfs_inode);
4149 u64 i_blocks = inode->i_blocks;
4150 struct super_block *sb = inode->i_sb;
4152 if (i_blocks <= ~0U) {
4154 * i_blocks can be represented in a 32 bit variable
4155 * as multiple of 512 bytes
4157 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4158 raw_inode->i_blocks_high = 0;
4159 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4162 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4165 if (i_blocks <= 0xffffffffffffULL) {
4167 * i_blocks can be represented in a 48 bit variable
4168 * as multiple of 512 bytes
4170 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4171 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4172 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4174 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4175 /* i_block is stored in file system block size */
4176 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4177 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4178 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4184 * Post the struct inode info into an on-disk inode location in the
4185 * buffer-cache. This gobbles the caller's reference to the
4186 * buffer_head in the inode location struct.
4188 * The caller must have write access to iloc->bh.
4190 static int ext4_do_update_inode(handle_t *handle,
4191 struct inode *inode,
4192 struct ext4_iloc *iloc)
4194 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4195 struct ext4_inode_info *ei = EXT4_I(inode);
4196 struct buffer_head *bh = iloc->bh;
4197 int err = 0, rc, block;
4198 int need_datasync = 0;
4202 /* For fields not not tracking in the in-memory inode,
4203 * initialise them to zero for new inodes. */
4204 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4205 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4207 ext4_get_inode_flags(ei);
4208 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4209 i_uid = i_uid_read(inode);
4210 i_gid = i_gid_read(inode);
4211 if (!(test_opt(inode->i_sb, NO_UID32))) {
4212 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4213 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4215 * Fix up interoperability with old kernels. Otherwise, old inodes get
4216 * re-used with the upper 16 bits of the uid/gid intact
4219 raw_inode->i_uid_high =
4220 cpu_to_le16(high_16_bits(i_uid));
4221 raw_inode->i_gid_high =
4222 cpu_to_le16(high_16_bits(i_gid));
4224 raw_inode->i_uid_high = 0;
4225 raw_inode->i_gid_high = 0;
4228 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4229 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4230 raw_inode->i_uid_high = 0;
4231 raw_inode->i_gid_high = 0;
4233 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4235 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4236 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4237 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4238 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4240 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4242 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4243 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4244 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4245 cpu_to_le32(EXT4_OS_HURD))
4246 raw_inode->i_file_acl_high =
4247 cpu_to_le16(ei->i_file_acl >> 32);
4248 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4249 if (ei->i_disksize != ext4_isize(raw_inode)) {
4250 ext4_isize_set(raw_inode, ei->i_disksize);
4253 if (ei->i_disksize > 0x7fffffffULL) {
4254 struct super_block *sb = inode->i_sb;
4255 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4256 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4257 EXT4_SB(sb)->s_es->s_rev_level ==
4258 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4259 /* If this is the first large file
4260 * created, add a flag to the superblock.
4262 err = ext4_journal_get_write_access(handle,
4263 EXT4_SB(sb)->s_sbh);
4266 ext4_update_dynamic_rev(sb);
4267 EXT4_SET_RO_COMPAT_FEATURE(sb,
4268 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4269 ext4_handle_sync(handle);
4270 err = ext4_handle_dirty_super(handle, sb);
4273 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4274 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4275 if (old_valid_dev(inode->i_rdev)) {
4276 raw_inode->i_block[0] =
4277 cpu_to_le32(old_encode_dev(inode->i_rdev));
4278 raw_inode->i_block[1] = 0;
4280 raw_inode->i_block[0] = 0;
4281 raw_inode->i_block[1] =
4282 cpu_to_le32(new_encode_dev(inode->i_rdev));
4283 raw_inode->i_block[2] = 0;
4285 } else if (!ext4_has_inline_data(inode)) {
4286 for (block = 0; block < EXT4_N_BLOCKS; block++)
4287 raw_inode->i_block[block] = ei->i_data[block];
4290 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4291 if (ei->i_extra_isize) {
4292 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4293 raw_inode->i_version_hi =
4294 cpu_to_le32(inode->i_version >> 32);
4295 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4298 ext4_inode_csum_set(inode, raw_inode, ei);
4300 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4301 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4304 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4306 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4309 ext4_std_error(inode->i_sb, err);
4314 * ext4_write_inode()
4316 * We are called from a few places:
4318 * - Within generic_file_write() for O_SYNC files.
4319 * Here, there will be no transaction running. We wait for any running
4320 * transaction to commit.
4322 * - Within sys_sync(), kupdate and such.
4323 * We wait on commit, if tol to.
4325 * - Within prune_icache() (PF_MEMALLOC == true)
4326 * Here we simply return. We can't afford to block kswapd on the
4329 * In all cases it is actually safe for us to return without doing anything,
4330 * because the inode has been copied into a raw inode buffer in
4331 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4334 * Note that we are absolutely dependent upon all inode dirtiers doing the
4335 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4336 * which we are interested.
4338 * It would be a bug for them to not do this. The code:
4340 * mark_inode_dirty(inode)
4342 * inode->i_size = expr;
4344 * is in error because a kswapd-driven write_inode() could occur while
4345 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4346 * will no longer be on the superblock's dirty inode list.
4348 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4352 if (current->flags & PF_MEMALLOC)
4355 if (EXT4_SB(inode->i_sb)->s_journal) {
4356 if (ext4_journal_current_handle()) {
4357 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4362 if (wbc->sync_mode != WB_SYNC_ALL)
4365 err = ext4_force_commit(inode->i_sb);
4367 struct ext4_iloc iloc;
4369 err = __ext4_get_inode_loc(inode, &iloc, 0);
4372 if (wbc->sync_mode == WB_SYNC_ALL)
4373 sync_dirty_buffer(iloc.bh);
4374 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4375 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4376 "IO error syncing inode");
4385 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4386 * buffers that are attached to a page stradding i_size and are undergoing
4387 * commit. In that case we have to wait for commit to finish and try again.
4389 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4393 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4394 tid_t commit_tid = 0;
4397 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4399 * All buffers in the last page remain valid? Then there's nothing to
4400 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4403 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4406 page = find_lock_page(inode->i_mapping,
4407 inode->i_size >> PAGE_CACHE_SHIFT);
4410 ret = __ext4_journalled_invalidatepage(page, offset,
4411 PAGE_CACHE_SIZE - offset);
4413 page_cache_release(page);
4417 read_lock(&journal->j_state_lock);
4418 if (journal->j_committing_transaction)
4419 commit_tid = journal->j_committing_transaction->t_tid;
4420 read_unlock(&journal->j_state_lock);
4422 jbd2_log_wait_commit(journal, commit_tid);
4429 * Called from notify_change.
4431 * We want to trap VFS attempts to truncate the file as soon as
4432 * possible. In particular, we want to make sure that when the VFS
4433 * shrinks i_size, we put the inode on the orphan list and modify
4434 * i_disksize immediately, so that during the subsequent flushing of
4435 * dirty pages and freeing of disk blocks, we can guarantee that any
4436 * commit will leave the blocks being flushed in an unused state on
4437 * disk. (On recovery, the inode will get truncated and the blocks will
4438 * be freed, so we have a strong guarantee that no future commit will
4439 * leave these blocks visible to the user.)
4441 * Another thing we have to assure is that if we are in ordered mode
4442 * and inode is still attached to the committing transaction, we must
4443 * we start writeout of all the dirty pages which are being truncated.
4444 * This way we are sure that all the data written in the previous
4445 * transaction are already on disk (truncate waits for pages under
4448 * Called with inode->i_mutex down.
4450 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4452 struct inode *inode = dentry->d_inode;
4455 const unsigned int ia_valid = attr->ia_valid;
4457 error = inode_change_ok(inode, attr);
4461 if (is_quota_modification(inode, attr))
4462 dquot_initialize(inode);
4463 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4464 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4467 /* (user+group)*(old+new) structure, inode write (sb,
4468 * inode block, ? - but truncate inode update has it) */
4469 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4470 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4471 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4472 if (IS_ERR(handle)) {
4473 error = PTR_ERR(handle);
4476 error = dquot_transfer(inode, attr);
4478 ext4_journal_stop(handle);
4481 /* Update corresponding info in inode so that everything is in
4482 * one transaction */
4483 if (attr->ia_valid & ATTR_UID)
4484 inode->i_uid = attr->ia_uid;
4485 if (attr->ia_valid & ATTR_GID)
4486 inode->i_gid = attr->ia_gid;
4487 error = ext4_mark_inode_dirty(handle, inode);
4488 ext4_journal_stop(handle);
4491 if (attr->ia_valid & ATTR_SIZE) {
4493 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4494 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4496 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4501 if (S_ISREG(inode->i_mode) &&
4502 attr->ia_valid & ATTR_SIZE &&
4503 (attr->ia_size < inode->i_size)) {
4506 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4507 if (IS_ERR(handle)) {
4508 error = PTR_ERR(handle);
4511 if (ext4_handle_valid(handle)) {
4512 error = ext4_orphan_add(handle, inode);
4515 EXT4_I(inode)->i_disksize = attr->ia_size;
4516 rc = ext4_mark_inode_dirty(handle, inode);
4519 ext4_journal_stop(handle);
4521 if (ext4_should_order_data(inode)) {
4522 error = ext4_begin_ordered_truncate(inode,
4525 /* Do as much error cleanup as possible */
4526 handle = ext4_journal_start(inode,
4528 if (IS_ERR(handle)) {
4529 ext4_orphan_del(NULL, inode);
4532 ext4_orphan_del(handle, inode);
4534 ext4_journal_stop(handle);
4540 if (attr->ia_valid & ATTR_SIZE) {
4541 if (attr->ia_size != inode->i_size) {
4542 loff_t oldsize = inode->i_size;
4544 i_size_write(inode, attr->ia_size);
4546 * Blocks are going to be removed from the inode. Wait
4547 * for dio in flight. Temporarily disable
4548 * dioread_nolock to prevent livelock.
4551 if (!ext4_should_journal_data(inode)) {
4552 ext4_inode_block_unlocked_dio(inode);
4553 inode_dio_wait(inode);
4554 ext4_inode_resume_unlocked_dio(inode);
4556 ext4_wait_for_tail_page_commit(inode);
4559 * Truncate pagecache after we've waited for commit
4560 * in data=journal mode to make pages freeable.
4562 truncate_pagecache(inode, oldsize, inode->i_size);
4564 ext4_truncate(inode);
4568 setattr_copy(inode, attr);
4569 mark_inode_dirty(inode);
4573 * If the call to ext4_truncate failed to get a transaction handle at
4574 * all, we need to clean up the in-core orphan list manually.
4576 if (orphan && inode->i_nlink)
4577 ext4_orphan_del(NULL, inode);
4579 if (!rc && (ia_valid & ATTR_MODE))
4580 rc = ext4_acl_chmod(inode);
4583 ext4_std_error(inode->i_sb, error);
4589 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4592 struct inode *inode;
4593 unsigned long long delalloc_blocks;
4595 inode = dentry->d_inode;
4596 generic_fillattr(inode, stat);
4599 * We can't update i_blocks if the block allocation is delayed
4600 * otherwise in the case of system crash before the real block
4601 * allocation is done, we will have i_blocks inconsistent with
4602 * on-disk file blocks.
4603 * We always keep i_blocks updated together with real
4604 * allocation. But to not confuse with user, stat
4605 * will return the blocks that include the delayed allocation
4606 * blocks for this file.
4608 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4609 EXT4_I(inode)->i_reserved_data_blocks);
4611 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4615 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4618 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4619 return ext4_ind_trans_blocks(inode, lblocks);
4620 return ext4_ext_index_trans_blocks(inode, pextents);
4624 * Account for index blocks, block groups bitmaps and block group
4625 * descriptor blocks if modify datablocks and index blocks
4626 * worse case, the indexs blocks spread over different block groups
4628 * If datablocks are discontiguous, they are possible to spread over
4629 * different block groups too. If they are contiguous, with flexbg,
4630 * they could still across block group boundary.
4632 * Also account for superblock, inode, quota and xattr blocks
4634 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4637 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4643 * How many index blocks need to touch to map @lblocks logical blocks
4644 * to @pextents physical extents?
4646 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
4651 * Now let's see how many group bitmaps and group descriptors need
4654 groups = idxblocks + pextents;
4656 if (groups > ngroups)
4658 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4659 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4661 /* bitmaps and block group descriptor blocks */
4662 ret += groups + gdpblocks;
4664 /* Blocks for super block, inode, quota and xattr blocks */
4665 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4671 * Calculate the total number of credits to reserve to fit
4672 * the modification of a single pages into a single transaction,
4673 * which may include multiple chunks of block allocations.
4675 * This could be called via ext4_write_begin()
4677 * We need to consider the worse case, when
4678 * one new block per extent.
4680 int ext4_writepage_trans_blocks(struct inode *inode)
4682 int bpp = ext4_journal_blocks_per_page(inode);
4685 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
4687 /* Account for data blocks for journalled mode */
4688 if (ext4_should_journal_data(inode))
4694 * Calculate the journal credits for a chunk of data modification.
4696 * This is called from DIO, fallocate or whoever calling
4697 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4699 * journal buffers for data blocks are not included here, as DIO
4700 * and fallocate do no need to journal data buffers.
4702 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4704 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4708 * The caller must have previously called ext4_reserve_inode_write().
4709 * Give this, we know that the caller already has write access to iloc->bh.
4711 int ext4_mark_iloc_dirty(handle_t *handle,
4712 struct inode *inode, struct ext4_iloc *iloc)
4716 if (IS_I_VERSION(inode))
4717 inode_inc_iversion(inode);
4719 /* the do_update_inode consumes one bh->b_count */
4722 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4723 err = ext4_do_update_inode(handle, inode, iloc);
4729 * On success, We end up with an outstanding reference count against
4730 * iloc->bh. This _must_ be cleaned up later.
4734 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4735 struct ext4_iloc *iloc)
4739 err = ext4_get_inode_loc(inode, iloc);
4741 BUFFER_TRACE(iloc->bh, "get_write_access");
4742 err = ext4_journal_get_write_access(handle, iloc->bh);
4748 ext4_std_error(inode->i_sb, err);
4753 * Expand an inode by new_extra_isize bytes.
4754 * Returns 0 on success or negative error number on failure.
4756 static int ext4_expand_extra_isize(struct inode *inode,
4757 unsigned int new_extra_isize,
4758 struct ext4_iloc iloc,
4761 struct ext4_inode *raw_inode;
4762 struct ext4_xattr_ibody_header *header;
4764 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4767 raw_inode = ext4_raw_inode(&iloc);
4769 header = IHDR(inode, raw_inode);
4771 /* No extended attributes present */
4772 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4773 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4774 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4776 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4780 /* try to expand with EAs present */
4781 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4786 * What we do here is to mark the in-core inode as clean with respect to inode
4787 * dirtiness (it may still be data-dirty).
4788 * This means that the in-core inode may be reaped by prune_icache
4789 * without having to perform any I/O. This is a very good thing,
4790 * because *any* task may call prune_icache - even ones which
4791 * have a transaction open against a different journal.
4793 * Is this cheating? Not really. Sure, we haven't written the
4794 * inode out, but prune_icache isn't a user-visible syncing function.
4795 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4796 * we start and wait on commits.
4798 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4800 struct ext4_iloc iloc;
4801 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4802 static unsigned int mnt_count;
4806 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4807 err = ext4_reserve_inode_write(handle, inode, &iloc);
4808 if (ext4_handle_valid(handle) &&
4809 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4810 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4812 * We need extra buffer credits since we may write into EA block
4813 * with this same handle. If journal_extend fails, then it will
4814 * only result in a minor loss of functionality for that inode.
4815 * If this is felt to be critical, then e2fsck should be run to
4816 * force a large enough s_min_extra_isize.
4818 if ((jbd2_journal_extend(handle,
4819 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4820 ret = ext4_expand_extra_isize(inode,
4821 sbi->s_want_extra_isize,
4824 ext4_set_inode_state(inode,
4825 EXT4_STATE_NO_EXPAND);
4827 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4828 ext4_warning(inode->i_sb,
4829 "Unable to expand inode %lu. Delete"
4830 " some EAs or run e2fsck.",
4833 le16_to_cpu(sbi->s_es->s_mnt_count);
4839 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4844 * ext4_dirty_inode() is called from __mark_inode_dirty()
4846 * We're really interested in the case where a file is being extended.
4847 * i_size has been changed by generic_commit_write() and we thus need
4848 * to include the updated inode in the current transaction.
4850 * Also, dquot_alloc_block() will always dirty the inode when blocks
4851 * are allocated to the file.
4853 * If the inode is marked synchronous, we don't honour that here - doing
4854 * so would cause a commit on atime updates, which we don't bother doing.
4855 * We handle synchronous inodes at the highest possible level.
4857 void ext4_dirty_inode(struct inode *inode, int flags)
4861 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4865 ext4_mark_inode_dirty(handle, inode);
4867 ext4_journal_stop(handle);
4874 * Bind an inode's backing buffer_head into this transaction, to prevent
4875 * it from being flushed to disk early. Unlike
4876 * ext4_reserve_inode_write, this leaves behind no bh reference and
4877 * returns no iloc structure, so the caller needs to repeat the iloc
4878 * lookup to mark the inode dirty later.
4880 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4882 struct ext4_iloc iloc;
4886 err = ext4_get_inode_loc(inode, &iloc);
4888 BUFFER_TRACE(iloc.bh, "get_write_access");
4889 err = jbd2_journal_get_write_access(handle, iloc.bh);
4891 err = ext4_handle_dirty_metadata(handle,
4897 ext4_std_error(inode->i_sb, err);
4902 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4909 * We have to be very careful here: changing a data block's
4910 * journaling status dynamically is dangerous. If we write a
4911 * data block to the journal, change the status and then delete
4912 * that block, we risk forgetting to revoke the old log record
4913 * from the journal and so a subsequent replay can corrupt data.
4914 * So, first we make sure that the journal is empty and that
4915 * nobody is changing anything.
4918 journal = EXT4_JOURNAL(inode);
4921 if (is_journal_aborted(journal))
4923 /* We have to allocate physical blocks for delalloc blocks
4924 * before flushing journal. otherwise delalloc blocks can not
4925 * be allocated any more. even more truncate on delalloc blocks
4926 * could trigger BUG by flushing delalloc blocks in journal.
4927 * There is no delalloc block in non-journal data mode.
4929 if (val && test_opt(inode->i_sb, DELALLOC)) {
4930 err = ext4_alloc_da_blocks(inode);
4935 /* Wait for all existing dio workers */
4936 ext4_inode_block_unlocked_dio(inode);
4937 inode_dio_wait(inode);
4939 jbd2_journal_lock_updates(journal);
4942 * OK, there are no updates running now, and all cached data is
4943 * synced to disk. We are now in a completely consistent state
4944 * which doesn't have anything in the journal, and we know that
4945 * no filesystem updates are running, so it is safe to modify
4946 * the inode's in-core data-journaling state flag now.
4950 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4952 jbd2_journal_flush(journal);
4953 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4955 ext4_set_aops(inode);
4957 jbd2_journal_unlock_updates(journal);
4958 ext4_inode_resume_unlocked_dio(inode);
4960 /* Finally we can mark the inode as dirty. */
4962 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
4964 return PTR_ERR(handle);
4966 err = ext4_mark_inode_dirty(handle, inode);
4967 ext4_handle_sync(handle);
4968 ext4_journal_stop(handle);
4969 ext4_std_error(inode->i_sb, err);
4974 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4976 return !buffer_mapped(bh);
4979 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4981 struct page *page = vmf->page;
4985 struct file *file = vma->vm_file;
4986 struct inode *inode = file_inode(file);
4987 struct address_space *mapping = inode->i_mapping;
4989 get_block_t *get_block;
4992 sb_start_pagefault(inode->i_sb);
4993 file_update_time(vma->vm_file);
4994 /* Delalloc case is easy... */
4995 if (test_opt(inode->i_sb, DELALLOC) &&
4996 !ext4_should_journal_data(inode) &&
4997 !ext4_nonda_switch(inode->i_sb)) {
4999 ret = __block_page_mkwrite(vma, vmf,
5000 ext4_da_get_block_prep);
5001 } while (ret == -ENOSPC &&
5002 ext4_should_retry_alloc(inode->i_sb, &retries));
5007 size = i_size_read(inode);
5008 /* Page got truncated from under us? */
5009 if (page->mapping != mapping || page_offset(page) > size) {
5011 ret = VM_FAULT_NOPAGE;
5015 if (page->index == size >> PAGE_CACHE_SHIFT)
5016 len = size & ~PAGE_CACHE_MASK;
5018 len = PAGE_CACHE_SIZE;
5020 * Return if we have all the buffers mapped. This avoids the need to do
5021 * journal_start/journal_stop which can block and take a long time
5023 if (page_has_buffers(page)) {
5024 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5026 ext4_bh_unmapped)) {
5027 /* Wait so that we don't change page under IO */
5028 wait_for_stable_page(page);
5029 ret = VM_FAULT_LOCKED;
5034 /* OK, we need to fill the hole... */
5035 if (ext4_should_dioread_nolock(inode))
5036 get_block = ext4_get_block_write;
5038 get_block = ext4_get_block;
5040 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5041 ext4_writepage_trans_blocks(inode));
5042 if (IS_ERR(handle)) {
5043 ret = VM_FAULT_SIGBUS;
5046 ret = __block_page_mkwrite(vma, vmf, get_block);
5047 if (!ret && ext4_should_journal_data(inode)) {
5048 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5049 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5051 ret = VM_FAULT_SIGBUS;
5052 ext4_journal_stop(handle);
5055 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5057 ext4_journal_stop(handle);
5058 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5061 ret = block_page_mkwrite_return(ret);
5063 sb_end_pagefault(inode->i_sb);