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);
141 * Test whether an inode is a fast symlink.
143 static int ext4_inode_is_fast_symlink(struct inode *inode)
145 int ea_blocks = EXT4_I(inode)->i_file_acl ?
146 (inode->i_sb->s_blocksize >> 9) : 0;
148 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
152 * Restart the transaction associated with *handle. This does a commit,
153 * so before we call here everything must be consistently dirtied against
156 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
162 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
163 * moment, get_block can be called only for blocks inside i_size since
164 * page cache has been already dropped and writes are blocked by
165 * i_mutex. So we can safely drop the i_data_sem here.
167 BUG_ON(EXT4_JOURNAL(inode) == NULL);
168 jbd_debug(2, "restarting handle %p\n", handle);
169 up_write(&EXT4_I(inode)->i_data_sem);
170 ret = ext4_journal_restart(handle, nblocks);
171 down_write(&EXT4_I(inode)->i_data_sem);
172 ext4_discard_preallocations(inode);
178 * Called at the last iput() if i_nlink is zero.
180 void ext4_evict_inode(struct inode *inode)
185 trace_ext4_evict_inode(inode);
187 if (inode->i_nlink) {
189 * When journalling data dirty buffers are tracked only in the
190 * journal. So although mm thinks everything is clean and
191 * ready for reaping the inode might still have some pages to
192 * write in the running transaction or waiting to be
193 * checkpointed. Thus calling jbd2_journal_invalidatepage()
194 * (via truncate_inode_pages()) to discard these buffers can
195 * cause data loss. Also even if we did not discard these
196 * buffers, we would have no way to find them after the inode
197 * is reaped and thus user could see stale data if he tries to
198 * read them before the transaction is checkpointed. So be
199 * careful and force everything to disk here... We use
200 * ei->i_datasync_tid to store the newest transaction
201 * containing inode's data.
203 * Note that directories do not have this problem because they
204 * don't use page cache.
206 if (ext4_should_journal_data(inode) &&
207 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
208 inode->i_ino != EXT4_JOURNAL_INO) {
209 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
210 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
212 jbd2_complete_transaction(journal, commit_tid);
213 filemap_write_and_wait(&inode->i_data);
215 truncate_inode_pages(&inode->i_data, 0);
216 ext4_ioend_shutdown(inode);
220 if (!is_bad_inode(inode))
221 dquot_initialize(inode);
223 if (ext4_should_order_data(inode))
224 ext4_begin_ordered_truncate(inode, 0);
225 truncate_inode_pages(&inode->i_data, 0);
226 ext4_ioend_shutdown(inode);
228 if (is_bad_inode(inode))
232 * Protect us against freezing - iput() caller didn't have to have any
233 * protection against it
235 sb_start_intwrite(inode->i_sb);
236 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
237 ext4_blocks_for_truncate(inode)+3);
238 if (IS_ERR(handle)) {
239 ext4_std_error(inode->i_sb, PTR_ERR(handle));
241 * If we're going to skip the normal cleanup, we still need to
242 * make sure that the in-core orphan linked list is properly
245 ext4_orphan_del(NULL, inode);
246 sb_end_intwrite(inode->i_sb);
251 ext4_handle_sync(handle);
253 err = ext4_mark_inode_dirty(handle, inode);
255 ext4_warning(inode->i_sb,
256 "couldn't mark inode dirty (err %d)", err);
260 ext4_truncate(inode);
263 * ext4_ext_truncate() doesn't reserve any slop when it
264 * restarts journal transactions; therefore there may not be
265 * enough credits left in the handle to remove the inode from
266 * the orphan list and set the dtime field.
268 if (!ext4_handle_has_enough_credits(handle, 3)) {
269 err = ext4_journal_extend(handle, 3);
271 err = ext4_journal_restart(handle, 3);
273 ext4_warning(inode->i_sb,
274 "couldn't extend journal (err %d)", err);
276 ext4_journal_stop(handle);
277 ext4_orphan_del(NULL, inode);
278 sb_end_intwrite(inode->i_sb);
284 * Kill off the orphan record which ext4_truncate created.
285 * AKPM: I think this can be inside the above `if'.
286 * Note that ext4_orphan_del() has to be able to cope with the
287 * deletion of a non-existent orphan - this is because we don't
288 * know if ext4_truncate() actually created an orphan record.
289 * (Well, we could do this if we need to, but heck - it works)
291 ext4_orphan_del(handle, inode);
292 EXT4_I(inode)->i_dtime = get_seconds();
295 * One subtle ordering requirement: if anything has gone wrong
296 * (transaction abort, IO errors, whatever), then we can still
297 * do these next steps (the fs will already have been marked as
298 * having errors), but we can't free the inode if the mark_dirty
301 if (ext4_mark_inode_dirty(handle, inode))
302 /* If that failed, just do the required in-core inode clear. */
303 ext4_clear_inode(inode);
305 ext4_free_inode(handle, inode);
306 ext4_journal_stop(handle);
307 sb_end_intwrite(inode->i_sb);
310 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
314 qsize_t *ext4_get_reserved_space(struct inode *inode)
316 return &EXT4_I(inode)->i_reserved_quota;
321 * Calculate the number of metadata blocks need to reserve
322 * to allocate a block located at @lblock
324 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
326 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
327 return ext4_ext_calc_metadata_amount(inode, lblock);
329 return ext4_ind_calc_metadata_amount(inode, lblock);
333 * Called with i_data_sem down, which is important since we can call
334 * ext4_discard_preallocations() from here.
336 void ext4_da_update_reserve_space(struct inode *inode,
337 int used, int quota_claim)
339 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
340 struct ext4_inode_info *ei = EXT4_I(inode);
342 spin_lock(&ei->i_block_reservation_lock);
343 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
344 if (unlikely(used > ei->i_reserved_data_blocks)) {
345 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
346 "with only %d reserved data blocks",
347 __func__, inode->i_ino, used,
348 ei->i_reserved_data_blocks);
350 used = ei->i_reserved_data_blocks;
353 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
354 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
355 "with only %d reserved metadata blocks "
356 "(releasing %d blocks with reserved %d data blocks)",
357 inode->i_ino, ei->i_allocated_meta_blocks,
358 ei->i_reserved_meta_blocks, used,
359 ei->i_reserved_data_blocks);
361 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
364 /* Update per-inode reservations */
365 ei->i_reserved_data_blocks -= used;
366 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
367 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
368 used + ei->i_allocated_meta_blocks);
369 ei->i_allocated_meta_blocks = 0;
371 if (ei->i_reserved_data_blocks == 0) {
373 * We can release all of the reserved metadata blocks
374 * only when we have written all of the delayed
377 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
378 ei->i_reserved_meta_blocks);
379 ei->i_reserved_meta_blocks = 0;
380 ei->i_da_metadata_calc_len = 0;
382 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
384 /* Update quota subsystem for data blocks */
386 dquot_claim_block(inode, EXT4_C2B(sbi, used));
389 * We did fallocate with an offset that is already delayed
390 * allocated. So on delayed allocated writeback we should
391 * not re-claim the quota for fallocated blocks.
393 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
397 * If we have done all the pending block allocations and if
398 * there aren't any writers on the inode, we can discard the
399 * inode's preallocations.
401 if ((ei->i_reserved_data_blocks == 0) &&
402 (atomic_read(&inode->i_writecount) == 0))
403 ext4_discard_preallocations(inode);
406 static int __check_block_validity(struct inode *inode, const char *func,
408 struct ext4_map_blocks *map)
410 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
412 ext4_error_inode(inode, func, line, map->m_pblk,
413 "lblock %lu mapped to illegal pblock "
414 "(length %d)", (unsigned long) map->m_lblk,
421 #define check_block_validity(inode, map) \
422 __check_block_validity((inode), __func__, __LINE__, (map))
424 #ifdef ES_AGGRESSIVE_TEST
425 static void ext4_map_blocks_es_recheck(handle_t *handle,
427 struct ext4_map_blocks *es_map,
428 struct ext4_map_blocks *map,
435 * There is a race window that the result is not the same.
436 * e.g. xfstests #223 when dioread_nolock enables. The reason
437 * is that we lookup a block mapping in extent status tree with
438 * out taking i_data_sem. So at the time the unwritten extent
439 * could be converted.
441 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
442 down_read((&EXT4_I(inode)->i_data_sem));
443 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
444 retval = ext4_ext_map_blocks(handle, inode, map, flags &
445 EXT4_GET_BLOCKS_KEEP_SIZE);
447 retval = ext4_ind_map_blocks(handle, inode, map, flags &
448 EXT4_GET_BLOCKS_KEEP_SIZE);
450 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
451 up_read((&EXT4_I(inode)->i_data_sem));
453 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
454 * because it shouldn't be marked in es_map->m_flags.
456 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
459 * We don't check m_len because extent will be collpased in status
460 * tree. So the m_len might not equal.
462 if (es_map->m_lblk != map->m_lblk ||
463 es_map->m_flags != map->m_flags ||
464 es_map->m_pblk != map->m_pblk) {
465 printk("ES cache assertation failed for inode: %lu "
466 "es_cached ex [%d/%d/%llu/%x] != "
467 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
468 inode->i_ino, es_map->m_lblk, es_map->m_len,
469 es_map->m_pblk, es_map->m_flags, map->m_lblk,
470 map->m_len, map->m_pblk, map->m_flags,
474 #endif /* ES_AGGRESSIVE_TEST */
477 * The ext4_map_blocks() function tries to look up the requested blocks,
478 * and returns if the blocks are already mapped.
480 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
481 * and store the allocated blocks in the result buffer head and mark it
484 * If file type is extents based, it will call ext4_ext_map_blocks(),
485 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
488 * On success, it returns the number of blocks being mapped or allocate.
489 * if create==0 and the blocks are pre-allocated and uninitialized block,
490 * the result buffer head is unmapped. If the create ==1, it will make sure
491 * the buffer head is mapped.
493 * It returns 0 if plain look up failed (blocks have not been allocated), in
494 * that case, buffer head is unmapped
496 * It returns the error in case of allocation failure.
498 int ext4_map_blocks(handle_t *handle, struct inode *inode,
499 struct ext4_map_blocks *map, int flags)
501 struct extent_status es;
503 #ifdef ES_AGGRESSIVE_TEST
504 struct ext4_map_blocks orig_map;
506 memcpy(&orig_map, map, sizeof(*map));
510 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
511 "logical block %lu\n", inode->i_ino, flags, map->m_len,
512 (unsigned long) map->m_lblk);
514 /* Lookup extent status tree firstly */
515 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
516 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
517 map->m_pblk = ext4_es_pblock(&es) +
518 map->m_lblk - es.es_lblk;
519 map->m_flags |= ext4_es_is_written(&es) ?
520 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
521 retval = es.es_len - (map->m_lblk - es.es_lblk);
522 if (retval > map->m_len)
525 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
530 #ifdef ES_AGGRESSIVE_TEST
531 ext4_map_blocks_es_recheck(handle, inode, map,
538 * Try to see if we can get the block without requesting a new
541 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
542 down_read((&EXT4_I(inode)->i_data_sem));
543 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
544 retval = ext4_ext_map_blocks(handle, inode, map, flags &
545 EXT4_GET_BLOCKS_KEEP_SIZE);
547 retval = ext4_ind_map_blocks(handle, inode, map, flags &
548 EXT4_GET_BLOCKS_KEEP_SIZE);
552 unsigned long long status;
554 #ifdef ES_AGGRESSIVE_TEST
555 if (retval != map->m_len) {
556 printk("ES len assertation failed for inode: %lu "
557 "retval %d != map->m_len %d "
558 "in %s (lookup)\n", inode->i_ino, retval,
559 map->m_len, __func__);
563 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
564 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
565 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
566 ext4_find_delalloc_range(inode, map->m_lblk,
567 map->m_lblk + map->m_len - 1))
568 status |= EXTENT_STATUS_DELAYED;
569 ret = ext4_es_insert_extent(inode, map->m_lblk,
570 map->m_len, map->m_pblk, status);
574 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
575 up_read((&EXT4_I(inode)->i_data_sem));
578 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
579 int ret = check_block_validity(inode, map);
584 /* If it is only a block(s) look up */
585 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
589 * Returns if the blocks have already allocated
591 * Note that if blocks have been preallocated
592 * ext4_ext_get_block() returns the create = 0
593 * with buffer head unmapped.
595 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
599 * Here we clear m_flags because after allocating an new extent,
600 * it will be set again.
602 map->m_flags &= ~EXT4_MAP_FLAGS;
605 * New blocks allocate and/or writing to uninitialized extent
606 * will possibly result in updating i_data, so we take
607 * the write lock of i_data_sem, and call get_blocks()
608 * with create == 1 flag.
610 down_write((&EXT4_I(inode)->i_data_sem));
613 * if the caller is from delayed allocation writeout path
614 * we have already reserved fs blocks for allocation
615 * let the underlying get_block() function know to
616 * avoid double accounting
618 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
619 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
621 * We need to check for EXT4 here because migrate
622 * could have changed the inode type in between
624 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
625 retval = ext4_ext_map_blocks(handle, inode, map, flags);
627 retval = ext4_ind_map_blocks(handle, inode, map, flags);
629 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
631 * We allocated new blocks which will result in
632 * i_data's format changing. Force the migrate
633 * to fail by clearing migrate flags
635 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
639 * Update reserved blocks/metadata blocks after successful
640 * block allocation which had been deferred till now. We don't
641 * support fallocate for non extent files. So we can update
642 * reserve space here.
645 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
646 ext4_da_update_reserve_space(inode, retval, 1);
648 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
649 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
653 unsigned long long status;
655 #ifdef ES_AGGRESSIVE_TEST
656 if (retval != map->m_len) {
657 printk("ES len assertation failed for inode: %lu "
658 "retval %d != map->m_len %d "
659 "in %s (allocation)\n", inode->i_ino, retval,
660 map->m_len, __func__);
665 * If the extent has been zeroed out, we don't need to update
666 * extent status tree.
668 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
669 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
670 if (ext4_es_is_written(&es))
673 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
674 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
675 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
676 ext4_find_delalloc_range(inode, map->m_lblk,
677 map->m_lblk + map->m_len - 1))
678 status |= EXTENT_STATUS_DELAYED;
679 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
680 map->m_pblk, status);
686 up_write((&EXT4_I(inode)->i_data_sem));
687 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
688 int ret = check_block_validity(inode, map);
695 /* Maximum number of blocks we map for direct IO at once. */
696 #define DIO_MAX_BLOCKS 4096
698 static int _ext4_get_block(struct inode *inode, sector_t iblock,
699 struct buffer_head *bh, int flags)
701 handle_t *handle = ext4_journal_current_handle();
702 struct ext4_map_blocks map;
703 int ret = 0, started = 0;
706 if (ext4_has_inline_data(inode))
710 map.m_len = bh->b_size >> inode->i_blkbits;
712 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
713 /* Direct IO write... */
714 if (map.m_len > DIO_MAX_BLOCKS)
715 map.m_len = DIO_MAX_BLOCKS;
716 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
717 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
719 if (IS_ERR(handle)) {
720 ret = PTR_ERR(handle);
726 ret = ext4_map_blocks(handle, inode, &map, flags);
728 map_bh(bh, inode->i_sb, map.m_pblk);
729 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
730 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
734 ext4_journal_stop(handle);
738 int ext4_get_block(struct inode *inode, sector_t iblock,
739 struct buffer_head *bh, int create)
741 return _ext4_get_block(inode, iblock, bh,
742 create ? EXT4_GET_BLOCKS_CREATE : 0);
746 * `handle' can be NULL if create is zero
748 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
749 ext4_lblk_t block, int create, int *errp)
751 struct ext4_map_blocks map;
752 struct buffer_head *bh;
755 J_ASSERT(handle != NULL || create == 0);
759 err = ext4_map_blocks(handle, inode, &map,
760 create ? EXT4_GET_BLOCKS_CREATE : 0);
762 /* ensure we send some value back into *errp */
765 if (create && err == 0)
766 err = -ENOSPC; /* should never happen */
772 bh = sb_getblk(inode->i_sb, map.m_pblk);
777 if (map.m_flags & EXT4_MAP_NEW) {
778 J_ASSERT(create != 0);
779 J_ASSERT(handle != NULL);
782 * Now that we do not always journal data, we should
783 * keep in mind whether this should always journal the
784 * new buffer as metadata. For now, regular file
785 * writes use ext4_get_block instead, so it's not a
789 BUFFER_TRACE(bh, "call get_create_access");
790 fatal = ext4_journal_get_create_access(handle, bh);
791 if (!fatal && !buffer_uptodate(bh)) {
792 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
793 set_buffer_uptodate(bh);
796 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
797 err = ext4_handle_dirty_metadata(handle, inode, bh);
801 BUFFER_TRACE(bh, "not a new buffer");
811 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
812 ext4_lblk_t block, int create, int *err)
814 struct buffer_head *bh;
816 bh = ext4_getblk(handle, inode, block, create, err);
819 if (buffer_uptodate(bh))
821 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
823 if (buffer_uptodate(bh))
830 int ext4_walk_page_buffers(handle_t *handle,
831 struct buffer_head *head,
835 int (*fn)(handle_t *handle,
836 struct buffer_head *bh))
838 struct buffer_head *bh;
839 unsigned block_start, block_end;
840 unsigned blocksize = head->b_size;
842 struct buffer_head *next;
844 for (bh = head, block_start = 0;
845 ret == 0 && (bh != head || !block_start);
846 block_start = block_end, bh = next) {
847 next = bh->b_this_page;
848 block_end = block_start + blocksize;
849 if (block_end <= from || block_start >= to) {
850 if (partial && !buffer_uptodate(bh))
854 err = (*fn)(handle, bh);
862 * To preserve ordering, it is essential that the hole instantiation and
863 * the data write be encapsulated in a single transaction. We cannot
864 * close off a transaction and start a new one between the ext4_get_block()
865 * and the commit_write(). So doing the jbd2_journal_start at the start of
866 * prepare_write() is the right place.
868 * Also, this function can nest inside ext4_writepage(). In that case, we
869 * *know* that ext4_writepage() has generated enough buffer credits to do the
870 * whole page. So we won't block on the journal in that case, which is good,
871 * because the caller may be PF_MEMALLOC.
873 * By accident, ext4 can be reentered when a transaction is open via
874 * quota file writes. If we were to commit the transaction while thus
875 * reentered, there can be a deadlock - we would be holding a quota
876 * lock, and the commit would never complete if another thread had a
877 * transaction open and was blocking on the quota lock - a ranking
880 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
881 * will _not_ run commit under these circumstances because handle->h_ref
882 * is elevated. We'll still have enough credits for the tiny quotafile
885 int do_journal_get_write_access(handle_t *handle,
886 struct buffer_head *bh)
888 int dirty = buffer_dirty(bh);
891 if (!buffer_mapped(bh) || buffer_freed(bh))
894 * __block_write_begin() could have dirtied some buffers. Clean
895 * the dirty bit as jbd2_journal_get_write_access() could complain
896 * otherwise about fs integrity issues. Setting of the dirty bit
897 * by __block_write_begin() isn't a real problem here as we clear
898 * the bit before releasing a page lock and thus writeback cannot
899 * ever write the buffer.
902 clear_buffer_dirty(bh);
903 ret = ext4_journal_get_write_access(handle, bh);
905 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
909 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
910 struct buffer_head *bh_result, int create);
911 static int ext4_write_begin(struct file *file, struct address_space *mapping,
912 loff_t pos, unsigned len, unsigned flags,
913 struct page **pagep, void **fsdata)
915 struct inode *inode = mapping->host;
916 int ret, needed_blocks;
923 trace_ext4_write_begin(inode, pos, len, flags);
925 * Reserve one block more for addition to orphan list in case
926 * we allocate blocks but write fails for some reason
928 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
929 index = pos >> PAGE_CACHE_SHIFT;
930 from = pos & (PAGE_CACHE_SIZE - 1);
933 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
934 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
943 * grab_cache_page_write_begin() can take a long time if the
944 * system is thrashing due to memory pressure, or if the page
945 * is being written back. So grab it first before we start
946 * the transaction handle. This also allows us to allocate
947 * the page (if needed) without using GFP_NOFS.
950 page = grab_cache_page_write_begin(mapping, index, flags);
956 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
957 if (IS_ERR(handle)) {
958 page_cache_release(page);
959 return PTR_ERR(handle);
963 if (page->mapping != mapping) {
964 /* The page got truncated from under us */
966 page_cache_release(page);
967 ext4_journal_stop(handle);
970 wait_on_page_writeback(page);
972 if (ext4_should_dioread_nolock(inode))
973 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
975 ret = __block_write_begin(page, pos, len, ext4_get_block);
977 if (!ret && ext4_should_journal_data(inode)) {
978 ret = ext4_walk_page_buffers(handle, page_buffers(page),
980 do_journal_get_write_access);
986 * __block_write_begin may have instantiated a few blocks
987 * outside i_size. Trim these off again. Don't need
988 * i_size_read because we hold i_mutex.
990 * Add inode to orphan list in case we crash before
993 if (pos + len > inode->i_size && ext4_can_truncate(inode))
994 ext4_orphan_add(handle, inode);
996 ext4_journal_stop(handle);
997 if (pos + len > inode->i_size) {
998 ext4_truncate_failed_write(inode);
1000 * If truncate failed early the inode might
1001 * still be on the orphan list; we need to
1002 * make sure the inode is removed from the
1003 * orphan list in that case.
1006 ext4_orphan_del(NULL, inode);
1009 if (ret == -ENOSPC &&
1010 ext4_should_retry_alloc(inode->i_sb, &retries))
1012 page_cache_release(page);
1019 /* For write_end() in data=journal mode */
1020 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1023 if (!buffer_mapped(bh) || buffer_freed(bh))
1025 set_buffer_uptodate(bh);
1026 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1027 clear_buffer_meta(bh);
1028 clear_buffer_prio(bh);
1033 * We need to pick up the new inode size which generic_commit_write gave us
1034 * `file' can be NULL - eg, when called from page_symlink().
1036 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1037 * buffers are managed internally.
1039 static int ext4_write_end(struct file *file,
1040 struct address_space *mapping,
1041 loff_t pos, unsigned len, unsigned copied,
1042 struct page *page, void *fsdata)
1044 handle_t *handle = ext4_journal_current_handle();
1045 struct inode *inode = mapping->host;
1047 int i_size_changed = 0;
1049 trace_ext4_write_end(inode, pos, len, copied);
1050 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1051 ret = ext4_jbd2_file_inode(handle, inode);
1054 page_cache_release(page);
1059 if (ext4_has_inline_data(inode))
1060 copied = ext4_write_inline_data_end(inode, pos, len,
1063 copied = block_write_end(file, mapping, pos,
1064 len, copied, page, fsdata);
1067 * No need to use i_size_read() here, the i_size
1068 * cannot change under us because we hole i_mutex.
1070 * But it's important to update i_size while still holding page lock:
1071 * page writeout could otherwise come in and zero beyond i_size.
1073 if (pos + copied > inode->i_size) {
1074 i_size_write(inode, pos + copied);
1078 if (pos + copied > EXT4_I(inode)->i_disksize) {
1079 /* We need to mark inode dirty even if
1080 * new_i_size is less that inode->i_size
1081 * but greater than i_disksize. (hint delalloc)
1083 ext4_update_i_disksize(inode, (pos + copied));
1087 page_cache_release(page);
1090 * Don't mark the inode dirty under page lock. First, it unnecessarily
1091 * makes the holding time of page lock longer. Second, it forces lock
1092 * ordering of page lock and transaction start for journaling
1096 ext4_mark_inode_dirty(handle, inode);
1100 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1101 /* if we have allocated more blocks and copied
1102 * less. We will have blocks allocated outside
1103 * inode->i_size. So truncate them
1105 ext4_orphan_add(handle, inode);
1107 ret2 = ext4_journal_stop(handle);
1111 if (pos + len > inode->i_size) {
1112 ext4_truncate_failed_write(inode);
1114 * If truncate failed early the inode might still be
1115 * on the orphan list; we need to make sure the inode
1116 * is removed from the orphan list in that case.
1119 ext4_orphan_del(NULL, inode);
1122 return ret ? ret : copied;
1125 static int ext4_journalled_write_end(struct file *file,
1126 struct address_space *mapping,
1127 loff_t pos, unsigned len, unsigned copied,
1128 struct page *page, void *fsdata)
1130 handle_t *handle = ext4_journal_current_handle();
1131 struct inode *inode = mapping->host;
1137 trace_ext4_journalled_write_end(inode, pos, len, copied);
1138 from = pos & (PAGE_CACHE_SIZE - 1);
1141 BUG_ON(!ext4_handle_valid(handle));
1143 if (ext4_has_inline_data(inode))
1144 copied = ext4_write_inline_data_end(inode, pos, len,
1148 if (!PageUptodate(page))
1150 page_zero_new_buffers(page, from+copied, to);
1153 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1154 to, &partial, write_end_fn);
1156 SetPageUptodate(page);
1158 new_i_size = pos + copied;
1159 if (new_i_size > inode->i_size)
1160 i_size_write(inode, pos+copied);
1161 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1162 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1163 if (new_i_size > EXT4_I(inode)->i_disksize) {
1164 ext4_update_i_disksize(inode, new_i_size);
1165 ret2 = ext4_mark_inode_dirty(handle, inode);
1171 page_cache_release(page);
1172 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1173 /* if we have allocated more blocks and copied
1174 * less. We will have blocks allocated outside
1175 * inode->i_size. So truncate them
1177 ext4_orphan_add(handle, inode);
1179 ret2 = ext4_journal_stop(handle);
1182 if (pos + len > inode->i_size) {
1183 ext4_truncate_failed_write(inode);
1185 * If truncate failed early the inode might still be
1186 * on the orphan list; we need to make sure the inode
1187 * is removed from the orphan list in that case.
1190 ext4_orphan_del(NULL, inode);
1193 return ret ? ret : copied;
1197 * Reserve a metadata for a single block located at lblock
1199 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1202 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1203 struct ext4_inode_info *ei = EXT4_I(inode);
1204 unsigned int md_needed;
1205 ext4_lblk_t save_last_lblock;
1209 * recalculate the amount of metadata blocks to reserve
1210 * in order to allocate nrblocks
1211 * worse case is one extent per block
1214 spin_lock(&ei->i_block_reservation_lock);
1216 * ext4_calc_metadata_amount() has side effects, which we have
1217 * to be prepared undo if we fail to claim space.
1219 save_len = ei->i_da_metadata_calc_len;
1220 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1221 md_needed = EXT4_NUM_B2C(sbi,
1222 ext4_calc_metadata_amount(inode, lblock));
1223 trace_ext4_da_reserve_space(inode, md_needed);
1226 * We do still charge estimated metadata to the sb though;
1227 * we cannot afford to run out of free blocks.
1229 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1230 ei->i_da_metadata_calc_len = save_len;
1231 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1232 spin_unlock(&ei->i_block_reservation_lock);
1233 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1239 ei->i_reserved_meta_blocks += md_needed;
1240 spin_unlock(&ei->i_block_reservation_lock);
1242 return 0; /* success */
1246 * Reserve a single cluster located at lblock
1248 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1251 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1252 struct ext4_inode_info *ei = EXT4_I(inode);
1253 unsigned int md_needed;
1255 ext4_lblk_t save_last_lblock;
1259 * We will charge metadata quota at writeout time; this saves
1260 * us from metadata over-estimation, though we may go over by
1261 * a small amount in the end. Here we just reserve for data.
1263 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1268 * recalculate the amount of metadata blocks to reserve
1269 * in order to allocate nrblocks
1270 * worse case is one extent per block
1273 spin_lock(&ei->i_block_reservation_lock);
1275 * ext4_calc_metadata_amount() has side effects, which we have
1276 * to be prepared undo if we fail to claim space.
1278 save_len = ei->i_da_metadata_calc_len;
1279 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1280 md_needed = EXT4_NUM_B2C(sbi,
1281 ext4_calc_metadata_amount(inode, lblock));
1282 trace_ext4_da_reserve_space(inode, md_needed);
1285 * We do still charge estimated metadata to the sb though;
1286 * we cannot afford to run out of free blocks.
1288 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1289 ei->i_da_metadata_calc_len = save_len;
1290 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1291 spin_unlock(&ei->i_block_reservation_lock);
1292 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1296 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1299 ei->i_reserved_data_blocks++;
1300 ei->i_reserved_meta_blocks += md_needed;
1301 spin_unlock(&ei->i_block_reservation_lock);
1303 return 0; /* success */
1306 static void ext4_da_release_space(struct inode *inode, int to_free)
1308 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1309 struct ext4_inode_info *ei = EXT4_I(inode);
1312 return; /* Nothing to release, exit */
1314 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1316 trace_ext4_da_release_space(inode, to_free);
1317 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1319 * if there aren't enough reserved blocks, then the
1320 * counter is messed up somewhere. Since this
1321 * function is called from invalidate page, it's
1322 * harmless to return without any action.
1324 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1325 "ino %lu, to_free %d with only %d reserved "
1326 "data blocks", inode->i_ino, to_free,
1327 ei->i_reserved_data_blocks);
1329 to_free = ei->i_reserved_data_blocks;
1331 ei->i_reserved_data_blocks -= to_free;
1333 if (ei->i_reserved_data_blocks == 0) {
1335 * We can release all of the reserved metadata blocks
1336 * only when we have written all of the delayed
1337 * allocation blocks.
1338 * Note that in case of bigalloc, i_reserved_meta_blocks,
1339 * i_reserved_data_blocks, etc. refer to number of clusters.
1341 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1342 ei->i_reserved_meta_blocks);
1343 ei->i_reserved_meta_blocks = 0;
1344 ei->i_da_metadata_calc_len = 0;
1347 /* update fs dirty data blocks counter */
1348 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1350 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1352 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1355 static void ext4_da_page_release_reservation(struct page *page,
1356 unsigned int offset,
1357 unsigned int length)
1360 struct buffer_head *head, *bh;
1361 unsigned int curr_off = 0;
1362 struct inode *inode = page->mapping->host;
1363 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1364 unsigned int stop = offset + length;
1368 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1370 head = page_buffers(page);
1373 unsigned int next_off = curr_off + bh->b_size;
1375 if (next_off > stop)
1378 if ((offset <= curr_off) && (buffer_delay(bh))) {
1380 clear_buffer_delay(bh);
1382 curr_off = next_off;
1383 } while ((bh = bh->b_this_page) != head);
1386 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1387 ext4_es_remove_extent(inode, lblk, to_release);
1390 /* If we have released all the blocks belonging to a cluster, then we
1391 * need to release the reserved space for that cluster. */
1392 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1393 while (num_clusters > 0) {
1394 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1395 ((num_clusters - 1) << sbi->s_cluster_bits);
1396 if (sbi->s_cluster_ratio == 1 ||
1397 !ext4_find_delalloc_cluster(inode, lblk))
1398 ext4_da_release_space(inode, 1);
1405 * Delayed allocation stuff
1408 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd);
1411 * mpage_da_submit_io - walks through extent of pages and try to write
1412 * them with writepage() call back
1414 * @mpd->inode: inode
1415 * @mpd->first_page: first page of the extent
1416 * @mpd->next_page: page after the last page of the extent
1418 * By the time mpage_da_submit_io() is called we expect all blocks
1419 * to be allocated. this may be wrong if allocation failed.
1421 * As pages are already locked by write_cache_pages(), we can't use it
1423 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1424 struct ext4_map_blocks *map)
1426 struct pagevec pvec;
1427 unsigned long index, end;
1428 int ret = 0, err, nr_pages, i;
1429 struct inode *inode = mpd->inode;
1430 struct address_space *mapping = inode->i_mapping;
1431 loff_t size = i_size_read(inode);
1432 unsigned int len, block_start;
1433 struct buffer_head *bh, *page_bufs = NULL;
1434 sector_t pblock = 0, cur_logical = 0;
1435 struct ext4_io_submit io_submit;
1437 BUG_ON(mpd->next_page <= mpd->first_page);
1438 ext4_io_submit_init(&io_submit, mpd->wbc);
1439 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1440 if (!io_submit.io_end) {
1441 ext4_da_block_invalidatepages(mpd);
1445 * We need to start from the first_page to the next_page - 1
1446 * to make sure we also write the mapped dirty buffer_heads.
1447 * If we look at mpd->b_blocknr we would only be looking
1448 * at the currently mapped buffer_heads.
1450 index = mpd->first_page;
1451 end = mpd->next_page - 1;
1453 pagevec_init(&pvec, 0);
1454 while (index <= end) {
1455 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1458 for (i = 0; i < nr_pages; i++) {
1460 struct page *page = pvec.pages[i];
1462 index = page->index;
1466 if (index == size >> PAGE_CACHE_SHIFT)
1467 len = size & ~PAGE_CACHE_MASK;
1469 len = PAGE_CACHE_SIZE;
1471 cur_logical = index << (PAGE_CACHE_SHIFT -
1473 pblock = map->m_pblk + (cur_logical -
1478 BUG_ON(!PageLocked(page));
1479 BUG_ON(PageWriteback(page));
1481 bh = page_bufs = page_buffers(page);
1484 if (map && (cur_logical >= map->m_lblk) &&
1485 (cur_logical <= (map->m_lblk +
1486 (map->m_len - 1)))) {
1487 if (buffer_delay(bh)) {
1488 clear_buffer_delay(bh);
1489 bh->b_blocknr = pblock;
1491 if (buffer_unwritten(bh) ||
1493 BUG_ON(bh->b_blocknr != pblock);
1494 if (map->m_flags & EXT4_MAP_UNINIT)
1495 set_buffer_uninit(bh);
1496 clear_buffer_unwritten(bh);
1500 * skip page if block allocation undone and
1503 if (ext4_bh_delay_or_unwritten(NULL, bh))
1505 bh = bh->b_this_page;
1506 block_start += bh->b_size;
1509 } while (bh != page_bufs);
1516 clear_page_dirty_for_io(page);
1517 err = ext4_bio_write_page(&io_submit, page, len,
1520 mpd->pages_written++;
1522 * In error case, we have to continue because
1523 * remaining pages are still locked
1528 pagevec_release(&pvec);
1530 ext4_io_submit(&io_submit);
1531 /* Drop io_end reference we got from init */
1532 ext4_put_io_end_defer(io_submit.io_end);
1536 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1540 struct pagevec pvec;
1541 struct inode *inode = mpd->inode;
1542 struct address_space *mapping = inode->i_mapping;
1543 ext4_lblk_t start, last;
1545 index = mpd->first_page;
1546 end = mpd->next_page - 1;
1548 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1549 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1550 ext4_es_remove_extent(inode, start, last - start + 1);
1552 pagevec_init(&pvec, 0);
1553 while (index <= end) {
1554 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1557 for (i = 0; i < nr_pages; i++) {
1558 struct page *page = pvec.pages[i];
1559 if (page->index > end)
1561 BUG_ON(!PageLocked(page));
1562 BUG_ON(PageWriteback(page));
1563 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1564 ClearPageUptodate(page);
1567 index = pvec.pages[nr_pages - 1]->index + 1;
1568 pagevec_release(&pvec);
1573 static void ext4_print_free_blocks(struct inode *inode)
1575 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1576 struct super_block *sb = inode->i_sb;
1577 struct ext4_inode_info *ei = EXT4_I(inode);
1579 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1580 EXT4_C2B(EXT4_SB(inode->i_sb),
1581 ext4_count_free_clusters(sb)));
1582 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1583 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1584 (long long) EXT4_C2B(EXT4_SB(sb),
1585 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1586 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1587 (long long) EXT4_C2B(EXT4_SB(sb),
1588 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1589 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1590 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1591 ei->i_reserved_data_blocks);
1592 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1593 ei->i_reserved_meta_blocks);
1594 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1595 ei->i_allocated_meta_blocks);
1600 * mpage_da_map_and_submit - go through given space, map them
1601 * if necessary, and then submit them for I/O
1603 * @mpd - bh describing space
1605 * The function skips space we know is already mapped to disk blocks.
1608 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1610 int err, blks, get_blocks_flags;
1611 struct ext4_map_blocks map, *mapp = NULL;
1612 sector_t next = mpd->b_blocknr;
1613 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1614 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1615 handle_t *handle = NULL;
1618 * If the blocks are mapped already, or we couldn't accumulate
1619 * any blocks, then proceed immediately to the submission stage.
1621 if ((mpd->b_size == 0) ||
1622 ((mpd->b_state & (1 << BH_Mapped)) &&
1623 !(mpd->b_state & (1 << BH_Delay)) &&
1624 !(mpd->b_state & (1 << BH_Unwritten))))
1627 handle = ext4_journal_current_handle();
1631 * Call ext4_map_blocks() to allocate any delayed allocation
1632 * blocks, or to convert an uninitialized extent to be
1633 * initialized (in the case where we have written into
1634 * one or more preallocated blocks).
1636 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1637 * indicate that we are on the delayed allocation path. This
1638 * affects functions in many different parts of the allocation
1639 * call path. This flag exists primarily because we don't
1640 * want to change *many* call functions, so ext4_map_blocks()
1641 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1642 * inode's allocation semaphore is taken.
1644 * If the blocks in questions were delalloc blocks, set
1645 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1646 * variables are updated after the blocks have been allocated.
1649 map.m_len = max_blocks;
1651 * We're in delalloc path and it is possible that we're going to
1652 * need more metadata blocks than previously reserved. However
1653 * we must not fail because we're in writeback and there is
1654 * nothing we can do about it so it might result in data loss.
1655 * So use reserved blocks to allocate metadata if possible.
1657 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1658 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1659 if (ext4_should_dioread_nolock(mpd->inode))
1660 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1661 if (mpd->b_state & (1 << BH_Delay))
1662 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1665 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1667 struct super_block *sb = mpd->inode->i_sb;
1671 * If get block returns EAGAIN or ENOSPC and there
1672 * appears to be free blocks we will just let
1673 * mpage_da_submit_io() unlock all of the pages.
1678 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1684 * get block failure will cause us to loop in
1685 * writepages, because a_ops->writepage won't be able
1686 * to make progress. The page will be redirtied by
1687 * writepage and writepages will again try to write
1690 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1691 ext4_msg(sb, KERN_CRIT,
1692 "delayed block allocation failed for inode %lu "
1693 "at logical offset %llu with max blocks %zd "
1694 "with error %d", mpd->inode->i_ino,
1695 (unsigned long long) next,
1696 mpd->b_size >> mpd->inode->i_blkbits, err);
1697 ext4_msg(sb, KERN_CRIT,
1698 "This should not happen!! Data will be lost");
1700 ext4_print_free_blocks(mpd->inode);
1702 /* invalidate all the pages */
1703 ext4_da_block_invalidatepages(mpd);
1705 /* Mark this page range as having been completed */
1712 if (map.m_flags & EXT4_MAP_NEW) {
1713 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1716 for (i = 0; i < map.m_len; i++)
1717 unmap_underlying_metadata(bdev, map.m_pblk + i);
1721 * Update on-disk size along with block allocation.
1723 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1724 if (disksize > i_size_read(mpd->inode))
1725 disksize = i_size_read(mpd->inode);
1726 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1727 ext4_update_i_disksize(mpd->inode, disksize);
1728 err = ext4_mark_inode_dirty(handle, mpd->inode);
1730 ext4_error(mpd->inode->i_sb,
1731 "Failed to mark inode %lu dirty",
1736 mpage_da_submit_io(mpd, mapp);
1740 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1741 (1 << BH_Delay) | (1 << BH_Unwritten))
1744 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1746 * @mpd->lbh - extent of blocks
1747 * @logical - logical number of the block in the file
1748 * @b_state - b_state of the buffer head added
1750 * the function is used to collect contig. blocks in same state
1752 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1753 unsigned long b_state)
1756 int blkbits = mpd->inode->i_blkbits;
1757 int nrblocks = mpd->b_size >> blkbits;
1760 * XXX Don't go larger than mballoc is willing to allocate
1761 * This is a stopgap solution. We eventually need to fold
1762 * mpage_da_submit_io() into this function and then call
1763 * ext4_map_blocks() multiple times in a loop
1765 if (nrblocks >= (8*1024*1024 >> blkbits))
1768 /* check if the reserved journal credits might overflow */
1769 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1770 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1772 * With non-extent format we are limited by the journal
1773 * credit available. Total credit needed to insert
1774 * nrblocks contiguous blocks is dependent on the
1775 * nrblocks. So limit nrblocks.
1781 * First block in the extent
1783 if (mpd->b_size == 0) {
1784 mpd->b_blocknr = logical;
1785 mpd->b_size = 1 << blkbits;
1786 mpd->b_state = b_state & BH_FLAGS;
1790 next = mpd->b_blocknr + nrblocks;
1792 * Can we merge the block to our big extent?
1794 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1795 mpd->b_size += 1 << blkbits;
1801 * We couldn't merge the block to our extent, so we
1802 * need to flush current extent and start new one
1804 mpage_da_map_and_submit(mpd);
1808 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1810 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1814 * This function is grabs code from the very beginning of
1815 * ext4_map_blocks, but assumes that the caller is from delayed write
1816 * time. This function looks up the requested blocks and sets the
1817 * buffer delay bit under the protection of i_data_sem.
1819 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1820 struct ext4_map_blocks *map,
1821 struct buffer_head *bh)
1823 struct extent_status es;
1825 sector_t invalid_block = ~((sector_t) 0xffff);
1826 #ifdef ES_AGGRESSIVE_TEST
1827 struct ext4_map_blocks orig_map;
1829 memcpy(&orig_map, map, sizeof(*map));
1832 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1836 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1837 "logical block %lu\n", inode->i_ino, map->m_len,
1838 (unsigned long) map->m_lblk);
1840 /* Lookup extent status tree firstly */
1841 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1843 if (ext4_es_is_hole(&es)) {
1845 down_read((&EXT4_I(inode)->i_data_sem));
1850 * Delayed extent could be allocated by fallocate.
1851 * So we need to check it.
1853 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1854 map_bh(bh, inode->i_sb, invalid_block);
1856 set_buffer_delay(bh);
1860 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1861 retval = es.es_len - (iblock - es.es_lblk);
1862 if (retval > map->m_len)
1863 retval = map->m_len;
1864 map->m_len = retval;
1865 if (ext4_es_is_written(&es))
1866 map->m_flags |= EXT4_MAP_MAPPED;
1867 else if (ext4_es_is_unwritten(&es))
1868 map->m_flags |= EXT4_MAP_UNWRITTEN;
1872 #ifdef ES_AGGRESSIVE_TEST
1873 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1879 * Try to see if we can get the block without requesting a new
1880 * file system block.
1882 down_read((&EXT4_I(inode)->i_data_sem));
1883 if (ext4_has_inline_data(inode)) {
1885 * We will soon create blocks for this page, and let
1886 * us pretend as if the blocks aren't allocated yet.
1887 * In case of clusters, we have to handle the work
1888 * of mapping from cluster so that the reserved space
1889 * is calculated properly.
1891 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1892 ext4_find_delalloc_cluster(inode, map->m_lblk))
1893 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1895 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1896 retval = ext4_ext_map_blocks(NULL, inode, map,
1897 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1899 retval = ext4_ind_map_blocks(NULL, inode, map,
1900 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1906 * XXX: __block_prepare_write() unmaps passed block,
1910 * If the block was allocated from previously allocated cluster,
1911 * then we don't need to reserve it again. However we still need
1912 * to reserve metadata for every block we're going to write.
1914 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1915 ret = ext4_da_reserve_space(inode, iblock);
1917 /* not enough space to reserve */
1922 ret = ext4_da_reserve_metadata(inode, iblock);
1924 /* not enough space to reserve */
1930 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1931 ~0, EXTENT_STATUS_DELAYED);
1937 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1938 * and it should not appear on the bh->b_state.
1940 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1942 map_bh(bh, inode->i_sb, invalid_block);
1944 set_buffer_delay(bh);
1945 } else if (retval > 0) {
1947 unsigned long long status;
1949 #ifdef ES_AGGRESSIVE_TEST
1950 if (retval != map->m_len) {
1951 printk("ES len assertation failed for inode: %lu "
1952 "retval %d != map->m_len %d "
1953 "in %s (lookup)\n", inode->i_ino, retval,
1954 map->m_len, __func__);
1958 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1959 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1960 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1961 map->m_pblk, status);
1967 up_read((&EXT4_I(inode)->i_data_sem));
1973 * This is a special get_blocks_t callback which is used by
1974 * ext4_da_write_begin(). It will either return mapped block or
1975 * reserve space for a single block.
1977 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1978 * We also have b_blocknr = -1 and b_bdev initialized properly
1980 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1981 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1982 * initialized properly.
1984 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1985 struct buffer_head *bh, int create)
1987 struct ext4_map_blocks map;
1990 BUG_ON(create == 0);
1991 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1993 map.m_lblk = iblock;
1997 * first, we need to know whether the block is allocated already
1998 * preallocated blocks are unmapped but should treated
1999 * the same as allocated blocks.
2001 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2005 map_bh(bh, inode->i_sb, map.m_pblk);
2006 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2008 if (buffer_unwritten(bh)) {
2009 /* A delayed write to unwritten bh should be marked
2010 * new and mapped. Mapped ensures that we don't do
2011 * get_block multiple times when we write to the same
2012 * offset and new ensures that we do proper zero out
2013 * for partial write.
2016 set_buffer_mapped(bh);
2021 static int bget_one(handle_t *handle, struct buffer_head *bh)
2027 static int bput_one(handle_t *handle, struct buffer_head *bh)
2033 static int __ext4_journalled_writepage(struct page *page,
2036 struct address_space *mapping = page->mapping;
2037 struct inode *inode = mapping->host;
2038 struct buffer_head *page_bufs = NULL;
2039 handle_t *handle = NULL;
2040 int ret = 0, err = 0;
2041 int inline_data = ext4_has_inline_data(inode);
2042 struct buffer_head *inode_bh = NULL;
2044 ClearPageChecked(page);
2047 BUG_ON(page->index != 0);
2048 BUG_ON(len > ext4_get_max_inline_size(inode));
2049 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2050 if (inode_bh == NULL)
2053 page_bufs = page_buffers(page);
2058 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2061 /* As soon as we unlock the page, it can go away, but we have
2062 * references to buffers so we are safe */
2065 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2066 ext4_writepage_trans_blocks(inode));
2067 if (IS_ERR(handle)) {
2068 ret = PTR_ERR(handle);
2072 BUG_ON(!ext4_handle_valid(handle));
2075 ret = ext4_journal_get_write_access(handle, inode_bh);
2077 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2080 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2081 do_journal_get_write_access);
2083 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2088 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2089 err = ext4_journal_stop(handle);
2093 if (!ext4_has_inline_data(inode))
2094 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2096 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2103 * Note that we don't need to start a transaction unless we're journaling data
2104 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2105 * need to file the inode to the transaction's list in ordered mode because if
2106 * we are writing back data added by write(), the inode is already there and if
2107 * we are writing back data modified via mmap(), no one guarantees in which
2108 * transaction the data will hit the disk. In case we are journaling data, we
2109 * cannot start transaction directly because transaction start ranks above page
2110 * lock so we have to do some magic.
2112 * This function can get called via...
2113 * - ext4_da_writepages after taking page lock (have journal handle)
2114 * - journal_submit_inode_data_buffers (no journal handle)
2115 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2116 * - grab_page_cache when doing write_begin (have journal handle)
2118 * We don't do any block allocation in this function. If we have page with
2119 * multiple blocks we need to write those buffer_heads that are mapped. This
2120 * is important for mmaped based write. So if we do with blocksize 1K
2121 * truncate(f, 1024);
2122 * a = mmap(f, 0, 4096);
2124 * truncate(f, 4096);
2125 * we have in the page first buffer_head mapped via page_mkwrite call back
2126 * but other buffer_heads would be unmapped but dirty (dirty done via the
2127 * do_wp_page). So writepage should write the first block. If we modify
2128 * the mmap area beyond 1024 we will again get a page_fault and the
2129 * page_mkwrite callback will do the block allocation and mark the
2130 * buffer_heads mapped.
2132 * We redirty the page if we have any buffer_heads that is either delay or
2133 * unwritten in the page.
2135 * We can get recursively called as show below.
2137 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2140 * But since we don't do any block allocation we should not deadlock.
2141 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2143 static int ext4_writepage(struct page *page,
2144 struct writeback_control *wbc)
2149 struct buffer_head *page_bufs = NULL;
2150 struct inode *inode = page->mapping->host;
2151 struct ext4_io_submit io_submit;
2153 trace_ext4_writepage(page);
2154 size = i_size_read(inode);
2155 if (page->index == size >> PAGE_CACHE_SHIFT)
2156 len = size & ~PAGE_CACHE_MASK;
2158 len = PAGE_CACHE_SIZE;
2160 page_bufs = page_buffers(page);
2162 * We cannot do block allocation or other extent handling in this
2163 * function. If there are buffers needing that, we have to redirty
2164 * the page. But we may reach here when we do a journal commit via
2165 * journal_submit_inode_data_buffers() and in that case we must write
2166 * allocated buffers to achieve data=ordered mode guarantees.
2168 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2169 ext4_bh_delay_or_unwritten)) {
2170 redirty_page_for_writepage(wbc, page);
2171 if (current->flags & PF_MEMALLOC) {
2173 * For memory cleaning there's no point in writing only
2174 * some buffers. So just bail out. Warn if we came here
2175 * from direct reclaim.
2177 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2184 if (PageChecked(page) && ext4_should_journal_data(inode))
2186 * It's mmapped pagecache. Add buffers and journal it. There
2187 * doesn't seem much point in redirtying the page here.
2189 return __ext4_journalled_writepage(page, len);
2191 ext4_io_submit_init(&io_submit, wbc);
2192 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2193 if (!io_submit.io_end) {
2194 redirty_page_for_writepage(wbc, page);
2198 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2199 ext4_io_submit(&io_submit);
2200 /* Drop io_end reference we got from init */
2201 ext4_put_io_end_defer(io_submit.io_end);
2206 * This is called via ext4_da_writepages() to
2207 * calculate the total number of credits to reserve to fit
2208 * a single extent allocation into a single transaction,
2209 * ext4_da_writpeages() will loop calling this before
2210 * the block allocation.
2213 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2215 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2218 * With non-extent format the journal credit needed to
2219 * insert nrblocks contiguous block is dependent on
2220 * number of contiguous block. So we will limit
2221 * number of contiguous block to a sane value
2223 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2224 (max_blocks > EXT4_MAX_TRANS_DATA))
2225 max_blocks = EXT4_MAX_TRANS_DATA;
2227 return ext4_chunk_trans_blocks(inode, max_blocks);
2231 * write_cache_pages_da - walk the list of dirty pages of the given
2232 * address space and accumulate pages that need writing, and call
2233 * mpage_da_map_and_submit to map a single contiguous memory region
2234 * and then write them.
2236 static int write_cache_pages_da(handle_t *handle,
2237 struct address_space *mapping,
2238 struct writeback_control *wbc,
2239 struct mpage_da_data *mpd,
2240 pgoff_t *done_index)
2242 struct buffer_head *bh, *head;
2243 struct inode *inode = mapping->host;
2244 struct pagevec pvec;
2245 unsigned int nr_pages;
2248 long nr_to_write = wbc->nr_to_write;
2249 int i, tag, ret = 0;
2251 memset(mpd, 0, sizeof(struct mpage_da_data));
2254 pagevec_init(&pvec, 0);
2255 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2256 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2258 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2259 tag = PAGECACHE_TAG_TOWRITE;
2261 tag = PAGECACHE_TAG_DIRTY;
2263 *done_index = 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 *done_index = page->index + 1;
2286 * If we can't merge this page, and we have
2287 * accumulated an contiguous region, write it
2289 if ((mpd->next_page != page->index) &&
2290 (mpd->next_page != mpd->first_page)) {
2291 mpage_da_map_and_submit(mpd);
2292 goto ret_extent_tail;
2298 * If the page is no longer dirty, or its
2299 * mapping no longer corresponds to inode we
2300 * are writing (which means it has been
2301 * truncated or invalidated), or the page is
2302 * already under writeback and we are not
2303 * doing a data integrity writeback, skip the page
2305 if (!PageDirty(page) ||
2306 (PageWriteback(page) &&
2307 (wbc->sync_mode == WB_SYNC_NONE)) ||
2308 unlikely(page->mapping != mapping)) {
2313 wait_on_page_writeback(page);
2314 BUG_ON(PageWriteback(page));
2317 * If we have inline data and arrive here, it means that
2318 * we will soon create the block for the 1st page, so
2319 * we'd better clear the inline data here.
2321 if (ext4_has_inline_data(inode)) {
2322 BUG_ON(ext4_test_inode_state(inode,
2323 EXT4_STATE_MAY_INLINE_DATA));
2324 ext4_destroy_inline_data(handle, inode);
2327 if (mpd->next_page != page->index)
2328 mpd->first_page = page->index;
2329 mpd->next_page = page->index + 1;
2330 logical = (sector_t) page->index <<
2331 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2333 /* Add all dirty buffers to mpd */
2334 head = page_buffers(page);
2337 BUG_ON(buffer_locked(bh));
2339 * We need to try to allocate unmapped blocks
2340 * in the same page. Otherwise we won't make
2341 * progress with the page in ext4_writepage
2343 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2344 mpage_add_bh_to_extent(mpd, logical,
2347 goto ret_extent_tail;
2348 } else if (buffer_dirty(bh) &&
2349 buffer_mapped(bh)) {
2351 * mapped dirty buffer. We need to
2352 * update the b_state because we look
2353 * at b_state in mpage_da_map_blocks.
2354 * We don't update b_size because if we
2355 * find an unmapped buffer_head later
2356 * we need to use the b_state flag of
2359 if (mpd->b_size == 0)
2361 bh->b_state & BH_FLAGS;
2364 } while ((bh = bh->b_this_page) != head);
2366 if (nr_to_write > 0) {
2368 if (nr_to_write == 0 &&
2369 wbc->sync_mode == WB_SYNC_NONE)
2371 * We stop writing back only if we are
2372 * not doing integrity sync. In case of
2373 * integrity sync we have to keep going
2374 * because someone may be concurrently
2375 * dirtying pages, and we might have
2376 * synced a lot of newly appeared dirty
2377 * pages, but have not synced all of the
2383 pagevec_release(&pvec);
2388 ret = MPAGE_DA_EXTENT_TAIL;
2390 pagevec_release(&pvec);
2396 static int ext4_da_writepages(struct address_space *mapping,
2397 struct writeback_control *wbc)
2400 int range_whole = 0;
2401 handle_t *handle = NULL;
2402 struct mpage_da_data mpd;
2403 struct inode *inode = mapping->host;
2404 int pages_written = 0;
2405 int range_cyclic, cycled = 1, io_done = 0;
2406 int needed_blocks, ret = 0;
2407 loff_t range_start = wbc->range_start;
2408 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2409 pgoff_t done_index = 0;
2411 struct blk_plug plug;
2413 trace_ext4_da_writepages(inode, wbc);
2416 * No pages to write? This is mainly a kludge to avoid starting
2417 * a transaction for special inodes like journal inode on last iput()
2418 * because that could violate lock ordering on umount
2420 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2424 * If the filesystem has aborted, it is read-only, so return
2425 * right away instead of dumping stack traces later on that
2426 * will obscure the real source of the problem. We test
2427 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2428 * the latter could be true if the filesystem is mounted
2429 * read-only, and in that case, ext4_da_writepages should
2430 * *never* be called, so if that ever happens, we would want
2433 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2436 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2439 range_cyclic = wbc->range_cyclic;
2440 if (wbc->range_cyclic) {
2441 index = mapping->writeback_index;
2444 wbc->range_start = index << PAGE_CACHE_SHIFT;
2445 wbc->range_end = LLONG_MAX;
2446 wbc->range_cyclic = 0;
2449 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2450 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2454 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2455 tag_pages_for_writeback(mapping, index, end);
2457 blk_start_plug(&plug);
2458 while (!ret && wbc->nr_to_write > 0) {
2461 * we insert one extent at a time. So we need
2462 * credit needed for single extent allocation.
2463 * journalled mode is currently not supported
2466 BUG_ON(ext4_should_journal_data(inode));
2467 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2469 /* start a new transaction*/
2470 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2472 if (IS_ERR(handle)) {
2473 ret = PTR_ERR(handle);
2474 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2475 "%ld pages, ino %lu; err %d", __func__,
2476 wbc->nr_to_write, inode->i_ino, ret);
2477 blk_finish_plug(&plug);
2478 goto out_writepages;
2482 * Now call write_cache_pages_da() to find the next
2483 * contiguous region of logical blocks that need
2484 * blocks to be allocated by ext4 and submit them.
2486 ret = write_cache_pages_da(handle, mapping,
2487 wbc, &mpd, &done_index);
2489 * If we have a contiguous extent of pages and we
2490 * haven't done the I/O yet, map the blocks and submit
2493 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2494 mpage_da_map_and_submit(&mpd);
2495 ret = MPAGE_DA_EXTENT_TAIL;
2497 trace_ext4_da_write_pages(inode, &mpd);
2498 wbc->nr_to_write -= mpd.pages_written;
2500 ext4_journal_stop(handle);
2502 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2503 /* commit the transaction which would
2504 * free blocks released in the transaction
2507 jbd2_journal_force_commit_nested(sbi->s_journal);
2509 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2511 * Got one extent now try with rest of the pages.
2512 * If mpd.retval is set -EIO, journal is aborted.
2513 * So we don't need to write any more.
2515 pages_written += mpd.pages_written;
2518 } else if (wbc->nr_to_write)
2520 * There is no more writeout needed
2521 * or we requested for a noblocking writeout
2522 * and we found the device congested
2526 blk_finish_plug(&plug);
2527 if (!io_done && !cycled) {
2530 wbc->range_start = index << PAGE_CACHE_SHIFT;
2531 wbc->range_end = mapping->writeback_index - 1;
2536 wbc->range_cyclic = range_cyclic;
2537 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2539 * set the writeback_index so that range_cyclic
2540 * mode will write it back later
2542 mapping->writeback_index = done_index;
2545 wbc->range_start = range_start;
2546 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2550 static int ext4_nonda_switch(struct super_block *sb)
2552 s64 free_clusters, dirty_clusters;
2553 struct ext4_sb_info *sbi = EXT4_SB(sb);
2556 * switch to non delalloc mode if we are running low
2557 * on free block. The free block accounting via percpu
2558 * counters can get slightly wrong with percpu_counter_batch getting
2559 * accumulated on each CPU without updating global counters
2560 * Delalloc need an accurate free block accounting. So switch
2561 * to non delalloc when we are near to error range.
2564 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2566 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2568 * Start pushing delalloc when 1/2 of free blocks are dirty.
2570 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2571 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2573 if (2 * free_clusters < 3 * dirty_clusters ||
2574 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2576 * free block count is less than 150% of dirty blocks
2577 * or free blocks is less than watermark
2584 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2585 loff_t pos, unsigned len, unsigned flags,
2586 struct page **pagep, void **fsdata)
2588 int ret, retries = 0;
2591 struct inode *inode = mapping->host;
2594 index = pos >> PAGE_CACHE_SHIFT;
2596 if (ext4_nonda_switch(inode->i_sb)) {
2597 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2598 return ext4_write_begin(file, mapping, pos,
2599 len, flags, pagep, fsdata);
2601 *fsdata = (void *)0;
2602 trace_ext4_da_write_begin(inode, pos, len, flags);
2604 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2605 ret = ext4_da_write_inline_data_begin(mapping, inode,
2615 * grab_cache_page_write_begin() can take a long time if the
2616 * system is thrashing due to memory pressure, or if the page
2617 * is being written back. So grab it first before we start
2618 * the transaction handle. This also allows us to allocate
2619 * the page (if needed) without using GFP_NOFS.
2622 page = grab_cache_page_write_begin(mapping, index, flags);
2628 * With delayed allocation, we don't log the i_disksize update
2629 * if there is delayed block allocation. But we still need
2630 * to journalling the i_disksize update if writes to the end
2631 * of file which has an already mapped buffer.
2634 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2635 if (IS_ERR(handle)) {
2636 page_cache_release(page);
2637 return PTR_ERR(handle);
2641 if (page->mapping != mapping) {
2642 /* The page got truncated from under us */
2644 page_cache_release(page);
2645 ext4_journal_stop(handle);
2648 /* In case writeback began while the page was unlocked */
2649 wait_on_page_writeback(page);
2651 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2654 ext4_journal_stop(handle);
2656 * block_write_begin may have instantiated a few blocks
2657 * outside i_size. Trim these off again. Don't need
2658 * i_size_read because we hold i_mutex.
2660 if (pos + len > inode->i_size)
2661 ext4_truncate_failed_write(inode);
2663 if (ret == -ENOSPC &&
2664 ext4_should_retry_alloc(inode->i_sb, &retries))
2667 page_cache_release(page);
2676 * Check if we should update i_disksize
2677 * when write to the end of file but not require block allocation
2679 static int ext4_da_should_update_i_disksize(struct page *page,
2680 unsigned long offset)
2682 struct buffer_head *bh;
2683 struct inode *inode = page->mapping->host;
2687 bh = page_buffers(page);
2688 idx = offset >> inode->i_blkbits;
2690 for (i = 0; i < idx; i++)
2691 bh = bh->b_this_page;
2693 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2698 static int ext4_da_write_end(struct file *file,
2699 struct address_space *mapping,
2700 loff_t pos, unsigned len, unsigned copied,
2701 struct page *page, void *fsdata)
2703 struct inode *inode = mapping->host;
2705 handle_t *handle = ext4_journal_current_handle();
2707 unsigned long start, end;
2708 int write_mode = (int)(unsigned long)fsdata;
2710 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2711 return ext4_write_end(file, mapping, pos,
2712 len, copied, page, fsdata);
2714 trace_ext4_da_write_end(inode, pos, len, copied);
2715 start = pos & (PAGE_CACHE_SIZE - 1);
2716 end = start + copied - 1;
2719 * generic_write_end() will run mark_inode_dirty() if i_size
2720 * changes. So let's piggyback the i_disksize mark_inode_dirty
2723 new_i_size = pos + copied;
2724 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2725 if (ext4_has_inline_data(inode) ||
2726 ext4_da_should_update_i_disksize(page, end)) {
2727 down_write(&EXT4_I(inode)->i_data_sem);
2728 if (new_i_size > EXT4_I(inode)->i_disksize)
2729 EXT4_I(inode)->i_disksize = new_i_size;
2730 up_write(&EXT4_I(inode)->i_data_sem);
2731 /* We need to mark inode dirty even if
2732 * new_i_size is less that inode->i_size
2733 * bu greater than i_disksize.(hint delalloc)
2735 ext4_mark_inode_dirty(handle, inode);
2739 if (write_mode != CONVERT_INLINE_DATA &&
2740 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2741 ext4_has_inline_data(inode))
2742 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2745 ret2 = generic_write_end(file, mapping, pos, len, copied,
2751 ret2 = ext4_journal_stop(handle);
2755 return ret ? ret : copied;
2758 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2759 unsigned int length)
2762 * Drop reserved blocks
2764 BUG_ON(!PageLocked(page));
2765 if (!page_has_buffers(page))
2768 ext4_da_page_release_reservation(page, offset, length);
2771 ext4_invalidatepage(page, offset, length);
2777 * Force all delayed allocation blocks to be allocated for a given inode.
2779 int ext4_alloc_da_blocks(struct inode *inode)
2781 trace_ext4_alloc_da_blocks(inode);
2783 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2784 !EXT4_I(inode)->i_reserved_meta_blocks)
2788 * We do something simple for now. The filemap_flush() will
2789 * also start triggering a write of the data blocks, which is
2790 * not strictly speaking necessary (and for users of
2791 * laptop_mode, not even desirable). However, to do otherwise
2792 * would require replicating code paths in:
2794 * ext4_da_writepages() ->
2795 * write_cache_pages() ---> (via passed in callback function)
2796 * __mpage_da_writepage() -->
2797 * mpage_add_bh_to_extent()
2798 * mpage_da_map_blocks()
2800 * The problem is that write_cache_pages(), located in
2801 * mm/page-writeback.c, marks pages clean in preparation for
2802 * doing I/O, which is not desirable if we're not planning on
2805 * We could call write_cache_pages(), and then redirty all of
2806 * the pages by calling redirty_page_for_writepage() but that
2807 * would be ugly in the extreme. So instead we would need to
2808 * replicate parts of the code in the above functions,
2809 * simplifying them because we wouldn't actually intend to
2810 * write out the pages, but rather only collect contiguous
2811 * logical block extents, call the multi-block allocator, and
2812 * then update the buffer heads with the block allocations.
2814 * For now, though, we'll cheat by calling filemap_flush(),
2815 * which will map the blocks, and start the I/O, but not
2816 * actually wait for the I/O to complete.
2818 return filemap_flush(inode->i_mapping);
2822 * bmap() is special. It gets used by applications such as lilo and by
2823 * the swapper to find the on-disk block of a specific piece of data.
2825 * Naturally, this is dangerous if the block concerned is still in the
2826 * journal. If somebody makes a swapfile on an ext4 data-journaling
2827 * filesystem and enables swap, then they may get a nasty shock when the
2828 * data getting swapped to that swapfile suddenly gets overwritten by
2829 * the original zero's written out previously to the journal and
2830 * awaiting writeback in the kernel's buffer cache.
2832 * So, if we see any bmap calls here on a modified, data-journaled file,
2833 * take extra steps to flush any blocks which might be in the cache.
2835 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2837 struct inode *inode = mapping->host;
2842 * We can get here for an inline file via the FIBMAP ioctl
2844 if (ext4_has_inline_data(inode))
2847 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2848 test_opt(inode->i_sb, DELALLOC)) {
2850 * With delalloc we want to sync the file
2851 * so that we can make sure we allocate
2854 filemap_write_and_wait(mapping);
2857 if (EXT4_JOURNAL(inode) &&
2858 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2860 * This is a REALLY heavyweight approach, but the use of
2861 * bmap on dirty files is expected to be extremely rare:
2862 * only if we run lilo or swapon on a freshly made file
2863 * do we expect this to happen.
2865 * (bmap requires CAP_SYS_RAWIO so this does not
2866 * represent an unprivileged user DOS attack --- we'd be
2867 * in trouble if mortal users could trigger this path at
2870 * NB. EXT4_STATE_JDATA is not set on files other than
2871 * regular files. If somebody wants to bmap a directory
2872 * or symlink and gets confused because the buffer
2873 * hasn't yet been flushed to disk, they deserve
2874 * everything they get.
2877 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2878 journal = EXT4_JOURNAL(inode);
2879 jbd2_journal_lock_updates(journal);
2880 err = jbd2_journal_flush(journal);
2881 jbd2_journal_unlock_updates(journal);
2887 return generic_block_bmap(mapping, block, ext4_get_block);
2890 static int ext4_readpage(struct file *file, struct page *page)
2893 struct inode *inode = page->mapping->host;
2895 trace_ext4_readpage(page);
2897 if (ext4_has_inline_data(inode))
2898 ret = ext4_readpage_inline(inode, page);
2901 return mpage_readpage(page, ext4_get_block);
2907 ext4_readpages(struct file *file, struct address_space *mapping,
2908 struct list_head *pages, unsigned nr_pages)
2910 struct inode *inode = mapping->host;
2912 /* If the file has inline data, no need to do readpages. */
2913 if (ext4_has_inline_data(inode))
2916 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2919 static void ext4_invalidatepage(struct page *page, unsigned int offset,
2920 unsigned int length)
2922 trace_ext4_invalidatepage(page, offset, length);
2924 /* No journalling happens on data buffers when this function is used */
2925 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2927 block_invalidatepage(page, offset, length);
2930 static int __ext4_journalled_invalidatepage(struct page *page,
2931 unsigned int offset,
2932 unsigned int length)
2934 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2936 trace_ext4_journalled_invalidatepage(page, offset, length);
2939 * If it's a full truncate we just forget about the pending dirtying
2941 if (offset == 0 && length == PAGE_CACHE_SIZE)
2942 ClearPageChecked(page);
2944 return jbd2_journal_invalidatepage(journal, page, offset, length);
2947 /* Wrapper for aops... */
2948 static void ext4_journalled_invalidatepage(struct page *page,
2949 unsigned int offset,
2950 unsigned int length)
2952 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
2955 static int ext4_releasepage(struct page *page, gfp_t wait)
2957 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2959 trace_ext4_releasepage(page);
2961 /* Page has dirty journalled data -> cannot release */
2962 if (PageChecked(page))
2965 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2967 return try_to_free_buffers(page);
2971 * ext4_get_block used when preparing for a DIO write or buffer write.
2972 * We allocate an uinitialized extent if blocks haven't been allocated.
2973 * The extent will be converted to initialized after the IO is complete.
2975 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2976 struct buffer_head *bh_result, int create)
2978 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2979 inode->i_ino, create);
2980 return _ext4_get_block(inode, iblock, bh_result,
2981 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2984 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2985 struct buffer_head *bh_result, int create)
2987 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2988 inode->i_ino, create);
2989 return _ext4_get_block(inode, iblock, bh_result,
2990 EXT4_GET_BLOCKS_NO_LOCK);
2993 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2994 ssize_t size, void *private, int ret,
2997 struct inode *inode = file_inode(iocb->ki_filp);
2998 ext4_io_end_t *io_end = iocb->private;
3000 /* if not async direct IO just return */
3002 inode_dio_done(inode);
3004 aio_complete(iocb, ret, 0);
3008 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3009 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3010 iocb->private, io_end->inode->i_ino, iocb, offset,
3013 iocb->private = NULL;
3014 io_end->offset = offset;
3015 io_end->size = size;
3017 io_end->iocb = iocb;
3018 io_end->result = ret;
3020 ext4_put_io_end_defer(io_end);
3024 * For ext4 extent files, ext4 will do direct-io write to holes,
3025 * preallocated extents, and those write extend the file, no need to
3026 * fall back to buffered IO.
3028 * For holes, we fallocate those blocks, mark them as uninitialized
3029 * If those blocks were preallocated, we mark sure they are split, but
3030 * still keep the range to write as uninitialized.
3032 * The unwritten extents will be converted to written when DIO is completed.
3033 * For async direct IO, since the IO may still pending when return, we
3034 * set up an end_io call back function, which will do the conversion
3035 * when async direct IO completed.
3037 * If the O_DIRECT write will extend the file then add this inode to the
3038 * orphan list. So recovery will truncate it back to the original size
3039 * if the machine crashes during the write.
3042 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3043 const struct iovec *iov, loff_t offset,
3044 unsigned long nr_segs)
3046 struct file *file = iocb->ki_filp;
3047 struct inode *inode = file->f_mapping->host;
3049 size_t count = iov_length(iov, nr_segs);
3051 get_block_t *get_block_func = NULL;
3053 loff_t final_size = offset + count;
3054 ext4_io_end_t *io_end = NULL;
3056 /* Use the old path for reads and writes beyond i_size. */
3057 if (rw != WRITE || final_size > inode->i_size)
3058 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3060 BUG_ON(iocb->private == NULL);
3062 /* If we do a overwrite dio, i_mutex locking can be released */
3063 overwrite = *((int *)iocb->private);
3066 atomic_inc(&inode->i_dio_count);
3067 down_read(&EXT4_I(inode)->i_data_sem);
3068 mutex_unlock(&inode->i_mutex);
3072 * We could direct write to holes and fallocate.
3074 * Allocated blocks to fill the hole are marked as
3075 * uninitialized to prevent parallel buffered read to expose
3076 * the stale data before DIO complete the data IO.
3078 * As to previously fallocated extents, ext4 get_block will
3079 * just simply mark the buffer mapped but still keep the
3080 * extents uninitialized.
3082 * For non AIO case, we will convert those unwritten extents
3083 * to written after return back from blockdev_direct_IO.
3085 * For async DIO, the conversion needs to be deferred when the
3086 * IO is completed. The ext4 end_io callback function will be
3087 * called to take care of the conversion work. Here for async
3088 * case, we allocate an io_end structure to hook to the iocb.
3090 iocb->private = NULL;
3091 ext4_inode_aio_set(inode, NULL);
3092 if (!is_sync_kiocb(iocb)) {
3093 io_end = ext4_init_io_end(inode, GFP_NOFS);
3098 io_end->flag |= EXT4_IO_END_DIRECT;
3100 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3102 iocb->private = ext4_get_io_end(io_end);
3104 * we save the io structure for current async direct
3105 * IO, so that later ext4_map_blocks() could flag the
3106 * io structure whether there is a unwritten extents
3107 * needs to be converted when IO is completed.
3109 ext4_inode_aio_set(inode, io_end);
3113 get_block_func = ext4_get_block_write_nolock;
3115 get_block_func = ext4_get_block_write;
3116 dio_flags = DIO_LOCKING;
3118 ret = __blockdev_direct_IO(rw, iocb, inode,
3119 inode->i_sb->s_bdev, iov,
3127 * Put our reference to io_end. This can free the io_end structure e.g.
3128 * in sync IO case or in case of error. It can even perform extent
3129 * conversion if all bios we submitted finished before we got here.
3130 * Note that in that case iocb->private can be already set to NULL
3134 ext4_inode_aio_set(inode, NULL);
3135 ext4_put_io_end(io_end);
3137 * When no IO was submitted ext4_end_io_dio() was not
3138 * called so we have to put iocb's reference.
3140 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3141 WARN_ON(iocb->private != io_end);
3142 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
3143 WARN_ON(io_end->iocb);
3145 * Generic code already did inode_dio_done() so we
3146 * have to clear EXT4_IO_END_DIRECT to not do it for
3150 ext4_put_io_end(io_end);
3151 iocb->private = NULL;
3154 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3155 EXT4_STATE_DIO_UNWRITTEN)) {
3158 * for non AIO case, since the IO is already
3159 * completed, we could do the conversion right here
3161 err = ext4_convert_unwritten_extents(inode,
3165 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3169 /* take i_mutex locking again if we do a ovewrite dio */
3171 inode_dio_done(inode);
3172 up_read(&EXT4_I(inode)->i_data_sem);
3173 mutex_lock(&inode->i_mutex);
3179 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3180 const struct iovec *iov, loff_t offset,
3181 unsigned long nr_segs)
3183 struct file *file = iocb->ki_filp;
3184 struct inode *inode = file->f_mapping->host;
3188 * If we are doing data journalling we don't support O_DIRECT
3190 if (ext4_should_journal_data(inode))
3193 /* Let buffer I/O handle the inline data case. */
3194 if (ext4_has_inline_data(inode))
3197 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3198 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3199 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3201 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3202 trace_ext4_direct_IO_exit(inode, offset,
3203 iov_length(iov, nr_segs), rw, ret);
3208 * Pages can be marked dirty completely asynchronously from ext4's journalling
3209 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3210 * much here because ->set_page_dirty is called under VFS locks. The page is
3211 * not necessarily locked.
3213 * We cannot just dirty the page and leave attached buffers clean, because the
3214 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3215 * or jbddirty because all the journalling code will explode.
3217 * So what we do is to mark the page "pending dirty" and next time writepage
3218 * is called, propagate that into the buffers appropriately.
3220 static int ext4_journalled_set_page_dirty(struct page *page)
3222 SetPageChecked(page);
3223 return __set_page_dirty_nobuffers(page);
3226 static const struct address_space_operations ext4_aops = {
3227 .readpage = ext4_readpage,
3228 .readpages = ext4_readpages,
3229 .writepage = ext4_writepage,
3230 .write_begin = ext4_write_begin,
3231 .write_end = ext4_write_end,
3233 .invalidatepage = ext4_invalidatepage,
3234 .releasepage = ext4_releasepage,
3235 .direct_IO = ext4_direct_IO,
3236 .migratepage = buffer_migrate_page,
3237 .is_partially_uptodate = block_is_partially_uptodate,
3238 .error_remove_page = generic_error_remove_page,
3241 static const struct address_space_operations ext4_journalled_aops = {
3242 .readpage = ext4_readpage,
3243 .readpages = ext4_readpages,
3244 .writepage = ext4_writepage,
3245 .write_begin = ext4_write_begin,
3246 .write_end = ext4_journalled_write_end,
3247 .set_page_dirty = ext4_journalled_set_page_dirty,
3249 .invalidatepage = ext4_journalled_invalidatepage,
3250 .releasepage = ext4_releasepage,
3251 .direct_IO = ext4_direct_IO,
3252 .is_partially_uptodate = block_is_partially_uptodate,
3253 .error_remove_page = generic_error_remove_page,
3256 static const struct address_space_operations ext4_da_aops = {
3257 .readpage = ext4_readpage,
3258 .readpages = ext4_readpages,
3259 .writepage = ext4_writepage,
3260 .writepages = ext4_da_writepages,
3261 .write_begin = ext4_da_write_begin,
3262 .write_end = ext4_da_write_end,
3264 .invalidatepage = ext4_da_invalidatepage,
3265 .releasepage = ext4_releasepage,
3266 .direct_IO = ext4_direct_IO,
3267 .migratepage = buffer_migrate_page,
3268 .is_partially_uptodate = block_is_partially_uptodate,
3269 .error_remove_page = generic_error_remove_page,
3272 void ext4_set_aops(struct inode *inode)
3274 switch (ext4_inode_journal_mode(inode)) {
3275 case EXT4_INODE_ORDERED_DATA_MODE:
3276 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3278 case EXT4_INODE_WRITEBACK_DATA_MODE:
3279 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3281 case EXT4_INODE_JOURNAL_DATA_MODE:
3282 inode->i_mapping->a_ops = &ext4_journalled_aops;
3287 if (test_opt(inode->i_sb, DELALLOC))
3288 inode->i_mapping->a_ops = &ext4_da_aops;
3290 inode->i_mapping->a_ops = &ext4_aops;
3294 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3295 * up to the end of the block which corresponds to `from'.
3296 * This required during truncate. We need to physically zero the tail end
3297 * of that block so it doesn't yield old data if the file is later grown.
3299 int ext4_block_truncate_page(handle_t *handle,
3300 struct address_space *mapping, loff_t from)
3302 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3305 struct inode *inode = mapping->host;
3307 blocksize = inode->i_sb->s_blocksize;
3308 length = blocksize - (offset & (blocksize - 1));
3310 return ext4_block_zero_page_range(handle, mapping, from, length);
3314 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3315 * starting from file offset 'from'. The range to be zero'd must
3316 * be contained with in one block. If the specified range exceeds
3317 * the end of the block it will be shortened to end of the block
3318 * that cooresponds to 'from'
3320 int ext4_block_zero_page_range(handle_t *handle,
3321 struct address_space *mapping, loff_t from, loff_t length)
3323 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3324 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3325 unsigned blocksize, max, pos;
3327 struct inode *inode = mapping->host;
3328 struct buffer_head *bh;
3332 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3333 mapping_gfp_mask(mapping) & ~__GFP_FS);
3337 blocksize = inode->i_sb->s_blocksize;
3338 max = blocksize - (offset & (blocksize - 1));
3341 * correct length if it does not fall between
3342 * 'from' and the end of the block
3344 if (length > max || length < 0)
3347 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3349 if (!page_has_buffers(page))
3350 create_empty_buffers(page, blocksize, 0);
3352 /* Find the buffer that contains "offset" */
3353 bh = page_buffers(page);
3355 while (offset >= pos) {
3356 bh = bh->b_this_page;
3362 if (buffer_freed(bh)) {
3363 BUFFER_TRACE(bh, "freed: skip");
3367 if (!buffer_mapped(bh)) {
3368 BUFFER_TRACE(bh, "unmapped");
3369 ext4_get_block(inode, iblock, bh, 0);
3370 /* unmapped? It's a hole - nothing to do */
3371 if (!buffer_mapped(bh)) {
3372 BUFFER_TRACE(bh, "still unmapped");
3377 /* Ok, it's mapped. Make sure it's up-to-date */
3378 if (PageUptodate(page))
3379 set_buffer_uptodate(bh);
3381 if (!buffer_uptodate(bh)) {
3383 ll_rw_block(READ, 1, &bh);
3385 /* Uhhuh. Read error. Complain and punt. */
3386 if (!buffer_uptodate(bh))
3390 if (ext4_should_journal_data(inode)) {
3391 BUFFER_TRACE(bh, "get write access");
3392 err = ext4_journal_get_write_access(handle, bh);
3397 zero_user(page, offset, length);
3399 BUFFER_TRACE(bh, "zeroed end of block");
3402 if (ext4_should_journal_data(inode)) {
3403 err = ext4_handle_dirty_metadata(handle, inode, bh);
3405 mark_buffer_dirty(bh);
3406 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3407 err = ext4_jbd2_file_inode(handle, inode);
3412 page_cache_release(page);
3416 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3417 loff_t lstart, loff_t length)
3419 struct super_block *sb = inode->i_sb;
3420 struct address_space *mapping = inode->i_mapping;
3421 unsigned partial = lstart & (sb->s_blocksize - 1);
3422 ext4_fsblk_t start, end;
3423 loff_t byte_end = (lstart + length - 1);
3426 start = lstart >> sb->s_blocksize_bits;
3427 end = byte_end >> sb->s_blocksize_bits;
3429 /* Handle partial zero within the single block */
3431 err = ext4_block_zero_page_range(handle, mapping,
3435 /* Handle partial zero out on the start of the range */
3437 err = ext4_block_zero_page_range(handle, mapping,
3438 lstart, sb->s_blocksize);
3442 /* Handle partial zero out on the end of the range */
3443 partial = byte_end & (sb->s_blocksize - 1);
3444 if (partial != sb->s_blocksize - 1)
3445 err = ext4_block_zero_page_range(handle, mapping,
3451 int ext4_can_truncate(struct inode *inode)
3453 if (S_ISREG(inode->i_mode))
3455 if (S_ISDIR(inode->i_mode))
3457 if (S_ISLNK(inode->i_mode))
3458 return !ext4_inode_is_fast_symlink(inode);
3463 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3464 * associated with the given offset and length
3466 * @inode: File inode
3467 * @offset: The offset where the hole will begin
3468 * @len: The length of the hole
3470 * Returns: 0 on success or negative on failure
3473 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3475 struct inode *inode = file_inode(file);
3476 struct super_block *sb = inode->i_sb;
3477 ext4_lblk_t first_block, stop_block;
3478 struct address_space *mapping = inode->i_mapping;
3479 loff_t first_block_offset, last_block_offset;
3481 unsigned int credits;
3484 if (!S_ISREG(inode->i_mode))
3487 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3488 /* TODO: Add support for bigalloc file systems */
3492 trace_ext4_punch_hole(inode, offset, length);
3495 * Write out all dirty pages to avoid race conditions
3496 * Then release them.
3498 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3499 ret = filemap_write_and_wait_range(mapping, offset,
3500 offset + length - 1);
3505 mutex_lock(&inode->i_mutex);
3506 /* It's not possible punch hole on append only file */
3507 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3511 if (IS_SWAPFILE(inode)) {
3516 /* No need to punch hole beyond i_size */
3517 if (offset >= inode->i_size)
3521 * If the hole extends beyond i_size, set the hole
3522 * to end after the page that contains i_size
3524 if (offset + length > inode->i_size) {
3525 length = inode->i_size +
3526 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3530 first_block_offset = round_up(offset, sb->s_blocksize);
3531 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3533 /* Now release the pages and zero block aligned part of pages*/
3534 if (last_block_offset > first_block_offset)
3535 truncate_pagecache_range(inode, first_block_offset,
3538 /* Wait all existing dio workers, newcomers will block on i_mutex */
3539 ext4_inode_block_unlocked_dio(inode);
3540 ret = ext4_flush_unwritten_io(inode);
3543 inode_dio_wait(inode);
3545 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3546 credits = ext4_writepage_trans_blocks(inode);
3548 credits = ext4_blocks_for_truncate(inode);
3549 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3550 if (IS_ERR(handle)) {
3551 ret = PTR_ERR(handle);
3552 ext4_std_error(sb, ret);
3556 ret = ext4_zero_partial_blocks(handle, inode, offset,
3561 first_block = (offset + sb->s_blocksize - 1) >>
3562 EXT4_BLOCK_SIZE_BITS(sb);
3563 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3565 /* If there are no blocks to remove, return now */
3566 if (first_block >= stop_block)
3569 down_write(&EXT4_I(inode)->i_data_sem);
3570 ext4_discard_preallocations(inode);
3572 ret = ext4_es_remove_extent(inode, first_block,
3573 stop_block - first_block);
3575 up_write(&EXT4_I(inode)->i_data_sem);
3579 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3580 ret = ext4_ext_remove_space(inode, first_block,
3583 ret = ext4_free_hole_blocks(handle, inode, first_block,
3586 ext4_discard_preallocations(inode);
3587 up_write(&EXT4_I(inode)->i_data_sem);
3589 ext4_handle_sync(handle);
3590 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3591 ext4_mark_inode_dirty(handle, inode);
3593 ext4_journal_stop(handle);
3595 ext4_inode_resume_unlocked_dio(inode);
3597 mutex_unlock(&inode->i_mutex);
3604 * We block out ext4_get_block() block instantiations across the entire
3605 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3606 * simultaneously on behalf of the same inode.
3608 * As we work through the truncate and commit bits of it to the journal there
3609 * is one core, guiding principle: the file's tree must always be consistent on
3610 * disk. We must be able to restart the truncate after a crash.
3612 * The file's tree may be transiently inconsistent in memory (although it
3613 * probably isn't), but whenever we close off and commit a journal transaction,
3614 * the contents of (the filesystem + the journal) must be consistent and
3615 * restartable. It's pretty simple, really: bottom up, right to left (although
3616 * left-to-right works OK too).
3618 * Note that at recovery time, journal replay occurs *before* the restart of
3619 * truncate against the orphan inode list.
3621 * The committed inode has the new, desired i_size (which is the same as
3622 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3623 * that this inode's truncate did not complete and it will again call
3624 * ext4_truncate() to have another go. So there will be instantiated blocks
3625 * to the right of the truncation point in a crashed ext4 filesystem. But
3626 * that's fine - as long as they are linked from the inode, the post-crash
3627 * ext4_truncate() run will find them and release them.
3629 void ext4_truncate(struct inode *inode)
3631 struct ext4_inode_info *ei = EXT4_I(inode);
3632 unsigned int credits;
3634 struct address_space *mapping = inode->i_mapping;
3637 * There is a possibility that we're either freeing the inode
3638 * or it completely new indode. In those cases we might not
3639 * have i_mutex locked because it's not necessary.
3641 if (!(inode->i_state & (I_NEW|I_FREEING)))
3642 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3643 trace_ext4_truncate_enter(inode);
3645 if (!ext4_can_truncate(inode))
3648 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3650 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3651 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3653 if (ext4_has_inline_data(inode)) {
3656 ext4_inline_data_truncate(inode, &has_inline);
3662 * finish any pending end_io work so we won't run the risk of
3663 * converting any truncated blocks to initialized later
3665 ext4_flush_unwritten_io(inode);
3667 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3668 credits = ext4_writepage_trans_blocks(inode);
3670 credits = ext4_blocks_for_truncate(inode);
3672 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3673 if (IS_ERR(handle)) {
3674 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3678 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3679 ext4_block_truncate_page(handle, mapping, inode->i_size);
3682 * We add the inode to the orphan list, so that if this
3683 * truncate spans multiple transactions, and we crash, we will
3684 * resume the truncate when the filesystem recovers. It also
3685 * marks the inode dirty, to catch the new size.
3687 * Implication: the file must always be in a sane, consistent
3688 * truncatable state while each transaction commits.
3690 if (ext4_orphan_add(handle, inode))
3693 down_write(&EXT4_I(inode)->i_data_sem);
3695 ext4_discard_preallocations(inode);
3697 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3698 ext4_ext_truncate(handle, inode);
3700 ext4_ind_truncate(handle, inode);
3702 up_write(&ei->i_data_sem);
3705 ext4_handle_sync(handle);
3709 * If this was a simple ftruncate() and the file will remain alive,
3710 * then we need to clear up the orphan record which we created above.
3711 * However, if this was a real unlink then we were called by
3712 * ext4_delete_inode(), and we allow that function to clean up the
3713 * orphan info for us.
3716 ext4_orphan_del(handle, inode);
3718 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3719 ext4_mark_inode_dirty(handle, inode);
3720 ext4_journal_stop(handle);
3722 trace_ext4_truncate_exit(inode);
3726 * ext4_get_inode_loc returns with an extra refcount against the inode's
3727 * underlying buffer_head on success. If 'in_mem' is true, we have all
3728 * data in memory that is needed to recreate the on-disk version of this
3731 static int __ext4_get_inode_loc(struct inode *inode,
3732 struct ext4_iloc *iloc, int in_mem)
3734 struct ext4_group_desc *gdp;
3735 struct buffer_head *bh;
3736 struct super_block *sb = inode->i_sb;
3738 int inodes_per_block, inode_offset;
3741 if (!ext4_valid_inum(sb, inode->i_ino))
3744 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3745 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3750 * Figure out the offset within the block group inode table
3752 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3753 inode_offset = ((inode->i_ino - 1) %
3754 EXT4_INODES_PER_GROUP(sb));
3755 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3756 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3758 bh = sb_getblk(sb, block);
3761 if (!buffer_uptodate(bh)) {
3765 * If the buffer has the write error flag, we have failed
3766 * to write out another inode in the same block. In this
3767 * case, we don't have to read the block because we may
3768 * read the old inode data successfully.
3770 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3771 set_buffer_uptodate(bh);
3773 if (buffer_uptodate(bh)) {
3774 /* someone brought it uptodate while we waited */
3780 * If we have all information of the inode in memory and this
3781 * is the only valid inode in the block, we need not read the
3785 struct buffer_head *bitmap_bh;
3788 start = inode_offset & ~(inodes_per_block - 1);
3790 /* Is the inode bitmap in cache? */
3791 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3792 if (unlikely(!bitmap_bh))
3796 * If the inode bitmap isn't in cache then the
3797 * optimisation may end up performing two reads instead
3798 * of one, so skip it.
3800 if (!buffer_uptodate(bitmap_bh)) {
3804 for (i = start; i < start + inodes_per_block; i++) {
3805 if (i == inode_offset)
3807 if (ext4_test_bit(i, bitmap_bh->b_data))
3811 if (i == start + inodes_per_block) {
3812 /* all other inodes are free, so skip I/O */
3813 memset(bh->b_data, 0, bh->b_size);
3814 set_buffer_uptodate(bh);
3822 * If we need to do any I/O, try to pre-readahead extra
3823 * blocks from the inode table.
3825 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3826 ext4_fsblk_t b, end, table;
3828 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3830 table = ext4_inode_table(sb, gdp);
3831 /* s_inode_readahead_blks is always a power of 2 */
3832 b = block & ~((ext4_fsblk_t) ra_blks - 1);
3836 num = EXT4_INODES_PER_GROUP(sb);
3837 if (ext4_has_group_desc_csum(sb))
3838 num -= ext4_itable_unused_count(sb, gdp);
3839 table += num / inodes_per_block;
3843 sb_breadahead(sb, b++);
3847 * There are other valid inodes in the buffer, this inode
3848 * has in-inode xattrs, or we don't have this inode in memory.
3849 * Read the block from disk.
3851 trace_ext4_load_inode(inode);
3853 bh->b_end_io = end_buffer_read_sync;
3854 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3856 if (!buffer_uptodate(bh)) {
3857 EXT4_ERROR_INODE_BLOCK(inode, block,
3858 "unable to read itable block");
3868 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3870 /* We have all inode data except xattrs in memory here. */
3871 return __ext4_get_inode_loc(inode, iloc,
3872 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3875 void ext4_set_inode_flags(struct inode *inode)
3877 unsigned int flags = EXT4_I(inode)->i_flags;
3879 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3880 if (flags & EXT4_SYNC_FL)
3881 inode->i_flags |= S_SYNC;
3882 if (flags & EXT4_APPEND_FL)
3883 inode->i_flags |= S_APPEND;
3884 if (flags & EXT4_IMMUTABLE_FL)
3885 inode->i_flags |= S_IMMUTABLE;
3886 if (flags & EXT4_NOATIME_FL)
3887 inode->i_flags |= S_NOATIME;
3888 if (flags & EXT4_DIRSYNC_FL)
3889 inode->i_flags |= S_DIRSYNC;
3892 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3893 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3895 unsigned int vfs_fl;
3896 unsigned long old_fl, new_fl;
3899 vfs_fl = ei->vfs_inode.i_flags;
3900 old_fl = ei->i_flags;
3901 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3902 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3904 if (vfs_fl & S_SYNC)
3905 new_fl |= EXT4_SYNC_FL;
3906 if (vfs_fl & S_APPEND)
3907 new_fl |= EXT4_APPEND_FL;
3908 if (vfs_fl & S_IMMUTABLE)
3909 new_fl |= EXT4_IMMUTABLE_FL;
3910 if (vfs_fl & S_NOATIME)
3911 new_fl |= EXT4_NOATIME_FL;
3912 if (vfs_fl & S_DIRSYNC)
3913 new_fl |= EXT4_DIRSYNC_FL;
3914 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3917 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3918 struct ext4_inode_info *ei)
3921 struct inode *inode = &(ei->vfs_inode);
3922 struct super_block *sb = inode->i_sb;
3924 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3925 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3926 /* we are using combined 48 bit field */
3927 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3928 le32_to_cpu(raw_inode->i_blocks_lo);
3929 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3930 /* i_blocks represent file system block size */
3931 return i_blocks << (inode->i_blkbits - 9);
3936 return le32_to_cpu(raw_inode->i_blocks_lo);
3940 static inline void ext4_iget_extra_inode(struct inode *inode,
3941 struct ext4_inode *raw_inode,
3942 struct ext4_inode_info *ei)
3944 __le32 *magic = (void *)raw_inode +
3945 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3946 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3947 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3948 ext4_find_inline_data_nolock(inode);
3950 EXT4_I(inode)->i_inline_off = 0;
3953 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3955 struct ext4_iloc iloc;
3956 struct ext4_inode *raw_inode;
3957 struct ext4_inode_info *ei;
3958 struct inode *inode;
3959 journal_t *journal = EXT4_SB(sb)->s_journal;
3965 inode = iget_locked(sb, ino);
3967 return ERR_PTR(-ENOMEM);
3968 if (!(inode->i_state & I_NEW))
3974 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3977 raw_inode = ext4_raw_inode(&iloc);
3979 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3980 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3981 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3982 EXT4_INODE_SIZE(inode->i_sb)) {
3983 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3984 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3985 EXT4_INODE_SIZE(inode->i_sb));
3990 ei->i_extra_isize = 0;
3992 /* Precompute checksum seed for inode metadata */
3993 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3994 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3995 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3997 __le32 inum = cpu_to_le32(inode->i_ino);
3998 __le32 gen = raw_inode->i_generation;
3999 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4001 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4005 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4006 EXT4_ERROR_INODE(inode, "checksum invalid");
4011 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4012 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4013 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4014 if (!(test_opt(inode->i_sb, NO_UID32))) {
4015 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4016 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4018 i_uid_write(inode, i_uid);
4019 i_gid_write(inode, i_gid);
4020 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4022 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4023 ei->i_inline_off = 0;
4024 ei->i_dir_start_lookup = 0;
4025 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4026 /* We now have enough fields to check if the inode was active or not.
4027 * This is needed because nfsd might try to access dead inodes
4028 * the test is that same one that e2fsck uses
4029 * NeilBrown 1999oct15
4031 if (inode->i_nlink == 0) {
4032 if ((inode->i_mode == 0 ||
4033 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4034 ino != EXT4_BOOT_LOADER_INO) {
4035 /* this inode is deleted */
4039 /* The only unlinked inodes we let through here have
4040 * valid i_mode and are being read by the orphan
4041 * recovery code: that's fine, we're about to complete
4042 * the process of deleting those.
4043 * OR it is the EXT4_BOOT_LOADER_INO which is
4044 * not initialized on a new filesystem. */
4046 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4047 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4048 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4049 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4051 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4052 inode->i_size = ext4_isize(raw_inode);
4053 ei->i_disksize = inode->i_size;
4055 ei->i_reserved_quota = 0;
4057 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4058 ei->i_block_group = iloc.block_group;
4059 ei->i_last_alloc_group = ~0;
4061 * NOTE! The in-memory inode i_data array is in little-endian order
4062 * even on big-endian machines: we do NOT byteswap the block numbers!
4064 for (block = 0; block < EXT4_N_BLOCKS; block++)
4065 ei->i_data[block] = raw_inode->i_block[block];
4066 INIT_LIST_HEAD(&ei->i_orphan);
4069 * Set transaction id's of transactions that have to be committed
4070 * to finish f[data]sync. We set them to currently running transaction
4071 * as we cannot be sure that the inode or some of its metadata isn't
4072 * part of the transaction - the inode could have been reclaimed and
4073 * now it is reread from disk.
4076 transaction_t *transaction;
4079 read_lock(&journal->j_state_lock);
4080 if (journal->j_running_transaction)
4081 transaction = journal->j_running_transaction;
4083 transaction = journal->j_committing_transaction;
4085 tid = transaction->t_tid;
4087 tid = journal->j_commit_sequence;
4088 read_unlock(&journal->j_state_lock);
4089 ei->i_sync_tid = tid;
4090 ei->i_datasync_tid = tid;
4093 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4094 if (ei->i_extra_isize == 0) {
4095 /* The extra space is currently unused. Use it. */
4096 ei->i_extra_isize = sizeof(struct ext4_inode) -
4097 EXT4_GOOD_OLD_INODE_SIZE;
4099 ext4_iget_extra_inode(inode, raw_inode, ei);
4103 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4104 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4105 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4106 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4108 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4109 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4110 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4112 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4116 if (ei->i_file_acl &&
4117 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4118 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4122 } else if (!ext4_has_inline_data(inode)) {
4123 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4124 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4125 (S_ISLNK(inode->i_mode) &&
4126 !ext4_inode_is_fast_symlink(inode))))
4127 /* Validate extent which is part of inode */
4128 ret = ext4_ext_check_inode(inode);
4129 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4130 (S_ISLNK(inode->i_mode) &&
4131 !ext4_inode_is_fast_symlink(inode))) {
4132 /* Validate block references which are part of inode */
4133 ret = ext4_ind_check_inode(inode);
4139 if (S_ISREG(inode->i_mode)) {
4140 inode->i_op = &ext4_file_inode_operations;
4141 inode->i_fop = &ext4_file_operations;
4142 ext4_set_aops(inode);
4143 } else if (S_ISDIR(inode->i_mode)) {
4144 inode->i_op = &ext4_dir_inode_operations;
4145 inode->i_fop = &ext4_dir_operations;
4146 } else if (S_ISLNK(inode->i_mode)) {
4147 if (ext4_inode_is_fast_symlink(inode)) {
4148 inode->i_op = &ext4_fast_symlink_inode_operations;
4149 nd_terminate_link(ei->i_data, inode->i_size,
4150 sizeof(ei->i_data) - 1);
4152 inode->i_op = &ext4_symlink_inode_operations;
4153 ext4_set_aops(inode);
4155 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4156 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4157 inode->i_op = &ext4_special_inode_operations;
4158 if (raw_inode->i_block[0])
4159 init_special_inode(inode, inode->i_mode,
4160 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4162 init_special_inode(inode, inode->i_mode,
4163 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4164 } else if (ino == EXT4_BOOT_LOADER_INO) {
4165 make_bad_inode(inode);
4168 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4172 ext4_set_inode_flags(inode);
4173 unlock_new_inode(inode);
4179 return ERR_PTR(ret);
4182 static int ext4_inode_blocks_set(handle_t *handle,
4183 struct ext4_inode *raw_inode,
4184 struct ext4_inode_info *ei)
4186 struct inode *inode = &(ei->vfs_inode);
4187 u64 i_blocks = inode->i_blocks;
4188 struct super_block *sb = inode->i_sb;
4190 if (i_blocks <= ~0U) {
4192 * i_blocks can be represented in a 32 bit variable
4193 * as multiple of 512 bytes
4195 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4196 raw_inode->i_blocks_high = 0;
4197 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4200 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4203 if (i_blocks <= 0xffffffffffffULL) {
4205 * i_blocks can be represented in a 48 bit variable
4206 * as multiple of 512 bytes
4208 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4209 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4210 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4212 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4213 /* i_block is stored in file system block size */
4214 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4215 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4216 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4222 * Post the struct inode info into an on-disk inode location in the
4223 * buffer-cache. This gobbles the caller's reference to the
4224 * buffer_head in the inode location struct.
4226 * The caller must have write access to iloc->bh.
4228 static int ext4_do_update_inode(handle_t *handle,
4229 struct inode *inode,
4230 struct ext4_iloc *iloc)
4232 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4233 struct ext4_inode_info *ei = EXT4_I(inode);
4234 struct buffer_head *bh = iloc->bh;
4235 int err = 0, rc, block;
4236 int need_datasync = 0;
4240 /* For fields not not tracking in the in-memory inode,
4241 * initialise them to zero for new inodes. */
4242 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4243 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4245 ext4_get_inode_flags(ei);
4246 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4247 i_uid = i_uid_read(inode);
4248 i_gid = i_gid_read(inode);
4249 if (!(test_opt(inode->i_sb, NO_UID32))) {
4250 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4251 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4253 * Fix up interoperability with old kernels. Otherwise, old inodes get
4254 * re-used with the upper 16 bits of the uid/gid intact
4257 raw_inode->i_uid_high =
4258 cpu_to_le16(high_16_bits(i_uid));
4259 raw_inode->i_gid_high =
4260 cpu_to_le16(high_16_bits(i_gid));
4262 raw_inode->i_uid_high = 0;
4263 raw_inode->i_gid_high = 0;
4266 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4267 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4268 raw_inode->i_uid_high = 0;
4269 raw_inode->i_gid_high = 0;
4271 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4273 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4274 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4275 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4276 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4278 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4280 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4281 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4282 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4283 cpu_to_le32(EXT4_OS_HURD))
4284 raw_inode->i_file_acl_high =
4285 cpu_to_le16(ei->i_file_acl >> 32);
4286 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4287 if (ei->i_disksize != ext4_isize(raw_inode)) {
4288 ext4_isize_set(raw_inode, ei->i_disksize);
4291 if (ei->i_disksize > 0x7fffffffULL) {
4292 struct super_block *sb = inode->i_sb;
4293 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4294 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4295 EXT4_SB(sb)->s_es->s_rev_level ==
4296 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4297 /* If this is the first large file
4298 * created, add a flag to the superblock.
4300 err = ext4_journal_get_write_access(handle,
4301 EXT4_SB(sb)->s_sbh);
4304 ext4_update_dynamic_rev(sb);
4305 EXT4_SET_RO_COMPAT_FEATURE(sb,
4306 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4307 ext4_handle_sync(handle);
4308 err = ext4_handle_dirty_super(handle, sb);
4311 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4312 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4313 if (old_valid_dev(inode->i_rdev)) {
4314 raw_inode->i_block[0] =
4315 cpu_to_le32(old_encode_dev(inode->i_rdev));
4316 raw_inode->i_block[1] = 0;
4318 raw_inode->i_block[0] = 0;
4319 raw_inode->i_block[1] =
4320 cpu_to_le32(new_encode_dev(inode->i_rdev));
4321 raw_inode->i_block[2] = 0;
4323 } else if (!ext4_has_inline_data(inode)) {
4324 for (block = 0; block < EXT4_N_BLOCKS; block++)
4325 raw_inode->i_block[block] = ei->i_data[block];
4328 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4329 if (ei->i_extra_isize) {
4330 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4331 raw_inode->i_version_hi =
4332 cpu_to_le32(inode->i_version >> 32);
4333 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4336 ext4_inode_csum_set(inode, raw_inode, ei);
4338 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4339 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4342 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4344 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4347 ext4_std_error(inode->i_sb, err);
4352 * ext4_write_inode()
4354 * We are called from a few places:
4356 * - Within generic_file_write() for O_SYNC files.
4357 * Here, there will be no transaction running. We wait for any running
4358 * transaction to commit.
4360 * - Within sys_sync(), kupdate and such.
4361 * We wait on commit, if tol to.
4363 * - Within prune_icache() (PF_MEMALLOC == true)
4364 * Here we simply return. We can't afford to block kswapd on the
4367 * In all cases it is actually safe for us to return without doing anything,
4368 * because the inode has been copied into a raw inode buffer in
4369 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4372 * Note that we are absolutely dependent upon all inode dirtiers doing the
4373 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4374 * which we are interested.
4376 * It would be a bug for them to not do this. The code:
4378 * mark_inode_dirty(inode)
4380 * inode->i_size = expr;
4382 * is in error because a kswapd-driven write_inode() could occur while
4383 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4384 * will no longer be on the superblock's dirty inode list.
4386 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4390 if (current->flags & PF_MEMALLOC)
4393 if (EXT4_SB(inode->i_sb)->s_journal) {
4394 if (ext4_journal_current_handle()) {
4395 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4400 if (wbc->sync_mode != WB_SYNC_ALL)
4403 err = ext4_force_commit(inode->i_sb);
4405 struct ext4_iloc iloc;
4407 err = __ext4_get_inode_loc(inode, &iloc, 0);
4410 if (wbc->sync_mode == WB_SYNC_ALL)
4411 sync_dirty_buffer(iloc.bh);
4412 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4413 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4414 "IO error syncing inode");
4423 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4424 * buffers that are attached to a page stradding i_size and are undergoing
4425 * commit. In that case we have to wait for commit to finish and try again.
4427 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4431 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4432 tid_t commit_tid = 0;
4435 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4437 * All buffers in the last page remain valid? Then there's nothing to
4438 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4441 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4444 page = find_lock_page(inode->i_mapping,
4445 inode->i_size >> PAGE_CACHE_SHIFT);
4448 ret = __ext4_journalled_invalidatepage(page, offset,
4449 PAGE_CACHE_SIZE - offset);
4451 page_cache_release(page);
4455 read_lock(&journal->j_state_lock);
4456 if (journal->j_committing_transaction)
4457 commit_tid = journal->j_committing_transaction->t_tid;
4458 read_unlock(&journal->j_state_lock);
4460 jbd2_log_wait_commit(journal, commit_tid);
4467 * Called from notify_change.
4469 * We want to trap VFS attempts to truncate the file as soon as
4470 * possible. In particular, we want to make sure that when the VFS
4471 * shrinks i_size, we put the inode on the orphan list and modify
4472 * i_disksize immediately, so that during the subsequent flushing of
4473 * dirty pages and freeing of disk blocks, we can guarantee that any
4474 * commit will leave the blocks being flushed in an unused state on
4475 * disk. (On recovery, the inode will get truncated and the blocks will
4476 * be freed, so we have a strong guarantee that no future commit will
4477 * leave these blocks visible to the user.)
4479 * Another thing we have to assure is that if we are in ordered mode
4480 * and inode is still attached to the committing transaction, we must
4481 * we start writeout of all the dirty pages which are being truncated.
4482 * This way we are sure that all the data written in the previous
4483 * transaction are already on disk (truncate waits for pages under
4486 * Called with inode->i_mutex down.
4488 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4490 struct inode *inode = dentry->d_inode;
4493 const unsigned int ia_valid = attr->ia_valid;
4495 error = inode_change_ok(inode, attr);
4499 if (is_quota_modification(inode, attr))
4500 dquot_initialize(inode);
4501 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4502 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4505 /* (user+group)*(old+new) structure, inode write (sb,
4506 * inode block, ? - but truncate inode update has it) */
4507 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4508 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4509 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4510 if (IS_ERR(handle)) {
4511 error = PTR_ERR(handle);
4514 error = dquot_transfer(inode, attr);
4516 ext4_journal_stop(handle);
4519 /* Update corresponding info in inode so that everything is in
4520 * one transaction */
4521 if (attr->ia_valid & ATTR_UID)
4522 inode->i_uid = attr->ia_uid;
4523 if (attr->ia_valid & ATTR_GID)
4524 inode->i_gid = attr->ia_gid;
4525 error = ext4_mark_inode_dirty(handle, inode);
4526 ext4_journal_stop(handle);
4529 if (attr->ia_valid & ATTR_SIZE) {
4531 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4532 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4534 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4539 if (S_ISREG(inode->i_mode) &&
4540 attr->ia_valid & ATTR_SIZE &&
4541 (attr->ia_size < inode->i_size)) {
4544 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4545 if (IS_ERR(handle)) {
4546 error = PTR_ERR(handle);
4549 if (ext4_handle_valid(handle)) {
4550 error = ext4_orphan_add(handle, inode);
4553 EXT4_I(inode)->i_disksize = attr->ia_size;
4554 rc = ext4_mark_inode_dirty(handle, inode);
4557 ext4_journal_stop(handle);
4559 if (ext4_should_order_data(inode)) {
4560 error = ext4_begin_ordered_truncate(inode,
4563 /* Do as much error cleanup as possible */
4564 handle = ext4_journal_start(inode,
4566 if (IS_ERR(handle)) {
4567 ext4_orphan_del(NULL, inode);
4570 ext4_orphan_del(handle, inode);
4572 ext4_journal_stop(handle);
4578 if (attr->ia_valid & ATTR_SIZE) {
4579 if (attr->ia_size != inode->i_size) {
4580 loff_t oldsize = inode->i_size;
4582 i_size_write(inode, attr->ia_size);
4584 * Blocks are going to be removed from the inode. Wait
4585 * for dio in flight. Temporarily disable
4586 * dioread_nolock to prevent livelock.
4589 if (!ext4_should_journal_data(inode)) {
4590 ext4_inode_block_unlocked_dio(inode);
4591 inode_dio_wait(inode);
4592 ext4_inode_resume_unlocked_dio(inode);
4594 ext4_wait_for_tail_page_commit(inode);
4597 * Truncate pagecache after we've waited for commit
4598 * in data=journal mode to make pages freeable.
4600 truncate_pagecache(inode, oldsize, inode->i_size);
4602 ext4_truncate(inode);
4606 setattr_copy(inode, attr);
4607 mark_inode_dirty(inode);
4611 * If the call to ext4_truncate failed to get a transaction handle at
4612 * all, we need to clean up the in-core orphan list manually.
4614 if (orphan && inode->i_nlink)
4615 ext4_orphan_del(NULL, inode);
4617 if (!rc && (ia_valid & ATTR_MODE))
4618 rc = ext4_acl_chmod(inode);
4621 ext4_std_error(inode->i_sb, error);
4627 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4630 struct inode *inode;
4631 unsigned long long delalloc_blocks;
4633 inode = dentry->d_inode;
4634 generic_fillattr(inode, stat);
4637 * We can't update i_blocks if the block allocation is delayed
4638 * otherwise in the case of system crash before the real block
4639 * allocation is done, we will have i_blocks inconsistent with
4640 * on-disk file blocks.
4641 * We always keep i_blocks updated together with real
4642 * allocation. But to not confuse with user, stat
4643 * will return the blocks that include the delayed allocation
4644 * blocks for this file.
4646 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4647 EXT4_I(inode)->i_reserved_data_blocks);
4649 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4653 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4655 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4656 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4657 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4661 * Account for index blocks, block groups bitmaps and block group
4662 * descriptor blocks if modify datablocks and index blocks
4663 * worse case, the indexs blocks spread over different block groups
4665 * If datablocks are discontiguous, they are possible to spread over
4666 * different block groups too. If they are contiguous, with flexbg,
4667 * they could still across block group boundary.
4669 * Also account for superblock, inode, quota and xattr blocks
4671 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4673 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4679 * How many index blocks need to touch to modify nrblocks?
4680 * The "Chunk" flag indicating whether the nrblocks is
4681 * physically contiguous on disk
4683 * For Direct IO and fallocate, they calls get_block to allocate
4684 * one single extent at a time, so they could set the "Chunk" flag
4686 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4691 * Now let's see how many group bitmaps and group descriptors need
4701 if (groups > ngroups)
4703 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4704 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4706 /* bitmaps and block group descriptor blocks */
4707 ret += groups + gdpblocks;
4709 /* Blocks for super block, inode, quota and xattr blocks */
4710 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4716 * Calculate the total number of credits to reserve to fit
4717 * the modification of a single pages into a single transaction,
4718 * which may include multiple chunks of block allocations.
4720 * This could be called via ext4_write_begin()
4722 * We need to consider the worse case, when
4723 * one new block per extent.
4725 int ext4_writepage_trans_blocks(struct inode *inode)
4727 int bpp = ext4_journal_blocks_per_page(inode);
4730 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4732 /* Account for data blocks for journalled mode */
4733 if (ext4_should_journal_data(inode))
4739 * Calculate the journal credits for a chunk of data modification.
4741 * This is called from DIO, fallocate or whoever calling
4742 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4744 * journal buffers for data blocks are not included here, as DIO
4745 * and fallocate do no need to journal data buffers.
4747 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4749 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4753 * The caller must have previously called ext4_reserve_inode_write().
4754 * Give this, we know that the caller already has write access to iloc->bh.
4756 int ext4_mark_iloc_dirty(handle_t *handle,
4757 struct inode *inode, struct ext4_iloc *iloc)
4761 if (IS_I_VERSION(inode))
4762 inode_inc_iversion(inode);
4764 /* the do_update_inode consumes one bh->b_count */
4767 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4768 err = ext4_do_update_inode(handle, inode, iloc);
4774 * On success, We end up with an outstanding reference count against
4775 * iloc->bh. This _must_ be cleaned up later.
4779 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4780 struct ext4_iloc *iloc)
4784 err = ext4_get_inode_loc(inode, iloc);
4786 BUFFER_TRACE(iloc->bh, "get_write_access");
4787 err = ext4_journal_get_write_access(handle, iloc->bh);
4793 ext4_std_error(inode->i_sb, err);
4798 * Expand an inode by new_extra_isize bytes.
4799 * Returns 0 on success or negative error number on failure.
4801 static int ext4_expand_extra_isize(struct inode *inode,
4802 unsigned int new_extra_isize,
4803 struct ext4_iloc iloc,
4806 struct ext4_inode *raw_inode;
4807 struct ext4_xattr_ibody_header *header;
4809 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4812 raw_inode = ext4_raw_inode(&iloc);
4814 header = IHDR(inode, raw_inode);
4816 /* No extended attributes present */
4817 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4818 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4819 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4821 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4825 /* try to expand with EAs present */
4826 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4831 * What we do here is to mark the in-core inode as clean with respect to inode
4832 * dirtiness (it may still be data-dirty).
4833 * This means that the in-core inode may be reaped by prune_icache
4834 * without having to perform any I/O. This is a very good thing,
4835 * because *any* task may call prune_icache - even ones which
4836 * have a transaction open against a different journal.
4838 * Is this cheating? Not really. Sure, we haven't written the
4839 * inode out, but prune_icache isn't a user-visible syncing function.
4840 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4841 * we start and wait on commits.
4843 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4845 struct ext4_iloc iloc;
4846 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4847 static unsigned int mnt_count;
4851 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4852 err = ext4_reserve_inode_write(handle, inode, &iloc);
4853 if (ext4_handle_valid(handle) &&
4854 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4855 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4857 * We need extra buffer credits since we may write into EA block
4858 * with this same handle. If journal_extend fails, then it will
4859 * only result in a minor loss of functionality for that inode.
4860 * If this is felt to be critical, then e2fsck should be run to
4861 * force a large enough s_min_extra_isize.
4863 if ((jbd2_journal_extend(handle,
4864 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4865 ret = ext4_expand_extra_isize(inode,
4866 sbi->s_want_extra_isize,
4869 ext4_set_inode_state(inode,
4870 EXT4_STATE_NO_EXPAND);
4872 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4873 ext4_warning(inode->i_sb,
4874 "Unable to expand inode %lu. Delete"
4875 " some EAs or run e2fsck.",
4878 le16_to_cpu(sbi->s_es->s_mnt_count);
4884 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4889 * ext4_dirty_inode() is called from __mark_inode_dirty()
4891 * We're really interested in the case where a file is being extended.
4892 * i_size has been changed by generic_commit_write() and we thus need
4893 * to include the updated inode in the current transaction.
4895 * Also, dquot_alloc_block() will always dirty the inode when blocks
4896 * are allocated to the file.
4898 * If the inode is marked synchronous, we don't honour that here - doing
4899 * so would cause a commit on atime updates, which we don't bother doing.
4900 * We handle synchronous inodes at the highest possible level.
4902 void ext4_dirty_inode(struct inode *inode, int flags)
4906 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4910 ext4_mark_inode_dirty(handle, inode);
4912 ext4_journal_stop(handle);
4919 * Bind an inode's backing buffer_head into this transaction, to prevent
4920 * it from being flushed to disk early. Unlike
4921 * ext4_reserve_inode_write, this leaves behind no bh reference and
4922 * returns no iloc structure, so the caller needs to repeat the iloc
4923 * lookup to mark the inode dirty later.
4925 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4927 struct ext4_iloc iloc;
4931 err = ext4_get_inode_loc(inode, &iloc);
4933 BUFFER_TRACE(iloc.bh, "get_write_access");
4934 err = jbd2_journal_get_write_access(handle, iloc.bh);
4936 err = ext4_handle_dirty_metadata(handle,
4942 ext4_std_error(inode->i_sb, err);
4947 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4954 * We have to be very careful here: changing a data block's
4955 * journaling status dynamically is dangerous. If we write a
4956 * data block to the journal, change the status and then delete
4957 * that block, we risk forgetting to revoke the old log record
4958 * from the journal and so a subsequent replay can corrupt data.
4959 * So, first we make sure that the journal is empty and that
4960 * nobody is changing anything.
4963 journal = EXT4_JOURNAL(inode);
4966 if (is_journal_aborted(journal))
4968 /* We have to allocate physical blocks for delalloc blocks
4969 * before flushing journal. otherwise delalloc blocks can not
4970 * be allocated any more. even more truncate on delalloc blocks
4971 * could trigger BUG by flushing delalloc blocks in journal.
4972 * There is no delalloc block in non-journal data mode.
4974 if (val && test_opt(inode->i_sb, DELALLOC)) {
4975 err = ext4_alloc_da_blocks(inode);
4980 /* Wait for all existing dio workers */
4981 ext4_inode_block_unlocked_dio(inode);
4982 inode_dio_wait(inode);
4984 jbd2_journal_lock_updates(journal);
4987 * OK, there are no updates running now, and all cached data is
4988 * synced to disk. We are now in a completely consistent state
4989 * which doesn't have anything in the journal, and we know that
4990 * no filesystem updates are running, so it is safe to modify
4991 * the inode's in-core data-journaling state flag now.
4995 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4997 jbd2_journal_flush(journal);
4998 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5000 ext4_set_aops(inode);
5002 jbd2_journal_unlock_updates(journal);
5003 ext4_inode_resume_unlocked_dio(inode);
5005 /* Finally we can mark the inode as dirty. */
5007 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5009 return PTR_ERR(handle);
5011 err = ext4_mark_inode_dirty(handle, inode);
5012 ext4_handle_sync(handle);
5013 ext4_journal_stop(handle);
5014 ext4_std_error(inode->i_sb, err);
5019 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5021 return !buffer_mapped(bh);
5024 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5026 struct page *page = vmf->page;
5030 struct file *file = vma->vm_file;
5031 struct inode *inode = file_inode(file);
5032 struct address_space *mapping = inode->i_mapping;
5034 get_block_t *get_block;
5037 sb_start_pagefault(inode->i_sb);
5038 file_update_time(vma->vm_file);
5039 /* Delalloc case is easy... */
5040 if (test_opt(inode->i_sb, DELALLOC) &&
5041 !ext4_should_journal_data(inode) &&
5042 !ext4_nonda_switch(inode->i_sb)) {
5044 ret = __block_page_mkwrite(vma, vmf,
5045 ext4_da_get_block_prep);
5046 } while (ret == -ENOSPC &&
5047 ext4_should_retry_alloc(inode->i_sb, &retries));
5052 size = i_size_read(inode);
5053 /* Page got truncated from under us? */
5054 if (page->mapping != mapping || page_offset(page) > size) {
5056 ret = VM_FAULT_NOPAGE;
5060 if (page->index == size >> PAGE_CACHE_SHIFT)
5061 len = size & ~PAGE_CACHE_MASK;
5063 len = PAGE_CACHE_SIZE;
5065 * Return if we have all the buffers mapped. This avoids the need to do
5066 * journal_start/journal_stop which can block and take a long time
5068 if (page_has_buffers(page)) {
5069 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5071 ext4_bh_unmapped)) {
5072 /* Wait so that we don't change page under IO */
5073 wait_for_stable_page(page);
5074 ret = VM_FAULT_LOCKED;
5079 /* OK, we need to fill the hole... */
5080 if (ext4_should_dioread_nolock(inode))
5081 get_block = ext4_get_block_write;
5083 get_block = ext4_get_block;
5085 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5086 ext4_writepage_trans_blocks(inode));
5087 if (IS_ERR(handle)) {
5088 ret = VM_FAULT_SIGBUS;
5091 ret = __block_page_mkwrite(vma, vmf, get_block);
5092 if (!ret && ext4_should_journal_data(inode)) {
5093 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5094 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5096 ret = VM_FAULT_SIGBUS;
5097 ext4_journal_stop(handle);
5100 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5102 ext4_journal_stop(handle);
5103 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5106 ret = block_page_mkwrite_return(ret);
5108 sb_end_pagefault(inode->i_sb);