2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/aio.h>
42 #include "ext4_jbd2.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 struct ext4_inode_info *ei)
54 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 csum_lo = le16_to_cpu(raw->i_checksum_lo);
60 raw->i_checksum_lo = 0;
61 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
62 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
63 csum_hi = le16_to_cpu(raw->i_checksum_hi);
64 raw->i_checksum_hi = 0;
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
68 EXT4_INODE_SIZE(inode->i_sb));
70 raw->i_checksum_lo = cpu_to_le16(csum_lo);
71 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
72 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
73 raw->i_checksum_hi = cpu_to_le16(csum_hi);
78 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
79 struct ext4_inode_info *ei)
81 __u32 provided, calculated;
83 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
84 cpu_to_le32(EXT4_OS_LINUX) ||
85 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
86 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
89 provided = le16_to_cpu(raw->i_checksum_lo);
90 calculated = ext4_inode_csum(inode, raw, ei);
91 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
92 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
93 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
97 return provided == calculated;
100 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
101 struct ext4_inode_info *ei)
105 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
106 cpu_to_le32(EXT4_OS_LINUX) ||
107 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
108 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
111 csum = ext4_inode_csum(inode, raw, ei);
112 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
113 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
114 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
115 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
118 static inline int ext4_begin_ordered_truncate(struct inode *inode,
121 trace_ext4_begin_ordered_truncate(inode, new_size);
123 * If jinode is zero, then we never opened the file for
124 * writing, so there's no need to call
125 * jbd2_journal_begin_ordered_truncate() since there's no
126 * outstanding writes we need to flush.
128 if (!EXT4_I(inode)->jinode)
130 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
131 EXT4_I(inode)->jinode,
135 static void ext4_invalidatepage(struct page *page, unsigned int offset,
136 unsigned int length);
137 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
138 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
139 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
140 struct inode *inode, struct page *page, loff_t from,
141 loff_t length, int flags);
144 * Test whether an inode is a fast symlink.
146 static int ext4_inode_is_fast_symlink(struct inode *inode)
148 int ea_blocks = EXT4_I(inode)->i_file_acl ?
149 (inode->i_sb->s_blocksize >> 9) : 0;
151 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
155 * Restart the transaction associated with *handle. This does a commit,
156 * so before we call here everything must be consistently dirtied against
159 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
165 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
166 * moment, get_block can be called only for blocks inside i_size since
167 * page cache has been already dropped and writes are blocked by
168 * i_mutex. So we can safely drop the i_data_sem here.
170 BUG_ON(EXT4_JOURNAL(inode) == NULL);
171 jbd_debug(2, "restarting handle %p\n", handle);
172 up_write(&EXT4_I(inode)->i_data_sem);
173 ret = ext4_journal_restart(handle, nblocks);
174 down_write(&EXT4_I(inode)->i_data_sem);
175 ext4_discard_preallocations(inode);
181 * Called at the last iput() if i_nlink is zero.
183 void ext4_evict_inode(struct inode *inode)
188 trace_ext4_evict_inode(inode);
190 if (inode->i_nlink) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode) &&
210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
211 inode->i_ino != EXT4_JOURNAL_INO) {
212 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
213 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
215 jbd2_complete_transaction(journal, commit_tid);
216 filemap_write_and_wait(&inode->i_data);
218 truncate_inode_pages(&inode->i_data, 0);
219 ext4_ioend_shutdown(inode);
223 if (!is_bad_inode(inode))
224 dquot_initialize(inode);
226 if (ext4_should_order_data(inode))
227 ext4_begin_ordered_truncate(inode, 0);
228 truncate_inode_pages(&inode->i_data, 0);
229 ext4_ioend_shutdown(inode);
231 if (is_bad_inode(inode))
235 * Protect us against freezing - iput() caller didn't have to have any
236 * protection against it
238 sb_start_intwrite(inode->i_sb);
239 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
240 ext4_blocks_for_truncate(inode)+3);
241 if (IS_ERR(handle)) {
242 ext4_std_error(inode->i_sb, PTR_ERR(handle));
244 * If we're going to skip the normal cleanup, we still need to
245 * make sure that the in-core orphan linked list is properly
248 ext4_orphan_del(NULL, inode);
249 sb_end_intwrite(inode->i_sb);
254 ext4_handle_sync(handle);
256 err = ext4_mark_inode_dirty(handle, inode);
258 ext4_warning(inode->i_sb,
259 "couldn't mark inode dirty (err %d)", err);
263 ext4_truncate(inode);
266 * ext4_ext_truncate() doesn't reserve any slop when it
267 * restarts journal transactions; therefore there may not be
268 * enough credits left in the handle to remove the inode from
269 * the orphan list and set the dtime field.
271 if (!ext4_handle_has_enough_credits(handle, 3)) {
272 err = ext4_journal_extend(handle, 3);
274 err = ext4_journal_restart(handle, 3);
276 ext4_warning(inode->i_sb,
277 "couldn't extend journal (err %d)", err);
279 ext4_journal_stop(handle);
280 ext4_orphan_del(NULL, inode);
281 sb_end_intwrite(inode->i_sb);
287 * Kill off the orphan record which ext4_truncate created.
288 * AKPM: I think this can be inside the above `if'.
289 * Note that ext4_orphan_del() has to be able to cope with the
290 * deletion of a non-existent orphan - this is because we don't
291 * know if ext4_truncate() actually created an orphan record.
292 * (Well, we could do this if we need to, but heck - it works)
294 ext4_orphan_del(handle, inode);
295 EXT4_I(inode)->i_dtime = get_seconds();
298 * One subtle ordering requirement: if anything has gone wrong
299 * (transaction abort, IO errors, whatever), then we can still
300 * do these next steps (the fs will already have been marked as
301 * having errors), but we can't free the inode if the mark_dirty
304 if (ext4_mark_inode_dirty(handle, inode))
305 /* If that failed, just do the required in-core inode clear. */
306 ext4_clear_inode(inode);
308 ext4_free_inode(handle, inode);
309 ext4_journal_stop(handle);
310 sb_end_intwrite(inode->i_sb);
313 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
317 qsize_t *ext4_get_reserved_space(struct inode *inode)
319 return &EXT4_I(inode)->i_reserved_quota;
324 * Calculate the number of metadata blocks need to reserve
325 * to allocate a block located at @lblock
327 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
329 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
330 return ext4_ext_calc_metadata_amount(inode, lblock);
332 return ext4_ind_calc_metadata_amount(inode, lblock);
336 * Called with i_data_sem down, which is important since we can call
337 * ext4_discard_preallocations() from here.
339 void ext4_da_update_reserve_space(struct inode *inode,
340 int used, int quota_claim)
342 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
343 struct ext4_inode_info *ei = EXT4_I(inode);
345 spin_lock(&ei->i_block_reservation_lock);
346 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
347 if (unlikely(used > ei->i_reserved_data_blocks)) {
348 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
349 "with only %d reserved data blocks",
350 __func__, inode->i_ino, used,
351 ei->i_reserved_data_blocks);
353 used = ei->i_reserved_data_blocks;
356 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
357 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
358 "with only %d reserved metadata blocks "
359 "(releasing %d blocks with reserved %d data blocks)",
360 inode->i_ino, ei->i_allocated_meta_blocks,
361 ei->i_reserved_meta_blocks, used,
362 ei->i_reserved_data_blocks);
364 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
367 /* Update per-inode reservations */
368 ei->i_reserved_data_blocks -= used;
369 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
370 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
371 used + ei->i_allocated_meta_blocks);
372 ei->i_allocated_meta_blocks = 0;
374 if (ei->i_reserved_data_blocks == 0) {
376 * We can release all of the reserved metadata blocks
377 * only when we have written all of the delayed
380 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
381 ei->i_reserved_meta_blocks);
382 ei->i_reserved_meta_blocks = 0;
383 ei->i_da_metadata_calc_len = 0;
385 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
387 /* Update quota subsystem for data blocks */
389 dquot_claim_block(inode, EXT4_C2B(sbi, used));
392 * We did fallocate with an offset that is already delayed
393 * allocated. So on delayed allocated writeback we should
394 * not re-claim the quota for fallocated blocks.
396 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
400 * If we have done all the pending block allocations and if
401 * there aren't any writers on the inode, we can discard the
402 * inode's preallocations.
404 if ((ei->i_reserved_data_blocks == 0) &&
405 (atomic_read(&inode->i_writecount) == 0))
406 ext4_discard_preallocations(inode);
409 static int __check_block_validity(struct inode *inode, const char *func,
411 struct ext4_map_blocks *map)
413 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
415 ext4_error_inode(inode, func, line, map->m_pblk,
416 "lblock %lu mapped to illegal pblock "
417 "(length %d)", (unsigned long) map->m_lblk,
424 #define check_block_validity(inode, map) \
425 __check_block_validity((inode), __func__, __LINE__, (map))
428 * Return the number of contiguous dirty pages in a given inode
429 * starting at page frame idx.
431 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
432 unsigned int max_pages)
434 struct address_space *mapping = inode->i_mapping;
438 int i, nr_pages, done = 0;
442 pagevec_init(&pvec, 0);
445 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
447 (pgoff_t)PAGEVEC_SIZE);
450 for (i = 0; i < nr_pages; i++) {
451 struct page *page = pvec.pages[i];
452 struct buffer_head *bh, *head;
455 if (unlikely(page->mapping != mapping) ||
457 PageWriteback(page) ||
458 page->index != idx) {
463 if (page_has_buffers(page)) {
464 bh = head = page_buffers(page);
466 if (!buffer_delay(bh) &&
467 !buffer_unwritten(bh))
469 bh = bh->b_this_page;
470 } while (!done && (bh != head));
477 if (num >= max_pages) {
482 pagevec_release(&pvec);
487 #ifdef ES_AGGRESSIVE_TEST
488 static void ext4_map_blocks_es_recheck(handle_t *handle,
490 struct ext4_map_blocks *es_map,
491 struct ext4_map_blocks *map,
498 * There is a race window that the result is not the same.
499 * e.g. xfstests #223 when dioread_nolock enables. The reason
500 * is that we lookup a block mapping in extent status tree with
501 * out taking i_data_sem. So at the time the unwritten extent
502 * could be converted.
504 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
505 down_read((&EXT4_I(inode)->i_data_sem));
506 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
507 retval = ext4_ext_map_blocks(handle, inode, map, flags &
508 EXT4_GET_BLOCKS_KEEP_SIZE);
510 retval = ext4_ind_map_blocks(handle, inode, map, flags &
511 EXT4_GET_BLOCKS_KEEP_SIZE);
513 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
514 up_read((&EXT4_I(inode)->i_data_sem));
516 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
517 * because it shouldn't be marked in es_map->m_flags.
519 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
522 * We don't check m_len because extent will be collpased in status
523 * tree. So the m_len might not equal.
525 if (es_map->m_lblk != map->m_lblk ||
526 es_map->m_flags != map->m_flags ||
527 es_map->m_pblk != map->m_pblk) {
528 printk("ES cache assertation failed for inode: %lu "
529 "es_cached ex [%d/%d/%llu/%x] != "
530 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
531 inode->i_ino, es_map->m_lblk, es_map->m_len,
532 es_map->m_pblk, es_map->m_flags, map->m_lblk,
533 map->m_len, map->m_pblk, map->m_flags,
537 #endif /* ES_AGGRESSIVE_TEST */
540 * The ext4_map_blocks() function tries to look up the requested blocks,
541 * and returns if the blocks are already mapped.
543 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
544 * and store the allocated blocks in the result buffer head and mark it
547 * If file type is extents based, it will call ext4_ext_map_blocks(),
548 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
551 * On success, it returns the number of blocks being mapped or allocate.
552 * if create==0 and the blocks are pre-allocated and uninitialized block,
553 * the result buffer head is unmapped. If the create ==1, it will make sure
554 * the buffer head is mapped.
556 * It returns 0 if plain look up failed (blocks have not been allocated), in
557 * that case, buffer head is unmapped
559 * It returns the error in case of allocation failure.
561 int ext4_map_blocks(handle_t *handle, struct inode *inode,
562 struct ext4_map_blocks *map, int flags)
564 struct extent_status es;
566 #ifdef ES_AGGRESSIVE_TEST
567 struct ext4_map_blocks orig_map;
569 memcpy(&orig_map, map, sizeof(*map));
573 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
574 "logical block %lu\n", inode->i_ino, flags, map->m_len,
575 (unsigned long) map->m_lblk);
577 /* Lookup extent status tree firstly */
578 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
579 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
580 map->m_pblk = ext4_es_pblock(&es) +
581 map->m_lblk - es.es_lblk;
582 map->m_flags |= ext4_es_is_written(&es) ?
583 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
584 retval = es.es_len - (map->m_lblk - es.es_lblk);
585 if (retval > map->m_len)
588 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
593 #ifdef ES_AGGRESSIVE_TEST
594 ext4_map_blocks_es_recheck(handle, inode, map,
601 * Try to see if we can get the block without requesting a new
604 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
605 down_read((&EXT4_I(inode)->i_data_sem));
606 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
607 retval = ext4_ext_map_blocks(handle, inode, map, flags &
608 EXT4_GET_BLOCKS_KEEP_SIZE);
610 retval = ext4_ind_map_blocks(handle, inode, map, flags &
611 EXT4_GET_BLOCKS_KEEP_SIZE);
615 unsigned long long status;
617 #ifdef ES_AGGRESSIVE_TEST
618 if (retval != map->m_len) {
619 printk("ES len assertation failed for inode: %lu "
620 "retval %d != map->m_len %d "
621 "in %s (lookup)\n", inode->i_ino, retval,
622 map->m_len, __func__);
626 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
627 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
628 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
629 ext4_find_delalloc_range(inode, map->m_lblk,
630 map->m_lblk + map->m_len - 1))
631 status |= EXTENT_STATUS_DELAYED;
632 ret = ext4_es_insert_extent(inode, map->m_lblk,
633 map->m_len, map->m_pblk, status);
637 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
638 up_read((&EXT4_I(inode)->i_data_sem));
641 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
642 int ret = check_block_validity(inode, map);
647 /* If it is only a block(s) look up */
648 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
652 * Returns if the blocks have already allocated
654 * Note that if blocks have been preallocated
655 * ext4_ext_get_block() returns the create = 0
656 * with buffer head unmapped.
658 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
662 * Here we clear m_flags because after allocating an new extent,
663 * it will be set again.
665 map->m_flags &= ~EXT4_MAP_FLAGS;
668 * New blocks allocate and/or writing to uninitialized extent
669 * will possibly result in updating i_data, so we take
670 * the write lock of i_data_sem, and call get_blocks()
671 * with create == 1 flag.
673 down_write((&EXT4_I(inode)->i_data_sem));
676 * if the caller is from delayed allocation writeout path
677 * we have already reserved fs blocks for allocation
678 * let the underlying get_block() function know to
679 * avoid double accounting
681 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
682 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
684 * We need to check for EXT4 here because migrate
685 * could have changed the inode type in between
687 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
688 retval = ext4_ext_map_blocks(handle, inode, map, flags);
690 retval = ext4_ind_map_blocks(handle, inode, map, flags);
692 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
694 * We allocated new blocks which will result in
695 * i_data's format changing. Force the migrate
696 * to fail by clearing migrate flags
698 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
702 * Update reserved blocks/metadata blocks after successful
703 * block allocation which had been deferred till now. We don't
704 * support fallocate for non extent files. So we can update
705 * reserve space here.
708 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
709 ext4_da_update_reserve_space(inode, retval, 1);
711 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
712 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
716 unsigned long long status;
718 #ifdef ES_AGGRESSIVE_TEST
719 if (retval != map->m_len) {
720 printk("ES len assertation failed for inode: %lu "
721 "retval %d != map->m_len %d "
722 "in %s (allocation)\n", inode->i_ino, retval,
723 map->m_len, __func__);
728 * If the extent has been zeroed out, we don't need to update
729 * extent status tree.
731 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
732 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
733 if (ext4_es_is_written(&es))
736 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
737 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
738 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
739 ext4_find_delalloc_range(inode, map->m_lblk,
740 map->m_lblk + map->m_len - 1))
741 status |= EXTENT_STATUS_DELAYED;
742 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
743 map->m_pblk, status);
749 up_write((&EXT4_I(inode)->i_data_sem));
750 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
751 int ret = check_block_validity(inode, map);
758 /* Maximum number of blocks we map for direct IO at once. */
759 #define DIO_MAX_BLOCKS 4096
761 static int _ext4_get_block(struct inode *inode, sector_t iblock,
762 struct buffer_head *bh, int flags)
764 handle_t *handle = ext4_journal_current_handle();
765 struct ext4_map_blocks map;
766 int ret = 0, started = 0;
769 if (ext4_has_inline_data(inode))
773 map.m_len = bh->b_size >> inode->i_blkbits;
775 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
776 /* Direct IO write... */
777 if (map.m_len > DIO_MAX_BLOCKS)
778 map.m_len = DIO_MAX_BLOCKS;
779 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
780 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
782 if (IS_ERR(handle)) {
783 ret = PTR_ERR(handle);
789 ret = ext4_map_blocks(handle, inode, &map, flags);
791 map_bh(bh, inode->i_sb, map.m_pblk);
792 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
793 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
797 ext4_journal_stop(handle);
801 int ext4_get_block(struct inode *inode, sector_t iblock,
802 struct buffer_head *bh, int create)
804 return _ext4_get_block(inode, iblock, bh,
805 create ? EXT4_GET_BLOCKS_CREATE : 0);
809 * `handle' can be NULL if create is zero
811 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
812 ext4_lblk_t block, int create, int *errp)
814 struct ext4_map_blocks map;
815 struct buffer_head *bh;
818 J_ASSERT(handle != NULL || create == 0);
822 err = ext4_map_blocks(handle, inode, &map,
823 create ? EXT4_GET_BLOCKS_CREATE : 0);
825 /* ensure we send some value back into *errp */
828 if (create && err == 0)
829 err = -ENOSPC; /* should never happen */
835 bh = sb_getblk(inode->i_sb, map.m_pblk);
840 if (map.m_flags & EXT4_MAP_NEW) {
841 J_ASSERT(create != 0);
842 J_ASSERT(handle != NULL);
845 * Now that we do not always journal data, we should
846 * keep in mind whether this should always journal the
847 * new buffer as metadata. For now, regular file
848 * writes use ext4_get_block instead, so it's not a
852 BUFFER_TRACE(bh, "call get_create_access");
853 fatal = ext4_journal_get_create_access(handle, bh);
854 if (!fatal && !buffer_uptodate(bh)) {
855 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
856 set_buffer_uptodate(bh);
859 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
860 err = ext4_handle_dirty_metadata(handle, inode, bh);
864 BUFFER_TRACE(bh, "not a new buffer");
874 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
875 ext4_lblk_t block, int create, int *err)
877 struct buffer_head *bh;
879 bh = ext4_getblk(handle, inode, block, create, err);
882 if (buffer_uptodate(bh))
884 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
886 if (buffer_uptodate(bh))
893 int ext4_walk_page_buffers(handle_t *handle,
894 struct buffer_head *head,
898 int (*fn)(handle_t *handle,
899 struct buffer_head *bh))
901 struct buffer_head *bh;
902 unsigned block_start, block_end;
903 unsigned blocksize = head->b_size;
905 struct buffer_head *next;
907 for (bh = head, block_start = 0;
908 ret == 0 && (bh != head || !block_start);
909 block_start = block_end, bh = next) {
910 next = bh->b_this_page;
911 block_end = block_start + blocksize;
912 if (block_end <= from || block_start >= to) {
913 if (partial && !buffer_uptodate(bh))
917 err = (*fn)(handle, bh);
925 * To preserve ordering, it is essential that the hole instantiation and
926 * the data write be encapsulated in a single transaction. We cannot
927 * close off a transaction and start a new one between the ext4_get_block()
928 * and the commit_write(). So doing the jbd2_journal_start at the start of
929 * prepare_write() is the right place.
931 * Also, this function can nest inside ext4_writepage(). In that case, we
932 * *know* that ext4_writepage() has generated enough buffer credits to do the
933 * whole page. So we won't block on the journal in that case, which is good,
934 * because the caller may be PF_MEMALLOC.
936 * By accident, ext4 can be reentered when a transaction is open via
937 * quota file writes. If we were to commit the transaction while thus
938 * reentered, there can be a deadlock - we would be holding a quota
939 * lock, and the commit would never complete if another thread had a
940 * transaction open and was blocking on the quota lock - a ranking
943 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
944 * will _not_ run commit under these circumstances because handle->h_ref
945 * is elevated. We'll still have enough credits for the tiny quotafile
948 int do_journal_get_write_access(handle_t *handle,
949 struct buffer_head *bh)
951 int dirty = buffer_dirty(bh);
954 if (!buffer_mapped(bh) || buffer_freed(bh))
957 * __block_write_begin() could have dirtied some buffers. Clean
958 * the dirty bit as jbd2_journal_get_write_access() could complain
959 * otherwise about fs integrity issues. Setting of the dirty bit
960 * by __block_write_begin() isn't a real problem here as we clear
961 * the bit before releasing a page lock and thus writeback cannot
962 * ever write the buffer.
965 clear_buffer_dirty(bh);
966 ret = ext4_journal_get_write_access(handle, bh);
968 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
972 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
973 struct buffer_head *bh_result, int create);
974 static int ext4_write_begin(struct file *file, struct address_space *mapping,
975 loff_t pos, unsigned len, unsigned flags,
976 struct page **pagep, void **fsdata)
978 struct inode *inode = mapping->host;
979 int ret, needed_blocks;
986 trace_ext4_write_begin(inode, pos, len, flags);
988 * Reserve one block more for addition to orphan list in case
989 * we allocate blocks but write fails for some reason
991 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
992 index = pos >> PAGE_CACHE_SHIFT;
993 from = pos & (PAGE_CACHE_SIZE - 1);
996 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
997 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1006 * grab_cache_page_write_begin() can take a long time if the
1007 * system is thrashing due to memory pressure, or if the page
1008 * is being written back. So grab it first before we start
1009 * the transaction handle. This also allows us to allocate
1010 * the page (if needed) without using GFP_NOFS.
1013 page = grab_cache_page_write_begin(mapping, index, flags);
1019 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1020 if (IS_ERR(handle)) {
1021 page_cache_release(page);
1022 return PTR_ERR(handle);
1026 if (page->mapping != mapping) {
1027 /* The page got truncated from under us */
1029 page_cache_release(page);
1030 ext4_journal_stop(handle);
1033 wait_on_page_writeback(page);
1035 if (ext4_should_dioread_nolock(inode))
1036 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1038 ret = __block_write_begin(page, pos, len, ext4_get_block);
1040 if (!ret && ext4_should_journal_data(inode)) {
1041 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1043 do_journal_get_write_access);
1049 * __block_write_begin may have instantiated a few blocks
1050 * outside i_size. Trim these off again. Don't need
1051 * i_size_read because we hold i_mutex.
1053 * Add inode to orphan list in case we crash before
1056 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1057 ext4_orphan_add(handle, inode);
1059 ext4_journal_stop(handle);
1060 if (pos + len > inode->i_size) {
1061 ext4_truncate_failed_write(inode);
1063 * If truncate failed early the inode might
1064 * still be on the orphan list; we need to
1065 * make sure the inode is removed from the
1066 * orphan list in that case.
1069 ext4_orphan_del(NULL, inode);
1072 if (ret == -ENOSPC &&
1073 ext4_should_retry_alloc(inode->i_sb, &retries))
1075 page_cache_release(page);
1082 /* For write_end() in data=journal mode */
1083 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1086 if (!buffer_mapped(bh) || buffer_freed(bh))
1088 set_buffer_uptodate(bh);
1089 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1090 clear_buffer_meta(bh);
1091 clear_buffer_prio(bh);
1096 * We need to pick up the new inode size which generic_commit_write gave us
1097 * `file' can be NULL - eg, when called from page_symlink().
1099 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1100 * buffers are managed internally.
1102 static int ext4_write_end(struct file *file,
1103 struct address_space *mapping,
1104 loff_t pos, unsigned len, unsigned copied,
1105 struct page *page, void *fsdata)
1107 handle_t *handle = ext4_journal_current_handle();
1108 struct inode *inode = mapping->host;
1110 int i_size_changed = 0;
1112 trace_ext4_write_end(inode, pos, len, copied);
1113 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1114 ret = ext4_jbd2_file_inode(handle, inode);
1117 page_cache_release(page);
1122 if (ext4_has_inline_data(inode))
1123 copied = ext4_write_inline_data_end(inode, pos, len,
1126 copied = block_write_end(file, mapping, pos,
1127 len, copied, page, fsdata);
1130 * No need to use i_size_read() here, the i_size
1131 * cannot change under us because we hole i_mutex.
1133 * But it's important to update i_size while still holding page lock:
1134 * page writeout could otherwise come in and zero beyond i_size.
1136 if (pos + copied > inode->i_size) {
1137 i_size_write(inode, pos + copied);
1141 if (pos + copied > EXT4_I(inode)->i_disksize) {
1142 /* We need to mark inode dirty even if
1143 * new_i_size is less that inode->i_size
1144 * but greater than i_disksize. (hint delalloc)
1146 ext4_update_i_disksize(inode, (pos + copied));
1150 page_cache_release(page);
1153 * Don't mark the inode dirty under page lock. First, it unnecessarily
1154 * makes the holding time of page lock longer. Second, it forces lock
1155 * ordering of page lock and transaction start for journaling
1159 ext4_mark_inode_dirty(handle, inode);
1163 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1164 /* if we have allocated more blocks and copied
1165 * less. We will have blocks allocated outside
1166 * inode->i_size. So truncate them
1168 ext4_orphan_add(handle, inode);
1170 ret2 = ext4_journal_stop(handle);
1174 if (pos + len > inode->i_size) {
1175 ext4_truncate_failed_write(inode);
1177 * If truncate failed early the inode might still be
1178 * on the orphan list; we need to make sure the inode
1179 * is removed from the orphan list in that case.
1182 ext4_orphan_del(NULL, inode);
1185 return ret ? ret : copied;
1188 static int ext4_journalled_write_end(struct file *file,
1189 struct address_space *mapping,
1190 loff_t pos, unsigned len, unsigned copied,
1191 struct page *page, void *fsdata)
1193 handle_t *handle = ext4_journal_current_handle();
1194 struct inode *inode = mapping->host;
1200 trace_ext4_journalled_write_end(inode, pos, len, copied);
1201 from = pos & (PAGE_CACHE_SIZE - 1);
1204 BUG_ON(!ext4_handle_valid(handle));
1206 if (ext4_has_inline_data(inode))
1207 copied = ext4_write_inline_data_end(inode, pos, len,
1211 if (!PageUptodate(page))
1213 page_zero_new_buffers(page, from+copied, to);
1216 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1217 to, &partial, write_end_fn);
1219 SetPageUptodate(page);
1221 new_i_size = pos + copied;
1222 if (new_i_size > inode->i_size)
1223 i_size_write(inode, pos+copied);
1224 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1225 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1226 if (new_i_size > EXT4_I(inode)->i_disksize) {
1227 ext4_update_i_disksize(inode, new_i_size);
1228 ret2 = ext4_mark_inode_dirty(handle, inode);
1234 page_cache_release(page);
1235 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1236 /* if we have allocated more blocks and copied
1237 * less. We will have blocks allocated outside
1238 * inode->i_size. So truncate them
1240 ext4_orphan_add(handle, inode);
1242 ret2 = ext4_journal_stop(handle);
1245 if (pos + len > inode->i_size) {
1246 ext4_truncate_failed_write(inode);
1248 * If truncate failed early the inode might still be
1249 * on the orphan list; we need to make sure the inode
1250 * is removed from the orphan list in that case.
1253 ext4_orphan_del(NULL, inode);
1256 return ret ? ret : copied;
1260 * Reserve a metadata for a single block located at lblock
1262 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1265 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1266 struct ext4_inode_info *ei = EXT4_I(inode);
1267 unsigned int md_needed;
1268 ext4_lblk_t save_last_lblock;
1272 * recalculate the amount of metadata blocks to reserve
1273 * in order to allocate nrblocks
1274 * worse case is one extent per block
1277 spin_lock(&ei->i_block_reservation_lock);
1279 * ext4_calc_metadata_amount() has side effects, which we have
1280 * to be prepared undo if we fail to claim space.
1282 save_len = ei->i_da_metadata_calc_len;
1283 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1284 md_needed = EXT4_NUM_B2C(sbi,
1285 ext4_calc_metadata_amount(inode, lblock));
1286 trace_ext4_da_reserve_space(inode, md_needed);
1289 * We do still charge estimated metadata to the sb though;
1290 * we cannot afford to run out of free blocks.
1292 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1293 ei->i_da_metadata_calc_len = save_len;
1294 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1295 spin_unlock(&ei->i_block_reservation_lock);
1296 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1302 ei->i_reserved_meta_blocks += md_needed;
1303 spin_unlock(&ei->i_block_reservation_lock);
1305 return 0; /* success */
1309 * Reserve a single cluster located at lblock
1311 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1314 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1315 struct ext4_inode_info *ei = EXT4_I(inode);
1316 unsigned int md_needed;
1318 ext4_lblk_t save_last_lblock;
1322 * We will charge metadata quota at writeout time; this saves
1323 * us from metadata over-estimation, though we may go over by
1324 * a small amount in the end. Here we just reserve for data.
1326 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1331 * recalculate the amount of metadata blocks to reserve
1332 * in order to allocate nrblocks
1333 * worse case is one extent per block
1336 spin_lock(&ei->i_block_reservation_lock);
1338 * ext4_calc_metadata_amount() has side effects, which we have
1339 * to be prepared undo if we fail to claim space.
1341 save_len = ei->i_da_metadata_calc_len;
1342 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1343 md_needed = EXT4_NUM_B2C(sbi,
1344 ext4_calc_metadata_amount(inode, lblock));
1345 trace_ext4_da_reserve_space(inode, md_needed);
1348 * We do still charge estimated metadata to the sb though;
1349 * we cannot afford to run out of free blocks.
1351 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1352 ei->i_da_metadata_calc_len = save_len;
1353 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1354 spin_unlock(&ei->i_block_reservation_lock);
1355 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1359 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1362 ei->i_reserved_data_blocks++;
1363 ei->i_reserved_meta_blocks += md_needed;
1364 spin_unlock(&ei->i_block_reservation_lock);
1366 return 0; /* success */
1369 static void ext4_da_release_space(struct inode *inode, int to_free)
1371 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1372 struct ext4_inode_info *ei = EXT4_I(inode);
1375 return; /* Nothing to release, exit */
1377 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1379 trace_ext4_da_release_space(inode, to_free);
1380 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1382 * if there aren't enough reserved blocks, then the
1383 * counter is messed up somewhere. Since this
1384 * function is called from invalidate page, it's
1385 * harmless to return without any action.
1387 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1388 "ino %lu, to_free %d with only %d reserved "
1389 "data blocks", inode->i_ino, to_free,
1390 ei->i_reserved_data_blocks);
1392 to_free = ei->i_reserved_data_blocks;
1394 ei->i_reserved_data_blocks -= to_free;
1396 if (ei->i_reserved_data_blocks == 0) {
1398 * We can release all of the reserved metadata blocks
1399 * only when we have written all of the delayed
1400 * allocation blocks.
1401 * Note that in case of bigalloc, i_reserved_meta_blocks,
1402 * i_reserved_data_blocks, etc. refer to number of clusters.
1404 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1405 ei->i_reserved_meta_blocks);
1406 ei->i_reserved_meta_blocks = 0;
1407 ei->i_da_metadata_calc_len = 0;
1410 /* update fs dirty data blocks counter */
1411 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1413 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1415 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1418 static void ext4_da_page_release_reservation(struct page *page,
1419 unsigned int offset,
1420 unsigned int length)
1423 struct buffer_head *head, *bh;
1424 unsigned int curr_off = 0;
1425 struct inode *inode = page->mapping->host;
1426 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1427 unsigned int stop = offset + length;
1431 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1433 head = page_buffers(page);
1436 unsigned int next_off = curr_off + bh->b_size;
1438 if (next_off > stop)
1441 if ((offset <= curr_off) && (buffer_delay(bh))) {
1443 clear_buffer_delay(bh);
1445 curr_off = next_off;
1446 } while ((bh = bh->b_this_page) != head);
1449 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1450 ext4_es_remove_extent(inode, lblk, to_release);
1453 /* If we have released all the blocks belonging to a cluster, then we
1454 * need to release the reserved space for that cluster. */
1455 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1456 while (num_clusters > 0) {
1457 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1458 ((num_clusters - 1) << sbi->s_cluster_bits);
1459 if (sbi->s_cluster_ratio == 1 ||
1460 !ext4_find_delalloc_cluster(inode, lblk))
1461 ext4_da_release_space(inode, 1);
1468 * Delayed allocation stuff
1472 * mpage_da_submit_io - walks through extent of pages and try to write
1473 * them with writepage() call back
1475 * @mpd->inode: inode
1476 * @mpd->first_page: first page of the extent
1477 * @mpd->next_page: page after the last page of the extent
1479 * By the time mpage_da_submit_io() is called we expect all blocks
1480 * to be allocated. this may be wrong if allocation failed.
1482 * As pages are already locked by write_cache_pages(), we can't use it
1484 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1485 struct ext4_map_blocks *map)
1487 struct pagevec pvec;
1488 unsigned long index, end;
1489 int ret = 0, err, nr_pages, i;
1490 struct inode *inode = mpd->inode;
1491 struct address_space *mapping = inode->i_mapping;
1492 loff_t size = i_size_read(inode);
1493 unsigned int len, block_start;
1494 struct buffer_head *bh, *page_bufs = NULL;
1495 sector_t pblock = 0, cur_logical = 0;
1496 struct ext4_io_submit io_submit;
1498 BUG_ON(mpd->next_page <= mpd->first_page);
1499 memset(&io_submit, 0, sizeof(io_submit));
1501 * We need to start from the first_page to the next_page - 1
1502 * to make sure we also write the mapped dirty buffer_heads.
1503 * If we look at mpd->b_blocknr we would only be looking
1504 * at the currently mapped buffer_heads.
1506 index = mpd->first_page;
1507 end = mpd->next_page - 1;
1509 pagevec_init(&pvec, 0);
1510 while (index <= end) {
1511 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1514 for (i = 0; i < nr_pages; i++) {
1516 struct page *page = pvec.pages[i];
1518 index = page->index;
1522 if (index == size >> PAGE_CACHE_SHIFT)
1523 len = size & ~PAGE_CACHE_MASK;
1525 len = PAGE_CACHE_SIZE;
1527 cur_logical = index << (PAGE_CACHE_SHIFT -
1529 pblock = map->m_pblk + (cur_logical -
1534 BUG_ON(!PageLocked(page));
1535 BUG_ON(PageWriteback(page));
1537 bh = page_bufs = page_buffers(page);
1540 if (map && (cur_logical >= map->m_lblk) &&
1541 (cur_logical <= (map->m_lblk +
1542 (map->m_len - 1)))) {
1543 if (buffer_delay(bh)) {
1544 clear_buffer_delay(bh);
1545 bh->b_blocknr = pblock;
1547 if (buffer_unwritten(bh) ||
1549 BUG_ON(bh->b_blocknr != pblock);
1550 if (map->m_flags & EXT4_MAP_UNINIT)
1551 set_buffer_uninit(bh);
1552 clear_buffer_unwritten(bh);
1556 * skip page if block allocation undone and
1559 if (ext4_bh_delay_or_unwritten(NULL, bh))
1561 bh = bh->b_this_page;
1562 block_start += bh->b_size;
1565 } while (bh != page_bufs);
1572 clear_page_dirty_for_io(page);
1573 err = ext4_bio_write_page(&io_submit, page, len,
1576 mpd->pages_written++;
1578 * In error case, we have to continue because
1579 * remaining pages are still locked
1584 pagevec_release(&pvec);
1586 ext4_io_submit(&io_submit);
1590 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1594 struct pagevec pvec;
1595 struct inode *inode = mpd->inode;
1596 struct address_space *mapping = inode->i_mapping;
1597 ext4_lblk_t start, last;
1599 index = mpd->first_page;
1600 end = mpd->next_page - 1;
1602 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1603 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1604 ext4_es_remove_extent(inode, start, last - start + 1);
1606 pagevec_init(&pvec, 0);
1607 while (index <= end) {
1608 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1611 for (i = 0; i < nr_pages; i++) {
1612 struct page *page = pvec.pages[i];
1613 if (page->index > end)
1615 BUG_ON(!PageLocked(page));
1616 BUG_ON(PageWriteback(page));
1617 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1618 ClearPageUptodate(page);
1621 index = pvec.pages[nr_pages - 1]->index + 1;
1622 pagevec_release(&pvec);
1627 static void ext4_print_free_blocks(struct inode *inode)
1629 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1630 struct super_block *sb = inode->i_sb;
1631 struct ext4_inode_info *ei = EXT4_I(inode);
1633 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1634 EXT4_C2B(EXT4_SB(inode->i_sb),
1635 ext4_count_free_clusters(sb)));
1636 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1637 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1638 (long long) EXT4_C2B(EXT4_SB(sb),
1639 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1640 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1641 (long long) EXT4_C2B(EXT4_SB(sb),
1642 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1643 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1644 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1645 ei->i_reserved_data_blocks);
1646 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1647 ei->i_reserved_meta_blocks);
1648 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1649 ei->i_allocated_meta_blocks);
1654 * mpage_da_map_and_submit - go through given space, map them
1655 * if necessary, and then submit them for I/O
1657 * @mpd - bh describing space
1659 * The function skips space we know is already mapped to disk blocks.
1662 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1664 int err, blks, get_blocks_flags;
1665 struct ext4_map_blocks map, *mapp = NULL;
1666 sector_t next = mpd->b_blocknr;
1667 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1668 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1669 handle_t *handle = NULL;
1672 * If the blocks are mapped already, or we couldn't accumulate
1673 * any blocks, then proceed immediately to the submission stage.
1675 if ((mpd->b_size == 0) ||
1676 ((mpd->b_state & (1 << BH_Mapped)) &&
1677 !(mpd->b_state & (1 << BH_Delay)) &&
1678 !(mpd->b_state & (1 << BH_Unwritten))))
1681 handle = ext4_journal_current_handle();
1685 * Call ext4_map_blocks() to allocate any delayed allocation
1686 * blocks, or to convert an uninitialized extent to be
1687 * initialized (in the case where we have written into
1688 * one or more preallocated blocks).
1690 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1691 * indicate that we are on the delayed allocation path. This
1692 * affects functions in many different parts of the allocation
1693 * call path. This flag exists primarily because we don't
1694 * want to change *many* call functions, so ext4_map_blocks()
1695 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1696 * inode's allocation semaphore is taken.
1698 * If the blocks in questions were delalloc blocks, set
1699 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1700 * variables are updated after the blocks have been allocated.
1703 map.m_len = max_blocks;
1705 * We're in delalloc path and it is possible that we're going to
1706 * need more metadata blocks than previously reserved. However
1707 * we must not fail because we're in writeback and there is
1708 * nothing we can do about it so it might result in data loss.
1709 * So use reserved blocks to allocate metadata if possible.
1711 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1712 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1713 if (ext4_should_dioread_nolock(mpd->inode))
1714 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1715 if (mpd->b_state & (1 << BH_Delay))
1716 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1719 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1721 struct super_block *sb = mpd->inode->i_sb;
1725 * If get block returns EAGAIN or ENOSPC and there
1726 * appears to be free blocks we will just let
1727 * mpage_da_submit_io() unlock all of the pages.
1732 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1738 * get block failure will cause us to loop in
1739 * writepages, because a_ops->writepage won't be able
1740 * to make progress. The page will be redirtied by
1741 * writepage and writepages will again try to write
1744 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1745 ext4_msg(sb, KERN_CRIT,
1746 "delayed block allocation failed for inode %lu "
1747 "at logical offset %llu with max blocks %zd "
1748 "with error %d", mpd->inode->i_ino,
1749 (unsigned long long) next,
1750 mpd->b_size >> mpd->inode->i_blkbits, err);
1751 ext4_msg(sb, KERN_CRIT,
1752 "This should not happen!! Data will be lost");
1754 ext4_print_free_blocks(mpd->inode);
1756 /* invalidate all the pages */
1757 ext4_da_block_invalidatepages(mpd);
1759 /* Mark this page range as having been completed */
1766 if (map.m_flags & EXT4_MAP_NEW) {
1767 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1770 for (i = 0; i < map.m_len; i++)
1771 unmap_underlying_metadata(bdev, map.m_pblk + i);
1775 * Update on-disk size along with block allocation.
1777 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1778 if (disksize > i_size_read(mpd->inode))
1779 disksize = i_size_read(mpd->inode);
1780 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1781 ext4_update_i_disksize(mpd->inode, disksize);
1782 err = ext4_mark_inode_dirty(handle, mpd->inode);
1784 ext4_error(mpd->inode->i_sb,
1785 "Failed to mark inode %lu dirty",
1790 mpage_da_submit_io(mpd, mapp);
1794 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1795 (1 << BH_Delay) | (1 << BH_Unwritten))
1798 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1800 * @mpd->lbh - extent of blocks
1801 * @logical - logical number of the block in the file
1802 * @b_state - b_state of the buffer head added
1804 * the function is used to collect contig. blocks in same state
1806 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1807 unsigned long b_state)
1810 int blkbits = mpd->inode->i_blkbits;
1811 int nrblocks = mpd->b_size >> blkbits;
1814 * XXX Don't go larger than mballoc is willing to allocate
1815 * This is a stopgap solution. We eventually need to fold
1816 * mpage_da_submit_io() into this function and then call
1817 * ext4_map_blocks() multiple times in a loop
1819 if (nrblocks >= (8*1024*1024 >> blkbits))
1822 /* check if the reserved journal credits might overflow */
1823 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1824 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1826 * With non-extent format we are limited by the journal
1827 * credit available. Total credit needed to insert
1828 * nrblocks contiguous blocks is dependent on the
1829 * nrblocks. So limit nrblocks.
1835 * First block in the extent
1837 if (mpd->b_size == 0) {
1838 mpd->b_blocknr = logical;
1839 mpd->b_size = 1 << blkbits;
1840 mpd->b_state = b_state & BH_FLAGS;
1844 next = mpd->b_blocknr + nrblocks;
1846 * Can we merge the block to our big extent?
1848 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1849 mpd->b_size += 1 << blkbits;
1855 * We couldn't merge the block to our extent, so we
1856 * need to flush current extent and start new one
1858 mpage_da_map_and_submit(mpd);
1862 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1864 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1868 * This function is grabs code from the very beginning of
1869 * ext4_map_blocks, but assumes that the caller is from delayed write
1870 * time. This function looks up the requested blocks and sets the
1871 * buffer delay bit under the protection of i_data_sem.
1873 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1874 struct ext4_map_blocks *map,
1875 struct buffer_head *bh)
1877 struct extent_status es;
1879 sector_t invalid_block = ~((sector_t) 0xffff);
1880 #ifdef ES_AGGRESSIVE_TEST
1881 struct ext4_map_blocks orig_map;
1883 memcpy(&orig_map, map, sizeof(*map));
1886 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1890 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1891 "logical block %lu\n", inode->i_ino, map->m_len,
1892 (unsigned long) map->m_lblk);
1894 /* Lookup extent status tree firstly */
1895 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1897 if (ext4_es_is_hole(&es)) {
1899 down_read((&EXT4_I(inode)->i_data_sem));
1904 * Delayed extent could be allocated by fallocate.
1905 * So we need to check it.
1907 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1908 map_bh(bh, inode->i_sb, invalid_block);
1910 set_buffer_delay(bh);
1914 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1915 retval = es.es_len - (iblock - es.es_lblk);
1916 if (retval > map->m_len)
1917 retval = map->m_len;
1918 map->m_len = retval;
1919 if (ext4_es_is_written(&es))
1920 map->m_flags |= EXT4_MAP_MAPPED;
1921 else if (ext4_es_is_unwritten(&es))
1922 map->m_flags |= EXT4_MAP_UNWRITTEN;
1926 #ifdef ES_AGGRESSIVE_TEST
1927 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1933 * Try to see if we can get the block without requesting a new
1934 * file system block.
1936 down_read((&EXT4_I(inode)->i_data_sem));
1937 if (ext4_has_inline_data(inode)) {
1939 * We will soon create blocks for this page, and let
1940 * us pretend as if the blocks aren't allocated yet.
1941 * In case of clusters, we have to handle the work
1942 * of mapping from cluster so that the reserved space
1943 * is calculated properly.
1945 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1946 ext4_find_delalloc_cluster(inode, map->m_lblk))
1947 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1949 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1950 retval = ext4_ext_map_blocks(NULL, inode, map,
1951 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1953 retval = ext4_ind_map_blocks(NULL, inode, map,
1954 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1960 * XXX: __block_prepare_write() unmaps passed block,
1964 * If the block was allocated from previously allocated cluster,
1965 * then we don't need to reserve it again. However we still need
1966 * to reserve metadata for every block we're going to write.
1968 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1969 ret = ext4_da_reserve_space(inode, iblock);
1971 /* not enough space to reserve */
1976 ret = ext4_da_reserve_metadata(inode, iblock);
1978 /* not enough space to reserve */
1984 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1985 ~0, EXTENT_STATUS_DELAYED);
1991 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1992 * and it should not appear on the bh->b_state.
1994 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1996 map_bh(bh, inode->i_sb, invalid_block);
1998 set_buffer_delay(bh);
1999 } else if (retval > 0) {
2001 unsigned long long status;
2003 #ifdef ES_AGGRESSIVE_TEST
2004 if (retval != map->m_len) {
2005 printk("ES len assertation failed for inode: %lu "
2006 "retval %d != map->m_len %d "
2007 "in %s (lookup)\n", inode->i_ino, retval,
2008 map->m_len, __func__);
2012 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
2013 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
2014 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2015 map->m_pblk, status);
2021 up_read((&EXT4_I(inode)->i_data_sem));
2027 * This is a special get_blocks_t callback which is used by
2028 * ext4_da_write_begin(). It will either return mapped block or
2029 * reserve space for a single block.
2031 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2032 * We also have b_blocknr = -1 and b_bdev initialized properly
2034 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2035 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2036 * initialized properly.
2038 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2039 struct buffer_head *bh, int create)
2041 struct ext4_map_blocks map;
2044 BUG_ON(create == 0);
2045 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2047 map.m_lblk = iblock;
2051 * first, we need to know whether the block is allocated already
2052 * preallocated blocks are unmapped but should treated
2053 * the same as allocated blocks.
2055 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2059 map_bh(bh, inode->i_sb, map.m_pblk);
2060 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2062 if (buffer_unwritten(bh)) {
2063 /* A delayed write to unwritten bh should be marked
2064 * new and mapped. Mapped ensures that we don't do
2065 * get_block multiple times when we write to the same
2066 * offset and new ensures that we do proper zero out
2067 * for partial write.
2070 set_buffer_mapped(bh);
2075 static int bget_one(handle_t *handle, struct buffer_head *bh)
2081 static int bput_one(handle_t *handle, struct buffer_head *bh)
2087 static int __ext4_journalled_writepage(struct page *page,
2090 struct address_space *mapping = page->mapping;
2091 struct inode *inode = mapping->host;
2092 struct buffer_head *page_bufs = NULL;
2093 handle_t *handle = NULL;
2094 int ret = 0, err = 0;
2095 int inline_data = ext4_has_inline_data(inode);
2096 struct buffer_head *inode_bh = NULL;
2098 ClearPageChecked(page);
2101 BUG_ON(page->index != 0);
2102 BUG_ON(len > ext4_get_max_inline_size(inode));
2103 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2104 if (inode_bh == NULL)
2107 page_bufs = page_buffers(page);
2112 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2115 /* As soon as we unlock the page, it can go away, but we have
2116 * references to buffers so we are safe */
2119 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2120 ext4_writepage_trans_blocks(inode));
2121 if (IS_ERR(handle)) {
2122 ret = PTR_ERR(handle);
2126 BUG_ON(!ext4_handle_valid(handle));
2129 ret = ext4_journal_get_write_access(handle, inode_bh);
2131 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2134 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2135 do_journal_get_write_access);
2137 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2142 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2143 err = ext4_journal_stop(handle);
2147 if (!ext4_has_inline_data(inode))
2148 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2150 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2157 * Note that we don't need to start a transaction unless we're journaling data
2158 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2159 * need to file the inode to the transaction's list in ordered mode because if
2160 * we are writing back data added by write(), the inode is already there and if
2161 * we are writing back data modified via mmap(), no one guarantees in which
2162 * transaction the data will hit the disk. In case we are journaling data, we
2163 * cannot start transaction directly because transaction start ranks above page
2164 * lock so we have to do some magic.
2166 * This function can get called via...
2167 * - ext4_da_writepages after taking page lock (have journal handle)
2168 * - journal_submit_inode_data_buffers (no journal handle)
2169 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2170 * - grab_page_cache when doing write_begin (have journal handle)
2172 * We don't do any block allocation in this function. If we have page with
2173 * multiple blocks we need to write those buffer_heads that are mapped. This
2174 * is important for mmaped based write. So if we do with blocksize 1K
2175 * truncate(f, 1024);
2176 * a = mmap(f, 0, 4096);
2178 * truncate(f, 4096);
2179 * we have in the page first buffer_head mapped via page_mkwrite call back
2180 * but other buffer_heads would be unmapped but dirty (dirty done via the
2181 * do_wp_page). So writepage should write the first block. If we modify
2182 * the mmap area beyond 1024 we will again get a page_fault and the
2183 * page_mkwrite callback will do the block allocation and mark the
2184 * buffer_heads mapped.
2186 * We redirty the page if we have any buffer_heads that is either delay or
2187 * unwritten in the page.
2189 * We can get recursively called as show below.
2191 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2194 * But since we don't do any block allocation we should not deadlock.
2195 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2197 static int ext4_writepage(struct page *page,
2198 struct writeback_control *wbc)
2203 struct buffer_head *page_bufs = NULL;
2204 struct inode *inode = page->mapping->host;
2205 struct ext4_io_submit io_submit;
2207 trace_ext4_writepage(page);
2208 size = i_size_read(inode);
2209 if (page->index == size >> PAGE_CACHE_SHIFT)
2210 len = size & ~PAGE_CACHE_MASK;
2212 len = PAGE_CACHE_SIZE;
2214 page_bufs = page_buffers(page);
2216 * We cannot do block allocation or other extent handling in this
2217 * function. If there are buffers needing that, we have to redirty
2218 * the page. But we may reach here when we do a journal commit via
2219 * journal_submit_inode_data_buffers() and in that case we must write
2220 * allocated buffers to achieve data=ordered mode guarantees.
2222 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2223 ext4_bh_delay_or_unwritten)) {
2224 redirty_page_for_writepage(wbc, page);
2225 if (current->flags & PF_MEMALLOC) {
2227 * For memory cleaning there's no point in writing only
2228 * some buffers. So just bail out. Warn if we came here
2229 * from direct reclaim.
2231 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2238 if (PageChecked(page) && ext4_should_journal_data(inode))
2240 * It's mmapped pagecache. Add buffers and journal it. There
2241 * doesn't seem much point in redirtying the page here.
2243 return __ext4_journalled_writepage(page, len);
2245 memset(&io_submit, 0, sizeof(io_submit));
2246 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2247 ext4_io_submit(&io_submit);
2252 * This is called via ext4_da_writepages() to
2253 * calculate the total number of credits to reserve to fit
2254 * a single extent allocation into a single transaction,
2255 * ext4_da_writpeages() will loop calling this before
2256 * the block allocation.
2259 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2261 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2264 * With non-extent format the journal credit needed to
2265 * insert nrblocks contiguous block is dependent on
2266 * number of contiguous block. So we will limit
2267 * number of contiguous block to a sane value
2269 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2270 (max_blocks > EXT4_MAX_TRANS_DATA))
2271 max_blocks = EXT4_MAX_TRANS_DATA;
2273 return ext4_chunk_trans_blocks(inode, max_blocks);
2277 * write_cache_pages_da - walk the list of dirty pages of the given
2278 * address space and accumulate pages that need writing, and call
2279 * mpage_da_map_and_submit to map a single contiguous memory region
2280 * and then write them.
2282 static int write_cache_pages_da(handle_t *handle,
2283 struct address_space *mapping,
2284 struct writeback_control *wbc,
2285 struct mpage_da_data *mpd,
2286 pgoff_t *done_index)
2288 struct buffer_head *bh, *head;
2289 struct inode *inode = mapping->host;
2290 struct pagevec pvec;
2291 unsigned int nr_pages;
2294 long nr_to_write = wbc->nr_to_write;
2295 int i, tag, ret = 0;
2297 memset(mpd, 0, sizeof(struct mpage_da_data));
2300 pagevec_init(&pvec, 0);
2301 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2302 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2304 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2305 tag = PAGECACHE_TAG_TOWRITE;
2307 tag = PAGECACHE_TAG_DIRTY;
2309 *done_index = index;
2310 while (index <= end) {
2311 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2312 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2316 for (i = 0; i < nr_pages; i++) {
2317 struct page *page = pvec.pages[i];
2320 * At this point, the page may be truncated or
2321 * invalidated (changing page->mapping to NULL), or
2322 * even swizzled back from swapper_space to tmpfs file
2323 * mapping. However, page->index will not change
2324 * because we have a reference on the page.
2326 if (page->index > end)
2329 *done_index = page->index + 1;
2332 * If we can't merge this page, and we have
2333 * accumulated an contiguous region, write it
2335 if ((mpd->next_page != page->index) &&
2336 (mpd->next_page != mpd->first_page)) {
2337 mpage_da_map_and_submit(mpd);
2338 goto ret_extent_tail;
2344 * If the page is no longer dirty, or its
2345 * mapping no longer corresponds to inode we
2346 * are writing (which means it has been
2347 * truncated or invalidated), or the page is
2348 * already under writeback and we are not
2349 * doing a data integrity writeback, skip the page
2351 if (!PageDirty(page) ||
2352 (PageWriteback(page) &&
2353 (wbc->sync_mode == WB_SYNC_NONE)) ||
2354 unlikely(page->mapping != mapping)) {
2359 wait_on_page_writeback(page);
2360 BUG_ON(PageWriteback(page));
2363 * If we have inline data and arrive here, it means that
2364 * we will soon create the block for the 1st page, so
2365 * we'd better clear the inline data here.
2367 if (ext4_has_inline_data(inode)) {
2368 BUG_ON(ext4_test_inode_state(inode,
2369 EXT4_STATE_MAY_INLINE_DATA));
2370 ext4_destroy_inline_data(handle, inode);
2373 if (mpd->next_page != page->index)
2374 mpd->first_page = page->index;
2375 mpd->next_page = page->index + 1;
2376 logical = (sector_t) page->index <<
2377 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2379 /* Add all dirty buffers to mpd */
2380 head = page_buffers(page);
2383 BUG_ON(buffer_locked(bh));
2385 * We need to try to allocate unmapped blocks
2386 * in the same page. Otherwise we won't make
2387 * progress with the page in ext4_writepage
2389 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2390 mpage_add_bh_to_extent(mpd, logical,
2393 goto ret_extent_tail;
2394 } else if (buffer_dirty(bh) &&
2395 buffer_mapped(bh)) {
2397 * mapped dirty buffer. We need to
2398 * update the b_state because we look
2399 * at b_state in mpage_da_map_blocks.
2400 * We don't update b_size because if we
2401 * find an unmapped buffer_head later
2402 * we need to use the b_state flag of
2405 if (mpd->b_size == 0)
2407 bh->b_state & BH_FLAGS;
2410 } while ((bh = bh->b_this_page) != head);
2412 if (nr_to_write > 0) {
2414 if (nr_to_write == 0 &&
2415 wbc->sync_mode == WB_SYNC_NONE)
2417 * We stop writing back only if we are
2418 * not doing integrity sync. In case of
2419 * integrity sync we have to keep going
2420 * because someone may be concurrently
2421 * dirtying pages, and we might have
2422 * synced a lot of newly appeared dirty
2423 * pages, but have not synced all of the
2429 pagevec_release(&pvec);
2434 ret = MPAGE_DA_EXTENT_TAIL;
2436 pagevec_release(&pvec);
2442 static int ext4_da_writepages(struct address_space *mapping,
2443 struct writeback_control *wbc)
2446 int range_whole = 0;
2447 handle_t *handle = NULL;
2448 struct mpage_da_data mpd;
2449 struct inode *inode = mapping->host;
2450 int pages_written = 0;
2451 unsigned int max_pages;
2452 int range_cyclic, cycled = 1, io_done = 0;
2453 int needed_blocks, ret = 0;
2454 long desired_nr_to_write, nr_to_writebump = 0;
2455 loff_t range_start = wbc->range_start;
2456 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2457 pgoff_t done_index = 0;
2459 struct blk_plug plug;
2461 trace_ext4_da_writepages(inode, wbc);
2464 * No pages to write? This is mainly a kludge to avoid starting
2465 * a transaction for special inodes like journal inode on last iput()
2466 * because that could violate lock ordering on umount
2468 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2472 * If the filesystem has aborted, it is read-only, so return
2473 * right away instead of dumping stack traces later on that
2474 * will obscure the real source of the problem. We test
2475 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2476 * the latter could be true if the filesystem is mounted
2477 * read-only, and in that case, ext4_da_writepages should
2478 * *never* be called, so if that ever happens, we would want
2481 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2484 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2487 range_cyclic = wbc->range_cyclic;
2488 if (wbc->range_cyclic) {
2489 index = mapping->writeback_index;
2492 wbc->range_start = index << PAGE_CACHE_SHIFT;
2493 wbc->range_end = LLONG_MAX;
2494 wbc->range_cyclic = 0;
2497 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2498 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2502 * This works around two forms of stupidity. The first is in
2503 * the writeback code, which caps the maximum number of pages
2504 * written to be 1024 pages. This is wrong on multiple
2505 * levels; different architectues have a different page size,
2506 * which changes the maximum amount of data which gets
2507 * written. Secondly, 4 megabytes is way too small. XFS
2508 * forces this value to be 16 megabytes by multiplying
2509 * nr_to_write parameter by four, and then relies on its
2510 * allocator to allocate larger extents to make them
2511 * contiguous. Unfortunately this brings us to the second
2512 * stupidity, which is that ext4's mballoc code only allocates
2513 * at most 2048 blocks. So we force contiguous writes up to
2514 * the number of dirty blocks in the inode, or
2515 * sbi->max_writeback_mb_bump whichever is smaller.
2517 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2518 if (!range_cyclic && range_whole) {
2519 if (wbc->nr_to_write == LONG_MAX)
2520 desired_nr_to_write = wbc->nr_to_write;
2522 desired_nr_to_write = wbc->nr_to_write * 8;
2524 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2526 if (desired_nr_to_write > max_pages)
2527 desired_nr_to_write = max_pages;
2529 if (wbc->nr_to_write < desired_nr_to_write) {
2530 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2531 wbc->nr_to_write = desired_nr_to_write;
2535 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2536 tag_pages_for_writeback(mapping, index, end);
2538 blk_start_plug(&plug);
2539 while (!ret && wbc->nr_to_write > 0) {
2542 * we insert one extent at a time. So we need
2543 * credit needed for single extent allocation.
2544 * journalled mode is currently not supported
2547 BUG_ON(ext4_should_journal_data(inode));
2548 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2550 /* start a new transaction*/
2551 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2553 if (IS_ERR(handle)) {
2554 ret = PTR_ERR(handle);
2555 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2556 "%ld pages, ino %lu; err %d", __func__,
2557 wbc->nr_to_write, inode->i_ino, ret);
2558 blk_finish_plug(&plug);
2559 goto out_writepages;
2563 * Now call write_cache_pages_da() to find the next
2564 * contiguous region of logical blocks that need
2565 * blocks to be allocated by ext4 and submit them.
2567 ret = write_cache_pages_da(handle, mapping,
2568 wbc, &mpd, &done_index);
2570 * If we have a contiguous extent of pages and we
2571 * haven't done the I/O yet, map the blocks and submit
2574 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2575 mpage_da_map_and_submit(&mpd);
2576 ret = MPAGE_DA_EXTENT_TAIL;
2578 trace_ext4_da_write_pages(inode, &mpd);
2579 wbc->nr_to_write -= mpd.pages_written;
2581 ext4_journal_stop(handle);
2583 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2584 /* commit the transaction which would
2585 * free blocks released in the transaction
2588 jbd2_journal_force_commit_nested(sbi->s_journal);
2590 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2592 * Got one extent now try with rest of the pages.
2593 * If mpd.retval is set -EIO, journal is aborted.
2594 * So we don't need to write any more.
2596 pages_written += mpd.pages_written;
2599 } else if (wbc->nr_to_write)
2601 * There is no more writeout needed
2602 * or we requested for a noblocking writeout
2603 * and we found the device congested
2607 blk_finish_plug(&plug);
2608 if (!io_done && !cycled) {
2611 wbc->range_start = index << PAGE_CACHE_SHIFT;
2612 wbc->range_end = mapping->writeback_index - 1;
2617 wbc->range_cyclic = range_cyclic;
2618 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2620 * set the writeback_index so that range_cyclic
2621 * mode will write it back later
2623 mapping->writeback_index = done_index;
2626 wbc->nr_to_write -= nr_to_writebump;
2627 wbc->range_start = range_start;
2628 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2632 static int ext4_nonda_switch(struct super_block *sb)
2634 s64 free_clusters, dirty_clusters;
2635 struct ext4_sb_info *sbi = EXT4_SB(sb);
2638 * switch to non delalloc mode if we are running low
2639 * on free block. The free block accounting via percpu
2640 * counters can get slightly wrong with percpu_counter_batch getting
2641 * accumulated on each CPU without updating global counters
2642 * Delalloc need an accurate free block accounting. So switch
2643 * to non delalloc when we are near to error range.
2646 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2648 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2650 * Start pushing delalloc when 1/2 of free blocks are dirty.
2652 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2653 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2655 if (2 * free_clusters < 3 * dirty_clusters ||
2656 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2658 * free block count is less than 150% of dirty blocks
2659 * or free blocks is less than watermark
2666 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2667 loff_t pos, unsigned len, unsigned flags,
2668 struct page **pagep, void **fsdata)
2670 int ret, retries = 0;
2673 struct inode *inode = mapping->host;
2676 index = pos >> PAGE_CACHE_SHIFT;
2678 if (ext4_nonda_switch(inode->i_sb)) {
2679 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2680 return ext4_write_begin(file, mapping, pos,
2681 len, flags, pagep, fsdata);
2683 *fsdata = (void *)0;
2684 trace_ext4_da_write_begin(inode, pos, len, flags);
2686 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2687 ret = ext4_da_write_inline_data_begin(mapping, inode,
2697 * grab_cache_page_write_begin() can take a long time if the
2698 * system is thrashing due to memory pressure, or if the page
2699 * is being written back. So grab it first before we start
2700 * the transaction handle. This also allows us to allocate
2701 * the page (if needed) without using GFP_NOFS.
2704 page = grab_cache_page_write_begin(mapping, index, flags);
2710 * With delayed allocation, we don't log the i_disksize update
2711 * if there is delayed block allocation. But we still need
2712 * to journalling the i_disksize update if writes to the end
2713 * of file which has an already mapped buffer.
2716 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2717 if (IS_ERR(handle)) {
2718 page_cache_release(page);
2719 return PTR_ERR(handle);
2723 if (page->mapping != mapping) {
2724 /* The page got truncated from under us */
2726 page_cache_release(page);
2727 ext4_journal_stop(handle);
2730 /* In case writeback began while the page was unlocked */
2731 wait_on_page_writeback(page);
2733 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2736 ext4_journal_stop(handle);
2738 * block_write_begin may have instantiated a few blocks
2739 * outside i_size. Trim these off again. Don't need
2740 * i_size_read because we hold i_mutex.
2742 if (pos + len > inode->i_size)
2743 ext4_truncate_failed_write(inode);
2745 if (ret == -ENOSPC &&
2746 ext4_should_retry_alloc(inode->i_sb, &retries))
2749 page_cache_release(page);
2758 * Check if we should update i_disksize
2759 * when write to the end of file but not require block allocation
2761 static int ext4_da_should_update_i_disksize(struct page *page,
2762 unsigned long offset)
2764 struct buffer_head *bh;
2765 struct inode *inode = page->mapping->host;
2769 bh = page_buffers(page);
2770 idx = offset >> inode->i_blkbits;
2772 for (i = 0; i < idx; i++)
2773 bh = bh->b_this_page;
2775 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2780 static int ext4_da_write_end(struct file *file,
2781 struct address_space *mapping,
2782 loff_t pos, unsigned len, unsigned copied,
2783 struct page *page, void *fsdata)
2785 struct inode *inode = mapping->host;
2787 handle_t *handle = ext4_journal_current_handle();
2789 unsigned long start, end;
2790 int write_mode = (int)(unsigned long)fsdata;
2792 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2793 return ext4_write_end(file, mapping, pos,
2794 len, copied, page, fsdata);
2796 trace_ext4_da_write_end(inode, pos, len, copied);
2797 start = pos & (PAGE_CACHE_SIZE - 1);
2798 end = start + copied - 1;
2801 * generic_write_end() will run mark_inode_dirty() if i_size
2802 * changes. So let's piggyback the i_disksize mark_inode_dirty
2805 new_i_size = pos + copied;
2806 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2807 if (ext4_has_inline_data(inode) ||
2808 ext4_da_should_update_i_disksize(page, end)) {
2809 down_write(&EXT4_I(inode)->i_data_sem);
2810 if (new_i_size > EXT4_I(inode)->i_disksize)
2811 EXT4_I(inode)->i_disksize = new_i_size;
2812 up_write(&EXT4_I(inode)->i_data_sem);
2813 /* We need to mark inode dirty even if
2814 * new_i_size is less that inode->i_size
2815 * bu greater than i_disksize.(hint delalloc)
2817 ext4_mark_inode_dirty(handle, inode);
2821 if (write_mode != CONVERT_INLINE_DATA &&
2822 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2823 ext4_has_inline_data(inode))
2824 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2827 ret2 = generic_write_end(file, mapping, pos, len, copied,
2833 ret2 = ext4_journal_stop(handle);
2837 return ret ? ret : copied;
2840 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2841 unsigned int length)
2844 * Drop reserved blocks
2846 BUG_ON(!PageLocked(page));
2847 if (!page_has_buffers(page))
2850 ext4_da_page_release_reservation(page, offset, length);
2853 ext4_invalidatepage(page, offset, length);
2859 * Force all delayed allocation blocks to be allocated for a given inode.
2861 int ext4_alloc_da_blocks(struct inode *inode)
2863 trace_ext4_alloc_da_blocks(inode);
2865 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2866 !EXT4_I(inode)->i_reserved_meta_blocks)
2870 * We do something simple for now. The filemap_flush() will
2871 * also start triggering a write of the data blocks, which is
2872 * not strictly speaking necessary (and for users of
2873 * laptop_mode, not even desirable). However, to do otherwise
2874 * would require replicating code paths in:
2876 * ext4_da_writepages() ->
2877 * write_cache_pages() ---> (via passed in callback function)
2878 * __mpage_da_writepage() -->
2879 * mpage_add_bh_to_extent()
2880 * mpage_da_map_blocks()
2882 * The problem is that write_cache_pages(), located in
2883 * mm/page-writeback.c, marks pages clean in preparation for
2884 * doing I/O, which is not desirable if we're not planning on
2887 * We could call write_cache_pages(), and then redirty all of
2888 * the pages by calling redirty_page_for_writepage() but that
2889 * would be ugly in the extreme. So instead we would need to
2890 * replicate parts of the code in the above functions,
2891 * simplifying them because we wouldn't actually intend to
2892 * write out the pages, but rather only collect contiguous
2893 * logical block extents, call the multi-block allocator, and
2894 * then update the buffer heads with the block allocations.
2896 * For now, though, we'll cheat by calling filemap_flush(),
2897 * which will map the blocks, and start the I/O, but not
2898 * actually wait for the I/O to complete.
2900 return filemap_flush(inode->i_mapping);
2904 * bmap() is special. It gets used by applications such as lilo and by
2905 * the swapper to find the on-disk block of a specific piece of data.
2907 * Naturally, this is dangerous if the block concerned is still in the
2908 * journal. If somebody makes a swapfile on an ext4 data-journaling
2909 * filesystem and enables swap, then they may get a nasty shock when the
2910 * data getting swapped to that swapfile suddenly gets overwritten by
2911 * the original zero's written out previously to the journal and
2912 * awaiting writeback in the kernel's buffer cache.
2914 * So, if we see any bmap calls here on a modified, data-journaled file,
2915 * take extra steps to flush any blocks which might be in the cache.
2917 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2919 struct inode *inode = mapping->host;
2924 * We can get here for an inline file via the FIBMAP ioctl
2926 if (ext4_has_inline_data(inode))
2929 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2930 test_opt(inode->i_sb, DELALLOC)) {
2932 * With delalloc we want to sync the file
2933 * so that we can make sure we allocate
2936 filemap_write_and_wait(mapping);
2939 if (EXT4_JOURNAL(inode) &&
2940 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2942 * This is a REALLY heavyweight approach, but the use of
2943 * bmap on dirty files is expected to be extremely rare:
2944 * only if we run lilo or swapon on a freshly made file
2945 * do we expect this to happen.
2947 * (bmap requires CAP_SYS_RAWIO so this does not
2948 * represent an unprivileged user DOS attack --- we'd be
2949 * in trouble if mortal users could trigger this path at
2952 * NB. EXT4_STATE_JDATA is not set on files other than
2953 * regular files. If somebody wants to bmap a directory
2954 * or symlink and gets confused because the buffer
2955 * hasn't yet been flushed to disk, they deserve
2956 * everything they get.
2959 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2960 journal = EXT4_JOURNAL(inode);
2961 jbd2_journal_lock_updates(journal);
2962 err = jbd2_journal_flush(journal);
2963 jbd2_journal_unlock_updates(journal);
2969 return generic_block_bmap(mapping, block, ext4_get_block);
2972 static int ext4_readpage(struct file *file, struct page *page)
2975 struct inode *inode = page->mapping->host;
2977 trace_ext4_readpage(page);
2979 if (ext4_has_inline_data(inode))
2980 ret = ext4_readpage_inline(inode, page);
2983 return mpage_readpage(page, ext4_get_block);
2989 ext4_readpages(struct file *file, struct address_space *mapping,
2990 struct list_head *pages, unsigned nr_pages)
2992 struct inode *inode = mapping->host;
2994 /* If the file has inline data, no need to do readpages. */
2995 if (ext4_has_inline_data(inode))
2998 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3001 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3002 unsigned int length)
3004 trace_ext4_invalidatepage(page, offset, length);
3006 /* No journalling happens on data buffers when this function is used */
3007 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3009 block_invalidatepage(page, offset, length);
3012 static int __ext4_journalled_invalidatepage(struct page *page,
3013 unsigned int offset,
3014 unsigned int length)
3016 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3018 trace_ext4_journalled_invalidatepage(page, offset, length);
3021 * If it's a full truncate we just forget about the pending dirtying
3023 if (offset == 0 && length == PAGE_CACHE_SIZE)
3024 ClearPageChecked(page);
3026 return jbd2_journal_invalidatepage(journal, page, offset, length);
3029 /* Wrapper for aops... */
3030 static void ext4_journalled_invalidatepage(struct page *page,
3031 unsigned int offset,
3032 unsigned int length)
3034 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3037 static int ext4_releasepage(struct page *page, gfp_t wait)
3039 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3041 trace_ext4_releasepage(page);
3043 /* Page has dirty journalled data -> cannot release */
3044 if (PageChecked(page))
3047 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3049 return try_to_free_buffers(page);
3053 * ext4_get_block used when preparing for a DIO write or buffer write.
3054 * We allocate an uinitialized extent if blocks haven't been allocated.
3055 * The extent will be converted to initialized after the IO is complete.
3057 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3058 struct buffer_head *bh_result, int create)
3060 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3061 inode->i_ino, create);
3062 return _ext4_get_block(inode, iblock, bh_result,
3063 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3066 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3067 struct buffer_head *bh_result, int create)
3069 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3070 inode->i_ino, create);
3071 return _ext4_get_block(inode, iblock, bh_result,
3072 EXT4_GET_BLOCKS_NO_LOCK);
3075 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3076 ssize_t size, void *private, int ret,
3079 struct inode *inode = file_inode(iocb->ki_filp);
3080 ext4_io_end_t *io_end = iocb->private;
3082 /* if not async direct IO or dio with 0 bytes write, just return */
3083 if (!io_end || !size)
3086 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3087 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3088 iocb->private, io_end->inode->i_ino, iocb, offset,
3091 iocb->private = NULL;
3093 /* if not aio dio with unwritten extents, just free io and return */
3094 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3095 ext4_free_io_end(io_end);
3097 inode_dio_done(inode);
3099 aio_complete(iocb, ret, 0);
3103 io_end->offset = offset;
3104 io_end->size = size;
3106 io_end->iocb = iocb;
3107 io_end->result = ret;
3110 ext4_add_complete_io(io_end);
3114 * For ext4 extent files, ext4 will do direct-io write to holes,
3115 * preallocated extents, and those write extend the file, no need to
3116 * fall back to buffered IO.
3118 * For holes, we fallocate those blocks, mark them as uninitialized
3119 * If those blocks were preallocated, we mark sure they are split, but
3120 * still keep the range to write as uninitialized.
3122 * The unwritten extents will be converted to written when DIO is completed.
3123 * For async direct IO, since the IO may still pending when return, we
3124 * set up an end_io call back function, which will do the conversion
3125 * when async direct IO completed.
3127 * If the O_DIRECT write will extend the file then add this inode to the
3128 * orphan list. So recovery will truncate it back to the original size
3129 * if the machine crashes during the write.
3132 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3133 const struct iovec *iov, loff_t offset,
3134 unsigned long nr_segs)
3136 struct file *file = iocb->ki_filp;
3137 struct inode *inode = file->f_mapping->host;
3139 size_t count = iov_length(iov, nr_segs);
3141 get_block_t *get_block_func = NULL;
3143 loff_t final_size = offset + count;
3145 /* Use the old path for reads and writes beyond i_size. */
3146 if (rw != WRITE || final_size > inode->i_size)
3147 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3149 BUG_ON(iocb->private == NULL);
3151 /* If we do a overwrite dio, i_mutex locking can be released */
3152 overwrite = *((int *)iocb->private);
3155 atomic_inc(&inode->i_dio_count);
3156 down_read(&EXT4_I(inode)->i_data_sem);
3157 mutex_unlock(&inode->i_mutex);
3161 * We could direct write to holes and fallocate.
3163 * Allocated blocks to fill the hole are marked as
3164 * uninitialized to prevent parallel buffered read to expose
3165 * the stale data before DIO complete the data IO.
3167 * As to previously fallocated extents, ext4 get_block will
3168 * just simply mark the buffer mapped but still keep the
3169 * extents uninitialized.
3171 * For non AIO case, we will convert those unwritten extents
3172 * to written after return back from blockdev_direct_IO.
3174 * For async DIO, the conversion needs to be deferred when the
3175 * IO is completed. The ext4 end_io callback function will be
3176 * called to take care of the conversion work. Here for async
3177 * case, we allocate an io_end structure to hook to the iocb.
3179 iocb->private = NULL;
3180 ext4_inode_aio_set(inode, NULL);
3181 if (!is_sync_kiocb(iocb)) {
3182 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3187 io_end->flag |= EXT4_IO_END_DIRECT;
3188 iocb->private = io_end;
3190 * we save the io structure for current async direct
3191 * IO, so that later ext4_map_blocks() could flag the
3192 * io structure whether there is a unwritten extents
3193 * needs to be converted when IO is completed.
3195 ext4_inode_aio_set(inode, io_end);
3199 get_block_func = ext4_get_block_write_nolock;
3201 get_block_func = ext4_get_block_write;
3202 dio_flags = DIO_LOCKING;
3204 ret = __blockdev_direct_IO(rw, iocb, inode,
3205 inode->i_sb->s_bdev, iov,
3213 ext4_inode_aio_set(inode, NULL);
3215 * The io_end structure takes a reference to the inode, that
3216 * structure needs to be destroyed and the reference to the
3217 * inode need to be dropped, when IO is complete, even with 0
3218 * byte write, or failed.
3220 * In the successful AIO DIO case, the io_end structure will
3221 * be destroyed and the reference to the inode will be dropped
3222 * after the end_io call back function is called.
3224 * In the case there is 0 byte write, or error case, since VFS
3225 * direct IO won't invoke the end_io call back function, we
3226 * need to free the end_io structure here.
3228 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3229 ext4_free_io_end(iocb->private);
3230 iocb->private = NULL;
3231 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3232 EXT4_STATE_DIO_UNWRITTEN)) {
3235 * for non AIO case, since the IO is already
3236 * completed, we could do the conversion right here
3238 err = ext4_convert_unwritten_extents(inode,
3242 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3246 /* take i_mutex locking again if we do a ovewrite dio */
3248 inode_dio_done(inode);
3249 up_read(&EXT4_I(inode)->i_data_sem);
3250 mutex_lock(&inode->i_mutex);
3256 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3257 const struct iovec *iov, loff_t offset,
3258 unsigned long nr_segs)
3260 struct file *file = iocb->ki_filp;
3261 struct inode *inode = file->f_mapping->host;
3265 * If we are doing data journalling we don't support O_DIRECT
3267 if (ext4_should_journal_data(inode))
3270 /* Let buffer I/O handle the inline data case. */
3271 if (ext4_has_inline_data(inode))
3274 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3275 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3276 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3278 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3279 trace_ext4_direct_IO_exit(inode, offset,
3280 iov_length(iov, nr_segs), rw, ret);
3285 * Pages can be marked dirty completely asynchronously from ext4's journalling
3286 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3287 * much here because ->set_page_dirty is called under VFS locks. The page is
3288 * not necessarily locked.
3290 * We cannot just dirty the page and leave attached buffers clean, because the
3291 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3292 * or jbddirty because all the journalling code will explode.
3294 * So what we do is to mark the page "pending dirty" and next time writepage
3295 * is called, propagate that into the buffers appropriately.
3297 static int ext4_journalled_set_page_dirty(struct page *page)
3299 SetPageChecked(page);
3300 return __set_page_dirty_nobuffers(page);
3303 static const struct address_space_operations ext4_aops = {
3304 .readpage = ext4_readpage,
3305 .readpages = ext4_readpages,
3306 .writepage = ext4_writepage,
3307 .write_begin = ext4_write_begin,
3308 .write_end = ext4_write_end,
3310 .invalidatepage = ext4_invalidatepage,
3311 .releasepage = ext4_releasepage,
3312 .direct_IO = ext4_direct_IO,
3313 .migratepage = buffer_migrate_page,
3314 .is_partially_uptodate = block_is_partially_uptodate,
3315 .error_remove_page = generic_error_remove_page,
3318 static const struct address_space_operations ext4_journalled_aops = {
3319 .readpage = ext4_readpage,
3320 .readpages = ext4_readpages,
3321 .writepage = ext4_writepage,
3322 .write_begin = ext4_write_begin,
3323 .write_end = ext4_journalled_write_end,
3324 .set_page_dirty = ext4_journalled_set_page_dirty,
3326 .invalidatepage = ext4_journalled_invalidatepage,
3327 .releasepage = ext4_releasepage,
3328 .direct_IO = ext4_direct_IO,
3329 .is_partially_uptodate = block_is_partially_uptodate,
3330 .error_remove_page = generic_error_remove_page,
3333 static const struct address_space_operations ext4_da_aops = {
3334 .readpage = ext4_readpage,
3335 .readpages = ext4_readpages,
3336 .writepage = ext4_writepage,
3337 .writepages = ext4_da_writepages,
3338 .write_begin = ext4_da_write_begin,
3339 .write_end = ext4_da_write_end,
3341 .invalidatepage = ext4_da_invalidatepage,
3342 .releasepage = ext4_releasepage,
3343 .direct_IO = ext4_direct_IO,
3344 .migratepage = buffer_migrate_page,
3345 .is_partially_uptodate = block_is_partially_uptodate,
3346 .error_remove_page = generic_error_remove_page,
3349 void ext4_set_aops(struct inode *inode)
3351 switch (ext4_inode_journal_mode(inode)) {
3352 case EXT4_INODE_ORDERED_DATA_MODE:
3353 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3355 case EXT4_INODE_WRITEBACK_DATA_MODE:
3356 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3358 case EXT4_INODE_JOURNAL_DATA_MODE:
3359 inode->i_mapping->a_ops = &ext4_journalled_aops;
3364 if (test_opt(inode->i_sb, DELALLOC))
3365 inode->i_mapping->a_ops = &ext4_da_aops;
3367 inode->i_mapping->a_ops = &ext4_aops;
3372 * ext4_discard_partial_page_buffers()
3373 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3374 * This function finds and locks the page containing the offset
3375 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3376 * Calling functions that already have the page locked should call
3377 * ext4_discard_partial_page_buffers_no_lock directly.
3379 int ext4_discard_partial_page_buffers(handle_t *handle,
3380 struct address_space *mapping, loff_t from,
3381 loff_t length, int flags)
3383 struct inode *inode = mapping->host;
3387 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3388 mapping_gfp_mask(mapping) & ~__GFP_FS);
3392 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3393 from, length, flags);
3396 page_cache_release(page);
3401 * ext4_discard_partial_page_buffers_no_lock()
3402 * Zeros a page range of length 'length' starting from offset 'from'.
3403 * Buffer heads that correspond to the block aligned regions of the
3404 * zeroed range will be unmapped. Unblock aligned regions
3405 * will have the corresponding buffer head mapped if needed so that
3406 * that region of the page can be updated with the partial zero out.
3408 * This function assumes that the page has already been locked. The
3409 * The range to be discarded must be contained with in the given page.
3410 * If the specified range exceeds the end of the page it will be shortened
3411 * to the end of the page that corresponds to 'from'. This function is
3412 * appropriate for updating a page and it buffer heads to be unmapped and
3413 * zeroed for blocks that have been either released, or are going to be
3416 * handle: The journal handle
3417 * inode: The files inode
3418 * page: A locked page that contains the offset "from"
3419 * from: The starting byte offset (from the beginning of the file)
3420 * to begin discarding
3421 * len: The length of bytes to discard
3422 * flags: Optional flags that may be used:
3424 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3425 * Only zero the regions of the page whose buffer heads
3426 * have already been unmapped. This flag is appropriate
3427 * for updating the contents of a page whose blocks may
3428 * have already been released, and we only want to zero
3429 * out the regions that correspond to those released blocks.
3431 * Returns zero on success or negative on failure.
3433 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3434 struct inode *inode, struct page *page, loff_t from,
3435 loff_t length, int flags)
3437 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3438 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3439 unsigned int blocksize, max, pos;
3441 struct buffer_head *bh;
3444 blocksize = inode->i_sb->s_blocksize;
3445 max = PAGE_CACHE_SIZE - offset;
3447 if (index != page->index)
3451 * correct length if it does not fall between
3452 * 'from' and the end of the page
3454 if (length > max || length < 0)
3457 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3459 if (!page_has_buffers(page))
3460 create_empty_buffers(page, blocksize, 0);
3462 /* Find the buffer that contains "offset" */
3463 bh = page_buffers(page);
3465 while (offset >= pos) {
3466 bh = bh->b_this_page;
3472 while (pos < offset + length) {
3473 unsigned int end_of_block, range_to_discard;
3477 /* The length of space left to zero and unmap */
3478 range_to_discard = offset + length - pos;
3480 /* The length of space until the end of the block */
3481 end_of_block = blocksize - (pos & (blocksize-1));
3484 * Do not unmap or zero past end of block
3485 * for this buffer head
3487 if (range_to_discard > end_of_block)
3488 range_to_discard = end_of_block;
3492 * Skip this buffer head if we are only zeroing unampped
3493 * regions of the page
3495 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3499 /* If the range is block aligned, unmap */
3500 if (range_to_discard == blocksize) {
3501 clear_buffer_dirty(bh);
3503 clear_buffer_mapped(bh);
3504 clear_buffer_req(bh);
3505 clear_buffer_new(bh);
3506 clear_buffer_delay(bh);
3507 clear_buffer_unwritten(bh);
3508 clear_buffer_uptodate(bh);
3509 zero_user(page, pos, range_to_discard);
3510 BUFFER_TRACE(bh, "Buffer discarded");
3515 * If this block is not completely contained in the range
3516 * to be discarded, then it is not going to be released. Because
3517 * we need to keep this block, we need to make sure this part
3518 * of the page is uptodate before we modify it by writeing
3519 * partial zeros on it.
3521 if (!buffer_mapped(bh)) {
3523 * Buffer head must be mapped before we can read
3526 BUFFER_TRACE(bh, "unmapped");
3527 ext4_get_block(inode, iblock, bh, 0);
3528 /* unmapped? It's a hole - nothing to do */
3529 if (!buffer_mapped(bh)) {
3530 BUFFER_TRACE(bh, "still unmapped");
3535 /* Ok, it's mapped. Make sure it's up-to-date */
3536 if (PageUptodate(page))
3537 set_buffer_uptodate(bh);
3539 if (!buffer_uptodate(bh)) {
3541 ll_rw_block(READ, 1, &bh);
3543 /* Uhhuh. Read error. Complain and punt.*/
3544 if (!buffer_uptodate(bh))
3548 if (ext4_should_journal_data(inode)) {
3549 BUFFER_TRACE(bh, "get write access");
3550 err = ext4_journal_get_write_access(handle, bh);
3555 zero_user(page, pos, range_to_discard);
3558 if (ext4_should_journal_data(inode)) {
3559 err = ext4_handle_dirty_metadata(handle, inode, bh);
3561 mark_buffer_dirty(bh);
3563 BUFFER_TRACE(bh, "Partial buffer zeroed");
3565 bh = bh->b_this_page;
3567 pos += range_to_discard;
3574 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3575 * up to the end of the block which corresponds to `from'.
3576 * This required during truncate. We need to physically zero the tail end
3577 * of that block so it doesn't yield old data if the file is later grown.
3579 int ext4_block_truncate_page(handle_t *handle,
3580 struct address_space *mapping, loff_t from)
3582 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3585 struct inode *inode = mapping->host;
3587 blocksize = inode->i_sb->s_blocksize;
3588 length = blocksize - (offset & (blocksize - 1));
3590 return ext4_block_zero_page_range(handle, mapping, from, length);
3594 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3595 * starting from file offset 'from'. The range to be zero'd must
3596 * be contained with in one block. If the specified range exceeds
3597 * the end of the block it will be shortened to end of the block
3598 * that cooresponds to 'from'
3600 int ext4_block_zero_page_range(handle_t *handle,
3601 struct address_space *mapping, loff_t from, loff_t length)
3603 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3604 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3605 unsigned blocksize, max, pos;
3607 struct inode *inode = mapping->host;
3608 struct buffer_head *bh;
3612 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3613 mapping_gfp_mask(mapping) & ~__GFP_FS);
3617 blocksize = inode->i_sb->s_blocksize;
3618 max = blocksize - (offset & (blocksize - 1));
3621 * correct length if it does not fall between
3622 * 'from' and the end of the block
3624 if (length > max || length < 0)
3627 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3629 if (!page_has_buffers(page))
3630 create_empty_buffers(page, blocksize, 0);
3632 /* Find the buffer that contains "offset" */
3633 bh = page_buffers(page);
3635 while (offset >= pos) {
3636 bh = bh->b_this_page;
3642 if (buffer_freed(bh)) {
3643 BUFFER_TRACE(bh, "freed: skip");
3647 if (!buffer_mapped(bh)) {
3648 BUFFER_TRACE(bh, "unmapped");
3649 ext4_get_block(inode, iblock, bh, 0);
3650 /* unmapped? It's a hole - nothing to do */
3651 if (!buffer_mapped(bh)) {
3652 BUFFER_TRACE(bh, "still unmapped");
3657 /* Ok, it's mapped. Make sure it's up-to-date */
3658 if (PageUptodate(page))
3659 set_buffer_uptodate(bh);
3661 if (!buffer_uptodate(bh)) {
3663 ll_rw_block(READ, 1, &bh);
3665 /* Uhhuh. Read error. Complain and punt. */
3666 if (!buffer_uptodate(bh))
3670 if (ext4_should_journal_data(inode)) {
3671 BUFFER_TRACE(bh, "get write access");
3672 err = ext4_journal_get_write_access(handle, bh);
3677 zero_user(page, offset, length);
3679 BUFFER_TRACE(bh, "zeroed end of block");
3682 if (ext4_should_journal_data(inode)) {
3683 err = ext4_handle_dirty_metadata(handle, inode, bh);
3685 mark_buffer_dirty(bh);
3686 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3687 err = ext4_jbd2_file_inode(handle, inode);
3692 page_cache_release(page);
3696 int ext4_can_truncate(struct inode *inode)
3698 if (S_ISREG(inode->i_mode))
3700 if (S_ISDIR(inode->i_mode))
3702 if (S_ISLNK(inode->i_mode))
3703 return !ext4_inode_is_fast_symlink(inode);
3708 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3709 * associated with the given offset and length
3711 * @inode: File inode
3712 * @offset: The offset where the hole will begin
3713 * @len: The length of the hole
3715 * Returns: 0 on success or negative on failure
3718 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3720 struct inode *inode = file_inode(file);
3721 struct super_block *sb = inode->i_sb;
3722 ext4_lblk_t first_block, stop_block;
3723 struct address_space *mapping = inode->i_mapping;
3724 loff_t first_page, last_page, page_len;
3725 loff_t first_page_offset, last_page_offset;
3727 unsigned int credits;
3730 if (!S_ISREG(inode->i_mode))
3733 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3734 /* TODO: Add support for bigalloc file systems */
3738 trace_ext4_punch_hole(inode, offset, length);
3741 * Write out all dirty pages to avoid race conditions
3742 * Then release them.
3744 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3745 ret = filemap_write_and_wait_range(mapping, offset,
3746 offset + length - 1);
3751 mutex_lock(&inode->i_mutex);
3752 /* It's not possible punch hole on append only file */
3753 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3757 if (IS_SWAPFILE(inode)) {
3762 /* No need to punch hole beyond i_size */
3763 if (offset >= inode->i_size)
3767 * If the hole extends beyond i_size, set the hole
3768 * to end after the page that contains i_size
3770 if (offset + length > inode->i_size) {
3771 length = inode->i_size +
3772 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3776 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
3777 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
3779 first_page_offset = first_page << PAGE_CACHE_SHIFT;
3780 last_page_offset = last_page << PAGE_CACHE_SHIFT;
3782 /* Now release the pages */
3783 if (last_page_offset > first_page_offset) {
3784 truncate_pagecache_range(inode, first_page_offset,
3785 last_page_offset - 1);
3788 /* Wait all existing dio workers, newcomers will block on i_mutex */
3789 ext4_inode_block_unlocked_dio(inode);
3790 ret = ext4_flush_unwritten_io(inode);
3793 inode_dio_wait(inode);
3795 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3796 credits = ext4_writepage_trans_blocks(inode);
3798 credits = ext4_blocks_for_truncate(inode);
3799 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3800 if (IS_ERR(handle)) {
3801 ret = PTR_ERR(handle);
3802 ext4_std_error(sb, ret);
3807 * Now we need to zero out the non-page-aligned data in the
3808 * pages at the start and tail of the hole, and unmap the
3809 * buffer heads for the block aligned regions of the page that
3810 * were completely zeroed.
3812 if (first_page > last_page) {
3814 * If the file space being truncated is contained
3815 * within a page just zero out and unmap the middle of
3818 ret = ext4_discard_partial_page_buffers(handle,
3819 mapping, offset, length, 0);
3825 * zero out and unmap the partial page that contains
3826 * the start of the hole
3828 page_len = first_page_offset - offset;
3830 ret = ext4_discard_partial_page_buffers(handle, mapping,
3831 offset, page_len, 0);
3837 * zero out and unmap the partial page that contains
3838 * the end of the hole
3840 page_len = offset + length - last_page_offset;
3842 ret = ext4_discard_partial_page_buffers(handle, mapping,
3843 last_page_offset, page_len, 0);
3850 * If i_size is contained in the last page, we need to
3851 * unmap and zero the partial page after i_size
3853 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
3854 inode->i_size % PAGE_CACHE_SIZE != 0) {
3855 page_len = PAGE_CACHE_SIZE -
3856 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3859 ret = ext4_discard_partial_page_buffers(handle,
3860 mapping, inode->i_size, page_len, 0);
3867 first_block = (offset + sb->s_blocksize - 1) >>
3868 EXT4_BLOCK_SIZE_BITS(sb);
3869 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3871 /* If there are no blocks to remove, return now */
3872 if (first_block >= stop_block)
3875 down_write(&EXT4_I(inode)->i_data_sem);
3876 ext4_discard_preallocations(inode);
3878 ret = ext4_es_remove_extent(inode, first_block,
3879 stop_block - first_block);
3881 up_write(&EXT4_I(inode)->i_data_sem);
3885 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3886 ret = ext4_ext_remove_space(inode, first_block,
3889 ret = ext4_free_hole_blocks(handle, inode, first_block,
3892 ext4_discard_preallocations(inode);
3893 up_write(&EXT4_I(inode)->i_data_sem);
3895 ext4_handle_sync(handle);
3896 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3897 ext4_mark_inode_dirty(handle, inode);
3899 ext4_journal_stop(handle);
3901 ext4_inode_resume_unlocked_dio(inode);
3903 mutex_unlock(&inode->i_mutex);
3910 * We block out ext4_get_block() block instantiations across the entire
3911 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3912 * simultaneously on behalf of the same inode.
3914 * As we work through the truncate and commit bits of it to the journal there
3915 * is one core, guiding principle: the file's tree must always be consistent on
3916 * disk. We must be able to restart the truncate after a crash.
3918 * The file's tree may be transiently inconsistent in memory (although it
3919 * probably isn't), but whenever we close off and commit a journal transaction,
3920 * the contents of (the filesystem + the journal) must be consistent and
3921 * restartable. It's pretty simple, really: bottom up, right to left (although
3922 * left-to-right works OK too).
3924 * Note that at recovery time, journal replay occurs *before* the restart of
3925 * truncate against the orphan inode list.
3927 * The committed inode has the new, desired i_size (which is the same as
3928 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3929 * that this inode's truncate did not complete and it will again call
3930 * ext4_truncate() to have another go. So there will be instantiated blocks
3931 * to the right of the truncation point in a crashed ext4 filesystem. But
3932 * that's fine - as long as they are linked from the inode, the post-crash
3933 * ext4_truncate() run will find them and release them.
3935 void ext4_truncate(struct inode *inode)
3937 struct ext4_inode_info *ei = EXT4_I(inode);
3938 unsigned int credits;
3940 struct address_space *mapping = inode->i_mapping;
3944 * There is a possibility that we're either freeing the inode
3945 * or it completely new indode. In those cases we might not
3946 * have i_mutex locked because it's not necessary.
3948 if (!(inode->i_state & (I_NEW|I_FREEING)))
3949 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3950 trace_ext4_truncate_enter(inode);
3952 if (!ext4_can_truncate(inode))
3955 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3957 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3958 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3960 if (ext4_has_inline_data(inode)) {
3963 ext4_inline_data_truncate(inode, &has_inline);
3969 * finish any pending end_io work so we won't run the risk of
3970 * converting any truncated blocks to initialized later
3972 ext4_flush_unwritten_io(inode);
3974 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3975 credits = ext4_writepage_trans_blocks(inode);
3977 credits = ext4_blocks_for_truncate(inode);
3979 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3980 if (IS_ERR(handle)) {
3981 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3985 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3986 page_len = PAGE_CACHE_SIZE -
3987 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3989 if (ext4_discard_partial_page_buffers(handle,
3990 mapping, inode->i_size, page_len, 0))
3995 * We add the inode to the orphan list, so that if this
3996 * truncate spans multiple transactions, and we crash, we will
3997 * resume the truncate when the filesystem recovers. It also
3998 * marks the inode dirty, to catch the new size.
4000 * Implication: the file must always be in a sane, consistent
4001 * truncatable state while each transaction commits.
4003 if (ext4_orphan_add(handle, inode))
4006 down_write(&EXT4_I(inode)->i_data_sem);
4008 ext4_discard_preallocations(inode);
4010 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4011 ext4_ext_truncate(handle, inode);
4013 ext4_ind_truncate(handle, inode);
4015 up_write(&ei->i_data_sem);
4018 ext4_handle_sync(handle);
4022 * If this was a simple ftruncate() and the file will remain alive,
4023 * then we need to clear up the orphan record which we created above.
4024 * However, if this was a real unlink then we were called by
4025 * ext4_delete_inode(), and we allow that function to clean up the
4026 * orphan info for us.
4029 ext4_orphan_del(handle, inode);
4031 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4032 ext4_mark_inode_dirty(handle, inode);
4033 ext4_journal_stop(handle);
4035 trace_ext4_truncate_exit(inode);
4039 * ext4_get_inode_loc returns with an extra refcount against the inode's
4040 * underlying buffer_head on success. If 'in_mem' is true, we have all
4041 * data in memory that is needed to recreate the on-disk version of this
4044 static int __ext4_get_inode_loc(struct inode *inode,
4045 struct ext4_iloc *iloc, int in_mem)
4047 struct ext4_group_desc *gdp;
4048 struct buffer_head *bh;
4049 struct super_block *sb = inode->i_sb;
4051 int inodes_per_block, inode_offset;
4054 if (!ext4_valid_inum(sb, inode->i_ino))
4057 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4058 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4063 * Figure out the offset within the block group inode table
4065 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4066 inode_offset = ((inode->i_ino - 1) %
4067 EXT4_INODES_PER_GROUP(sb));
4068 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4069 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4071 bh = sb_getblk(sb, block);
4074 if (!buffer_uptodate(bh)) {
4078 * If the buffer has the write error flag, we have failed
4079 * to write out another inode in the same block. In this
4080 * case, we don't have to read the block because we may
4081 * read the old inode data successfully.
4083 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4084 set_buffer_uptodate(bh);
4086 if (buffer_uptodate(bh)) {
4087 /* someone brought it uptodate while we waited */
4093 * If we have all information of the inode in memory and this
4094 * is the only valid inode in the block, we need not read the
4098 struct buffer_head *bitmap_bh;
4101 start = inode_offset & ~(inodes_per_block - 1);
4103 /* Is the inode bitmap in cache? */
4104 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4105 if (unlikely(!bitmap_bh))
4109 * If the inode bitmap isn't in cache then the
4110 * optimisation may end up performing two reads instead
4111 * of one, so skip it.
4113 if (!buffer_uptodate(bitmap_bh)) {
4117 for (i = start; i < start + inodes_per_block; i++) {
4118 if (i == inode_offset)
4120 if (ext4_test_bit(i, bitmap_bh->b_data))
4124 if (i == start + inodes_per_block) {
4125 /* all other inodes are free, so skip I/O */
4126 memset(bh->b_data, 0, bh->b_size);
4127 set_buffer_uptodate(bh);
4135 * If we need to do any I/O, try to pre-readahead extra
4136 * blocks from the inode table.
4138 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4139 ext4_fsblk_t b, end, table;
4141 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4143 table = ext4_inode_table(sb, gdp);
4144 /* s_inode_readahead_blks is always a power of 2 */
4145 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4149 num = EXT4_INODES_PER_GROUP(sb);
4150 if (ext4_has_group_desc_csum(sb))
4151 num -= ext4_itable_unused_count(sb, gdp);
4152 table += num / inodes_per_block;
4156 sb_breadahead(sb, b++);
4160 * There are other valid inodes in the buffer, this inode
4161 * has in-inode xattrs, or we don't have this inode in memory.
4162 * Read the block from disk.
4164 trace_ext4_load_inode(inode);
4166 bh->b_end_io = end_buffer_read_sync;
4167 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4169 if (!buffer_uptodate(bh)) {
4170 EXT4_ERROR_INODE_BLOCK(inode, block,
4171 "unable to read itable block");
4181 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4183 /* We have all inode data except xattrs in memory here. */
4184 return __ext4_get_inode_loc(inode, iloc,
4185 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4188 void ext4_set_inode_flags(struct inode *inode)
4190 unsigned int flags = EXT4_I(inode)->i_flags;
4192 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4193 if (flags & EXT4_SYNC_FL)
4194 inode->i_flags |= S_SYNC;
4195 if (flags & EXT4_APPEND_FL)
4196 inode->i_flags |= S_APPEND;
4197 if (flags & EXT4_IMMUTABLE_FL)
4198 inode->i_flags |= S_IMMUTABLE;
4199 if (flags & EXT4_NOATIME_FL)
4200 inode->i_flags |= S_NOATIME;
4201 if (flags & EXT4_DIRSYNC_FL)
4202 inode->i_flags |= S_DIRSYNC;
4205 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4206 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4208 unsigned int vfs_fl;
4209 unsigned long old_fl, new_fl;
4212 vfs_fl = ei->vfs_inode.i_flags;
4213 old_fl = ei->i_flags;
4214 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4215 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4217 if (vfs_fl & S_SYNC)
4218 new_fl |= EXT4_SYNC_FL;
4219 if (vfs_fl & S_APPEND)
4220 new_fl |= EXT4_APPEND_FL;
4221 if (vfs_fl & S_IMMUTABLE)
4222 new_fl |= EXT4_IMMUTABLE_FL;
4223 if (vfs_fl & S_NOATIME)
4224 new_fl |= EXT4_NOATIME_FL;
4225 if (vfs_fl & S_DIRSYNC)
4226 new_fl |= EXT4_DIRSYNC_FL;
4227 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4230 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4231 struct ext4_inode_info *ei)
4234 struct inode *inode = &(ei->vfs_inode);
4235 struct super_block *sb = inode->i_sb;
4237 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4238 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4239 /* we are using combined 48 bit field */
4240 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4241 le32_to_cpu(raw_inode->i_blocks_lo);
4242 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4243 /* i_blocks represent file system block size */
4244 return i_blocks << (inode->i_blkbits - 9);
4249 return le32_to_cpu(raw_inode->i_blocks_lo);
4253 static inline void ext4_iget_extra_inode(struct inode *inode,
4254 struct ext4_inode *raw_inode,
4255 struct ext4_inode_info *ei)
4257 __le32 *magic = (void *)raw_inode +
4258 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4259 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4260 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4261 ext4_find_inline_data_nolock(inode);
4263 EXT4_I(inode)->i_inline_off = 0;
4266 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4268 struct ext4_iloc iloc;
4269 struct ext4_inode *raw_inode;
4270 struct ext4_inode_info *ei;
4271 struct inode *inode;
4272 journal_t *journal = EXT4_SB(sb)->s_journal;
4278 inode = iget_locked(sb, ino);
4280 return ERR_PTR(-ENOMEM);
4281 if (!(inode->i_state & I_NEW))
4287 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4290 raw_inode = ext4_raw_inode(&iloc);
4292 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4293 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4294 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4295 EXT4_INODE_SIZE(inode->i_sb)) {
4296 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4297 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4298 EXT4_INODE_SIZE(inode->i_sb));
4303 ei->i_extra_isize = 0;
4305 /* Precompute checksum seed for inode metadata */
4306 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4307 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4308 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4310 __le32 inum = cpu_to_le32(inode->i_ino);
4311 __le32 gen = raw_inode->i_generation;
4312 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4314 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4318 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4319 EXT4_ERROR_INODE(inode, "checksum invalid");
4324 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4325 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4326 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4327 if (!(test_opt(inode->i_sb, NO_UID32))) {
4328 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4329 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4331 i_uid_write(inode, i_uid);
4332 i_gid_write(inode, i_gid);
4333 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4335 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4336 ei->i_inline_off = 0;
4337 ei->i_dir_start_lookup = 0;
4338 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4339 /* We now have enough fields to check if the inode was active or not.
4340 * This is needed because nfsd might try to access dead inodes
4341 * the test is that same one that e2fsck uses
4342 * NeilBrown 1999oct15
4344 if (inode->i_nlink == 0) {
4345 if ((inode->i_mode == 0 ||
4346 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4347 ino != EXT4_BOOT_LOADER_INO) {
4348 /* this inode is deleted */
4352 /* The only unlinked inodes we let through here have
4353 * valid i_mode and are being read by the orphan
4354 * recovery code: that's fine, we're about to complete
4355 * the process of deleting those.
4356 * OR it is the EXT4_BOOT_LOADER_INO which is
4357 * not initialized on a new filesystem. */
4359 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4360 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4361 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4362 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4364 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4365 inode->i_size = ext4_isize(raw_inode);
4366 ei->i_disksize = inode->i_size;
4368 ei->i_reserved_quota = 0;
4370 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4371 ei->i_block_group = iloc.block_group;
4372 ei->i_last_alloc_group = ~0;
4374 * NOTE! The in-memory inode i_data array is in little-endian order
4375 * even on big-endian machines: we do NOT byteswap the block numbers!
4377 for (block = 0; block < EXT4_N_BLOCKS; block++)
4378 ei->i_data[block] = raw_inode->i_block[block];
4379 INIT_LIST_HEAD(&ei->i_orphan);
4382 * Set transaction id's of transactions that have to be committed
4383 * to finish f[data]sync. We set them to currently running transaction
4384 * as we cannot be sure that the inode or some of its metadata isn't
4385 * part of the transaction - the inode could have been reclaimed and
4386 * now it is reread from disk.
4389 transaction_t *transaction;
4392 read_lock(&journal->j_state_lock);
4393 if (journal->j_running_transaction)
4394 transaction = journal->j_running_transaction;
4396 transaction = journal->j_committing_transaction;
4398 tid = transaction->t_tid;
4400 tid = journal->j_commit_sequence;
4401 read_unlock(&journal->j_state_lock);
4402 ei->i_sync_tid = tid;
4403 ei->i_datasync_tid = tid;
4406 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4407 if (ei->i_extra_isize == 0) {
4408 /* The extra space is currently unused. Use it. */
4409 ei->i_extra_isize = sizeof(struct ext4_inode) -
4410 EXT4_GOOD_OLD_INODE_SIZE;
4412 ext4_iget_extra_inode(inode, raw_inode, ei);
4416 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4417 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4418 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4419 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4421 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4422 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4423 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4425 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4429 if (ei->i_file_acl &&
4430 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4431 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4435 } else if (!ext4_has_inline_data(inode)) {
4436 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4437 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4438 (S_ISLNK(inode->i_mode) &&
4439 !ext4_inode_is_fast_symlink(inode))))
4440 /* Validate extent which is part of inode */
4441 ret = ext4_ext_check_inode(inode);
4442 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4443 (S_ISLNK(inode->i_mode) &&
4444 !ext4_inode_is_fast_symlink(inode))) {
4445 /* Validate block references which are part of inode */
4446 ret = ext4_ind_check_inode(inode);
4452 if (S_ISREG(inode->i_mode)) {
4453 inode->i_op = &ext4_file_inode_operations;
4454 inode->i_fop = &ext4_file_operations;
4455 ext4_set_aops(inode);
4456 } else if (S_ISDIR(inode->i_mode)) {
4457 inode->i_op = &ext4_dir_inode_operations;
4458 inode->i_fop = &ext4_dir_operations;
4459 } else if (S_ISLNK(inode->i_mode)) {
4460 if (ext4_inode_is_fast_symlink(inode)) {
4461 inode->i_op = &ext4_fast_symlink_inode_operations;
4462 nd_terminate_link(ei->i_data, inode->i_size,
4463 sizeof(ei->i_data) - 1);
4465 inode->i_op = &ext4_symlink_inode_operations;
4466 ext4_set_aops(inode);
4468 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4469 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4470 inode->i_op = &ext4_special_inode_operations;
4471 if (raw_inode->i_block[0])
4472 init_special_inode(inode, inode->i_mode,
4473 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4475 init_special_inode(inode, inode->i_mode,
4476 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4477 } else if (ino == EXT4_BOOT_LOADER_INO) {
4478 make_bad_inode(inode);
4481 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4485 ext4_set_inode_flags(inode);
4486 unlock_new_inode(inode);
4492 return ERR_PTR(ret);
4495 static int ext4_inode_blocks_set(handle_t *handle,
4496 struct ext4_inode *raw_inode,
4497 struct ext4_inode_info *ei)
4499 struct inode *inode = &(ei->vfs_inode);
4500 u64 i_blocks = inode->i_blocks;
4501 struct super_block *sb = inode->i_sb;
4503 if (i_blocks <= ~0U) {
4505 * i_blocks can be represented in a 32 bit variable
4506 * as multiple of 512 bytes
4508 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4509 raw_inode->i_blocks_high = 0;
4510 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4513 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4516 if (i_blocks <= 0xffffffffffffULL) {
4518 * i_blocks can be represented in a 48 bit variable
4519 * as multiple of 512 bytes
4521 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4522 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4523 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4525 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4526 /* i_block is stored in file system block size */
4527 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4528 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4529 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4535 * Post the struct inode info into an on-disk inode location in the
4536 * buffer-cache. This gobbles the caller's reference to the
4537 * buffer_head in the inode location struct.
4539 * The caller must have write access to iloc->bh.
4541 static int ext4_do_update_inode(handle_t *handle,
4542 struct inode *inode,
4543 struct ext4_iloc *iloc)
4545 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4546 struct ext4_inode_info *ei = EXT4_I(inode);
4547 struct buffer_head *bh = iloc->bh;
4548 int err = 0, rc, block;
4549 int need_datasync = 0;
4553 /* For fields not not tracking in the in-memory inode,
4554 * initialise them to zero for new inodes. */
4555 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4556 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4558 ext4_get_inode_flags(ei);
4559 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4560 i_uid = i_uid_read(inode);
4561 i_gid = i_gid_read(inode);
4562 if (!(test_opt(inode->i_sb, NO_UID32))) {
4563 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4564 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4566 * Fix up interoperability with old kernels. Otherwise, old inodes get
4567 * re-used with the upper 16 bits of the uid/gid intact
4570 raw_inode->i_uid_high =
4571 cpu_to_le16(high_16_bits(i_uid));
4572 raw_inode->i_gid_high =
4573 cpu_to_le16(high_16_bits(i_gid));
4575 raw_inode->i_uid_high = 0;
4576 raw_inode->i_gid_high = 0;
4579 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4580 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4581 raw_inode->i_uid_high = 0;
4582 raw_inode->i_gid_high = 0;
4584 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4586 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4587 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4588 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4589 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4591 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4593 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4594 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4595 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4596 cpu_to_le32(EXT4_OS_HURD))
4597 raw_inode->i_file_acl_high =
4598 cpu_to_le16(ei->i_file_acl >> 32);
4599 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4600 if (ei->i_disksize != ext4_isize(raw_inode)) {
4601 ext4_isize_set(raw_inode, ei->i_disksize);
4604 if (ei->i_disksize > 0x7fffffffULL) {
4605 struct super_block *sb = inode->i_sb;
4606 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4607 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4608 EXT4_SB(sb)->s_es->s_rev_level ==
4609 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4610 /* If this is the first large file
4611 * created, add a flag to the superblock.
4613 err = ext4_journal_get_write_access(handle,
4614 EXT4_SB(sb)->s_sbh);
4617 ext4_update_dynamic_rev(sb);
4618 EXT4_SET_RO_COMPAT_FEATURE(sb,
4619 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4620 ext4_handle_sync(handle);
4621 err = ext4_handle_dirty_super(handle, sb);
4624 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4625 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4626 if (old_valid_dev(inode->i_rdev)) {
4627 raw_inode->i_block[0] =
4628 cpu_to_le32(old_encode_dev(inode->i_rdev));
4629 raw_inode->i_block[1] = 0;
4631 raw_inode->i_block[0] = 0;
4632 raw_inode->i_block[1] =
4633 cpu_to_le32(new_encode_dev(inode->i_rdev));
4634 raw_inode->i_block[2] = 0;
4636 } else if (!ext4_has_inline_data(inode)) {
4637 for (block = 0; block < EXT4_N_BLOCKS; block++)
4638 raw_inode->i_block[block] = ei->i_data[block];
4641 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4642 if (ei->i_extra_isize) {
4643 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4644 raw_inode->i_version_hi =
4645 cpu_to_le32(inode->i_version >> 32);
4646 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4649 ext4_inode_csum_set(inode, raw_inode, ei);
4651 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4652 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4655 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4657 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4660 ext4_std_error(inode->i_sb, err);
4665 * ext4_write_inode()
4667 * We are called from a few places:
4669 * - Within generic_file_write() for O_SYNC files.
4670 * Here, there will be no transaction running. We wait for any running
4671 * transaction to commit.
4673 * - Within sys_sync(), kupdate and such.
4674 * We wait on commit, if tol to.
4676 * - Within prune_icache() (PF_MEMALLOC == true)
4677 * Here we simply return. We can't afford to block kswapd on the
4680 * In all cases it is actually safe for us to return without doing anything,
4681 * because the inode has been copied into a raw inode buffer in
4682 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4685 * Note that we are absolutely dependent upon all inode dirtiers doing the
4686 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4687 * which we are interested.
4689 * It would be a bug for them to not do this. The code:
4691 * mark_inode_dirty(inode)
4693 * inode->i_size = expr;
4695 * is in error because a kswapd-driven write_inode() could occur while
4696 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4697 * will no longer be on the superblock's dirty inode list.
4699 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4703 if (current->flags & PF_MEMALLOC)
4706 if (EXT4_SB(inode->i_sb)->s_journal) {
4707 if (ext4_journal_current_handle()) {
4708 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4713 if (wbc->sync_mode != WB_SYNC_ALL)
4716 err = ext4_force_commit(inode->i_sb);
4718 struct ext4_iloc iloc;
4720 err = __ext4_get_inode_loc(inode, &iloc, 0);
4723 if (wbc->sync_mode == WB_SYNC_ALL)
4724 sync_dirty_buffer(iloc.bh);
4725 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4726 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4727 "IO error syncing inode");
4736 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4737 * buffers that are attached to a page stradding i_size and are undergoing
4738 * commit. In that case we have to wait for commit to finish and try again.
4740 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4744 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4745 tid_t commit_tid = 0;
4748 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4750 * All buffers in the last page remain valid? Then there's nothing to
4751 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4754 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4757 page = find_lock_page(inode->i_mapping,
4758 inode->i_size >> PAGE_CACHE_SHIFT);
4761 ret = __ext4_journalled_invalidatepage(page, offset,
4762 PAGE_CACHE_SIZE - offset);
4764 page_cache_release(page);
4768 read_lock(&journal->j_state_lock);
4769 if (journal->j_committing_transaction)
4770 commit_tid = journal->j_committing_transaction->t_tid;
4771 read_unlock(&journal->j_state_lock);
4773 jbd2_log_wait_commit(journal, commit_tid);
4780 * Called from notify_change.
4782 * We want to trap VFS attempts to truncate the file as soon as
4783 * possible. In particular, we want to make sure that when the VFS
4784 * shrinks i_size, we put the inode on the orphan list and modify
4785 * i_disksize immediately, so that during the subsequent flushing of
4786 * dirty pages and freeing of disk blocks, we can guarantee that any
4787 * commit will leave the blocks being flushed in an unused state on
4788 * disk. (On recovery, the inode will get truncated and the blocks will
4789 * be freed, so we have a strong guarantee that no future commit will
4790 * leave these blocks visible to the user.)
4792 * Another thing we have to assure is that if we are in ordered mode
4793 * and inode is still attached to the committing transaction, we must
4794 * we start writeout of all the dirty pages which are being truncated.
4795 * This way we are sure that all the data written in the previous
4796 * transaction are already on disk (truncate waits for pages under
4799 * Called with inode->i_mutex down.
4801 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4803 struct inode *inode = dentry->d_inode;
4806 const unsigned int ia_valid = attr->ia_valid;
4808 error = inode_change_ok(inode, attr);
4812 if (is_quota_modification(inode, attr))
4813 dquot_initialize(inode);
4814 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4815 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4818 /* (user+group)*(old+new) structure, inode write (sb,
4819 * inode block, ? - but truncate inode update has it) */
4820 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4821 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4822 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4823 if (IS_ERR(handle)) {
4824 error = PTR_ERR(handle);
4827 error = dquot_transfer(inode, attr);
4829 ext4_journal_stop(handle);
4832 /* Update corresponding info in inode so that everything is in
4833 * one transaction */
4834 if (attr->ia_valid & ATTR_UID)
4835 inode->i_uid = attr->ia_uid;
4836 if (attr->ia_valid & ATTR_GID)
4837 inode->i_gid = attr->ia_gid;
4838 error = ext4_mark_inode_dirty(handle, inode);
4839 ext4_journal_stop(handle);
4842 if (attr->ia_valid & ATTR_SIZE) {
4844 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4845 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4847 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4852 if (S_ISREG(inode->i_mode) &&
4853 attr->ia_valid & ATTR_SIZE &&
4854 (attr->ia_size < inode->i_size)) {
4857 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4858 if (IS_ERR(handle)) {
4859 error = PTR_ERR(handle);
4862 if (ext4_handle_valid(handle)) {
4863 error = ext4_orphan_add(handle, inode);
4866 EXT4_I(inode)->i_disksize = attr->ia_size;
4867 rc = ext4_mark_inode_dirty(handle, inode);
4870 ext4_journal_stop(handle);
4872 if (ext4_should_order_data(inode)) {
4873 error = ext4_begin_ordered_truncate(inode,
4876 /* Do as much error cleanup as possible */
4877 handle = ext4_journal_start(inode,
4879 if (IS_ERR(handle)) {
4880 ext4_orphan_del(NULL, inode);
4883 ext4_orphan_del(handle, inode);
4885 ext4_journal_stop(handle);
4891 if (attr->ia_valid & ATTR_SIZE) {
4892 if (attr->ia_size != inode->i_size) {
4893 loff_t oldsize = inode->i_size;
4895 i_size_write(inode, attr->ia_size);
4897 * Blocks are going to be removed from the inode. Wait
4898 * for dio in flight. Temporarily disable
4899 * dioread_nolock to prevent livelock.
4902 if (!ext4_should_journal_data(inode)) {
4903 ext4_inode_block_unlocked_dio(inode);
4904 inode_dio_wait(inode);
4905 ext4_inode_resume_unlocked_dio(inode);
4907 ext4_wait_for_tail_page_commit(inode);
4910 * Truncate pagecache after we've waited for commit
4911 * in data=journal mode to make pages freeable.
4913 truncate_pagecache(inode, oldsize, inode->i_size);
4915 ext4_truncate(inode);
4919 setattr_copy(inode, attr);
4920 mark_inode_dirty(inode);
4924 * If the call to ext4_truncate failed to get a transaction handle at
4925 * all, we need to clean up the in-core orphan list manually.
4927 if (orphan && inode->i_nlink)
4928 ext4_orphan_del(NULL, inode);
4930 if (!rc && (ia_valid & ATTR_MODE))
4931 rc = ext4_acl_chmod(inode);
4934 ext4_std_error(inode->i_sb, error);
4940 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4943 struct inode *inode;
4944 unsigned long delalloc_blocks;
4946 inode = dentry->d_inode;
4947 generic_fillattr(inode, stat);
4950 * We can't update i_blocks if the block allocation is delayed
4951 * otherwise in the case of system crash before the real block
4952 * allocation is done, we will have i_blocks inconsistent with
4953 * on-disk file blocks.
4954 * We always keep i_blocks updated together with real
4955 * allocation. But to not confuse with user, stat
4956 * will return the blocks that include the delayed allocation
4957 * blocks for this file.
4959 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4960 EXT4_I(inode)->i_reserved_data_blocks);
4962 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4966 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4968 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4969 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4970 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4974 * Account for index blocks, block groups bitmaps and block group
4975 * descriptor blocks if modify datablocks and index blocks
4976 * worse case, the indexs blocks spread over different block groups
4978 * If datablocks are discontiguous, they are possible to spread over
4979 * different block groups too. If they are contiguous, with flexbg,
4980 * they could still across block group boundary.
4982 * Also account for superblock, inode, quota and xattr blocks
4984 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4986 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4992 * How many index blocks need to touch to modify nrblocks?
4993 * The "Chunk" flag indicating whether the nrblocks is
4994 * physically contiguous on disk
4996 * For Direct IO and fallocate, they calls get_block to allocate
4997 * one single extent at a time, so they could set the "Chunk" flag
4999 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5004 * Now let's see how many group bitmaps and group descriptors need
5014 if (groups > ngroups)
5016 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5017 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5019 /* bitmaps and block group descriptor blocks */
5020 ret += groups + gdpblocks;
5022 /* Blocks for super block, inode, quota and xattr blocks */
5023 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5029 * Calculate the total number of credits to reserve to fit
5030 * the modification of a single pages into a single transaction,
5031 * which may include multiple chunks of block allocations.
5033 * This could be called via ext4_write_begin()
5035 * We need to consider the worse case, when
5036 * one new block per extent.
5038 int ext4_writepage_trans_blocks(struct inode *inode)
5040 int bpp = ext4_journal_blocks_per_page(inode);
5043 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5045 /* Account for data blocks for journalled mode */
5046 if (ext4_should_journal_data(inode))
5052 * Calculate the journal credits for a chunk of data modification.
5054 * This is called from DIO, fallocate or whoever calling
5055 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5057 * journal buffers for data blocks are not included here, as DIO
5058 * and fallocate do no need to journal data buffers.
5060 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5062 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5066 * The caller must have previously called ext4_reserve_inode_write().
5067 * Give this, we know that the caller already has write access to iloc->bh.
5069 int ext4_mark_iloc_dirty(handle_t *handle,
5070 struct inode *inode, struct ext4_iloc *iloc)
5074 if (IS_I_VERSION(inode))
5075 inode_inc_iversion(inode);
5077 /* the do_update_inode consumes one bh->b_count */
5080 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5081 err = ext4_do_update_inode(handle, inode, iloc);
5087 * On success, We end up with an outstanding reference count against
5088 * iloc->bh. This _must_ be cleaned up later.
5092 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5093 struct ext4_iloc *iloc)
5097 err = ext4_get_inode_loc(inode, iloc);
5099 BUFFER_TRACE(iloc->bh, "get_write_access");
5100 err = ext4_journal_get_write_access(handle, iloc->bh);
5106 ext4_std_error(inode->i_sb, err);
5111 * Expand an inode by new_extra_isize bytes.
5112 * Returns 0 on success or negative error number on failure.
5114 static int ext4_expand_extra_isize(struct inode *inode,
5115 unsigned int new_extra_isize,
5116 struct ext4_iloc iloc,
5119 struct ext4_inode *raw_inode;
5120 struct ext4_xattr_ibody_header *header;
5122 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5125 raw_inode = ext4_raw_inode(&iloc);
5127 header = IHDR(inode, raw_inode);
5129 /* No extended attributes present */
5130 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5131 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5132 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5134 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5138 /* try to expand with EAs present */
5139 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5144 * What we do here is to mark the in-core inode as clean with respect to inode
5145 * dirtiness (it may still be data-dirty).
5146 * This means that the in-core inode may be reaped by prune_icache
5147 * without having to perform any I/O. This is a very good thing,
5148 * because *any* task may call prune_icache - even ones which
5149 * have a transaction open against a different journal.
5151 * Is this cheating? Not really. Sure, we haven't written the
5152 * inode out, but prune_icache isn't a user-visible syncing function.
5153 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5154 * we start and wait on commits.
5156 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5158 struct ext4_iloc iloc;
5159 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5160 static unsigned int mnt_count;
5164 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5165 err = ext4_reserve_inode_write(handle, inode, &iloc);
5166 if (ext4_handle_valid(handle) &&
5167 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5168 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5170 * We need extra buffer credits since we may write into EA block
5171 * with this same handle. If journal_extend fails, then it will
5172 * only result in a minor loss of functionality for that inode.
5173 * If this is felt to be critical, then e2fsck should be run to
5174 * force a large enough s_min_extra_isize.
5176 if ((jbd2_journal_extend(handle,
5177 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5178 ret = ext4_expand_extra_isize(inode,
5179 sbi->s_want_extra_isize,
5182 ext4_set_inode_state(inode,
5183 EXT4_STATE_NO_EXPAND);
5185 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5186 ext4_warning(inode->i_sb,
5187 "Unable to expand inode %lu. Delete"
5188 " some EAs or run e2fsck.",
5191 le16_to_cpu(sbi->s_es->s_mnt_count);
5197 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5202 * ext4_dirty_inode() is called from __mark_inode_dirty()
5204 * We're really interested in the case where a file is being extended.
5205 * i_size has been changed by generic_commit_write() and we thus need
5206 * to include the updated inode in the current transaction.
5208 * Also, dquot_alloc_block() will always dirty the inode when blocks
5209 * are allocated to the file.
5211 * If the inode is marked synchronous, we don't honour that here - doing
5212 * so would cause a commit on atime updates, which we don't bother doing.
5213 * We handle synchronous inodes at the highest possible level.
5215 void ext4_dirty_inode(struct inode *inode, int flags)
5219 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5223 ext4_mark_inode_dirty(handle, inode);
5225 ext4_journal_stop(handle);
5232 * Bind an inode's backing buffer_head into this transaction, to prevent
5233 * it from being flushed to disk early. Unlike
5234 * ext4_reserve_inode_write, this leaves behind no bh reference and
5235 * returns no iloc structure, so the caller needs to repeat the iloc
5236 * lookup to mark the inode dirty later.
5238 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5240 struct ext4_iloc iloc;
5244 err = ext4_get_inode_loc(inode, &iloc);
5246 BUFFER_TRACE(iloc.bh, "get_write_access");
5247 err = jbd2_journal_get_write_access(handle, iloc.bh);
5249 err = ext4_handle_dirty_metadata(handle,
5255 ext4_std_error(inode->i_sb, err);
5260 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5267 * We have to be very careful here: changing a data block's
5268 * journaling status dynamically is dangerous. If we write a
5269 * data block to the journal, change the status and then delete
5270 * that block, we risk forgetting to revoke the old log record
5271 * from the journal and so a subsequent replay can corrupt data.
5272 * So, first we make sure that the journal is empty and that
5273 * nobody is changing anything.
5276 journal = EXT4_JOURNAL(inode);
5279 if (is_journal_aborted(journal))
5281 /* We have to allocate physical blocks for delalloc blocks
5282 * before flushing journal. otherwise delalloc blocks can not
5283 * be allocated any more. even more truncate on delalloc blocks
5284 * could trigger BUG by flushing delalloc blocks in journal.
5285 * There is no delalloc block in non-journal data mode.
5287 if (val && test_opt(inode->i_sb, DELALLOC)) {
5288 err = ext4_alloc_da_blocks(inode);
5293 /* Wait for all existing dio workers */
5294 ext4_inode_block_unlocked_dio(inode);
5295 inode_dio_wait(inode);
5297 jbd2_journal_lock_updates(journal);
5300 * OK, there are no updates running now, and all cached data is
5301 * synced to disk. We are now in a completely consistent state
5302 * which doesn't have anything in the journal, and we know that
5303 * no filesystem updates are running, so it is safe to modify
5304 * the inode's in-core data-journaling state flag now.
5308 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5310 jbd2_journal_flush(journal);
5311 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5313 ext4_set_aops(inode);
5315 jbd2_journal_unlock_updates(journal);
5316 ext4_inode_resume_unlocked_dio(inode);
5318 /* Finally we can mark the inode as dirty. */
5320 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5322 return PTR_ERR(handle);
5324 err = ext4_mark_inode_dirty(handle, inode);
5325 ext4_handle_sync(handle);
5326 ext4_journal_stop(handle);
5327 ext4_std_error(inode->i_sb, err);
5332 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5334 return !buffer_mapped(bh);
5337 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5339 struct page *page = vmf->page;
5343 struct file *file = vma->vm_file;
5344 struct inode *inode = file_inode(file);
5345 struct address_space *mapping = inode->i_mapping;
5347 get_block_t *get_block;
5350 sb_start_pagefault(inode->i_sb);
5351 file_update_time(vma->vm_file);
5352 /* Delalloc case is easy... */
5353 if (test_opt(inode->i_sb, DELALLOC) &&
5354 !ext4_should_journal_data(inode) &&
5355 !ext4_nonda_switch(inode->i_sb)) {
5357 ret = __block_page_mkwrite(vma, vmf,
5358 ext4_da_get_block_prep);
5359 } while (ret == -ENOSPC &&
5360 ext4_should_retry_alloc(inode->i_sb, &retries));
5365 size = i_size_read(inode);
5366 /* Page got truncated from under us? */
5367 if (page->mapping != mapping || page_offset(page) > size) {
5369 ret = VM_FAULT_NOPAGE;
5373 if (page->index == size >> PAGE_CACHE_SHIFT)
5374 len = size & ~PAGE_CACHE_MASK;
5376 len = PAGE_CACHE_SIZE;
5378 * Return if we have all the buffers mapped. This avoids the need to do
5379 * journal_start/journal_stop which can block and take a long time
5381 if (page_has_buffers(page)) {
5382 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5384 ext4_bh_unmapped)) {
5385 /* Wait so that we don't change page under IO */
5386 wait_for_stable_page(page);
5387 ret = VM_FAULT_LOCKED;
5392 /* OK, we need to fill the hole... */
5393 if (ext4_should_dioread_nolock(inode))
5394 get_block = ext4_get_block_write;
5396 get_block = ext4_get_block;
5398 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5399 ext4_writepage_trans_blocks(inode));
5400 if (IS_ERR(handle)) {
5401 ret = VM_FAULT_SIGBUS;
5404 ret = __block_page_mkwrite(vma, vmf, get_block);
5405 if (!ret && ext4_should_journal_data(inode)) {
5406 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5407 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5409 ret = VM_FAULT_SIGBUS;
5410 ext4_journal_stop(handle);
5413 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5415 ext4_journal_stop(handle);
5416 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5419 ret = block_page_mkwrite_return(ret);
5421 sb_end_pagefault(inode->i_sb);