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>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
96 return provided == calculated;
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 trace_ext4_begin_ordered_truncate(inode, new_size);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode)->jinode)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
134 static void ext4_invalidatepage(struct page *page, unsigned long offset);
135 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
136 struct buffer_head *bh_result, int create);
137 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
138 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
139 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
140 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
141 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
142 struct inode *inode, struct page *page, loff_t from,
143 loff_t length, int flags);
146 * Test whether an inode is a fast symlink.
148 static int ext4_inode_is_fast_symlink(struct inode *inode)
150 int ea_blocks = EXT4_I(inode)->i_file_acl ?
151 (inode->i_sb->s_blocksize >> 9) : 0;
153 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
157 * Restart the transaction associated with *handle. This does a commit,
158 * so before we call here everything must be consistently dirtied against
161 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
167 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
168 * moment, get_block can be called only for blocks inside i_size since
169 * page cache has been already dropped and writes are blocked by
170 * i_mutex. So we can safely drop the i_data_sem here.
172 BUG_ON(EXT4_JOURNAL(inode) == NULL);
173 jbd_debug(2, "restarting handle %p\n", handle);
174 up_write(&EXT4_I(inode)->i_data_sem);
175 ret = ext4_journal_restart(handle, nblocks);
176 down_write(&EXT4_I(inode)->i_data_sem);
177 ext4_discard_preallocations(inode);
183 * Called at the last iput() if i_nlink is zero.
185 void ext4_evict_inode(struct inode *inode)
190 trace_ext4_evict_inode(inode);
192 ext4_ioend_wait(inode);
194 if (inode->i_nlink) {
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
210 * Note that directories do not have this problem because they
211 * don't use page cache.
213 if (ext4_should_journal_data(inode) &&
214 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
218 jbd2_log_start_commit(journal, commit_tid);
219 jbd2_log_wait_commit(journal, commit_tid);
220 filemap_write_and_wait(&inode->i_data);
222 truncate_inode_pages(&inode->i_data, 0);
226 if (!is_bad_inode(inode))
227 dquot_initialize(inode);
229 if (ext4_should_order_data(inode))
230 ext4_begin_ordered_truncate(inode, 0);
231 truncate_inode_pages(&inode->i_data, 0);
233 if (is_bad_inode(inode))
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
240 sb_start_intwrite(inode->i_sb);
241 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
242 if (IS_ERR(handle)) {
243 ext4_std_error(inode->i_sb, PTR_ERR(handle));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL, inode);
250 sb_end_intwrite(inode->i_sb);
255 ext4_handle_sync(handle);
257 err = ext4_mark_inode_dirty(handle, inode);
259 ext4_warning(inode->i_sb,
260 "couldn't mark inode dirty (err %d)", err);
264 ext4_truncate(inode);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle, 3)) {
273 err = ext4_journal_extend(handle, 3);
275 err = ext4_journal_restart(handle, 3);
277 ext4_warning(inode->i_sb,
278 "couldn't extend journal (err %d)", err);
280 ext4_journal_stop(handle);
281 ext4_orphan_del(NULL, inode);
282 sb_end_intwrite(inode->i_sb);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle, inode);
296 EXT4_I(inode)->i_dtime = get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle, inode))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode);
309 ext4_free_inode(handle, inode);
310 ext4_journal_stop(handle);
311 sb_end_intwrite(inode->i_sb);
314 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
318 qsize_t *ext4_get_reserved_space(struct inode *inode)
320 return &EXT4_I(inode)->i_reserved_quota;
325 * Calculate the number of metadata blocks need to reserve
326 * to allocate a block located at @lblock
328 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
330 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
331 return ext4_ext_calc_metadata_amount(inode, lblock);
333 return ext4_ind_calc_metadata_amount(inode, lblock);
337 * Called with i_data_sem down, which is important since we can call
338 * ext4_discard_preallocations() from here.
340 void ext4_da_update_reserve_space(struct inode *inode,
341 int used, int quota_claim)
343 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
344 struct ext4_inode_info *ei = EXT4_I(inode);
346 spin_lock(&ei->i_block_reservation_lock);
347 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
348 if (unlikely(used > ei->i_reserved_data_blocks)) {
349 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
350 "with only %d reserved data blocks",
351 __func__, inode->i_ino, used,
352 ei->i_reserved_data_blocks);
354 used = ei->i_reserved_data_blocks;
357 /* Update per-inode reservations */
358 ei->i_reserved_data_blocks -= used;
359 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
360 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
361 used + ei->i_allocated_meta_blocks);
362 ei->i_allocated_meta_blocks = 0;
364 if (ei->i_reserved_data_blocks == 0) {
366 * We can release all of the reserved metadata blocks
367 * only when we have written all of the delayed
370 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
371 ei->i_reserved_meta_blocks);
372 ei->i_reserved_meta_blocks = 0;
373 ei->i_da_metadata_calc_len = 0;
375 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
377 /* Update quota subsystem for data blocks */
379 dquot_claim_block(inode, EXT4_C2B(sbi, used));
382 * We did fallocate with an offset that is already delayed
383 * allocated. So on delayed allocated writeback we should
384 * not re-claim the quota for fallocated blocks.
386 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
390 * If we have done all the pending block allocations and if
391 * there aren't any writers on the inode, we can discard the
392 * inode's preallocations.
394 if ((ei->i_reserved_data_blocks == 0) &&
395 (atomic_read(&inode->i_writecount) == 0))
396 ext4_discard_preallocations(inode);
399 static int __check_block_validity(struct inode *inode, const char *func,
401 struct ext4_map_blocks *map)
403 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
405 ext4_error_inode(inode, func, line, map->m_pblk,
406 "lblock %lu mapped to illegal pblock "
407 "(length %d)", (unsigned long) map->m_lblk,
414 #define check_block_validity(inode, map) \
415 __check_block_validity((inode), __func__, __LINE__, (map))
418 * Return the number of contiguous dirty pages in a given inode
419 * starting at page frame idx.
421 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
422 unsigned int max_pages)
424 struct address_space *mapping = inode->i_mapping;
428 int i, nr_pages, done = 0;
432 pagevec_init(&pvec, 0);
435 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
437 (pgoff_t)PAGEVEC_SIZE);
440 for (i = 0; i < nr_pages; i++) {
441 struct page *page = pvec.pages[i];
442 struct buffer_head *bh, *head;
445 if (unlikely(page->mapping != mapping) ||
447 PageWriteback(page) ||
448 page->index != idx) {
453 if (page_has_buffers(page)) {
454 bh = head = page_buffers(page);
456 if (!buffer_delay(bh) &&
457 !buffer_unwritten(bh))
459 bh = bh->b_this_page;
460 } while (!done && (bh != head));
467 if (num >= max_pages) {
472 pagevec_release(&pvec);
478 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
480 static void set_buffers_da_mapped(struct inode *inode,
481 struct ext4_map_blocks *map)
483 struct address_space *mapping = inode->i_mapping;
488 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
489 end = (map->m_lblk + map->m_len - 1) >>
490 (PAGE_CACHE_SHIFT - inode->i_blkbits);
492 pagevec_init(&pvec, 0);
493 while (index <= end) {
494 nr_pages = pagevec_lookup(&pvec, mapping, index,
496 (pgoff_t)PAGEVEC_SIZE));
499 for (i = 0; i < nr_pages; i++) {
500 struct page *page = pvec.pages[i];
501 struct buffer_head *bh, *head;
503 if (unlikely(page->mapping != mapping) ||
507 if (page_has_buffers(page)) {
508 bh = head = page_buffers(page);
510 set_buffer_da_mapped(bh);
511 bh = bh->b_this_page;
512 } while (bh != head);
516 pagevec_release(&pvec);
521 * The ext4_map_blocks() function tries to look up the requested blocks,
522 * and returns if the blocks are already mapped.
524 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
525 * and store the allocated blocks in the result buffer head and mark it
528 * If file type is extents based, it will call ext4_ext_map_blocks(),
529 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
532 * On success, it returns the number of blocks being mapped or allocate.
533 * if create==0 and the blocks are pre-allocated and uninitialized block,
534 * the result buffer head is unmapped. If the create ==1, it will make sure
535 * the buffer head is mapped.
537 * It returns 0 if plain look up failed (blocks have not been allocated), in
538 * that case, buffer head is unmapped
540 * It returns the error in case of allocation failure.
542 int ext4_map_blocks(handle_t *handle, struct inode *inode,
543 struct ext4_map_blocks *map, int flags)
548 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
549 "logical block %lu\n", inode->i_ino, flags, map->m_len,
550 (unsigned long) map->m_lblk);
552 * Try to see if we can get the block without requesting a new
555 down_read((&EXT4_I(inode)->i_data_sem));
556 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
557 retval = ext4_ext_map_blocks(handle, inode, map, flags &
558 EXT4_GET_BLOCKS_KEEP_SIZE);
560 retval = ext4_ind_map_blocks(handle, inode, map, flags &
561 EXT4_GET_BLOCKS_KEEP_SIZE);
563 up_read((&EXT4_I(inode)->i_data_sem));
565 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
566 int ret = check_block_validity(inode, map);
571 /* If it is only a block(s) look up */
572 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
576 * Returns if the blocks have already allocated
578 * Note that if blocks have been preallocated
579 * ext4_ext_get_block() returns the create = 0
580 * with buffer head unmapped.
582 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
586 * When we call get_blocks without the create flag, the
587 * BH_Unwritten flag could have gotten set if the blocks
588 * requested were part of a uninitialized extent. We need to
589 * clear this flag now that we are committed to convert all or
590 * part of the uninitialized extent to be an initialized
591 * extent. This is because we need to avoid the combination
592 * of BH_Unwritten and BH_Mapped flags being simultaneously
593 * set on the buffer_head.
595 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
598 * New blocks allocate and/or writing to uninitialized extent
599 * will possibly result in updating i_data, so we take
600 * the write lock of i_data_sem, and call get_blocks()
601 * with create == 1 flag.
603 down_write((&EXT4_I(inode)->i_data_sem));
606 * if the caller is from delayed allocation writeout path
607 * we have already reserved fs blocks for allocation
608 * let the underlying get_block() function know to
609 * avoid double accounting
611 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
612 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
614 * We need to check for EXT4 here because migrate
615 * could have changed the inode type in between
617 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
618 retval = ext4_ext_map_blocks(handle, inode, map, flags);
620 retval = ext4_ind_map_blocks(handle, inode, map, flags);
622 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
624 * We allocated new blocks which will result in
625 * i_data's format changing. Force the migrate
626 * to fail by clearing migrate flags
628 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
632 * Update reserved blocks/metadata blocks after successful
633 * block allocation which had been deferred till now. We don't
634 * support fallocate for non extent files. So we can update
635 * reserve space here.
638 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
639 ext4_da_update_reserve_space(inode, retval, 1);
641 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
642 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
644 /* If we have successfully mapped the delayed allocated blocks,
645 * set the BH_Da_Mapped bit on them. Its important to do this
646 * under the protection of i_data_sem.
648 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
649 set_buffers_da_mapped(inode, map);
652 up_write((&EXT4_I(inode)->i_data_sem));
653 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
654 int ret = check_block_validity(inode, map);
661 /* Maximum number of blocks we map for direct IO at once. */
662 #define DIO_MAX_BLOCKS 4096
664 static int _ext4_get_block(struct inode *inode, sector_t iblock,
665 struct buffer_head *bh, int flags)
667 handle_t *handle = ext4_journal_current_handle();
668 struct ext4_map_blocks map;
669 int ret = 0, started = 0;
673 map.m_len = bh->b_size >> inode->i_blkbits;
675 if (flags && !handle) {
676 /* Direct IO write... */
677 if (map.m_len > DIO_MAX_BLOCKS)
678 map.m_len = DIO_MAX_BLOCKS;
679 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
680 handle = ext4_journal_start(inode, dio_credits);
681 if (IS_ERR(handle)) {
682 ret = PTR_ERR(handle);
688 ret = ext4_map_blocks(handle, inode, &map, flags);
690 map_bh(bh, inode->i_sb, map.m_pblk);
691 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
692 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
696 ext4_journal_stop(handle);
700 int ext4_get_block(struct inode *inode, sector_t iblock,
701 struct buffer_head *bh, int create)
703 return _ext4_get_block(inode, iblock, bh,
704 create ? EXT4_GET_BLOCKS_CREATE : 0);
708 * `handle' can be NULL if create is zero
710 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
711 ext4_lblk_t block, int create, int *errp)
713 struct ext4_map_blocks map;
714 struct buffer_head *bh;
717 J_ASSERT(handle != NULL || create == 0);
721 err = ext4_map_blocks(handle, inode, &map,
722 create ? EXT4_GET_BLOCKS_CREATE : 0);
730 bh = sb_getblk(inode->i_sb, map.m_pblk);
735 if (map.m_flags & EXT4_MAP_NEW) {
736 J_ASSERT(create != 0);
737 J_ASSERT(handle != NULL);
740 * Now that we do not always journal data, we should
741 * keep in mind whether this should always journal the
742 * new buffer as metadata. For now, regular file
743 * writes use ext4_get_block instead, so it's not a
747 BUFFER_TRACE(bh, "call get_create_access");
748 fatal = ext4_journal_get_create_access(handle, bh);
749 if (!fatal && !buffer_uptodate(bh)) {
750 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
751 set_buffer_uptodate(bh);
754 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
755 err = ext4_handle_dirty_metadata(handle, inode, bh);
759 BUFFER_TRACE(bh, "not a new buffer");
769 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
770 ext4_lblk_t block, int create, int *err)
772 struct buffer_head *bh;
774 bh = ext4_getblk(handle, inode, block, create, err);
777 if (buffer_uptodate(bh))
779 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
781 if (buffer_uptodate(bh))
788 static int walk_page_buffers(handle_t *handle,
789 struct buffer_head *head,
793 int (*fn)(handle_t *handle,
794 struct buffer_head *bh))
796 struct buffer_head *bh;
797 unsigned block_start, block_end;
798 unsigned blocksize = head->b_size;
800 struct buffer_head *next;
802 for (bh = head, block_start = 0;
803 ret == 0 && (bh != head || !block_start);
804 block_start = block_end, bh = next) {
805 next = bh->b_this_page;
806 block_end = block_start + blocksize;
807 if (block_end <= from || block_start >= to) {
808 if (partial && !buffer_uptodate(bh))
812 err = (*fn)(handle, bh);
820 * To preserve ordering, it is essential that the hole instantiation and
821 * the data write be encapsulated in a single transaction. We cannot
822 * close off a transaction and start a new one between the ext4_get_block()
823 * and the commit_write(). So doing the jbd2_journal_start at the start of
824 * prepare_write() is the right place.
826 * Also, this function can nest inside ext4_writepage() ->
827 * block_write_full_page(). In that case, we *know* that ext4_writepage()
828 * has generated enough buffer credits to do the whole page. So we won't
829 * block on the journal in that case, which is good, because the caller may
832 * By accident, ext4 can be reentered when a transaction is open via
833 * quota file writes. If we were to commit the transaction while thus
834 * reentered, there can be a deadlock - we would be holding a quota
835 * lock, and the commit would never complete if another thread had a
836 * transaction open and was blocking on the quota lock - a ranking
839 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
840 * will _not_ run commit under these circumstances because handle->h_ref
841 * is elevated. We'll still have enough credits for the tiny quotafile
844 static int do_journal_get_write_access(handle_t *handle,
845 struct buffer_head *bh)
847 int dirty = buffer_dirty(bh);
850 if (!buffer_mapped(bh) || buffer_freed(bh))
853 * __block_write_begin() could have dirtied some buffers. Clean
854 * the dirty bit as jbd2_journal_get_write_access() could complain
855 * otherwise about fs integrity issues. Setting of the dirty bit
856 * by __block_write_begin() isn't a real problem here as we clear
857 * the bit before releasing a page lock and thus writeback cannot
858 * ever write the buffer.
861 clear_buffer_dirty(bh);
862 ret = ext4_journal_get_write_access(handle, bh);
864 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
868 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
869 struct buffer_head *bh_result, int create);
870 static int ext4_write_begin(struct file *file, struct address_space *mapping,
871 loff_t pos, unsigned len, unsigned flags,
872 struct page **pagep, void **fsdata)
874 struct inode *inode = mapping->host;
875 int ret, needed_blocks;
882 trace_ext4_write_begin(inode, pos, len, flags);
884 * Reserve one block more for addition to orphan list in case
885 * we allocate blocks but write fails for some reason
887 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
888 index = pos >> PAGE_CACHE_SHIFT;
889 from = pos & (PAGE_CACHE_SIZE - 1);
893 handle = ext4_journal_start(inode, needed_blocks);
894 if (IS_ERR(handle)) {
895 ret = PTR_ERR(handle);
899 /* We cannot recurse into the filesystem as the transaction is already
901 flags |= AOP_FLAG_NOFS;
903 page = grab_cache_page_write_begin(mapping, index, flags);
905 ext4_journal_stop(handle);
911 if (ext4_should_dioread_nolock(inode))
912 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
914 ret = __block_write_begin(page, pos, len, ext4_get_block);
916 if (!ret && ext4_should_journal_data(inode)) {
917 ret = walk_page_buffers(handle, page_buffers(page),
918 from, to, NULL, do_journal_get_write_access);
923 page_cache_release(page);
925 * __block_write_begin may have instantiated a few blocks
926 * outside i_size. Trim these off again. Don't need
927 * i_size_read because we hold i_mutex.
929 * Add inode to orphan list in case we crash before
932 if (pos + len > inode->i_size && ext4_can_truncate(inode))
933 ext4_orphan_add(handle, inode);
935 ext4_journal_stop(handle);
936 if (pos + len > inode->i_size) {
937 ext4_truncate_failed_write(inode);
939 * If truncate failed early the inode might
940 * still be on the orphan list; we need to
941 * make sure the inode is removed from the
942 * orphan list in that case.
945 ext4_orphan_del(NULL, inode);
949 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
955 /* For write_end() in data=journal mode */
956 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
958 if (!buffer_mapped(bh) || buffer_freed(bh))
960 set_buffer_uptodate(bh);
961 return ext4_handle_dirty_metadata(handle, NULL, bh);
964 static int ext4_generic_write_end(struct file *file,
965 struct address_space *mapping,
966 loff_t pos, unsigned len, unsigned copied,
967 struct page *page, void *fsdata)
969 int i_size_changed = 0;
970 struct inode *inode = mapping->host;
971 handle_t *handle = ext4_journal_current_handle();
973 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
976 * No need to use i_size_read() here, the i_size
977 * cannot change under us because we hold i_mutex.
979 * But it's important to update i_size while still holding page lock:
980 * page writeout could otherwise come in and zero beyond i_size.
982 if (pos + copied > inode->i_size) {
983 i_size_write(inode, pos + copied);
987 if (pos + copied > EXT4_I(inode)->i_disksize) {
988 /* We need to mark inode dirty even if
989 * new_i_size is less that inode->i_size
990 * bu greater than i_disksize.(hint delalloc)
992 ext4_update_i_disksize(inode, (pos + copied));
996 page_cache_release(page);
999 * Don't mark the inode dirty under page lock. First, it unnecessarily
1000 * makes the holding time of page lock longer. Second, it forces lock
1001 * ordering of page lock and transaction start for journaling
1005 ext4_mark_inode_dirty(handle, inode);
1011 * We need to pick up the new inode size which generic_commit_write gave us
1012 * `file' can be NULL - eg, when called from page_symlink().
1014 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1015 * buffers are managed internally.
1017 static int ext4_ordered_write_end(struct file *file,
1018 struct address_space *mapping,
1019 loff_t pos, unsigned len, unsigned copied,
1020 struct page *page, void *fsdata)
1022 handle_t *handle = ext4_journal_current_handle();
1023 struct inode *inode = mapping->host;
1026 trace_ext4_ordered_write_end(inode, pos, len, copied);
1027 ret = ext4_jbd2_file_inode(handle, inode);
1030 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1033 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1034 /* if we have allocated more blocks and copied
1035 * less. We will have blocks allocated outside
1036 * inode->i_size. So truncate them
1038 ext4_orphan_add(handle, inode);
1043 page_cache_release(page);
1046 ret2 = ext4_journal_stop(handle);
1050 if (pos + len > inode->i_size) {
1051 ext4_truncate_failed_write(inode);
1053 * If truncate failed early the inode might still be
1054 * on the orphan list; we need to make sure the inode
1055 * is removed from the orphan list in that case.
1058 ext4_orphan_del(NULL, inode);
1062 return ret ? ret : copied;
1065 static int ext4_writeback_write_end(struct file *file,
1066 struct address_space *mapping,
1067 loff_t pos, unsigned len, unsigned copied,
1068 struct page *page, void *fsdata)
1070 handle_t *handle = ext4_journal_current_handle();
1071 struct inode *inode = mapping->host;
1074 trace_ext4_writeback_write_end(inode, pos, len, copied);
1075 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1078 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1079 /* if we have allocated more blocks and copied
1080 * less. We will have blocks allocated outside
1081 * inode->i_size. So truncate them
1083 ext4_orphan_add(handle, inode);
1088 ret2 = ext4_journal_stop(handle);
1092 if (pos + len > inode->i_size) {
1093 ext4_truncate_failed_write(inode);
1095 * If truncate failed early the inode might still be
1096 * on the orphan list; we need to make sure the inode
1097 * is removed from the orphan list in that case.
1100 ext4_orphan_del(NULL, inode);
1103 return ret ? ret : copied;
1106 static int ext4_journalled_write_end(struct file *file,
1107 struct address_space *mapping,
1108 loff_t pos, unsigned len, unsigned copied,
1109 struct page *page, void *fsdata)
1111 handle_t *handle = ext4_journal_current_handle();
1112 struct inode *inode = mapping->host;
1118 trace_ext4_journalled_write_end(inode, pos, len, copied);
1119 from = pos & (PAGE_CACHE_SIZE - 1);
1122 BUG_ON(!ext4_handle_valid(handle));
1125 if (!PageUptodate(page))
1127 page_zero_new_buffers(page, from+copied, to);
1130 ret = walk_page_buffers(handle, page_buffers(page), from,
1131 to, &partial, write_end_fn);
1133 SetPageUptodate(page);
1134 new_i_size = pos + copied;
1135 if (new_i_size > inode->i_size)
1136 i_size_write(inode, pos+copied);
1137 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1138 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1139 if (new_i_size > EXT4_I(inode)->i_disksize) {
1140 ext4_update_i_disksize(inode, new_i_size);
1141 ret2 = ext4_mark_inode_dirty(handle, inode);
1147 page_cache_release(page);
1148 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1149 /* if we have allocated more blocks and copied
1150 * less. We will have blocks allocated outside
1151 * inode->i_size. So truncate them
1153 ext4_orphan_add(handle, inode);
1155 ret2 = ext4_journal_stop(handle);
1158 if (pos + len > inode->i_size) {
1159 ext4_truncate_failed_write(inode);
1161 * If truncate failed early the inode might still be
1162 * on the orphan list; we need to make sure the inode
1163 * is removed from the orphan list in that case.
1166 ext4_orphan_del(NULL, inode);
1169 return ret ? ret : copied;
1173 * Reserve a single cluster located at lblock
1175 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1178 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1179 struct ext4_inode_info *ei = EXT4_I(inode);
1180 unsigned int md_needed;
1184 * recalculate the amount of metadata blocks to reserve
1185 * in order to allocate nrblocks
1186 * worse case is one extent per block
1189 spin_lock(&ei->i_block_reservation_lock);
1190 md_needed = EXT4_NUM_B2C(sbi,
1191 ext4_calc_metadata_amount(inode, lblock));
1192 trace_ext4_da_reserve_space(inode, md_needed);
1193 spin_unlock(&ei->i_block_reservation_lock);
1196 * We will charge metadata quota at writeout time; this saves
1197 * us from metadata over-estimation, though we may go over by
1198 * a small amount in the end. Here we just reserve for data.
1200 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1204 * We do still charge estimated metadata to the sb though;
1205 * we cannot afford to run out of free blocks.
1207 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1208 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1209 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1215 spin_lock(&ei->i_block_reservation_lock);
1216 ei->i_reserved_data_blocks++;
1217 ei->i_reserved_meta_blocks += md_needed;
1218 spin_unlock(&ei->i_block_reservation_lock);
1220 return 0; /* success */
1223 static void ext4_da_release_space(struct inode *inode, int to_free)
1225 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1226 struct ext4_inode_info *ei = EXT4_I(inode);
1229 return; /* Nothing to release, exit */
1231 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1233 trace_ext4_da_release_space(inode, to_free);
1234 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1236 * if there aren't enough reserved blocks, then the
1237 * counter is messed up somewhere. Since this
1238 * function is called from invalidate page, it's
1239 * harmless to return without any action.
1241 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1242 "ino %lu, to_free %d with only %d reserved "
1243 "data blocks", inode->i_ino, to_free,
1244 ei->i_reserved_data_blocks);
1246 to_free = ei->i_reserved_data_blocks;
1248 ei->i_reserved_data_blocks -= to_free;
1250 if (ei->i_reserved_data_blocks == 0) {
1252 * We can release all of the reserved metadata blocks
1253 * only when we have written all of the delayed
1254 * allocation blocks.
1255 * Note that in case of bigalloc, i_reserved_meta_blocks,
1256 * i_reserved_data_blocks, etc. refer to number of clusters.
1258 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1259 ei->i_reserved_meta_blocks);
1260 ei->i_reserved_meta_blocks = 0;
1261 ei->i_da_metadata_calc_len = 0;
1264 /* update fs dirty data blocks counter */
1265 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1267 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1269 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1272 static void ext4_da_page_release_reservation(struct page *page,
1273 unsigned long offset)
1276 struct buffer_head *head, *bh;
1277 unsigned int curr_off = 0;
1278 struct inode *inode = page->mapping->host;
1279 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1282 head = page_buffers(page);
1285 unsigned int next_off = curr_off + bh->b_size;
1287 if ((offset <= curr_off) && (buffer_delay(bh))) {
1289 clear_buffer_delay(bh);
1290 clear_buffer_da_mapped(bh);
1292 curr_off = next_off;
1293 } while ((bh = bh->b_this_page) != head);
1295 /* If we have released all the blocks belonging to a cluster, then we
1296 * need to release the reserved space for that cluster. */
1297 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1298 while (num_clusters > 0) {
1300 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1301 ((num_clusters - 1) << sbi->s_cluster_bits);
1302 if (sbi->s_cluster_ratio == 1 ||
1303 !ext4_find_delalloc_cluster(inode, lblk, 1))
1304 ext4_da_release_space(inode, 1);
1311 * Delayed allocation stuff
1315 * mpage_da_submit_io - walks through extent of pages and try to write
1316 * them with writepage() call back
1318 * @mpd->inode: inode
1319 * @mpd->first_page: first page of the extent
1320 * @mpd->next_page: page after the last page of the extent
1322 * By the time mpage_da_submit_io() is called we expect all blocks
1323 * to be allocated. this may be wrong if allocation failed.
1325 * As pages are already locked by write_cache_pages(), we can't use it
1327 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1328 struct ext4_map_blocks *map)
1330 struct pagevec pvec;
1331 unsigned long index, end;
1332 int ret = 0, err, nr_pages, i;
1333 struct inode *inode = mpd->inode;
1334 struct address_space *mapping = inode->i_mapping;
1335 loff_t size = i_size_read(inode);
1336 unsigned int len, block_start;
1337 struct buffer_head *bh, *page_bufs = NULL;
1338 int journal_data = ext4_should_journal_data(inode);
1339 sector_t pblock = 0, cur_logical = 0;
1340 struct ext4_io_submit io_submit;
1342 BUG_ON(mpd->next_page <= mpd->first_page);
1343 memset(&io_submit, 0, sizeof(io_submit));
1345 * We need to start from the first_page to the next_page - 1
1346 * to make sure we also write the mapped dirty buffer_heads.
1347 * If we look at mpd->b_blocknr we would only be looking
1348 * at the currently mapped buffer_heads.
1350 index = mpd->first_page;
1351 end = mpd->next_page - 1;
1353 pagevec_init(&pvec, 0);
1354 while (index <= end) {
1355 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1358 for (i = 0; i < nr_pages; i++) {
1359 int commit_write = 0, skip_page = 0;
1360 struct page *page = pvec.pages[i];
1362 index = page->index;
1366 if (index == size >> PAGE_CACHE_SHIFT)
1367 len = size & ~PAGE_CACHE_MASK;
1369 len = PAGE_CACHE_SIZE;
1371 cur_logical = index << (PAGE_CACHE_SHIFT -
1373 pblock = map->m_pblk + (cur_logical -
1378 BUG_ON(!PageLocked(page));
1379 BUG_ON(PageWriteback(page));
1382 * If the page does not have buffers (for
1383 * whatever reason), try to create them using
1384 * __block_write_begin. If this fails,
1385 * skip the page and move on.
1387 if (!page_has_buffers(page)) {
1388 if (__block_write_begin(page, 0, len,
1389 noalloc_get_block_write)) {
1397 bh = page_bufs = page_buffers(page);
1402 if (map && (cur_logical >= map->m_lblk) &&
1403 (cur_logical <= (map->m_lblk +
1404 (map->m_len - 1)))) {
1405 if (buffer_delay(bh)) {
1406 clear_buffer_delay(bh);
1407 bh->b_blocknr = pblock;
1409 if (buffer_da_mapped(bh))
1410 clear_buffer_da_mapped(bh);
1411 if (buffer_unwritten(bh) ||
1413 BUG_ON(bh->b_blocknr != pblock);
1414 if (map->m_flags & EXT4_MAP_UNINIT)
1415 set_buffer_uninit(bh);
1416 clear_buffer_unwritten(bh);
1420 * skip page if block allocation undone and
1423 if (ext4_bh_delay_or_unwritten(NULL, bh))
1425 bh = bh->b_this_page;
1426 block_start += bh->b_size;
1429 } while (bh != page_bufs);
1435 /* mark the buffer_heads as dirty & uptodate */
1436 block_commit_write(page, 0, len);
1438 clear_page_dirty_for_io(page);
1440 * Delalloc doesn't support data journalling,
1441 * but eventually maybe we'll lift this
1444 if (unlikely(journal_data && PageChecked(page)))
1445 err = __ext4_journalled_writepage(page, len);
1446 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1447 err = ext4_bio_write_page(&io_submit, page,
1449 else if (buffer_uninit(page_bufs)) {
1450 ext4_set_bh_endio(page_bufs, inode);
1451 err = block_write_full_page_endio(page,
1452 noalloc_get_block_write,
1453 mpd->wbc, ext4_end_io_buffer_write);
1455 err = block_write_full_page(page,
1456 noalloc_get_block_write, mpd->wbc);
1459 mpd->pages_written++;
1461 * In error case, we have to continue because
1462 * remaining pages are still locked
1467 pagevec_release(&pvec);
1469 ext4_io_submit(&io_submit);
1473 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1477 struct pagevec pvec;
1478 struct inode *inode = mpd->inode;
1479 struct address_space *mapping = inode->i_mapping;
1481 index = mpd->first_page;
1482 end = mpd->next_page - 1;
1483 while (index <= end) {
1484 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1487 for (i = 0; i < nr_pages; i++) {
1488 struct page *page = pvec.pages[i];
1489 if (page->index > end)
1491 BUG_ON(!PageLocked(page));
1492 BUG_ON(PageWriteback(page));
1493 block_invalidatepage(page, 0);
1494 ClearPageUptodate(page);
1497 index = pvec.pages[nr_pages - 1]->index + 1;
1498 pagevec_release(&pvec);
1503 static void ext4_print_free_blocks(struct inode *inode)
1505 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1506 struct super_block *sb = inode->i_sb;
1508 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1509 EXT4_C2B(EXT4_SB(inode->i_sb),
1510 ext4_count_free_clusters(inode->i_sb)));
1511 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1512 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1513 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1514 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1515 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1516 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1517 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1518 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1519 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1520 EXT4_I(inode)->i_reserved_data_blocks);
1521 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1522 EXT4_I(inode)->i_reserved_meta_blocks);
1527 * mpage_da_map_and_submit - go through given space, map them
1528 * if necessary, and then submit them for I/O
1530 * @mpd - bh describing space
1532 * The function skips space we know is already mapped to disk blocks.
1535 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1537 int err, blks, get_blocks_flags;
1538 struct ext4_map_blocks map, *mapp = NULL;
1539 sector_t next = mpd->b_blocknr;
1540 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1541 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1542 handle_t *handle = NULL;
1545 * If the blocks are mapped already, or we couldn't accumulate
1546 * any blocks, then proceed immediately to the submission stage.
1548 if ((mpd->b_size == 0) ||
1549 ((mpd->b_state & (1 << BH_Mapped)) &&
1550 !(mpd->b_state & (1 << BH_Delay)) &&
1551 !(mpd->b_state & (1 << BH_Unwritten))))
1554 handle = ext4_journal_current_handle();
1558 * Call ext4_map_blocks() to allocate any delayed allocation
1559 * blocks, or to convert an uninitialized extent to be
1560 * initialized (in the case where we have written into
1561 * one or more preallocated blocks).
1563 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1564 * indicate that we are on the delayed allocation path. This
1565 * affects functions in many different parts of the allocation
1566 * call path. This flag exists primarily because we don't
1567 * want to change *many* call functions, so ext4_map_blocks()
1568 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1569 * inode's allocation semaphore is taken.
1571 * If the blocks in questions were delalloc blocks, set
1572 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1573 * variables are updated after the blocks have been allocated.
1576 map.m_len = max_blocks;
1577 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1578 if (ext4_should_dioread_nolock(mpd->inode))
1579 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1580 if (mpd->b_state & (1 << BH_Delay))
1581 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1583 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1585 struct super_block *sb = mpd->inode->i_sb;
1589 * If get block returns EAGAIN or ENOSPC and there
1590 * appears to be free blocks we will just let
1591 * mpage_da_submit_io() unlock all of the pages.
1596 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1602 * get block failure will cause us to loop in
1603 * writepages, because a_ops->writepage won't be able
1604 * to make progress. The page will be redirtied by
1605 * writepage and writepages will again try to write
1608 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1609 ext4_msg(sb, KERN_CRIT,
1610 "delayed block allocation failed for inode %lu "
1611 "at logical offset %llu with max blocks %zd "
1612 "with error %d", mpd->inode->i_ino,
1613 (unsigned long long) next,
1614 mpd->b_size >> mpd->inode->i_blkbits, err);
1615 ext4_msg(sb, KERN_CRIT,
1616 "This should not happen!! Data will be lost\n");
1618 ext4_print_free_blocks(mpd->inode);
1620 /* invalidate all the pages */
1621 ext4_da_block_invalidatepages(mpd);
1623 /* Mark this page range as having been completed */
1630 if (map.m_flags & EXT4_MAP_NEW) {
1631 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1634 for (i = 0; i < map.m_len; i++)
1635 unmap_underlying_metadata(bdev, map.m_pblk + i);
1637 if (ext4_should_order_data(mpd->inode)) {
1638 err = ext4_jbd2_file_inode(handle, mpd->inode);
1640 /* Only if the journal is aborted */
1648 * Update on-disk size along with block allocation.
1650 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1651 if (disksize > i_size_read(mpd->inode))
1652 disksize = i_size_read(mpd->inode);
1653 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1654 ext4_update_i_disksize(mpd->inode, disksize);
1655 err = ext4_mark_inode_dirty(handle, mpd->inode);
1657 ext4_error(mpd->inode->i_sb,
1658 "Failed to mark inode %lu dirty",
1663 mpage_da_submit_io(mpd, mapp);
1667 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1668 (1 << BH_Delay) | (1 << BH_Unwritten))
1671 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1673 * @mpd->lbh - extent of blocks
1674 * @logical - logical number of the block in the file
1675 * @bh - bh of the block (used to access block's state)
1677 * the function is used to collect contig. blocks in same state
1679 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1680 sector_t logical, size_t b_size,
1681 unsigned long b_state)
1684 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1687 * XXX Don't go larger than mballoc is willing to allocate
1688 * This is a stopgap solution. We eventually need to fold
1689 * mpage_da_submit_io() into this function and then call
1690 * ext4_map_blocks() multiple times in a loop
1692 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1695 /* check if thereserved journal credits might overflow */
1696 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1697 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1699 * With non-extent format we are limited by the journal
1700 * credit available. Total credit needed to insert
1701 * nrblocks contiguous blocks is dependent on the
1702 * nrblocks. So limit nrblocks.
1705 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1706 EXT4_MAX_TRANS_DATA) {
1708 * Adding the new buffer_head would make it cross the
1709 * allowed limit for which we have journal credit
1710 * reserved. So limit the new bh->b_size
1712 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1713 mpd->inode->i_blkbits;
1714 /* we will do mpage_da_submit_io in the next loop */
1718 * First block in the extent
1720 if (mpd->b_size == 0) {
1721 mpd->b_blocknr = logical;
1722 mpd->b_size = b_size;
1723 mpd->b_state = b_state & BH_FLAGS;
1727 next = mpd->b_blocknr + nrblocks;
1729 * Can we merge the block to our big extent?
1731 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1732 mpd->b_size += b_size;
1738 * We couldn't merge the block to our extent, so we
1739 * need to flush current extent and start new one
1741 mpage_da_map_and_submit(mpd);
1745 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1747 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1751 * This function is grabs code from the very beginning of
1752 * ext4_map_blocks, but assumes that the caller is from delayed write
1753 * time. This function looks up the requested blocks and sets the
1754 * buffer delay bit under the protection of i_data_sem.
1756 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1757 struct ext4_map_blocks *map,
1758 struct buffer_head *bh)
1761 sector_t invalid_block = ~((sector_t) 0xffff);
1763 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1767 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1768 "logical block %lu\n", inode->i_ino, map->m_len,
1769 (unsigned long) map->m_lblk);
1771 * Try to see if we can get the block without requesting a new
1772 * file system block.
1774 down_read((&EXT4_I(inode)->i_data_sem));
1775 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1776 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1778 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1782 * XXX: __block_prepare_write() unmaps passed block,
1785 /* If the block was allocated from previously allocated cluster,
1786 * then we dont need to reserve it again. */
1787 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1788 retval = ext4_da_reserve_space(inode, iblock);
1790 /* not enough space to reserve */
1794 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1795 * and it should not appear on the bh->b_state.
1797 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1799 map_bh(bh, inode->i_sb, invalid_block);
1801 set_buffer_delay(bh);
1805 up_read((&EXT4_I(inode)->i_data_sem));
1811 * This is a special get_blocks_t callback which is used by
1812 * ext4_da_write_begin(). It will either return mapped block or
1813 * reserve space for a single block.
1815 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1816 * We also have b_blocknr = -1 and b_bdev initialized properly
1818 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1819 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1820 * initialized properly.
1822 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1823 struct buffer_head *bh, int create)
1825 struct ext4_map_blocks map;
1828 BUG_ON(create == 0);
1829 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1831 map.m_lblk = iblock;
1835 * first, we need to know whether the block is allocated already
1836 * preallocated blocks are unmapped but should treated
1837 * the same as allocated blocks.
1839 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1843 map_bh(bh, inode->i_sb, map.m_pblk);
1844 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1846 if (buffer_unwritten(bh)) {
1847 /* A delayed write to unwritten bh should be marked
1848 * new and mapped. Mapped ensures that we don't do
1849 * get_block multiple times when we write to the same
1850 * offset and new ensures that we do proper zero out
1851 * for partial write.
1854 set_buffer_mapped(bh);
1860 * This function is used as a standard get_block_t calback function
1861 * when there is no desire to allocate any blocks. It is used as a
1862 * callback function for block_write_begin() and block_write_full_page().
1863 * These functions should only try to map a single block at a time.
1865 * Since this function doesn't do block allocations even if the caller
1866 * requests it by passing in create=1, it is critically important that
1867 * any caller checks to make sure that any buffer heads are returned
1868 * by this function are either all already mapped or marked for
1869 * delayed allocation before calling block_write_full_page(). Otherwise,
1870 * b_blocknr could be left unitialized, and the page write functions will
1871 * be taken by surprise.
1873 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1874 struct buffer_head *bh_result, int create)
1876 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1877 return _ext4_get_block(inode, iblock, bh_result, 0);
1880 static int bget_one(handle_t *handle, struct buffer_head *bh)
1886 static int bput_one(handle_t *handle, struct buffer_head *bh)
1892 static int __ext4_journalled_writepage(struct page *page,
1895 struct address_space *mapping = page->mapping;
1896 struct inode *inode = mapping->host;
1897 struct buffer_head *page_bufs;
1898 handle_t *handle = NULL;
1902 ClearPageChecked(page);
1903 page_bufs = page_buffers(page);
1905 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1906 /* As soon as we unlock the page, it can go away, but we have
1907 * references to buffers so we are safe */
1910 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1911 if (IS_ERR(handle)) {
1912 ret = PTR_ERR(handle);
1916 BUG_ON(!ext4_handle_valid(handle));
1918 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1919 do_journal_get_write_access);
1921 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1925 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1926 err = ext4_journal_stop(handle);
1930 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1931 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1936 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1937 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1940 * Note that we don't need to start a transaction unless we're journaling data
1941 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1942 * need to file the inode to the transaction's list in ordered mode because if
1943 * we are writing back data added by write(), the inode is already there and if
1944 * we are writing back data modified via mmap(), no one guarantees in which
1945 * transaction the data will hit the disk. In case we are journaling data, we
1946 * cannot start transaction directly because transaction start ranks above page
1947 * lock so we have to do some magic.
1949 * This function can get called via...
1950 * - ext4_da_writepages after taking page lock (have journal handle)
1951 * - journal_submit_inode_data_buffers (no journal handle)
1952 * - shrink_page_list via pdflush (no journal handle)
1953 * - grab_page_cache when doing write_begin (have journal handle)
1955 * We don't do any block allocation in this function. If we have page with
1956 * multiple blocks we need to write those buffer_heads that are mapped. This
1957 * is important for mmaped based write. So if we do with blocksize 1K
1958 * truncate(f, 1024);
1959 * a = mmap(f, 0, 4096);
1961 * truncate(f, 4096);
1962 * we have in the page first buffer_head mapped via page_mkwrite call back
1963 * but other buffer_heads would be unmapped but dirty (dirty done via the
1964 * do_wp_page). So writepage should write the first block. If we modify
1965 * the mmap area beyond 1024 we will again get a page_fault and the
1966 * page_mkwrite callback will do the block allocation and mark the
1967 * buffer_heads mapped.
1969 * We redirty the page if we have any buffer_heads that is either delay or
1970 * unwritten in the page.
1972 * We can get recursively called as show below.
1974 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1977 * But since we don't do any block allocation we should not deadlock.
1978 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1980 static int ext4_writepage(struct page *page,
1981 struct writeback_control *wbc)
1983 int ret = 0, commit_write = 0;
1986 struct buffer_head *page_bufs = NULL;
1987 struct inode *inode = page->mapping->host;
1989 trace_ext4_writepage(page);
1990 size = i_size_read(inode);
1991 if (page->index == size >> PAGE_CACHE_SHIFT)
1992 len = size & ~PAGE_CACHE_MASK;
1994 len = PAGE_CACHE_SIZE;
1997 * If the page does not have buffers (for whatever reason),
1998 * try to create them using __block_write_begin. If this
1999 * fails, redirty the page and move on.
2001 if (!page_has_buffers(page)) {
2002 if (__block_write_begin(page, 0, len,
2003 noalloc_get_block_write)) {
2005 redirty_page_for_writepage(wbc, page);
2011 page_bufs = page_buffers(page);
2012 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2013 ext4_bh_delay_or_unwritten)) {
2015 * We don't want to do block allocation, so redirty
2016 * the page and return. We may reach here when we do
2017 * a journal commit via journal_submit_inode_data_buffers.
2018 * We can also reach here via shrink_page_list but it
2019 * should never be for direct reclaim so warn if that
2022 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2027 /* now mark the buffer_heads as dirty and uptodate */
2028 block_commit_write(page, 0, len);
2030 if (PageChecked(page) && ext4_should_journal_data(inode))
2032 * It's mmapped pagecache. Add buffers and journal it. There
2033 * doesn't seem much point in redirtying the page here.
2035 return __ext4_journalled_writepage(page, len);
2037 if (buffer_uninit(page_bufs)) {
2038 ext4_set_bh_endio(page_bufs, inode);
2039 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2040 wbc, ext4_end_io_buffer_write);
2042 ret = block_write_full_page(page, noalloc_get_block_write,
2049 * This is called via ext4_da_writepages() to
2050 * calculate the total number of credits to reserve to fit
2051 * a single extent allocation into a single transaction,
2052 * ext4_da_writpeages() will loop calling this before
2053 * the block allocation.
2056 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2058 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2061 * With non-extent format the journal credit needed to
2062 * insert nrblocks contiguous block is dependent on
2063 * number of contiguous block. So we will limit
2064 * number of contiguous block to a sane value
2066 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2067 (max_blocks > EXT4_MAX_TRANS_DATA))
2068 max_blocks = EXT4_MAX_TRANS_DATA;
2070 return ext4_chunk_trans_blocks(inode, max_blocks);
2074 * write_cache_pages_da - walk the list of dirty pages of the given
2075 * address space and accumulate pages that need writing, and call
2076 * mpage_da_map_and_submit to map a single contiguous memory region
2077 * and then write them.
2079 static int write_cache_pages_da(struct address_space *mapping,
2080 struct writeback_control *wbc,
2081 struct mpage_da_data *mpd,
2082 pgoff_t *done_index)
2084 struct buffer_head *bh, *head;
2085 struct inode *inode = mapping->host;
2086 struct pagevec pvec;
2087 unsigned int nr_pages;
2090 long nr_to_write = wbc->nr_to_write;
2091 int i, tag, ret = 0;
2093 memset(mpd, 0, sizeof(struct mpage_da_data));
2096 pagevec_init(&pvec, 0);
2097 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2098 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2100 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2101 tag = PAGECACHE_TAG_TOWRITE;
2103 tag = PAGECACHE_TAG_DIRTY;
2105 *done_index = index;
2106 while (index <= end) {
2107 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2108 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2112 for (i = 0; i < nr_pages; i++) {
2113 struct page *page = pvec.pages[i];
2116 * At this point, the page may be truncated or
2117 * invalidated (changing page->mapping to NULL), or
2118 * even swizzled back from swapper_space to tmpfs file
2119 * mapping. However, page->index will not change
2120 * because we have a reference on the page.
2122 if (page->index > end)
2125 *done_index = page->index + 1;
2128 * If we can't merge this page, and we have
2129 * accumulated an contiguous region, write it
2131 if ((mpd->next_page != page->index) &&
2132 (mpd->next_page != mpd->first_page)) {
2133 mpage_da_map_and_submit(mpd);
2134 goto ret_extent_tail;
2140 * If the page is no longer dirty, or its
2141 * mapping no longer corresponds to inode we
2142 * are writing (which means it has been
2143 * truncated or invalidated), or the page is
2144 * already under writeback and we are not
2145 * doing a data integrity writeback, skip the page
2147 if (!PageDirty(page) ||
2148 (PageWriteback(page) &&
2149 (wbc->sync_mode == WB_SYNC_NONE)) ||
2150 unlikely(page->mapping != mapping)) {
2155 wait_on_page_writeback(page);
2156 BUG_ON(PageWriteback(page));
2158 if (mpd->next_page != page->index)
2159 mpd->first_page = page->index;
2160 mpd->next_page = page->index + 1;
2161 logical = (sector_t) page->index <<
2162 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2164 if (!page_has_buffers(page)) {
2165 mpage_add_bh_to_extent(mpd, logical,
2167 (1 << BH_Dirty) | (1 << BH_Uptodate));
2169 goto ret_extent_tail;
2172 * Page with regular buffer heads,
2173 * just add all dirty ones
2175 head = page_buffers(page);
2178 BUG_ON(buffer_locked(bh));
2180 * We need to try to allocate
2181 * unmapped blocks in the same page.
2182 * Otherwise we won't make progress
2183 * with the page in ext4_writepage
2185 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2186 mpage_add_bh_to_extent(mpd, logical,
2190 goto ret_extent_tail;
2191 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2193 * mapped dirty buffer. We need
2194 * to update the b_state
2195 * because we look at b_state
2196 * in mpage_da_map_blocks. We
2197 * don't update b_size because
2198 * if we find an unmapped
2199 * buffer_head later we need to
2200 * use the b_state flag of that
2203 if (mpd->b_size == 0)
2204 mpd->b_state = bh->b_state & BH_FLAGS;
2207 } while ((bh = bh->b_this_page) != head);
2210 if (nr_to_write > 0) {
2212 if (nr_to_write == 0 &&
2213 wbc->sync_mode == WB_SYNC_NONE)
2215 * We stop writing back only if we are
2216 * not doing integrity sync. In case of
2217 * integrity sync we have to keep going
2218 * because someone may be concurrently
2219 * dirtying pages, and we might have
2220 * synced a lot of newly appeared dirty
2221 * pages, but have not synced all of the
2227 pagevec_release(&pvec);
2232 ret = MPAGE_DA_EXTENT_TAIL;
2234 pagevec_release(&pvec);
2240 static int ext4_da_writepages(struct address_space *mapping,
2241 struct writeback_control *wbc)
2244 int range_whole = 0;
2245 handle_t *handle = NULL;
2246 struct mpage_da_data mpd;
2247 struct inode *inode = mapping->host;
2248 int pages_written = 0;
2249 unsigned int max_pages;
2250 int range_cyclic, cycled = 1, io_done = 0;
2251 int needed_blocks, ret = 0;
2252 long desired_nr_to_write, nr_to_writebump = 0;
2253 loff_t range_start = wbc->range_start;
2254 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2255 pgoff_t done_index = 0;
2257 struct blk_plug plug;
2259 trace_ext4_da_writepages(inode, wbc);
2262 * No pages to write? This is mainly a kludge to avoid starting
2263 * a transaction for special inodes like journal inode on last iput()
2264 * because that could violate lock ordering on umount
2266 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2270 * If the filesystem has aborted, it is read-only, so return
2271 * right away instead of dumping stack traces later on that
2272 * will obscure the real source of the problem. We test
2273 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2274 * the latter could be true if the filesystem is mounted
2275 * read-only, and in that case, ext4_da_writepages should
2276 * *never* be called, so if that ever happens, we would want
2279 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2282 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2285 range_cyclic = wbc->range_cyclic;
2286 if (wbc->range_cyclic) {
2287 index = mapping->writeback_index;
2290 wbc->range_start = index << PAGE_CACHE_SHIFT;
2291 wbc->range_end = LLONG_MAX;
2292 wbc->range_cyclic = 0;
2295 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2296 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2300 * This works around two forms of stupidity. The first is in
2301 * the writeback code, which caps the maximum number of pages
2302 * written to be 1024 pages. This is wrong on multiple
2303 * levels; different architectues have a different page size,
2304 * which changes the maximum amount of data which gets
2305 * written. Secondly, 4 megabytes is way too small. XFS
2306 * forces this value to be 16 megabytes by multiplying
2307 * nr_to_write parameter by four, and then relies on its
2308 * allocator to allocate larger extents to make them
2309 * contiguous. Unfortunately this brings us to the second
2310 * stupidity, which is that ext4's mballoc code only allocates
2311 * at most 2048 blocks. So we force contiguous writes up to
2312 * the number of dirty blocks in the inode, or
2313 * sbi->max_writeback_mb_bump whichever is smaller.
2315 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2316 if (!range_cyclic && range_whole) {
2317 if (wbc->nr_to_write == LONG_MAX)
2318 desired_nr_to_write = wbc->nr_to_write;
2320 desired_nr_to_write = wbc->nr_to_write * 8;
2322 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2324 if (desired_nr_to_write > max_pages)
2325 desired_nr_to_write = max_pages;
2327 if (wbc->nr_to_write < desired_nr_to_write) {
2328 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2329 wbc->nr_to_write = desired_nr_to_write;
2333 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2334 tag_pages_for_writeback(mapping, index, end);
2336 blk_start_plug(&plug);
2337 while (!ret && wbc->nr_to_write > 0) {
2340 * we insert one extent at a time. So we need
2341 * credit needed for single extent allocation.
2342 * journalled mode is currently not supported
2345 BUG_ON(ext4_should_journal_data(inode));
2346 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2348 /* start a new transaction*/
2349 handle = ext4_journal_start(inode, needed_blocks);
2350 if (IS_ERR(handle)) {
2351 ret = PTR_ERR(handle);
2352 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2353 "%ld pages, ino %lu; err %d", __func__,
2354 wbc->nr_to_write, inode->i_ino, ret);
2355 blk_finish_plug(&plug);
2356 goto out_writepages;
2360 * Now call write_cache_pages_da() to find the next
2361 * contiguous region of logical blocks that need
2362 * blocks to be allocated by ext4 and submit them.
2364 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2366 * If we have a contiguous extent of pages and we
2367 * haven't done the I/O yet, map the blocks and submit
2370 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2371 mpage_da_map_and_submit(&mpd);
2372 ret = MPAGE_DA_EXTENT_TAIL;
2374 trace_ext4_da_write_pages(inode, &mpd);
2375 wbc->nr_to_write -= mpd.pages_written;
2377 ext4_journal_stop(handle);
2379 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2380 /* commit the transaction which would
2381 * free blocks released in the transaction
2384 jbd2_journal_force_commit_nested(sbi->s_journal);
2386 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2388 * Got one extent now try with rest of the pages.
2389 * If mpd.retval is set -EIO, journal is aborted.
2390 * So we don't need to write any more.
2392 pages_written += mpd.pages_written;
2395 } else if (wbc->nr_to_write)
2397 * There is no more writeout needed
2398 * or we requested for a noblocking writeout
2399 * and we found the device congested
2403 blk_finish_plug(&plug);
2404 if (!io_done && !cycled) {
2407 wbc->range_start = index << PAGE_CACHE_SHIFT;
2408 wbc->range_end = mapping->writeback_index - 1;
2413 wbc->range_cyclic = range_cyclic;
2414 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2416 * set the writeback_index so that range_cyclic
2417 * mode will write it back later
2419 mapping->writeback_index = done_index;
2422 wbc->nr_to_write -= nr_to_writebump;
2423 wbc->range_start = range_start;
2424 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2428 #define FALL_BACK_TO_NONDELALLOC 1
2429 static int ext4_nonda_switch(struct super_block *sb)
2431 s64 free_blocks, dirty_blocks;
2432 struct ext4_sb_info *sbi = EXT4_SB(sb);
2435 * switch to non delalloc mode if we are running low
2436 * on free block. The free block accounting via percpu
2437 * counters can get slightly wrong with percpu_counter_batch getting
2438 * accumulated on each CPU without updating global counters
2439 * Delalloc need an accurate free block accounting. So switch
2440 * to non delalloc when we are near to error range.
2442 free_blocks = EXT4_C2B(sbi,
2443 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2444 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2445 if (2 * free_blocks < 3 * dirty_blocks ||
2446 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2448 * free block count is less than 150% of dirty blocks
2449 * or free blocks is less than watermark
2454 * Even if we don't switch but are nearing capacity,
2455 * start pushing delalloc when 1/2 of free blocks are dirty.
2457 if (free_blocks < 2 * dirty_blocks)
2458 writeback_inodes_sb_if_idle(sb, WB_REASON_FS_FREE_SPACE);
2463 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2464 loff_t pos, unsigned len, unsigned flags,
2465 struct page **pagep, void **fsdata)
2467 int ret, retries = 0;
2470 struct inode *inode = mapping->host;
2473 index = pos >> PAGE_CACHE_SHIFT;
2475 if (ext4_nonda_switch(inode->i_sb)) {
2476 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2477 return ext4_write_begin(file, mapping, pos,
2478 len, flags, pagep, fsdata);
2480 *fsdata = (void *)0;
2481 trace_ext4_da_write_begin(inode, pos, len, flags);
2484 * With delayed allocation, we don't log the i_disksize update
2485 * if there is delayed block allocation. But we still need
2486 * to journalling the i_disksize update if writes to the end
2487 * of file which has an already mapped buffer.
2489 handle = ext4_journal_start(inode, 1);
2490 if (IS_ERR(handle)) {
2491 ret = PTR_ERR(handle);
2494 /* We cannot recurse into the filesystem as the transaction is already
2496 flags |= AOP_FLAG_NOFS;
2498 page = grab_cache_page_write_begin(mapping, index, flags);
2500 ext4_journal_stop(handle);
2506 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2509 ext4_journal_stop(handle);
2510 page_cache_release(page);
2512 * block_write_begin may have instantiated a few blocks
2513 * outside i_size. Trim these off again. Don't need
2514 * i_size_read because we hold i_mutex.
2516 if (pos + len > inode->i_size)
2517 ext4_truncate_failed_write(inode);
2520 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2527 * Check if we should update i_disksize
2528 * when write to the end of file but not require block allocation
2530 static int ext4_da_should_update_i_disksize(struct page *page,
2531 unsigned long offset)
2533 struct buffer_head *bh;
2534 struct inode *inode = page->mapping->host;
2538 bh = page_buffers(page);
2539 idx = offset >> inode->i_blkbits;
2541 for (i = 0; i < idx; i++)
2542 bh = bh->b_this_page;
2544 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2549 static int ext4_da_write_end(struct file *file,
2550 struct address_space *mapping,
2551 loff_t pos, unsigned len, unsigned copied,
2552 struct page *page, void *fsdata)
2554 struct inode *inode = mapping->host;
2556 handle_t *handle = ext4_journal_current_handle();
2558 unsigned long start, end;
2559 int write_mode = (int)(unsigned long)fsdata;
2561 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2562 switch (ext4_inode_journal_mode(inode)) {
2563 case EXT4_INODE_ORDERED_DATA_MODE:
2564 return ext4_ordered_write_end(file, mapping, pos,
2565 len, copied, page, fsdata);
2566 case EXT4_INODE_WRITEBACK_DATA_MODE:
2567 return ext4_writeback_write_end(file, mapping, pos,
2568 len, copied, page, fsdata);
2574 trace_ext4_da_write_end(inode, pos, len, copied);
2575 start = pos & (PAGE_CACHE_SIZE - 1);
2576 end = start + copied - 1;
2579 * generic_write_end() will run mark_inode_dirty() if i_size
2580 * changes. So let's piggyback the i_disksize mark_inode_dirty
2584 new_i_size = pos + copied;
2585 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2586 if (ext4_da_should_update_i_disksize(page, end)) {
2587 down_write(&EXT4_I(inode)->i_data_sem);
2588 if (new_i_size > EXT4_I(inode)->i_disksize) {
2590 * Updating i_disksize when extending file
2591 * without needing block allocation
2593 if (ext4_should_order_data(inode))
2594 ret = ext4_jbd2_file_inode(handle,
2597 EXT4_I(inode)->i_disksize = new_i_size;
2599 up_write(&EXT4_I(inode)->i_data_sem);
2600 /* We need to mark inode dirty even if
2601 * new_i_size is less that inode->i_size
2602 * bu greater than i_disksize.(hint delalloc)
2604 ext4_mark_inode_dirty(handle, inode);
2607 ret2 = generic_write_end(file, mapping, pos, len, copied,
2612 ret2 = ext4_journal_stop(handle);
2616 return ret ? ret : copied;
2619 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2622 * Drop reserved blocks
2624 BUG_ON(!PageLocked(page));
2625 if (!page_has_buffers(page))
2628 ext4_da_page_release_reservation(page, offset);
2631 ext4_invalidatepage(page, offset);
2637 * Force all delayed allocation blocks to be allocated for a given inode.
2639 int ext4_alloc_da_blocks(struct inode *inode)
2641 trace_ext4_alloc_da_blocks(inode);
2643 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2644 !EXT4_I(inode)->i_reserved_meta_blocks)
2648 * We do something simple for now. The filemap_flush() will
2649 * also start triggering a write of the data blocks, which is
2650 * not strictly speaking necessary (and for users of
2651 * laptop_mode, not even desirable). However, to do otherwise
2652 * would require replicating code paths in:
2654 * ext4_da_writepages() ->
2655 * write_cache_pages() ---> (via passed in callback function)
2656 * __mpage_da_writepage() -->
2657 * mpage_add_bh_to_extent()
2658 * mpage_da_map_blocks()
2660 * The problem is that write_cache_pages(), located in
2661 * mm/page-writeback.c, marks pages clean in preparation for
2662 * doing I/O, which is not desirable if we're not planning on
2665 * We could call write_cache_pages(), and then redirty all of
2666 * the pages by calling redirty_page_for_writepage() but that
2667 * would be ugly in the extreme. So instead we would need to
2668 * replicate parts of the code in the above functions,
2669 * simplifying them because we wouldn't actually intend to
2670 * write out the pages, but rather only collect contiguous
2671 * logical block extents, call the multi-block allocator, and
2672 * then update the buffer heads with the block allocations.
2674 * For now, though, we'll cheat by calling filemap_flush(),
2675 * which will map the blocks, and start the I/O, but not
2676 * actually wait for the I/O to complete.
2678 return filemap_flush(inode->i_mapping);
2682 * bmap() is special. It gets used by applications such as lilo and by
2683 * the swapper to find the on-disk block of a specific piece of data.
2685 * Naturally, this is dangerous if the block concerned is still in the
2686 * journal. If somebody makes a swapfile on an ext4 data-journaling
2687 * filesystem and enables swap, then they may get a nasty shock when the
2688 * data getting swapped to that swapfile suddenly gets overwritten by
2689 * the original zero's written out previously to the journal and
2690 * awaiting writeback in the kernel's buffer cache.
2692 * So, if we see any bmap calls here on a modified, data-journaled file,
2693 * take extra steps to flush any blocks which might be in the cache.
2695 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2697 struct inode *inode = mapping->host;
2701 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2702 test_opt(inode->i_sb, DELALLOC)) {
2704 * With delalloc we want to sync the file
2705 * so that we can make sure we allocate
2708 filemap_write_and_wait(mapping);
2711 if (EXT4_JOURNAL(inode) &&
2712 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2714 * This is a REALLY heavyweight approach, but the use of
2715 * bmap on dirty files is expected to be extremely rare:
2716 * only if we run lilo or swapon on a freshly made file
2717 * do we expect this to happen.
2719 * (bmap requires CAP_SYS_RAWIO so this does not
2720 * represent an unprivileged user DOS attack --- we'd be
2721 * in trouble if mortal users could trigger this path at
2724 * NB. EXT4_STATE_JDATA is not set on files other than
2725 * regular files. If somebody wants to bmap a directory
2726 * or symlink and gets confused because the buffer
2727 * hasn't yet been flushed to disk, they deserve
2728 * everything they get.
2731 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2732 journal = EXT4_JOURNAL(inode);
2733 jbd2_journal_lock_updates(journal);
2734 err = jbd2_journal_flush(journal);
2735 jbd2_journal_unlock_updates(journal);
2741 return generic_block_bmap(mapping, block, ext4_get_block);
2744 static int ext4_readpage(struct file *file, struct page *page)
2746 trace_ext4_readpage(page);
2747 return mpage_readpage(page, ext4_get_block);
2751 ext4_readpages(struct file *file, struct address_space *mapping,
2752 struct list_head *pages, unsigned nr_pages)
2754 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2757 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2759 struct buffer_head *head, *bh;
2760 unsigned int curr_off = 0;
2762 if (!page_has_buffers(page))
2764 head = bh = page_buffers(page);
2766 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2768 ext4_free_io_end(bh->b_private);
2769 bh->b_private = NULL;
2770 bh->b_end_io = NULL;
2772 curr_off = curr_off + bh->b_size;
2773 bh = bh->b_this_page;
2774 } while (bh != head);
2777 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2779 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2781 trace_ext4_invalidatepage(page, offset);
2784 * free any io_end structure allocated for buffers to be discarded
2786 if (ext4_should_dioread_nolock(page->mapping->host))
2787 ext4_invalidatepage_free_endio(page, offset);
2789 * If it's a full truncate we just forget about the pending dirtying
2792 ClearPageChecked(page);
2795 jbd2_journal_invalidatepage(journal, page, offset);
2797 block_invalidatepage(page, offset);
2800 static int ext4_releasepage(struct page *page, gfp_t wait)
2802 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2804 trace_ext4_releasepage(page);
2806 WARN_ON(PageChecked(page));
2807 if (!page_has_buffers(page))
2810 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2812 return try_to_free_buffers(page);
2816 * ext4_get_block used when preparing for a DIO write or buffer write.
2817 * We allocate an uinitialized extent if blocks haven't been allocated.
2818 * The extent will be converted to initialized after the IO is complete.
2820 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2821 struct buffer_head *bh_result, int create)
2823 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2824 inode->i_ino, create);
2825 return _ext4_get_block(inode, iblock, bh_result,
2826 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2829 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2830 ssize_t size, void *private, int ret,
2833 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2834 ext4_io_end_t *io_end = iocb->private;
2835 struct workqueue_struct *wq;
2836 unsigned long flags;
2837 struct ext4_inode_info *ei;
2839 /* if not async direct IO or dio with 0 bytes write, just return */
2840 if (!io_end || !size)
2843 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2844 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2845 iocb->private, io_end->inode->i_ino, iocb, offset,
2848 iocb->private = NULL;
2850 /* if not aio dio with unwritten extents, just free io and return */
2851 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2852 ext4_free_io_end(io_end);
2855 aio_complete(iocb, ret, 0);
2856 inode_dio_done(inode);
2860 io_end->offset = offset;
2861 io_end->size = size;
2863 io_end->iocb = iocb;
2864 io_end->result = ret;
2866 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2868 /* Add the io_end to per-inode completed aio dio list*/
2869 ei = EXT4_I(io_end->inode);
2870 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2871 list_add_tail(&io_end->list, &ei->i_completed_io_list);
2872 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2874 /* queue the work to convert unwritten extents to written */
2875 queue_work(wq, &io_end->work);
2878 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2880 ext4_io_end_t *io_end = bh->b_private;
2881 struct workqueue_struct *wq;
2882 struct inode *inode;
2883 unsigned long flags;
2885 if (!test_clear_buffer_uninit(bh) || !io_end)
2888 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2889 ext4_msg(io_end->inode->i_sb, KERN_INFO,
2890 "sb umounted, discard end_io request for inode %lu",
2891 io_end->inode->i_ino);
2892 ext4_free_io_end(io_end);
2897 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2898 * but being more careful is always safe for the future change.
2900 inode = io_end->inode;
2901 ext4_set_io_unwritten_flag(inode, io_end);
2903 /* Add the io_end to per-inode completed io list*/
2904 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2905 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2906 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2908 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2909 /* queue the work to convert unwritten extents to written */
2910 queue_work(wq, &io_end->work);
2912 bh->b_private = NULL;
2913 bh->b_end_io = NULL;
2914 clear_buffer_uninit(bh);
2915 end_buffer_async_write(bh, uptodate);
2918 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2920 ext4_io_end_t *io_end;
2921 struct page *page = bh->b_page;
2922 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2923 size_t size = bh->b_size;
2926 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2928 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2932 io_end->offset = offset;
2933 io_end->size = size;
2935 * We need to hold a reference to the page to make sure it
2936 * doesn't get evicted before ext4_end_io_work() has a chance
2937 * to convert the extent from written to unwritten.
2939 io_end->page = page;
2940 get_page(io_end->page);
2942 bh->b_private = io_end;
2943 bh->b_end_io = ext4_end_io_buffer_write;
2948 * For ext4 extent files, ext4 will do direct-io write to holes,
2949 * preallocated extents, and those write extend the file, no need to
2950 * fall back to buffered IO.
2952 * For holes, we fallocate those blocks, mark them as uninitialized
2953 * If those blocks were preallocated, we mark sure they are splited, but
2954 * still keep the range to write as uninitialized.
2956 * The unwrritten extents will be converted to written when DIO is completed.
2957 * For async direct IO, since the IO may still pending when return, we
2958 * set up an end_io call back function, which will do the conversion
2959 * when async direct IO completed.
2961 * If the O_DIRECT write will extend the file then add this inode to the
2962 * orphan list. So recovery will truncate it back to the original size
2963 * if the machine crashes during the write.
2966 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2967 const struct iovec *iov, loff_t offset,
2968 unsigned long nr_segs)
2970 struct file *file = iocb->ki_filp;
2971 struct inode *inode = file->f_mapping->host;
2973 size_t count = iov_length(iov, nr_segs);
2975 loff_t final_size = offset + count;
2976 if (rw == WRITE && final_size <= inode->i_size) {
2978 * We could direct write to holes and fallocate.
2980 * Allocated blocks to fill the hole are marked as uninitialized
2981 * to prevent parallel buffered read to expose the stale data
2982 * before DIO complete the data IO.
2984 * As to previously fallocated extents, ext4 get_block
2985 * will just simply mark the buffer mapped but still
2986 * keep the extents uninitialized.
2988 * for non AIO case, we will convert those unwritten extents
2989 * to written after return back from blockdev_direct_IO.
2991 * for async DIO, the conversion needs to be defered when
2992 * the IO is completed. The ext4 end_io callback function
2993 * will be called to take care of the conversion work.
2994 * Here for async case, we allocate an io_end structure to
2997 iocb->private = NULL;
2998 EXT4_I(inode)->cur_aio_dio = NULL;
2999 if (!is_sync_kiocb(iocb)) {
3000 ext4_io_end_t *io_end =
3001 ext4_init_io_end(inode, GFP_NOFS);
3004 io_end->flag |= EXT4_IO_END_DIRECT;
3005 iocb->private = io_end;
3007 * we save the io structure for current async
3008 * direct IO, so that later ext4_map_blocks()
3009 * could flag the io structure whether there
3010 * is a unwritten extents needs to be converted
3011 * when IO is completed.
3013 EXT4_I(inode)->cur_aio_dio = iocb->private;
3016 ret = __blockdev_direct_IO(rw, iocb, inode,
3017 inode->i_sb->s_bdev, iov,
3019 ext4_get_block_write,
3024 EXT4_I(inode)->cur_aio_dio = NULL;
3026 * The io_end structure takes a reference to the inode,
3027 * that structure needs to be destroyed and the
3028 * reference to the inode need to be dropped, when IO is
3029 * complete, even with 0 byte write, or failed.
3031 * In the successful AIO DIO case, the io_end structure will be
3032 * desctroyed and the reference to the inode will be dropped
3033 * after the end_io call back function is called.
3035 * In the case there is 0 byte write, or error case, since
3036 * VFS direct IO won't invoke the end_io call back function,
3037 * we need to free the end_io structure here.
3039 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3040 ext4_free_io_end(iocb->private);
3041 iocb->private = NULL;
3042 } else if (ret > 0 && ext4_test_inode_state(inode,
3043 EXT4_STATE_DIO_UNWRITTEN)) {
3046 * for non AIO case, since the IO is already
3047 * completed, we could do the conversion right here
3049 err = ext4_convert_unwritten_extents(inode,
3053 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3058 /* for write the the end of file case, we fall back to old way */
3059 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3062 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3063 const struct iovec *iov, loff_t offset,
3064 unsigned long nr_segs)
3066 struct file *file = iocb->ki_filp;
3067 struct inode *inode = file->f_mapping->host;
3071 * If we are doing data journalling we don't support O_DIRECT
3073 if (ext4_should_journal_data(inode))
3076 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3077 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3078 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3080 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3081 trace_ext4_direct_IO_exit(inode, offset,
3082 iov_length(iov, nr_segs), rw, ret);
3087 * Pages can be marked dirty completely asynchronously from ext4's journalling
3088 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3089 * much here because ->set_page_dirty is called under VFS locks. The page is
3090 * not necessarily locked.
3092 * We cannot just dirty the page and leave attached buffers clean, because the
3093 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3094 * or jbddirty because all the journalling code will explode.
3096 * So what we do is to mark the page "pending dirty" and next time writepage
3097 * is called, propagate that into the buffers appropriately.
3099 static int ext4_journalled_set_page_dirty(struct page *page)
3101 SetPageChecked(page);
3102 return __set_page_dirty_nobuffers(page);
3105 static const struct address_space_operations ext4_ordered_aops = {
3106 .readpage = ext4_readpage,
3107 .readpages = ext4_readpages,
3108 .writepage = ext4_writepage,
3109 .write_begin = ext4_write_begin,
3110 .write_end = ext4_ordered_write_end,
3112 .invalidatepage = ext4_invalidatepage,
3113 .releasepage = ext4_releasepage,
3114 .direct_IO = ext4_direct_IO,
3115 .migratepage = buffer_migrate_page,
3116 .is_partially_uptodate = block_is_partially_uptodate,
3117 .error_remove_page = generic_error_remove_page,
3120 static const struct address_space_operations ext4_writeback_aops = {
3121 .readpage = ext4_readpage,
3122 .readpages = ext4_readpages,
3123 .writepage = ext4_writepage,
3124 .write_begin = ext4_write_begin,
3125 .write_end = ext4_writeback_write_end,
3127 .invalidatepage = ext4_invalidatepage,
3128 .releasepage = ext4_releasepage,
3129 .direct_IO = ext4_direct_IO,
3130 .migratepage = buffer_migrate_page,
3131 .is_partially_uptodate = block_is_partially_uptodate,
3132 .error_remove_page = generic_error_remove_page,
3135 static const struct address_space_operations ext4_journalled_aops = {
3136 .readpage = ext4_readpage,
3137 .readpages = ext4_readpages,
3138 .writepage = ext4_writepage,
3139 .write_begin = ext4_write_begin,
3140 .write_end = ext4_journalled_write_end,
3141 .set_page_dirty = ext4_journalled_set_page_dirty,
3143 .invalidatepage = ext4_invalidatepage,
3144 .releasepage = ext4_releasepage,
3145 .direct_IO = ext4_direct_IO,
3146 .is_partially_uptodate = block_is_partially_uptodate,
3147 .error_remove_page = generic_error_remove_page,
3150 static const struct address_space_operations ext4_da_aops = {
3151 .readpage = ext4_readpage,
3152 .readpages = ext4_readpages,
3153 .writepage = ext4_writepage,
3154 .writepages = ext4_da_writepages,
3155 .write_begin = ext4_da_write_begin,
3156 .write_end = ext4_da_write_end,
3158 .invalidatepage = ext4_da_invalidatepage,
3159 .releasepage = ext4_releasepage,
3160 .direct_IO = ext4_direct_IO,
3161 .migratepage = buffer_migrate_page,
3162 .is_partially_uptodate = block_is_partially_uptodate,
3163 .error_remove_page = generic_error_remove_page,
3166 void ext4_set_aops(struct inode *inode)
3168 switch (ext4_inode_journal_mode(inode)) {
3169 case EXT4_INODE_ORDERED_DATA_MODE:
3170 if (test_opt(inode->i_sb, DELALLOC))
3171 inode->i_mapping->a_ops = &ext4_da_aops;
3173 inode->i_mapping->a_ops = &ext4_ordered_aops;
3175 case EXT4_INODE_WRITEBACK_DATA_MODE:
3176 if (test_opt(inode->i_sb, DELALLOC))
3177 inode->i_mapping->a_ops = &ext4_da_aops;
3179 inode->i_mapping->a_ops = &ext4_writeback_aops;
3181 case EXT4_INODE_JOURNAL_DATA_MODE:
3182 inode->i_mapping->a_ops = &ext4_journalled_aops;
3191 * ext4_discard_partial_page_buffers()
3192 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3193 * This function finds and locks the page containing the offset
3194 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3195 * Calling functions that already have the page locked should call
3196 * ext4_discard_partial_page_buffers_no_lock directly.
3198 int ext4_discard_partial_page_buffers(handle_t *handle,
3199 struct address_space *mapping, loff_t from,
3200 loff_t length, int flags)
3202 struct inode *inode = mapping->host;
3206 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3207 mapping_gfp_mask(mapping) & ~__GFP_FS);
3211 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3212 from, length, flags);
3215 page_cache_release(page);
3220 * ext4_discard_partial_page_buffers_no_lock()
3221 * Zeros a page range of length 'length' starting from offset 'from'.
3222 * Buffer heads that correspond to the block aligned regions of the
3223 * zeroed range will be unmapped. Unblock aligned regions
3224 * will have the corresponding buffer head mapped if needed so that
3225 * that region of the page can be updated with the partial zero out.
3227 * This function assumes that the page has already been locked. The
3228 * The range to be discarded must be contained with in the given page.
3229 * If the specified range exceeds the end of the page it will be shortened
3230 * to the end of the page that corresponds to 'from'. This function is
3231 * appropriate for updating a page and it buffer heads to be unmapped and
3232 * zeroed for blocks that have been either released, or are going to be
3235 * handle: The journal handle
3236 * inode: The files inode
3237 * page: A locked page that contains the offset "from"
3238 * from: The starting byte offset (from the begining of the file)
3239 * to begin discarding
3240 * len: The length of bytes to discard
3241 * flags: Optional flags that may be used:
3243 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3244 * Only zero the regions of the page whose buffer heads
3245 * have already been unmapped. This flag is appropriate
3246 * for updateing the contents of a page whose blocks may
3247 * have already been released, and we only want to zero
3248 * out the regions that correspond to those released blocks.
3250 * Returns zero on sucess or negative on failure.
3252 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3253 struct inode *inode, struct page *page, loff_t from,
3254 loff_t length, int flags)
3256 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3257 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3258 unsigned int blocksize, max, pos;
3260 struct buffer_head *bh;
3263 blocksize = inode->i_sb->s_blocksize;
3264 max = PAGE_CACHE_SIZE - offset;
3266 if (index != page->index)
3270 * correct length if it does not fall between
3271 * 'from' and the end of the page
3273 if (length > max || length < 0)
3276 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3278 if (!page_has_buffers(page))
3279 create_empty_buffers(page, blocksize, 0);
3281 /* Find the buffer that contains "offset" */
3282 bh = page_buffers(page);
3284 while (offset >= pos) {
3285 bh = bh->b_this_page;
3291 while (pos < offset + length) {
3292 unsigned int end_of_block, range_to_discard;
3296 /* The length of space left to zero and unmap */
3297 range_to_discard = offset + length - pos;
3299 /* The length of space until the end of the block */
3300 end_of_block = blocksize - (pos & (blocksize-1));
3303 * Do not unmap or zero past end of block
3304 * for this buffer head
3306 if (range_to_discard > end_of_block)
3307 range_to_discard = end_of_block;
3311 * Skip this buffer head if we are only zeroing unampped
3312 * regions of the page
3314 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3318 /* If the range is block aligned, unmap */
3319 if (range_to_discard == blocksize) {
3320 clear_buffer_dirty(bh);
3322 clear_buffer_mapped(bh);
3323 clear_buffer_req(bh);
3324 clear_buffer_new(bh);
3325 clear_buffer_delay(bh);
3326 clear_buffer_unwritten(bh);
3327 clear_buffer_uptodate(bh);
3328 zero_user(page, pos, range_to_discard);
3329 BUFFER_TRACE(bh, "Buffer discarded");
3334 * If this block is not completely contained in the range
3335 * to be discarded, then it is not going to be released. Because
3336 * we need to keep this block, we need to make sure this part
3337 * of the page is uptodate before we modify it by writeing
3338 * partial zeros on it.
3340 if (!buffer_mapped(bh)) {
3342 * Buffer head must be mapped before we can read
3345 BUFFER_TRACE(bh, "unmapped");
3346 ext4_get_block(inode, iblock, bh, 0);
3347 /* unmapped? It's a hole - nothing to do */
3348 if (!buffer_mapped(bh)) {
3349 BUFFER_TRACE(bh, "still unmapped");
3354 /* Ok, it's mapped. Make sure it's up-to-date */
3355 if (PageUptodate(page))
3356 set_buffer_uptodate(bh);
3358 if (!buffer_uptodate(bh)) {
3360 ll_rw_block(READ, 1, &bh);
3362 /* Uhhuh. Read error. Complain and punt.*/
3363 if (!buffer_uptodate(bh))
3367 if (ext4_should_journal_data(inode)) {
3368 BUFFER_TRACE(bh, "get write access");
3369 err = ext4_journal_get_write_access(handle, bh);
3374 zero_user(page, pos, range_to_discard);
3377 if (ext4_should_journal_data(inode)) {
3378 err = ext4_handle_dirty_metadata(handle, inode, bh);
3380 mark_buffer_dirty(bh);
3382 BUFFER_TRACE(bh, "Partial buffer zeroed");
3384 bh = bh->b_this_page;
3386 pos += range_to_discard;
3392 int ext4_can_truncate(struct inode *inode)
3394 if (S_ISREG(inode->i_mode))
3396 if (S_ISDIR(inode->i_mode))
3398 if (S_ISLNK(inode->i_mode))
3399 return !ext4_inode_is_fast_symlink(inode);
3404 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3405 * associated with the given offset and length
3407 * @inode: File inode
3408 * @offset: The offset where the hole will begin
3409 * @len: The length of the hole
3411 * Returns: 0 on sucess or negative on failure
3414 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3416 struct inode *inode = file->f_path.dentry->d_inode;
3417 if (!S_ISREG(inode->i_mode))
3420 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3421 /* TODO: Add support for non extent hole punching */
3425 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3426 /* TODO: Add support for bigalloc file systems */
3430 return ext4_ext_punch_hole(file, offset, length);
3436 * We block out ext4_get_block() block instantiations across the entire
3437 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3438 * simultaneously on behalf of the same inode.
3440 * As we work through the truncate and commit bits of it to the journal there
3441 * is one core, guiding principle: the file's tree must always be consistent on
3442 * disk. We must be able to restart the truncate after a crash.
3444 * The file's tree may be transiently inconsistent in memory (although it
3445 * probably isn't), but whenever we close off and commit a journal transaction,
3446 * the contents of (the filesystem + the journal) must be consistent and
3447 * restartable. It's pretty simple, really: bottom up, right to left (although
3448 * left-to-right works OK too).
3450 * Note that at recovery time, journal replay occurs *before* the restart of
3451 * truncate against the orphan inode list.
3453 * The committed inode has the new, desired i_size (which is the same as
3454 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3455 * that this inode's truncate did not complete and it will again call
3456 * ext4_truncate() to have another go. So there will be instantiated blocks
3457 * to the right of the truncation point in a crashed ext4 filesystem. But
3458 * that's fine - as long as they are linked from the inode, the post-crash
3459 * ext4_truncate() run will find them and release them.
3461 void ext4_truncate(struct inode *inode)
3463 trace_ext4_truncate_enter(inode);
3465 if (!ext4_can_truncate(inode))
3468 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3470 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3471 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3473 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3474 ext4_ext_truncate(inode);
3476 ext4_ind_truncate(inode);
3478 trace_ext4_truncate_exit(inode);
3482 * ext4_get_inode_loc returns with an extra refcount against the inode's
3483 * underlying buffer_head on success. If 'in_mem' is true, we have all
3484 * data in memory that is needed to recreate the on-disk version of this
3487 static int __ext4_get_inode_loc(struct inode *inode,
3488 struct ext4_iloc *iloc, int in_mem)
3490 struct ext4_group_desc *gdp;
3491 struct buffer_head *bh;
3492 struct super_block *sb = inode->i_sb;
3494 int inodes_per_block, inode_offset;
3497 if (!ext4_valid_inum(sb, inode->i_ino))
3500 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3501 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3506 * Figure out the offset within the block group inode table
3508 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3509 inode_offset = ((inode->i_ino - 1) %
3510 EXT4_INODES_PER_GROUP(sb));
3511 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3512 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3514 bh = sb_getblk(sb, block);
3516 EXT4_ERROR_INODE_BLOCK(inode, block,
3517 "unable to read itable block");
3520 if (!buffer_uptodate(bh)) {
3524 * If the buffer has the write error flag, we have failed
3525 * to write out another inode in the same block. In this
3526 * case, we don't have to read the block because we may
3527 * read the old inode data successfully.
3529 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3530 set_buffer_uptodate(bh);
3532 if (buffer_uptodate(bh)) {
3533 /* someone brought it uptodate while we waited */
3539 * If we have all information of the inode in memory and this
3540 * is the only valid inode in the block, we need not read the
3544 struct buffer_head *bitmap_bh;
3547 start = inode_offset & ~(inodes_per_block - 1);
3549 /* Is the inode bitmap in cache? */
3550 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3555 * If the inode bitmap isn't in cache then the
3556 * optimisation may end up performing two reads instead
3557 * of one, so skip it.
3559 if (!buffer_uptodate(bitmap_bh)) {
3563 for (i = start; i < start + inodes_per_block; i++) {
3564 if (i == inode_offset)
3566 if (ext4_test_bit(i, bitmap_bh->b_data))
3570 if (i == start + inodes_per_block) {
3571 /* all other inodes are free, so skip I/O */
3572 memset(bh->b_data, 0, bh->b_size);
3573 set_buffer_uptodate(bh);
3581 * If we need to do any I/O, try to pre-readahead extra
3582 * blocks from the inode table.
3584 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3585 ext4_fsblk_t b, end, table;
3588 table = ext4_inode_table(sb, gdp);
3589 /* s_inode_readahead_blks is always a power of 2 */
3590 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3593 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3594 num = EXT4_INODES_PER_GROUP(sb);
3595 if (ext4_has_group_desc_csum(sb))
3596 num -= ext4_itable_unused_count(sb, gdp);
3597 table += num / inodes_per_block;
3601 sb_breadahead(sb, b++);
3605 * There are other valid inodes in the buffer, this inode
3606 * has in-inode xattrs, or we don't have this inode in memory.
3607 * Read the block from disk.
3609 trace_ext4_load_inode(inode);
3611 bh->b_end_io = end_buffer_read_sync;
3612 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3614 if (!buffer_uptodate(bh)) {
3615 EXT4_ERROR_INODE_BLOCK(inode, block,
3616 "unable to read itable block");
3626 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3628 /* We have all inode data except xattrs in memory here. */
3629 return __ext4_get_inode_loc(inode, iloc,
3630 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3633 void ext4_set_inode_flags(struct inode *inode)
3635 unsigned int flags = EXT4_I(inode)->i_flags;
3637 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3638 if (flags & EXT4_SYNC_FL)
3639 inode->i_flags |= S_SYNC;
3640 if (flags & EXT4_APPEND_FL)
3641 inode->i_flags |= S_APPEND;
3642 if (flags & EXT4_IMMUTABLE_FL)
3643 inode->i_flags |= S_IMMUTABLE;
3644 if (flags & EXT4_NOATIME_FL)
3645 inode->i_flags |= S_NOATIME;
3646 if (flags & EXT4_DIRSYNC_FL)
3647 inode->i_flags |= S_DIRSYNC;
3650 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3651 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3653 unsigned int vfs_fl;
3654 unsigned long old_fl, new_fl;
3657 vfs_fl = ei->vfs_inode.i_flags;
3658 old_fl = ei->i_flags;
3659 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3660 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3662 if (vfs_fl & S_SYNC)
3663 new_fl |= EXT4_SYNC_FL;
3664 if (vfs_fl & S_APPEND)
3665 new_fl |= EXT4_APPEND_FL;
3666 if (vfs_fl & S_IMMUTABLE)
3667 new_fl |= EXT4_IMMUTABLE_FL;
3668 if (vfs_fl & S_NOATIME)
3669 new_fl |= EXT4_NOATIME_FL;
3670 if (vfs_fl & S_DIRSYNC)
3671 new_fl |= EXT4_DIRSYNC_FL;
3672 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3675 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3676 struct ext4_inode_info *ei)
3679 struct inode *inode = &(ei->vfs_inode);
3680 struct super_block *sb = inode->i_sb;
3682 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3683 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3684 /* we are using combined 48 bit field */
3685 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3686 le32_to_cpu(raw_inode->i_blocks_lo);
3687 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3688 /* i_blocks represent file system block size */
3689 return i_blocks << (inode->i_blkbits - 9);
3694 return le32_to_cpu(raw_inode->i_blocks_lo);
3698 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3700 struct ext4_iloc iloc;
3701 struct ext4_inode *raw_inode;
3702 struct ext4_inode_info *ei;
3703 struct inode *inode;
3704 journal_t *journal = EXT4_SB(sb)->s_journal;
3710 inode = iget_locked(sb, ino);
3712 return ERR_PTR(-ENOMEM);
3713 if (!(inode->i_state & I_NEW))
3719 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3722 raw_inode = ext4_raw_inode(&iloc);
3724 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3725 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3726 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3727 EXT4_INODE_SIZE(inode->i_sb)) {
3728 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3729 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3730 EXT4_INODE_SIZE(inode->i_sb));
3735 ei->i_extra_isize = 0;
3737 /* Precompute checksum seed for inode metadata */
3738 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3739 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3740 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3742 __le32 inum = cpu_to_le32(inode->i_ino);
3743 __le32 gen = raw_inode->i_generation;
3744 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3746 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3750 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3751 EXT4_ERROR_INODE(inode, "checksum invalid");
3756 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3757 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3758 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3759 if (!(test_opt(inode->i_sb, NO_UID32))) {
3760 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3761 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3763 i_uid_write(inode, i_uid);
3764 i_gid_write(inode, i_gid);
3765 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3767 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3768 ei->i_dir_start_lookup = 0;
3769 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3770 /* We now have enough fields to check if the inode was active or not.
3771 * This is needed because nfsd might try to access dead inodes
3772 * the test is that same one that e2fsck uses
3773 * NeilBrown 1999oct15
3775 if (inode->i_nlink == 0) {
3776 if (inode->i_mode == 0 ||
3777 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3778 /* this inode is deleted */
3782 /* The only unlinked inodes we let through here have
3783 * valid i_mode and are being read by the orphan
3784 * recovery code: that's fine, we're about to complete
3785 * the process of deleting those. */
3787 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3788 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3789 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3790 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3792 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3793 inode->i_size = ext4_isize(raw_inode);
3794 ei->i_disksize = inode->i_size;
3796 ei->i_reserved_quota = 0;
3798 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3799 ei->i_block_group = iloc.block_group;
3800 ei->i_last_alloc_group = ~0;
3802 * NOTE! The in-memory inode i_data array is in little-endian order
3803 * even on big-endian machines: we do NOT byteswap the block numbers!
3805 for (block = 0; block < EXT4_N_BLOCKS; block++)
3806 ei->i_data[block] = raw_inode->i_block[block];
3807 INIT_LIST_HEAD(&ei->i_orphan);
3810 * Set transaction id's of transactions that have to be committed
3811 * to finish f[data]sync. We set them to currently running transaction
3812 * as we cannot be sure that the inode or some of its metadata isn't
3813 * part of the transaction - the inode could have been reclaimed and
3814 * now it is reread from disk.
3817 transaction_t *transaction;
3820 read_lock(&journal->j_state_lock);
3821 if (journal->j_running_transaction)
3822 transaction = journal->j_running_transaction;
3824 transaction = journal->j_committing_transaction;
3826 tid = transaction->t_tid;
3828 tid = journal->j_commit_sequence;
3829 read_unlock(&journal->j_state_lock);
3830 ei->i_sync_tid = tid;
3831 ei->i_datasync_tid = tid;
3834 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3835 if (ei->i_extra_isize == 0) {
3836 /* The extra space is currently unused. Use it. */
3837 ei->i_extra_isize = sizeof(struct ext4_inode) -
3838 EXT4_GOOD_OLD_INODE_SIZE;
3840 __le32 *magic = (void *)raw_inode +
3841 EXT4_GOOD_OLD_INODE_SIZE +
3843 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3844 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3848 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3849 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3850 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3851 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3853 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3854 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3855 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3857 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3861 if (ei->i_file_acl &&
3862 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3863 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3867 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3868 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3869 (S_ISLNK(inode->i_mode) &&
3870 !ext4_inode_is_fast_symlink(inode)))
3871 /* Validate extent which is part of inode */
3872 ret = ext4_ext_check_inode(inode);
3873 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3874 (S_ISLNK(inode->i_mode) &&
3875 !ext4_inode_is_fast_symlink(inode))) {
3876 /* Validate block references which are part of inode */
3877 ret = ext4_ind_check_inode(inode);
3882 if (S_ISREG(inode->i_mode)) {
3883 inode->i_op = &ext4_file_inode_operations;
3884 inode->i_fop = &ext4_file_operations;
3885 ext4_set_aops(inode);
3886 } else if (S_ISDIR(inode->i_mode)) {
3887 inode->i_op = &ext4_dir_inode_operations;
3888 inode->i_fop = &ext4_dir_operations;
3889 } else if (S_ISLNK(inode->i_mode)) {
3890 if (ext4_inode_is_fast_symlink(inode)) {
3891 inode->i_op = &ext4_fast_symlink_inode_operations;
3892 nd_terminate_link(ei->i_data, inode->i_size,
3893 sizeof(ei->i_data) - 1);
3895 inode->i_op = &ext4_symlink_inode_operations;
3896 ext4_set_aops(inode);
3898 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3899 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3900 inode->i_op = &ext4_special_inode_operations;
3901 if (raw_inode->i_block[0])
3902 init_special_inode(inode, inode->i_mode,
3903 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3905 init_special_inode(inode, inode->i_mode,
3906 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3909 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3913 ext4_set_inode_flags(inode);
3914 unlock_new_inode(inode);
3920 return ERR_PTR(ret);
3923 static int ext4_inode_blocks_set(handle_t *handle,
3924 struct ext4_inode *raw_inode,
3925 struct ext4_inode_info *ei)
3927 struct inode *inode = &(ei->vfs_inode);
3928 u64 i_blocks = inode->i_blocks;
3929 struct super_block *sb = inode->i_sb;
3931 if (i_blocks <= ~0U) {
3933 * i_blocks can be represnted in a 32 bit variable
3934 * as multiple of 512 bytes
3936 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3937 raw_inode->i_blocks_high = 0;
3938 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3941 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3944 if (i_blocks <= 0xffffffffffffULL) {
3946 * i_blocks can be represented in a 48 bit variable
3947 * as multiple of 512 bytes
3949 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3950 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3951 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3953 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3954 /* i_block is stored in file system block size */
3955 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3956 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3957 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3963 * Post the struct inode info into an on-disk inode location in the
3964 * buffer-cache. This gobbles the caller's reference to the
3965 * buffer_head in the inode location struct.
3967 * The caller must have write access to iloc->bh.
3969 static int ext4_do_update_inode(handle_t *handle,
3970 struct inode *inode,
3971 struct ext4_iloc *iloc)
3973 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3974 struct ext4_inode_info *ei = EXT4_I(inode);
3975 struct buffer_head *bh = iloc->bh;
3976 int err = 0, rc, block;
3980 /* For fields not not tracking in the in-memory inode,
3981 * initialise them to zero for new inodes. */
3982 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
3983 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3985 ext4_get_inode_flags(ei);
3986 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3987 i_uid = i_uid_read(inode);
3988 i_gid = i_gid_read(inode);
3989 if (!(test_opt(inode->i_sb, NO_UID32))) {
3990 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3991 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
3993 * Fix up interoperability with old kernels. Otherwise, old inodes get
3994 * re-used with the upper 16 bits of the uid/gid intact
3997 raw_inode->i_uid_high =
3998 cpu_to_le16(high_16_bits(i_uid));
3999 raw_inode->i_gid_high =
4000 cpu_to_le16(high_16_bits(i_gid));
4002 raw_inode->i_uid_high = 0;
4003 raw_inode->i_gid_high = 0;
4006 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4007 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4008 raw_inode->i_uid_high = 0;
4009 raw_inode->i_gid_high = 0;
4011 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4013 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4014 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4015 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4016 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4018 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4020 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4021 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4022 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4023 cpu_to_le32(EXT4_OS_HURD))
4024 raw_inode->i_file_acl_high =
4025 cpu_to_le16(ei->i_file_acl >> 32);
4026 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4027 ext4_isize_set(raw_inode, ei->i_disksize);
4028 if (ei->i_disksize > 0x7fffffffULL) {
4029 struct super_block *sb = inode->i_sb;
4030 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4031 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4032 EXT4_SB(sb)->s_es->s_rev_level ==
4033 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4034 /* If this is the first large file
4035 * created, add a flag to the superblock.
4037 err = ext4_journal_get_write_access(handle,
4038 EXT4_SB(sb)->s_sbh);
4041 ext4_update_dynamic_rev(sb);
4042 EXT4_SET_RO_COMPAT_FEATURE(sb,
4043 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4044 ext4_handle_sync(handle);
4045 err = ext4_handle_dirty_super_now(handle, sb);
4048 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4049 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4050 if (old_valid_dev(inode->i_rdev)) {
4051 raw_inode->i_block[0] =
4052 cpu_to_le32(old_encode_dev(inode->i_rdev));
4053 raw_inode->i_block[1] = 0;
4055 raw_inode->i_block[0] = 0;
4056 raw_inode->i_block[1] =
4057 cpu_to_le32(new_encode_dev(inode->i_rdev));
4058 raw_inode->i_block[2] = 0;
4061 for (block = 0; block < EXT4_N_BLOCKS; block++)
4062 raw_inode->i_block[block] = ei->i_data[block];
4064 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4065 if (ei->i_extra_isize) {
4066 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4067 raw_inode->i_version_hi =
4068 cpu_to_le32(inode->i_version >> 32);
4069 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4072 ext4_inode_csum_set(inode, raw_inode, ei);
4074 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4075 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4078 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4080 ext4_update_inode_fsync_trans(handle, inode, 0);
4083 ext4_std_error(inode->i_sb, err);
4088 * ext4_write_inode()
4090 * We are called from a few places:
4092 * - Within generic_file_write() for O_SYNC files.
4093 * Here, there will be no transaction running. We wait for any running
4094 * trasnaction to commit.
4096 * - Within sys_sync(), kupdate and such.
4097 * We wait on commit, if tol to.
4099 * - Within prune_icache() (PF_MEMALLOC == true)
4100 * Here we simply return. We can't afford to block kswapd on the
4103 * In all cases it is actually safe for us to return without doing anything,
4104 * because the inode has been copied into a raw inode buffer in
4105 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4108 * Note that we are absolutely dependent upon all inode dirtiers doing the
4109 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4110 * which we are interested.
4112 * It would be a bug for them to not do this. The code:
4114 * mark_inode_dirty(inode)
4116 * inode->i_size = expr;
4118 * is in error because a kswapd-driven write_inode() could occur while
4119 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4120 * will no longer be on the superblock's dirty inode list.
4122 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4126 if (current->flags & PF_MEMALLOC)
4129 if (EXT4_SB(inode->i_sb)->s_journal) {
4130 if (ext4_journal_current_handle()) {
4131 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4136 if (wbc->sync_mode != WB_SYNC_ALL)
4139 err = ext4_force_commit(inode->i_sb);
4141 struct ext4_iloc iloc;
4143 err = __ext4_get_inode_loc(inode, &iloc, 0);
4146 if (wbc->sync_mode == WB_SYNC_ALL)
4147 sync_dirty_buffer(iloc.bh);
4148 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4149 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4150 "IO error syncing inode");
4161 * Called from notify_change.
4163 * We want to trap VFS attempts to truncate the file as soon as
4164 * possible. In particular, we want to make sure that when the VFS
4165 * shrinks i_size, we put the inode on the orphan list and modify
4166 * i_disksize immediately, so that during the subsequent flushing of
4167 * dirty pages and freeing of disk blocks, we can guarantee that any
4168 * commit will leave the blocks being flushed in an unused state on
4169 * disk. (On recovery, the inode will get truncated and the blocks will
4170 * be freed, so we have a strong guarantee that no future commit will
4171 * leave these blocks visible to the user.)
4173 * Another thing we have to assure is that if we are in ordered mode
4174 * and inode is still attached to the committing transaction, we must
4175 * we start writeout of all the dirty pages which are being truncated.
4176 * This way we are sure that all the data written in the previous
4177 * transaction are already on disk (truncate waits for pages under
4180 * Called with inode->i_mutex down.
4182 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4184 struct inode *inode = dentry->d_inode;
4187 const unsigned int ia_valid = attr->ia_valid;
4189 error = inode_change_ok(inode, attr);
4193 if (is_quota_modification(inode, attr))
4194 dquot_initialize(inode);
4195 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4196 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4199 /* (user+group)*(old+new) structure, inode write (sb,
4200 * inode block, ? - but truncate inode update has it) */
4201 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4202 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4203 if (IS_ERR(handle)) {
4204 error = PTR_ERR(handle);
4207 error = dquot_transfer(inode, attr);
4209 ext4_journal_stop(handle);
4212 /* Update corresponding info in inode so that everything is in
4213 * one transaction */
4214 if (attr->ia_valid & ATTR_UID)
4215 inode->i_uid = attr->ia_uid;
4216 if (attr->ia_valid & ATTR_GID)
4217 inode->i_gid = attr->ia_gid;
4218 error = ext4_mark_inode_dirty(handle, inode);
4219 ext4_journal_stop(handle);
4222 if (attr->ia_valid & ATTR_SIZE) {
4223 inode_dio_wait(inode);
4225 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4226 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4228 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4233 if (S_ISREG(inode->i_mode) &&
4234 attr->ia_valid & ATTR_SIZE &&
4235 (attr->ia_size < inode->i_size)) {
4238 handle = ext4_journal_start(inode, 3);
4239 if (IS_ERR(handle)) {
4240 error = PTR_ERR(handle);
4243 if (ext4_handle_valid(handle)) {
4244 error = ext4_orphan_add(handle, inode);
4247 EXT4_I(inode)->i_disksize = attr->ia_size;
4248 rc = ext4_mark_inode_dirty(handle, inode);
4251 ext4_journal_stop(handle);
4253 if (ext4_should_order_data(inode)) {
4254 error = ext4_begin_ordered_truncate(inode,
4257 /* Do as much error cleanup as possible */
4258 handle = ext4_journal_start(inode, 3);
4259 if (IS_ERR(handle)) {
4260 ext4_orphan_del(NULL, inode);
4263 ext4_orphan_del(handle, inode);
4265 ext4_journal_stop(handle);
4271 if (attr->ia_valid & ATTR_SIZE) {
4272 if (attr->ia_size != i_size_read(inode))
4273 truncate_setsize(inode, attr->ia_size);
4274 ext4_truncate(inode);
4278 setattr_copy(inode, attr);
4279 mark_inode_dirty(inode);
4283 * If the call to ext4_truncate failed to get a transaction handle at
4284 * all, we need to clean up the in-core orphan list manually.
4286 if (orphan && inode->i_nlink)
4287 ext4_orphan_del(NULL, inode);
4289 if (!rc && (ia_valid & ATTR_MODE))
4290 rc = ext4_acl_chmod(inode);
4293 ext4_std_error(inode->i_sb, error);
4299 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4302 struct inode *inode;
4303 unsigned long delalloc_blocks;
4305 inode = dentry->d_inode;
4306 generic_fillattr(inode, stat);
4309 * We can't update i_blocks if the block allocation is delayed
4310 * otherwise in the case of system crash before the real block
4311 * allocation is done, we will have i_blocks inconsistent with
4312 * on-disk file blocks.
4313 * We always keep i_blocks updated together with real
4314 * allocation. But to not confuse with user, stat
4315 * will return the blocks that include the delayed allocation
4316 * blocks for this file.
4318 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4319 EXT4_I(inode)->i_reserved_data_blocks);
4321 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4325 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4327 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4328 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4329 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4333 * Account for index blocks, block groups bitmaps and block group
4334 * descriptor blocks if modify datablocks and index blocks
4335 * worse case, the indexs blocks spread over different block groups
4337 * If datablocks are discontiguous, they are possible to spread over
4338 * different block groups too. If they are contiuguous, with flexbg,
4339 * they could still across block group boundary.
4341 * Also account for superblock, inode, quota and xattr blocks
4343 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4345 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4351 * How many index blocks need to touch to modify nrblocks?
4352 * The "Chunk" flag indicating whether the nrblocks is
4353 * physically contiguous on disk
4355 * For Direct IO and fallocate, they calls get_block to allocate
4356 * one single extent at a time, so they could set the "Chunk" flag
4358 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4363 * Now let's see how many group bitmaps and group descriptors need
4373 if (groups > ngroups)
4375 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4376 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4378 /* bitmaps and block group descriptor blocks */
4379 ret += groups + gdpblocks;
4381 /* Blocks for super block, inode, quota and xattr blocks */
4382 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4388 * Calculate the total number of credits to reserve to fit
4389 * the modification of a single pages into a single transaction,
4390 * which may include multiple chunks of block allocations.
4392 * This could be called via ext4_write_begin()
4394 * We need to consider the worse case, when
4395 * one new block per extent.
4397 int ext4_writepage_trans_blocks(struct inode *inode)
4399 int bpp = ext4_journal_blocks_per_page(inode);
4402 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4404 /* Account for data blocks for journalled mode */
4405 if (ext4_should_journal_data(inode))
4411 * Calculate the journal credits for a chunk of data modification.
4413 * This is called from DIO, fallocate or whoever calling
4414 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4416 * journal buffers for data blocks are not included here, as DIO
4417 * and fallocate do no need to journal data buffers.
4419 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4421 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4425 * The caller must have previously called ext4_reserve_inode_write().
4426 * Give this, we know that the caller already has write access to iloc->bh.
4428 int ext4_mark_iloc_dirty(handle_t *handle,
4429 struct inode *inode, struct ext4_iloc *iloc)
4433 if (IS_I_VERSION(inode))
4434 inode_inc_iversion(inode);
4436 /* the do_update_inode consumes one bh->b_count */
4439 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4440 err = ext4_do_update_inode(handle, inode, iloc);
4446 * On success, We end up with an outstanding reference count against
4447 * iloc->bh. This _must_ be cleaned up later.
4451 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4452 struct ext4_iloc *iloc)
4456 err = ext4_get_inode_loc(inode, iloc);
4458 BUFFER_TRACE(iloc->bh, "get_write_access");
4459 err = ext4_journal_get_write_access(handle, iloc->bh);
4465 ext4_std_error(inode->i_sb, err);
4470 * Expand an inode by new_extra_isize bytes.
4471 * Returns 0 on success or negative error number on failure.
4473 static int ext4_expand_extra_isize(struct inode *inode,
4474 unsigned int new_extra_isize,
4475 struct ext4_iloc iloc,
4478 struct ext4_inode *raw_inode;
4479 struct ext4_xattr_ibody_header *header;
4481 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4484 raw_inode = ext4_raw_inode(&iloc);
4486 header = IHDR(inode, raw_inode);
4488 /* No extended attributes present */
4489 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4490 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4491 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4493 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4497 /* try to expand with EAs present */
4498 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4503 * What we do here is to mark the in-core inode as clean with respect to inode
4504 * dirtiness (it may still be data-dirty).
4505 * This means that the in-core inode may be reaped by prune_icache
4506 * without having to perform any I/O. This is a very good thing,
4507 * because *any* task may call prune_icache - even ones which
4508 * have a transaction open against a different journal.
4510 * Is this cheating? Not really. Sure, we haven't written the
4511 * inode out, but prune_icache isn't a user-visible syncing function.
4512 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4513 * we start and wait on commits.
4515 * Is this efficient/effective? Well, we're being nice to the system
4516 * by cleaning up our inodes proactively so they can be reaped
4517 * without I/O. But we are potentially leaving up to five seconds'
4518 * worth of inodes floating about which prune_icache wants us to
4519 * write out. One way to fix that would be to get prune_icache()
4520 * to do a write_super() to free up some memory. It has the desired
4523 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4525 struct ext4_iloc iloc;
4526 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4527 static unsigned int mnt_count;
4531 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4532 err = ext4_reserve_inode_write(handle, inode, &iloc);
4533 if (ext4_handle_valid(handle) &&
4534 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4535 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4537 * We need extra buffer credits since we may write into EA block
4538 * with this same handle. If journal_extend fails, then it will
4539 * only result in a minor loss of functionality for that inode.
4540 * If this is felt to be critical, then e2fsck should be run to
4541 * force a large enough s_min_extra_isize.
4543 if ((jbd2_journal_extend(handle,
4544 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4545 ret = ext4_expand_extra_isize(inode,
4546 sbi->s_want_extra_isize,
4549 ext4_set_inode_state(inode,
4550 EXT4_STATE_NO_EXPAND);
4552 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4553 ext4_warning(inode->i_sb,
4554 "Unable to expand inode %lu. Delete"
4555 " some EAs or run e2fsck.",
4558 le16_to_cpu(sbi->s_es->s_mnt_count);
4564 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4569 * ext4_dirty_inode() is called from __mark_inode_dirty()
4571 * We're really interested in the case where a file is being extended.
4572 * i_size has been changed by generic_commit_write() and we thus need
4573 * to include the updated inode in the current transaction.
4575 * Also, dquot_alloc_block() will always dirty the inode when blocks
4576 * are allocated to the file.
4578 * If the inode is marked synchronous, we don't honour that here - doing
4579 * so would cause a commit on atime updates, which we don't bother doing.
4580 * We handle synchronous inodes at the highest possible level.
4582 void ext4_dirty_inode(struct inode *inode, int flags)
4586 handle = ext4_journal_start(inode, 2);
4590 ext4_mark_inode_dirty(handle, inode);
4592 ext4_journal_stop(handle);
4599 * Bind an inode's backing buffer_head into this transaction, to prevent
4600 * it from being flushed to disk early. Unlike
4601 * ext4_reserve_inode_write, this leaves behind no bh reference and
4602 * returns no iloc structure, so the caller needs to repeat the iloc
4603 * lookup to mark the inode dirty later.
4605 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4607 struct ext4_iloc iloc;
4611 err = ext4_get_inode_loc(inode, &iloc);
4613 BUFFER_TRACE(iloc.bh, "get_write_access");
4614 err = jbd2_journal_get_write_access(handle, iloc.bh);
4616 err = ext4_handle_dirty_metadata(handle,
4622 ext4_std_error(inode->i_sb, err);
4627 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4634 * We have to be very careful here: changing a data block's
4635 * journaling status dynamically is dangerous. If we write a
4636 * data block to the journal, change the status and then delete
4637 * that block, we risk forgetting to revoke the old log record
4638 * from the journal and so a subsequent replay can corrupt data.
4639 * So, first we make sure that the journal is empty and that
4640 * nobody is changing anything.
4643 journal = EXT4_JOURNAL(inode);
4646 if (is_journal_aborted(journal))
4648 /* We have to allocate physical blocks for delalloc blocks
4649 * before flushing journal. otherwise delalloc blocks can not
4650 * be allocated any more. even more truncate on delalloc blocks
4651 * could trigger BUG by flushing delalloc blocks in journal.
4652 * There is no delalloc block in non-journal data mode.
4654 if (val && test_opt(inode->i_sb, DELALLOC)) {
4655 err = ext4_alloc_da_blocks(inode);
4660 jbd2_journal_lock_updates(journal);
4663 * OK, there are no updates running now, and all cached data is
4664 * synced to disk. We are now in a completely consistent state
4665 * which doesn't have anything in the journal, and we know that
4666 * no filesystem updates are running, so it is safe to modify
4667 * the inode's in-core data-journaling state flag now.
4671 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4673 jbd2_journal_flush(journal);
4674 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4676 ext4_set_aops(inode);
4678 jbd2_journal_unlock_updates(journal);
4680 /* Finally we can mark the inode as dirty. */
4682 handle = ext4_journal_start(inode, 1);
4684 return PTR_ERR(handle);
4686 err = ext4_mark_inode_dirty(handle, inode);
4687 ext4_handle_sync(handle);
4688 ext4_journal_stop(handle);
4689 ext4_std_error(inode->i_sb, err);
4694 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4696 return !buffer_mapped(bh);
4699 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4701 struct page *page = vmf->page;
4705 struct file *file = vma->vm_file;
4706 struct inode *inode = file->f_path.dentry->d_inode;
4707 struct address_space *mapping = inode->i_mapping;
4709 get_block_t *get_block;
4712 sb_start_pagefault(inode->i_sb);
4713 /* Delalloc case is easy... */
4714 if (test_opt(inode->i_sb, DELALLOC) &&
4715 !ext4_should_journal_data(inode) &&
4716 !ext4_nonda_switch(inode->i_sb)) {
4718 ret = __block_page_mkwrite(vma, vmf,
4719 ext4_da_get_block_prep);
4720 } while (ret == -ENOSPC &&
4721 ext4_should_retry_alloc(inode->i_sb, &retries));
4726 size = i_size_read(inode);
4727 /* Page got truncated from under us? */
4728 if (page->mapping != mapping || page_offset(page) > size) {
4730 ret = VM_FAULT_NOPAGE;
4734 if (page->index == size >> PAGE_CACHE_SHIFT)
4735 len = size & ~PAGE_CACHE_MASK;
4737 len = PAGE_CACHE_SIZE;
4739 * Return if we have all the buffers mapped. This avoids the need to do
4740 * journal_start/journal_stop which can block and take a long time
4742 if (page_has_buffers(page)) {
4743 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4744 ext4_bh_unmapped)) {
4745 /* Wait so that we don't change page under IO */
4746 wait_on_page_writeback(page);
4747 ret = VM_FAULT_LOCKED;
4752 /* OK, we need to fill the hole... */
4753 if (ext4_should_dioread_nolock(inode))
4754 get_block = ext4_get_block_write;
4756 get_block = ext4_get_block;
4758 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4759 if (IS_ERR(handle)) {
4760 ret = VM_FAULT_SIGBUS;
4763 ret = __block_page_mkwrite(vma, vmf, get_block);
4764 if (!ret && ext4_should_journal_data(inode)) {
4765 if (walk_page_buffers(handle, page_buffers(page), 0,
4766 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4768 ret = VM_FAULT_SIGBUS;
4769 ext4_journal_stop(handle);
4772 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4774 ext4_journal_stop(handle);
4775 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4778 ret = block_page_mkwrite_return(ret);
4780 sb_end_pagefault(inode->i_sb);