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 __ext4_journalled_writepage(struct page *page, unsigned int len);
136 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
137 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
138 struct inode *inode, struct page *page, loff_t from,
139 loff_t length, int flags);
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode *inode)
146 int ea_blocks = EXT4_I(inode)->i_file_acl ?
147 (inode->i_sb->s_blocksize >> 9) : 0;
149 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
157 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
168 BUG_ON(EXT4_JOURNAL(inode) == NULL);
169 jbd_debug(2, "restarting handle %p\n", handle);
170 up_write(&EXT4_I(inode)->i_data_sem);
171 ret = ext4_journal_restart(handle, nblocks);
172 down_write(&EXT4_I(inode)->i_data_sem);
173 ext4_discard_preallocations(inode);
179 * Called at the last iput() if i_nlink is zero.
181 void ext4_evict_inode(struct inode *inode)
186 trace_ext4_evict_inode(inode);
188 ext4_ioend_wait(inode);
190 if (inode->i_nlink) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode) &&
210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_log_start_commit(journal, commit_tid);
215 jbd2_log_wait_commit(journal, commit_tid);
216 filemap_write_and_wait(&inode->i_data);
218 truncate_inode_pages(&inode->i_data, 0);
222 if (!is_bad_inode(inode))
223 dquot_initialize(inode);
225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
227 truncate_inode_pages(&inode->i_data, 0);
229 if (is_bad_inode(inode))
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
236 sb_start_intwrite(inode->i_sb);
237 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
238 if (IS_ERR(handle)) {
239 ext4_std_error(inode->i_sb, PTR_ERR(handle));
241 * If we're going to skip the normal cleanup, we still need to
242 * make sure that the in-core orphan linked list is properly
245 ext4_orphan_del(NULL, inode);
246 sb_end_intwrite(inode->i_sb);
251 ext4_handle_sync(handle);
253 err = ext4_mark_inode_dirty(handle, inode);
255 ext4_warning(inode->i_sb,
256 "couldn't mark inode dirty (err %d)", err);
260 ext4_truncate(inode);
263 * ext4_ext_truncate() doesn't reserve any slop when it
264 * restarts journal transactions; therefore there may not be
265 * enough credits left in the handle to remove the inode from
266 * the orphan list and set the dtime field.
268 if (!ext4_handle_has_enough_credits(handle, 3)) {
269 err = ext4_journal_extend(handle, 3);
271 err = ext4_journal_restart(handle, 3);
273 ext4_warning(inode->i_sb,
274 "couldn't extend journal (err %d)", err);
276 ext4_journal_stop(handle);
277 ext4_orphan_del(NULL, inode);
278 sb_end_intwrite(inode->i_sb);
284 * Kill off the orphan record which ext4_truncate created.
285 * AKPM: I think this can be inside the above `if'.
286 * Note that ext4_orphan_del() has to be able to cope with the
287 * deletion of a non-existent orphan - this is because we don't
288 * know if ext4_truncate() actually created an orphan record.
289 * (Well, we could do this if we need to, but heck - it works)
291 ext4_orphan_del(handle, inode);
292 EXT4_I(inode)->i_dtime = get_seconds();
295 * One subtle ordering requirement: if anything has gone wrong
296 * (transaction abort, IO errors, whatever), then we can still
297 * do these next steps (the fs will already have been marked as
298 * having errors), but we can't free the inode if the mark_dirty
301 if (ext4_mark_inode_dirty(handle, inode))
302 /* If that failed, just do the required in-core inode clear. */
303 ext4_clear_inode(inode);
305 ext4_free_inode(handle, inode);
306 ext4_journal_stop(handle);
307 sb_end_intwrite(inode->i_sb);
310 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
314 qsize_t *ext4_get_reserved_space(struct inode *inode)
316 return &EXT4_I(inode)->i_reserved_quota;
321 * Calculate the number of metadata blocks need to reserve
322 * to allocate a block located at @lblock
324 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
326 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
327 return ext4_ext_calc_metadata_amount(inode, lblock);
329 return ext4_ind_calc_metadata_amount(inode, lblock);
333 * Called with i_data_sem down, which is important since we can call
334 * ext4_discard_preallocations() from here.
336 void ext4_da_update_reserve_space(struct inode *inode,
337 int used, int quota_claim)
339 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
340 struct ext4_inode_info *ei = EXT4_I(inode);
342 spin_lock(&ei->i_block_reservation_lock);
343 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
344 if (unlikely(used > ei->i_reserved_data_blocks)) {
345 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
346 "with only %d reserved data blocks",
347 __func__, inode->i_ino, used,
348 ei->i_reserved_data_blocks);
350 used = ei->i_reserved_data_blocks;
353 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
354 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
355 "with only %d reserved metadata blocks\n", __func__,
356 inode->i_ino, ei->i_allocated_meta_blocks,
357 ei->i_reserved_meta_blocks);
359 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
362 /* Update per-inode reservations */
363 ei->i_reserved_data_blocks -= used;
364 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
365 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
366 used + ei->i_allocated_meta_blocks);
367 ei->i_allocated_meta_blocks = 0;
369 if (ei->i_reserved_data_blocks == 0) {
371 * We can release all of the reserved metadata blocks
372 * only when we have written all of the delayed
375 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
376 ei->i_reserved_meta_blocks);
377 ei->i_reserved_meta_blocks = 0;
378 ei->i_da_metadata_calc_len = 0;
380 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
382 /* Update quota subsystem for data blocks */
384 dquot_claim_block(inode, EXT4_C2B(sbi, used));
387 * We did fallocate with an offset that is already delayed
388 * allocated. So on delayed allocated writeback we should
389 * not re-claim the quota for fallocated blocks.
391 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
395 * If we have done all the pending block allocations and if
396 * there aren't any writers on the inode, we can discard the
397 * inode's preallocations.
399 if ((ei->i_reserved_data_blocks == 0) &&
400 (atomic_read(&inode->i_writecount) == 0))
401 ext4_discard_preallocations(inode);
404 static int __check_block_validity(struct inode *inode, const char *func,
406 struct ext4_map_blocks *map)
408 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
410 ext4_error_inode(inode, func, line, map->m_pblk,
411 "lblock %lu mapped to illegal pblock "
412 "(length %d)", (unsigned long) map->m_lblk,
419 #define check_block_validity(inode, map) \
420 __check_block_validity((inode), __func__, __LINE__, (map))
423 * Return the number of contiguous dirty pages in a given inode
424 * starting at page frame idx.
426 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
427 unsigned int max_pages)
429 struct address_space *mapping = inode->i_mapping;
433 int i, nr_pages, done = 0;
437 pagevec_init(&pvec, 0);
440 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
442 (pgoff_t)PAGEVEC_SIZE);
445 for (i = 0; i < nr_pages; i++) {
446 struct page *page = pvec.pages[i];
447 struct buffer_head *bh, *head;
450 if (unlikely(page->mapping != mapping) ||
452 PageWriteback(page) ||
453 page->index != idx) {
458 if (page_has_buffers(page)) {
459 bh = head = page_buffers(page);
461 if (!buffer_delay(bh) &&
462 !buffer_unwritten(bh))
464 bh = bh->b_this_page;
465 } while (!done && (bh != head));
472 if (num >= max_pages) {
477 pagevec_release(&pvec);
483 * The ext4_map_blocks() function tries to look up the requested blocks,
484 * and returns if the blocks are already mapped.
486 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
487 * and store the allocated blocks in the result buffer head and mark it
490 * If file type is extents based, it will call ext4_ext_map_blocks(),
491 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
494 * On success, it returns the number of blocks being mapped or allocate.
495 * if create==0 and the blocks are pre-allocated and uninitialized block,
496 * the result buffer head is unmapped. If the create ==1, it will make sure
497 * the buffer head is mapped.
499 * It returns 0 if plain look up failed (blocks have not been allocated), in
500 * that case, buffer head is unmapped
502 * It returns the error in case of allocation failure.
504 int ext4_map_blocks(handle_t *handle, struct inode *inode,
505 struct ext4_map_blocks *map, int flags)
510 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
511 "logical block %lu\n", inode->i_ino, flags, map->m_len,
512 (unsigned long) map->m_lblk);
514 * Try to see if we can get the block without requesting a new
517 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
518 down_read((&EXT4_I(inode)->i_data_sem));
519 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
520 retval = ext4_ext_map_blocks(handle, inode, map, flags &
521 EXT4_GET_BLOCKS_KEEP_SIZE);
523 retval = ext4_ind_map_blocks(handle, inode, map, flags &
524 EXT4_GET_BLOCKS_KEEP_SIZE);
526 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
527 up_read((&EXT4_I(inode)->i_data_sem));
529 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
531 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
532 /* delayed alloc may be allocated by fallocate and
533 * coverted to initialized by directIO.
534 * we need to handle delayed extent here.
536 down_write((&EXT4_I(inode)->i_data_sem));
539 ret = check_block_validity(inode, map);
544 /* If it is only a block(s) look up */
545 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
549 * Returns if the blocks have already allocated
551 * Note that if blocks have been preallocated
552 * ext4_ext_get_block() returns the create = 0
553 * with buffer head unmapped.
555 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
559 * When we call get_blocks without the create flag, the
560 * BH_Unwritten flag could have gotten set if the blocks
561 * requested were part of a uninitialized extent. We need to
562 * clear this flag now that we are committed to convert all or
563 * part of the uninitialized extent to be an initialized
564 * extent. This is because we need to avoid the combination
565 * of BH_Unwritten and BH_Mapped flags being simultaneously
566 * set on the buffer_head.
568 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
571 * New blocks allocate and/or writing to uninitialized extent
572 * will possibly result in updating i_data, so we take
573 * the write lock of i_data_sem, and call get_blocks()
574 * with create == 1 flag.
576 down_write((&EXT4_I(inode)->i_data_sem));
579 * if the caller is from delayed allocation writeout path
580 * we have already reserved fs blocks for allocation
581 * let the underlying get_block() function know to
582 * avoid double accounting
584 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
585 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
587 * We need to check for EXT4 here because migrate
588 * could have changed the inode type in between
590 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
591 retval = ext4_ext_map_blocks(handle, inode, map, flags);
593 retval = ext4_ind_map_blocks(handle, inode, map, flags);
595 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
597 * We allocated new blocks which will result in
598 * i_data's format changing. Force the migrate
599 * to fail by clearing migrate flags
601 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
605 * Update reserved blocks/metadata blocks after successful
606 * block allocation which had been deferred till now. We don't
607 * support fallocate for non extent files. So we can update
608 * reserve space here.
611 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
612 ext4_da_update_reserve_space(inode, retval, 1);
614 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
615 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
617 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
620 /* delayed allocation blocks has been allocated */
621 ret = ext4_es_remove_extent(inode, map->m_lblk,
628 up_write((&EXT4_I(inode)->i_data_sem));
629 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
630 int ret = check_block_validity(inode, map);
637 /* Maximum number of blocks we map for direct IO at once. */
638 #define DIO_MAX_BLOCKS 4096
640 static int _ext4_get_block(struct inode *inode, sector_t iblock,
641 struct buffer_head *bh, int flags)
643 handle_t *handle = ext4_journal_current_handle();
644 struct ext4_map_blocks map;
645 int ret = 0, started = 0;
648 if (ext4_has_inline_data(inode))
652 map.m_len = bh->b_size >> inode->i_blkbits;
654 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
655 /* Direct IO write... */
656 if (map.m_len > DIO_MAX_BLOCKS)
657 map.m_len = DIO_MAX_BLOCKS;
658 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
659 handle = ext4_journal_start(inode, dio_credits);
660 if (IS_ERR(handle)) {
661 ret = PTR_ERR(handle);
667 ret = ext4_map_blocks(handle, inode, &map, flags);
669 map_bh(bh, inode->i_sb, map.m_pblk);
670 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
671 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
675 ext4_journal_stop(handle);
679 int ext4_get_block(struct inode *inode, sector_t iblock,
680 struct buffer_head *bh, int create)
682 return _ext4_get_block(inode, iblock, bh,
683 create ? EXT4_GET_BLOCKS_CREATE : 0);
687 * `handle' can be NULL if create is zero
689 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
690 ext4_lblk_t block, int create, int *errp)
692 struct ext4_map_blocks map;
693 struct buffer_head *bh;
696 J_ASSERT(handle != NULL || create == 0);
700 err = ext4_map_blocks(handle, inode, &map,
701 create ? EXT4_GET_BLOCKS_CREATE : 0);
703 /* ensure we send some value back into *errp */
711 bh = sb_getblk(inode->i_sb, map.m_pblk);
716 if (map.m_flags & EXT4_MAP_NEW) {
717 J_ASSERT(create != 0);
718 J_ASSERT(handle != NULL);
721 * Now that we do not always journal data, we should
722 * keep in mind whether this should always journal the
723 * new buffer as metadata. For now, regular file
724 * writes use ext4_get_block instead, so it's not a
728 BUFFER_TRACE(bh, "call get_create_access");
729 fatal = ext4_journal_get_create_access(handle, bh);
730 if (!fatal && !buffer_uptodate(bh)) {
731 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
732 set_buffer_uptodate(bh);
735 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
736 err = ext4_handle_dirty_metadata(handle, inode, bh);
740 BUFFER_TRACE(bh, "not a new buffer");
750 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
751 ext4_lblk_t block, int create, int *err)
753 struct buffer_head *bh;
755 bh = ext4_getblk(handle, inode, block, create, err);
758 if (buffer_uptodate(bh))
760 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
762 if (buffer_uptodate(bh))
769 int ext4_walk_page_buffers(handle_t *handle,
770 struct buffer_head *head,
774 int (*fn)(handle_t *handle,
775 struct buffer_head *bh))
777 struct buffer_head *bh;
778 unsigned block_start, block_end;
779 unsigned blocksize = head->b_size;
781 struct buffer_head *next;
783 for (bh = head, block_start = 0;
784 ret == 0 && (bh != head || !block_start);
785 block_start = block_end, bh = next) {
786 next = bh->b_this_page;
787 block_end = block_start + blocksize;
788 if (block_end <= from || block_start >= to) {
789 if (partial && !buffer_uptodate(bh))
793 err = (*fn)(handle, bh);
801 * To preserve ordering, it is essential that the hole instantiation and
802 * the data write be encapsulated in a single transaction. We cannot
803 * close off a transaction and start a new one between the ext4_get_block()
804 * and the commit_write(). So doing the jbd2_journal_start at the start of
805 * prepare_write() is the right place.
807 * Also, this function can nest inside ext4_writepage(). In that case, we
808 * *know* that ext4_writepage() has generated enough buffer credits to do the
809 * whole page. So we won't block on the journal in that case, which is good,
810 * because the caller may be PF_MEMALLOC.
812 * By accident, ext4 can be reentered when a transaction is open via
813 * quota file writes. If we were to commit the transaction while thus
814 * reentered, there can be a deadlock - we would be holding a quota
815 * lock, and the commit would never complete if another thread had a
816 * transaction open and was blocking on the quota lock - a ranking
819 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
820 * will _not_ run commit under these circumstances because handle->h_ref
821 * is elevated. We'll still have enough credits for the tiny quotafile
824 int do_journal_get_write_access(handle_t *handle,
825 struct buffer_head *bh)
827 int dirty = buffer_dirty(bh);
830 if (!buffer_mapped(bh) || buffer_freed(bh))
833 * __block_write_begin() could have dirtied some buffers. Clean
834 * the dirty bit as jbd2_journal_get_write_access() could complain
835 * otherwise about fs integrity issues. Setting of the dirty bit
836 * by __block_write_begin() isn't a real problem here as we clear
837 * the bit before releasing a page lock and thus writeback cannot
838 * ever write the buffer.
841 clear_buffer_dirty(bh);
842 ret = ext4_journal_get_write_access(handle, bh);
844 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
848 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
849 struct buffer_head *bh_result, int create);
850 static int ext4_write_begin(struct file *file, struct address_space *mapping,
851 loff_t pos, unsigned len, unsigned flags,
852 struct page **pagep, void **fsdata)
854 struct inode *inode = mapping->host;
855 int ret, needed_blocks;
862 trace_ext4_write_begin(inode, pos, len, flags);
864 * Reserve one block more for addition to orphan list in case
865 * we allocate blocks but write fails for some reason
867 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
868 index = pos >> PAGE_CACHE_SHIFT;
869 from = pos & (PAGE_CACHE_SIZE - 1);
872 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
873 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
884 handle = ext4_journal_start(inode, needed_blocks);
885 if (IS_ERR(handle)) {
886 ret = PTR_ERR(handle);
890 /* We cannot recurse into the filesystem as the transaction is already
892 flags |= AOP_FLAG_NOFS;
894 page = grab_cache_page_write_begin(mapping, index, flags);
896 ext4_journal_stop(handle);
903 if (ext4_should_dioread_nolock(inode))
904 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
906 ret = __block_write_begin(page, pos, len, ext4_get_block);
908 if (!ret && ext4_should_journal_data(inode)) {
909 ret = ext4_walk_page_buffers(handle, page_buffers(page),
911 do_journal_get_write_access);
916 page_cache_release(page);
918 * __block_write_begin may have instantiated a few blocks
919 * outside i_size. Trim these off again. Don't need
920 * i_size_read because we hold i_mutex.
922 * Add inode to orphan list in case we crash before
925 if (pos + len > inode->i_size && ext4_can_truncate(inode))
926 ext4_orphan_add(handle, inode);
928 ext4_journal_stop(handle);
929 if (pos + len > inode->i_size) {
930 ext4_truncate_failed_write(inode);
932 * If truncate failed early the inode might
933 * still be on the orphan list; we need to
934 * make sure the inode is removed from the
935 * orphan list in that case.
938 ext4_orphan_del(NULL, inode);
942 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
948 /* For write_end() in data=journal mode */
949 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
951 if (!buffer_mapped(bh) || buffer_freed(bh))
953 set_buffer_uptodate(bh);
954 return ext4_handle_dirty_metadata(handle, NULL, bh);
957 static int ext4_generic_write_end(struct file *file,
958 struct address_space *mapping,
959 loff_t pos, unsigned len, unsigned copied,
960 struct page *page, void *fsdata)
962 int i_size_changed = 0;
963 struct inode *inode = mapping->host;
964 handle_t *handle = ext4_journal_current_handle();
966 if (ext4_has_inline_data(inode))
967 copied = ext4_write_inline_data_end(inode, pos, len,
970 copied = block_write_end(file, mapping, pos,
971 len, copied, page, fsdata);
974 * No need to use i_size_read() here, the i_size
975 * cannot change under us because we hold i_mutex.
977 * But it's important to update i_size while still holding page lock:
978 * page writeout could otherwise come in and zero beyond i_size.
980 if (pos + copied > inode->i_size) {
981 i_size_write(inode, pos + copied);
985 if (pos + copied > EXT4_I(inode)->i_disksize) {
986 /* We need to mark inode dirty even if
987 * new_i_size is less that inode->i_size
988 * bu greater than i_disksize.(hint delalloc)
990 ext4_update_i_disksize(inode, (pos + copied));
994 page_cache_release(page);
997 * Don't mark the inode dirty under page lock. First, it unnecessarily
998 * makes the holding time of page lock longer. Second, it forces lock
999 * ordering of page lock and transaction start for journaling
1003 ext4_mark_inode_dirty(handle, inode);
1009 * We need to pick up the new inode size which generic_commit_write gave us
1010 * `file' can be NULL - eg, when called from page_symlink().
1012 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1013 * buffers are managed internally.
1015 static int ext4_ordered_write_end(struct file *file,
1016 struct address_space *mapping,
1017 loff_t pos, unsigned len, unsigned copied,
1018 struct page *page, void *fsdata)
1020 handle_t *handle = ext4_journal_current_handle();
1021 struct inode *inode = mapping->host;
1024 trace_ext4_ordered_write_end(inode, pos, len, copied);
1025 ret = ext4_jbd2_file_inode(handle, inode);
1028 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1031 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1032 /* if we have allocated more blocks and copied
1033 * less. We will have blocks allocated outside
1034 * inode->i_size. So truncate them
1036 ext4_orphan_add(handle, inode);
1041 page_cache_release(page);
1044 ret2 = ext4_journal_stop(handle);
1048 if (pos + len > inode->i_size) {
1049 ext4_truncate_failed_write(inode);
1051 * If truncate failed early the inode might still be
1052 * on the orphan list; we need to make sure the inode
1053 * is removed from the orphan list in that case.
1056 ext4_orphan_del(NULL, inode);
1060 return ret ? ret : copied;
1063 static int ext4_writeback_write_end(struct file *file,
1064 struct address_space *mapping,
1065 loff_t pos, unsigned len, unsigned copied,
1066 struct page *page, void *fsdata)
1068 handle_t *handle = ext4_journal_current_handle();
1069 struct inode *inode = mapping->host;
1072 trace_ext4_writeback_write_end(inode, pos, len, copied);
1073 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1076 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1077 /* if we have allocated more blocks and copied
1078 * less. We will have blocks allocated outside
1079 * inode->i_size. So truncate them
1081 ext4_orphan_add(handle, inode);
1086 ret2 = ext4_journal_stop(handle);
1090 if (pos + len > inode->i_size) {
1091 ext4_truncate_failed_write(inode);
1093 * If truncate failed early the inode might still be
1094 * on the orphan list; we need to make sure the inode
1095 * is removed from the orphan list in that case.
1098 ext4_orphan_del(NULL, inode);
1101 return ret ? ret : copied;
1104 static int ext4_journalled_write_end(struct file *file,
1105 struct address_space *mapping,
1106 loff_t pos, unsigned len, unsigned copied,
1107 struct page *page, void *fsdata)
1109 handle_t *handle = ext4_journal_current_handle();
1110 struct inode *inode = mapping->host;
1116 trace_ext4_journalled_write_end(inode, pos, len, copied);
1117 from = pos & (PAGE_CACHE_SIZE - 1);
1120 BUG_ON(!ext4_handle_valid(handle));
1122 if (ext4_has_inline_data(inode))
1123 copied = ext4_write_inline_data_end(inode, pos, len,
1127 if (!PageUptodate(page))
1129 page_zero_new_buffers(page, from+copied, to);
1132 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1133 to, &partial, write_end_fn);
1135 SetPageUptodate(page);
1137 new_i_size = pos + copied;
1138 if (new_i_size > inode->i_size)
1139 i_size_write(inode, pos+copied);
1140 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1141 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1142 if (new_i_size > EXT4_I(inode)->i_disksize) {
1143 ext4_update_i_disksize(inode, new_i_size);
1144 ret2 = ext4_mark_inode_dirty(handle, inode);
1150 page_cache_release(page);
1151 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1152 /* if we have allocated more blocks and copied
1153 * less. We will have blocks allocated outside
1154 * inode->i_size. So truncate them
1156 ext4_orphan_add(handle, inode);
1158 ret2 = ext4_journal_stop(handle);
1161 if (pos + len > inode->i_size) {
1162 ext4_truncate_failed_write(inode);
1164 * If truncate failed early the inode might still be
1165 * on the orphan list; we need to make sure the inode
1166 * is removed from the orphan list in that case.
1169 ext4_orphan_del(NULL, inode);
1172 return ret ? ret : copied;
1176 * Reserve a single cluster located at lblock
1178 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1181 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1182 struct ext4_inode_info *ei = EXT4_I(inode);
1183 unsigned int md_needed;
1185 ext4_lblk_t save_last_lblock;
1189 * We will charge metadata quota at writeout time; this saves
1190 * us from metadata over-estimation, though we may go over by
1191 * a small amount in the end. Here we just reserve for data.
1193 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1198 * recalculate the amount of metadata blocks to reserve
1199 * in order to allocate nrblocks
1200 * worse case is one extent per block
1203 spin_lock(&ei->i_block_reservation_lock);
1205 * ext4_calc_metadata_amount() has side effects, which we have
1206 * to be prepared undo if we fail to claim space.
1208 save_len = ei->i_da_metadata_calc_len;
1209 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1210 md_needed = EXT4_NUM_B2C(sbi,
1211 ext4_calc_metadata_amount(inode, lblock));
1212 trace_ext4_da_reserve_space(inode, md_needed);
1215 * We do still charge estimated metadata to the sb though;
1216 * we cannot afford to run out of free blocks.
1218 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1219 ei->i_da_metadata_calc_len = save_len;
1220 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1221 spin_unlock(&ei->i_block_reservation_lock);
1222 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1226 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1229 ei->i_reserved_data_blocks++;
1230 ei->i_reserved_meta_blocks += md_needed;
1231 spin_unlock(&ei->i_block_reservation_lock);
1233 return 0; /* success */
1236 static void ext4_da_release_space(struct inode *inode, int to_free)
1238 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1239 struct ext4_inode_info *ei = EXT4_I(inode);
1242 return; /* Nothing to release, exit */
1244 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1246 trace_ext4_da_release_space(inode, to_free);
1247 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1249 * if there aren't enough reserved blocks, then the
1250 * counter is messed up somewhere. Since this
1251 * function is called from invalidate page, it's
1252 * harmless to return without any action.
1254 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1255 "ino %lu, to_free %d with only %d reserved "
1256 "data blocks", inode->i_ino, to_free,
1257 ei->i_reserved_data_blocks);
1259 to_free = ei->i_reserved_data_blocks;
1261 ei->i_reserved_data_blocks -= to_free;
1263 if (ei->i_reserved_data_blocks == 0) {
1265 * We can release all of the reserved metadata blocks
1266 * only when we have written all of the delayed
1267 * allocation blocks.
1268 * Note that in case of bigalloc, i_reserved_meta_blocks,
1269 * i_reserved_data_blocks, etc. refer to number of clusters.
1271 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1272 ei->i_reserved_meta_blocks);
1273 ei->i_reserved_meta_blocks = 0;
1274 ei->i_da_metadata_calc_len = 0;
1277 /* update fs dirty data blocks counter */
1278 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1280 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1282 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1285 static void ext4_da_page_release_reservation(struct page *page,
1286 unsigned long offset)
1289 struct buffer_head *head, *bh;
1290 unsigned int curr_off = 0;
1291 struct inode *inode = page->mapping->host;
1292 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1296 head = page_buffers(page);
1299 unsigned int next_off = curr_off + bh->b_size;
1301 if ((offset <= curr_off) && (buffer_delay(bh))) {
1303 clear_buffer_delay(bh);
1305 curr_off = next_off;
1306 } while ((bh = bh->b_this_page) != head);
1309 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1310 ext4_es_remove_extent(inode, lblk, to_release);
1313 /* If we have released all the blocks belonging to a cluster, then we
1314 * need to release the reserved space for that cluster. */
1315 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1316 while (num_clusters > 0) {
1317 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1318 ((num_clusters - 1) << sbi->s_cluster_bits);
1319 if (sbi->s_cluster_ratio == 1 ||
1320 !ext4_find_delalloc_cluster(inode, lblk))
1321 ext4_da_release_space(inode, 1);
1328 * Delayed allocation stuff
1332 * mpage_da_submit_io - walks through extent of pages and try to write
1333 * them with writepage() call back
1335 * @mpd->inode: inode
1336 * @mpd->first_page: first page of the extent
1337 * @mpd->next_page: page after the last page of the extent
1339 * By the time mpage_da_submit_io() is called we expect all blocks
1340 * to be allocated. this may be wrong if allocation failed.
1342 * As pages are already locked by write_cache_pages(), we can't use it
1344 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1345 struct ext4_map_blocks *map)
1347 struct pagevec pvec;
1348 unsigned long index, end;
1349 int ret = 0, err, nr_pages, i;
1350 struct inode *inode = mpd->inode;
1351 struct address_space *mapping = inode->i_mapping;
1352 loff_t size = i_size_read(inode);
1353 unsigned int len, block_start;
1354 struct buffer_head *bh, *page_bufs = NULL;
1355 sector_t pblock = 0, cur_logical = 0;
1356 struct ext4_io_submit io_submit;
1358 BUG_ON(mpd->next_page <= mpd->first_page);
1359 memset(&io_submit, 0, sizeof(io_submit));
1361 * We need to start from the first_page to the next_page - 1
1362 * to make sure we also write the mapped dirty buffer_heads.
1363 * If we look at mpd->b_blocknr we would only be looking
1364 * at the currently mapped buffer_heads.
1366 index = mpd->first_page;
1367 end = mpd->next_page - 1;
1369 pagevec_init(&pvec, 0);
1370 while (index <= end) {
1371 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1374 for (i = 0; i < nr_pages; i++) {
1376 struct page *page = pvec.pages[i];
1378 index = page->index;
1382 if (index == size >> PAGE_CACHE_SHIFT)
1383 len = size & ~PAGE_CACHE_MASK;
1385 len = PAGE_CACHE_SIZE;
1387 cur_logical = index << (PAGE_CACHE_SHIFT -
1389 pblock = map->m_pblk + (cur_logical -
1394 BUG_ON(!PageLocked(page));
1395 BUG_ON(PageWriteback(page));
1397 bh = page_bufs = page_buffers(page);
1400 if (map && (cur_logical >= map->m_lblk) &&
1401 (cur_logical <= (map->m_lblk +
1402 (map->m_len - 1)))) {
1403 if (buffer_delay(bh)) {
1404 clear_buffer_delay(bh);
1405 bh->b_blocknr = pblock;
1407 if (buffer_unwritten(bh) ||
1409 BUG_ON(bh->b_blocknr != pblock);
1410 if (map->m_flags & EXT4_MAP_UNINIT)
1411 set_buffer_uninit(bh);
1412 clear_buffer_unwritten(bh);
1416 * skip page if block allocation undone and
1419 if (ext4_bh_delay_or_unwritten(NULL, bh))
1421 bh = bh->b_this_page;
1422 block_start += bh->b_size;
1425 } while (bh != page_bufs);
1432 clear_page_dirty_for_io(page);
1433 err = ext4_bio_write_page(&io_submit, page, len,
1436 mpd->pages_written++;
1438 * In error case, we have to continue because
1439 * remaining pages are still locked
1444 pagevec_release(&pvec);
1446 ext4_io_submit(&io_submit);
1450 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1454 struct pagevec pvec;
1455 struct inode *inode = mpd->inode;
1456 struct address_space *mapping = inode->i_mapping;
1457 ext4_lblk_t start, last;
1459 index = mpd->first_page;
1460 end = mpd->next_page - 1;
1462 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1463 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1464 ext4_es_remove_extent(inode, start, last - start + 1);
1466 pagevec_init(&pvec, 0);
1467 while (index <= end) {
1468 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1471 for (i = 0; i < nr_pages; i++) {
1472 struct page *page = pvec.pages[i];
1473 if (page->index > end)
1475 BUG_ON(!PageLocked(page));
1476 BUG_ON(PageWriteback(page));
1477 block_invalidatepage(page, 0);
1478 ClearPageUptodate(page);
1481 index = pvec.pages[nr_pages - 1]->index + 1;
1482 pagevec_release(&pvec);
1487 static void ext4_print_free_blocks(struct inode *inode)
1489 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1490 struct super_block *sb = inode->i_sb;
1492 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1493 EXT4_C2B(EXT4_SB(inode->i_sb),
1494 ext4_count_free_clusters(inode->i_sb)));
1495 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1496 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1497 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1498 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1499 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1500 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1501 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1502 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1503 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1504 EXT4_I(inode)->i_reserved_data_blocks);
1505 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1506 EXT4_I(inode)->i_reserved_meta_blocks);
1511 * mpage_da_map_and_submit - go through given space, map them
1512 * if necessary, and then submit them for I/O
1514 * @mpd - bh describing space
1516 * The function skips space we know is already mapped to disk blocks.
1519 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1521 int err, blks, get_blocks_flags;
1522 struct ext4_map_blocks map, *mapp = NULL;
1523 sector_t next = mpd->b_blocknr;
1524 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1525 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1526 handle_t *handle = NULL;
1529 * If the blocks are mapped already, or we couldn't accumulate
1530 * any blocks, then proceed immediately to the submission stage.
1532 if ((mpd->b_size == 0) ||
1533 ((mpd->b_state & (1 << BH_Mapped)) &&
1534 !(mpd->b_state & (1 << BH_Delay)) &&
1535 !(mpd->b_state & (1 << BH_Unwritten))))
1538 handle = ext4_journal_current_handle();
1542 * Call ext4_map_blocks() to allocate any delayed allocation
1543 * blocks, or to convert an uninitialized extent to be
1544 * initialized (in the case where we have written into
1545 * one or more preallocated blocks).
1547 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1548 * indicate that we are on the delayed allocation path. This
1549 * affects functions in many different parts of the allocation
1550 * call path. This flag exists primarily because we don't
1551 * want to change *many* call functions, so ext4_map_blocks()
1552 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1553 * inode's allocation semaphore is taken.
1555 * If the blocks in questions were delalloc blocks, set
1556 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1557 * variables are updated after the blocks have been allocated.
1560 map.m_len = max_blocks;
1561 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1562 if (ext4_should_dioread_nolock(mpd->inode))
1563 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1564 if (mpd->b_state & (1 << BH_Delay))
1565 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1567 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1569 struct super_block *sb = mpd->inode->i_sb;
1573 * If get block returns EAGAIN or ENOSPC and there
1574 * appears to be free blocks we will just let
1575 * mpage_da_submit_io() unlock all of the pages.
1580 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1586 * get block failure will cause us to loop in
1587 * writepages, because a_ops->writepage won't be able
1588 * to make progress. The page will be redirtied by
1589 * writepage and writepages will again try to write
1592 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1593 ext4_msg(sb, KERN_CRIT,
1594 "delayed block allocation failed for inode %lu "
1595 "at logical offset %llu with max blocks %zd "
1596 "with error %d", mpd->inode->i_ino,
1597 (unsigned long long) next,
1598 mpd->b_size >> mpd->inode->i_blkbits, err);
1599 ext4_msg(sb, KERN_CRIT,
1600 "This should not happen!! Data will be lost\n");
1602 ext4_print_free_blocks(mpd->inode);
1604 /* invalidate all the pages */
1605 ext4_da_block_invalidatepages(mpd);
1607 /* Mark this page range as having been completed */
1614 if (map.m_flags & EXT4_MAP_NEW) {
1615 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1618 for (i = 0; i < map.m_len; i++)
1619 unmap_underlying_metadata(bdev, map.m_pblk + i);
1623 * Update on-disk size along with block allocation.
1625 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1626 if (disksize > i_size_read(mpd->inode))
1627 disksize = i_size_read(mpd->inode);
1628 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1629 ext4_update_i_disksize(mpd->inode, disksize);
1630 err = ext4_mark_inode_dirty(handle, mpd->inode);
1632 ext4_error(mpd->inode->i_sb,
1633 "Failed to mark inode %lu dirty",
1638 mpage_da_submit_io(mpd, mapp);
1642 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1643 (1 << BH_Delay) | (1 << BH_Unwritten))
1646 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1648 * @mpd->lbh - extent of blocks
1649 * @logical - logical number of the block in the file
1650 * @b_state - b_state of the buffer head added
1652 * the function is used to collect contig. blocks in same state
1654 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1655 unsigned long b_state)
1658 int blkbits = mpd->inode->i_blkbits;
1659 int nrblocks = mpd->b_size >> blkbits;
1662 * XXX Don't go larger than mballoc is willing to allocate
1663 * This is a stopgap solution. We eventually need to fold
1664 * mpage_da_submit_io() into this function and then call
1665 * ext4_map_blocks() multiple times in a loop
1667 if (nrblocks >= (8*1024*1024 >> blkbits))
1670 /* check if the reserved journal credits might overflow */
1671 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1672 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1674 * With non-extent format we are limited by the journal
1675 * credit available. Total credit needed to insert
1676 * nrblocks contiguous blocks is dependent on the
1677 * nrblocks. So limit nrblocks.
1683 * First block in the extent
1685 if (mpd->b_size == 0) {
1686 mpd->b_blocknr = logical;
1687 mpd->b_size = 1 << blkbits;
1688 mpd->b_state = b_state & BH_FLAGS;
1692 next = mpd->b_blocknr + nrblocks;
1694 * Can we merge the block to our big extent?
1696 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1697 mpd->b_size += 1 << blkbits;
1703 * We couldn't merge the block to our extent, so we
1704 * need to flush current extent and start new one
1706 mpage_da_map_and_submit(mpd);
1710 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1712 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1716 * This function is grabs code from the very beginning of
1717 * ext4_map_blocks, but assumes that the caller is from delayed write
1718 * time. This function looks up the requested blocks and sets the
1719 * buffer delay bit under the protection of i_data_sem.
1721 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1722 struct ext4_map_blocks *map,
1723 struct buffer_head *bh)
1726 sector_t invalid_block = ~((sector_t) 0xffff);
1728 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1732 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1733 "logical block %lu\n", inode->i_ino, map->m_len,
1734 (unsigned long) map->m_lblk);
1736 * Try to see if we can get the block without requesting a new
1737 * file system block.
1739 down_read((&EXT4_I(inode)->i_data_sem));
1740 if (ext4_has_inline_data(inode)) {
1742 * We will soon create blocks for this page, and let
1743 * us pretend as if the blocks aren't allocated yet.
1744 * In case of clusters, we have to handle the work
1745 * of mapping from cluster so that the reserved space
1746 * is calculated properly.
1748 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1749 ext4_find_delalloc_cluster(inode, map->m_lblk))
1750 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1752 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1753 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1755 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1759 * XXX: __block_prepare_write() unmaps passed block,
1762 /* If the block was allocated from previously allocated cluster,
1763 * then we dont need to reserve it again. */
1764 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1765 retval = ext4_da_reserve_space(inode, iblock);
1767 /* not enough space to reserve */
1771 retval = ext4_es_insert_extent(inode, map->m_lblk, map->m_len);
1775 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1776 * and it should not appear on the bh->b_state.
1778 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1780 map_bh(bh, inode->i_sb, invalid_block);
1782 set_buffer_delay(bh);
1786 up_read((&EXT4_I(inode)->i_data_sem));
1792 * This is a special get_blocks_t callback which is used by
1793 * ext4_da_write_begin(). It will either return mapped block or
1794 * reserve space for a single block.
1796 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1797 * We also have b_blocknr = -1 and b_bdev initialized properly
1799 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1800 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1801 * initialized properly.
1803 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1804 struct buffer_head *bh, int create)
1806 struct ext4_map_blocks map;
1809 BUG_ON(create == 0);
1810 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1812 map.m_lblk = iblock;
1816 * first, we need to know whether the block is allocated already
1817 * preallocated blocks are unmapped but should treated
1818 * the same as allocated blocks.
1820 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1824 map_bh(bh, inode->i_sb, map.m_pblk);
1825 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1827 if (buffer_unwritten(bh)) {
1828 /* A delayed write to unwritten bh should be marked
1829 * new and mapped. Mapped ensures that we don't do
1830 * get_block multiple times when we write to the same
1831 * offset and new ensures that we do proper zero out
1832 * for partial write.
1835 set_buffer_mapped(bh);
1840 static int bget_one(handle_t *handle, struct buffer_head *bh)
1846 static int bput_one(handle_t *handle, struct buffer_head *bh)
1852 static int __ext4_journalled_writepage(struct page *page,
1855 struct address_space *mapping = page->mapping;
1856 struct inode *inode = mapping->host;
1857 struct buffer_head *page_bufs = NULL;
1858 handle_t *handle = NULL;
1859 int ret = 0, err = 0;
1860 int inline_data = ext4_has_inline_data(inode);
1861 struct buffer_head *inode_bh = NULL;
1863 ClearPageChecked(page);
1866 BUG_ON(page->index != 0);
1867 BUG_ON(len > ext4_get_max_inline_size(inode));
1868 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1869 if (inode_bh == NULL)
1872 page_bufs = page_buffers(page);
1877 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1880 /* As soon as we unlock the page, it can go away, but we have
1881 * references to buffers so we are safe */
1884 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1885 if (IS_ERR(handle)) {
1886 ret = PTR_ERR(handle);
1890 BUG_ON(!ext4_handle_valid(handle));
1893 ret = ext4_journal_get_write_access(handle, inode_bh);
1895 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1898 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1899 do_journal_get_write_access);
1901 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1906 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1907 err = ext4_journal_stop(handle);
1911 if (!ext4_has_inline_data(inode))
1912 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1914 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1921 * Note that we don't need to start a transaction unless we're journaling data
1922 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1923 * need to file the inode to the transaction's list in ordered mode because if
1924 * we are writing back data added by write(), the inode is already there and if
1925 * we are writing back data modified via mmap(), no one guarantees in which
1926 * transaction the data will hit the disk. In case we are journaling data, we
1927 * cannot start transaction directly because transaction start ranks above page
1928 * lock so we have to do some magic.
1930 * This function can get called via...
1931 * - ext4_da_writepages after taking page lock (have journal handle)
1932 * - journal_submit_inode_data_buffers (no journal handle)
1933 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1934 * - grab_page_cache when doing write_begin (have journal handle)
1936 * We don't do any block allocation in this function. If we have page with
1937 * multiple blocks we need to write those buffer_heads that are mapped. This
1938 * is important for mmaped based write. So if we do with blocksize 1K
1939 * truncate(f, 1024);
1940 * a = mmap(f, 0, 4096);
1942 * truncate(f, 4096);
1943 * we have in the page first buffer_head mapped via page_mkwrite call back
1944 * but other buffer_heads would be unmapped but dirty (dirty done via the
1945 * do_wp_page). So writepage should write the first block. If we modify
1946 * the mmap area beyond 1024 we will again get a page_fault and the
1947 * page_mkwrite callback will do the block allocation and mark the
1948 * buffer_heads mapped.
1950 * We redirty the page if we have any buffer_heads that is either delay or
1951 * unwritten in the page.
1953 * We can get recursively called as show below.
1955 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1958 * But since we don't do any block allocation we should not deadlock.
1959 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1961 static int ext4_writepage(struct page *page,
1962 struct writeback_control *wbc)
1967 struct buffer_head *page_bufs = NULL;
1968 struct inode *inode = page->mapping->host;
1969 struct ext4_io_submit io_submit;
1971 trace_ext4_writepage(page);
1972 size = i_size_read(inode);
1973 if (page->index == size >> PAGE_CACHE_SHIFT)
1974 len = size & ~PAGE_CACHE_MASK;
1976 len = PAGE_CACHE_SIZE;
1978 page_bufs = page_buffers(page);
1979 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1980 ext4_bh_delay_or_unwritten)) {
1982 * We don't want to do block allocation, so redirty
1983 * the page and return. We may reach here when we do
1984 * a journal commit via journal_submit_inode_data_buffers.
1985 * We can also reach here via shrink_page_list but it
1986 * should never be for direct reclaim so warn if that
1989 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1991 redirty_page_for_writepage(wbc, page);
1996 if (PageChecked(page) && ext4_should_journal_data(inode))
1998 * It's mmapped pagecache. Add buffers and journal it. There
1999 * doesn't seem much point in redirtying the page here.
2001 return __ext4_journalled_writepage(page, len);
2003 memset(&io_submit, 0, sizeof(io_submit));
2004 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2005 ext4_io_submit(&io_submit);
2010 * This is called via ext4_da_writepages() to
2011 * calculate the total number of credits to reserve to fit
2012 * a single extent allocation into a single transaction,
2013 * ext4_da_writpeages() will loop calling this before
2014 * the block allocation.
2017 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2019 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2022 * With non-extent format the journal credit needed to
2023 * insert nrblocks contiguous block is dependent on
2024 * number of contiguous block. So we will limit
2025 * number of contiguous block to a sane value
2027 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2028 (max_blocks > EXT4_MAX_TRANS_DATA))
2029 max_blocks = EXT4_MAX_TRANS_DATA;
2031 return ext4_chunk_trans_blocks(inode, max_blocks);
2035 * write_cache_pages_da - walk the list of dirty pages of the given
2036 * address space and accumulate pages that need writing, and call
2037 * mpage_da_map_and_submit to map a single contiguous memory region
2038 * and then write them.
2040 static int write_cache_pages_da(handle_t *handle,
2041 struct address_space *mapping,
2042 struct writeback_control *wbc,
2043 struct mpage_da_data *mpd,
2044 pgoff_t *done_index)
2046 struct buffer_head *bh, *head;
2047 struct inode *inode = mapping->host;
2048 struct pagevec pvec;
2049 unsigned int nr_pages;
2052 long nr_to_write = wbc->nr_to_write;
2053 int i, tag, ret = 0;
2055 memset(mpd, 0, sizeof(struct mpage_da_data));
2058 pagevec_init(&pvec, 0);
2059 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2060 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2062 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2063 tag = PAGECACHE_TAG_TOWRITE;
2065 tag = PAGECACHE_TAG_DIRTY;
2067 *done_index = index;
2068 while (index <= end) {
2069 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2070 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2074 for (i = 0; i < nr_pages; i++) {
2075 struct page *page = pvec.pages[i];
2078 * At this point, the page may be truncated or
2079 * invalidated (changing page->mapping to NULL), or
2080 * even swizzled back from swapper_space to tmpfs file
2081 * mapping. However, page->index will not change
2082 * because we have a reference on the page.
2084 if (page->index > end)
2087 *done_index = page->index + 1;
2090 * If we can't merge this page, and we have
2091 * accumulated an contiguous region, write it
2093 if ((mpd->next_page != page->index) &&
2094 (mpd->next_page != mpd->first_page)) {
2095 mpage_da_map_and_submit(mpd);
2096 goto ret_extent_tail;
2102 * If the page is no longer dirty, or its
2103 * mapping no longer corresponds to inode we
2104 * are writing (which means it has been
2105 * truncated or invalidated), or the page is
2106 * already under writeback and we are not
2107 * doing a data integrity writeback, skip the page
2109 if (!PageDirty(page) ||
2110 (PageWriteback(page) &&
2111 (wbc->sync_mode == WB_SYNC_NONE)) ||
2112 unlikely(page->mapping != mapping)) {
2117 wait_on_page_writeback(page);
2118 BUG_ON(PageWriteback(page));
2121 * If we have inline data and arrive here, it means that
2122 * we will soon create the block for the 1st page, so
2123 * we'd better clear the inline data here.
2125 if (ext4_has_inline_data(inode)) {
2126 BUG_ON(ext4_test_inode_state(inode,
2127 EXT4_STATE_MAY_INLINE_DATA));
2128 ext4_destroy_inline_data(handle, inode);
2131 if (mpd->next_page != page->index)
2132 mpd->first_page = page->index;
2133 mpd->next_page = page->index + 1;
2134 logical = (sector_t) page->index <<
2135 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2137 /* Add all dirty buffers to mpd */
2138 head = page_buffers(page);
2141 BUG_ON(buffer_locked(bh));
2143 * We need to try to allocate unmapped blocks
2144 * in the same page. Otherwise we won't make
2145 * progress with the page in ext4_writepage
2147 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2148 mpage_add_bh_to_extent(mpd, logical,
2151 goto ret_extent_tail;
2152 } else if (buffer_dirty(bh) &&
2153 buffer_mapped(bh)) {
2155 * mapped dirty buffer. We need to
2156 * update the b_state because we look
2157 * at b_state in mpage_da_map_blocks.
2158 * We don't update b_size because if we
2159 * find an unmapped buffer_head later
2160 * we need to use the b_state flag of
2163 if (mpd->b_size == 0)
2165 bh->b_state & BH_FLAGS;
2168 } while ((bh = bh->b_this_page) != head);
2170 if (nr_to_write > 0) {
2172 if (nr_to_write == 0 &&
2173 wbc->sync_mode == WB_SYNC_NONE)
2175 * We stop writing back only if we are
2176 * not doing integrity sync. In case of
2177 * integrity sync we have to keep going
2178 * because someone may be concurrently
2179 * dirtying pages, and we might have
2180 * synced a lot of newly appeared dirty
2181 * pages, but have not synced all of the
2187 pagevec_release(&pvec);
2192 ret = MPAGE_DA_EXTENT_TAIL;
2194 pagevec_release(&pvec);
2200 static int ext4_da_writepages(struct address_space *mapping,
2201 struct writeback_control *wbc)
2204 int range_whole = 0;
2205 handle_t *handle = NULL;
2206 struct mpage_da_data mpd;
2207 struct inode *inode = mapping->host;
2208 int pages_written = 0;
2209 unsigned int max_pages;
2210 int range_cyclic, cycled = 1, io_done = 0;
2211 int needed_blocks, ret = 0;
2212 long desired_nr_to_write, nr_to_writebump = 0;
2213 loff_t range_start = wbc->range_start;
2214 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2215 pgoff_t done_index = 0;
2217 struct blk_plug plug;
2219 trace_ext4_da_writepages(inode, wbc);
2222 * No pages to write? This is mainly a kludge to avoid starting
2223 * a transaction for special inodes like journal inode on last iput()
2224 * because that could violate lock ordering on umount
2226 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2230 * If the filesystem has aborted, it is read-only, so return
2231 * right away instead of dumping stack traces later on that
2232 * will obscure the real source of the problem. We test
2233 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2234 * the latter could be true if the filesystem is mounted
2235 * read-only, and in that case, ext4_da_writepages should
2236 * *never* be called, so if that ever happens, we would want
2239 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2242 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2245 range_cyclic = wbc->range_cyclic;
2246 if (wbc->range_cyclic) {
2247 index = mapping->writeback_index;
2250 wbc->range_start = index << PAGE_CACHE_SHIFT;
2251 wbc->range_end = LLONG_MAX;
2252 wbc->range_cyclic = 0;
2255 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2256 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2260 * This works around two forms of stupidity. The first is in
2261 * the writeback code, which caps the maximum number of pages
2262 * written to be 1024 pages. This is wrong on multiple
2263 * levels; different architectues have a different page size,
2264 * which changes the maximum amount of data which gets
2265 * written. Secondly, 4 megabytes is way too small. XFS
2266 * forces this value to be 16 megabytes by multiplying
2267 * nr_to_write parameter by four, and then relies on its
2268 * allocator to allocate larger extents to make them
2269 * contiguous. Unfortunately this brings us to the second
2270 * stupidity, which is that ext4's mballoc code only allocates
2271 * at most 2048 blocks. So we force contiguous writes up to
2272 * the number of dirty blocks in the inode, or
2273 * sbi->max_writeback_mb_bump whichever is smaller.
2275 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2276 if (!range_cyclic && range_whole) {
2277 if (wbc->nr_to_write == LONG_MAX)
2278 desired_nr_to_write = wbc->nr_to_write;
2280 desired_nr_to_write = wbc->nr_to_write * 8;
2282 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2284 if (desired_nr_to_write > max_pages)
2285 desired_nr_to_write = max_pages;
2287 if (wbc->nr_to_write < desired_nr_to_write) {
2288 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2289 wbc->nr_to_write = desired_nr_to_write;
2293 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2294 tag_pages_for_writeback(mapping, index, end);
2296 blk_start_plug(&plug);
2297 while (!ret && wbc->nr_to_write > 0) {
2300 * we insert one extent at a time. So we need
2301 * credit needed for single extent allocation.
2302 * journalled mode is currently not supported
2305 BUG_ON(ext4_should_journal_data(inode));
2306 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2308 /* start a new transaction*/
2309 handle = ext4_journal_start(inode, needed_blocks);
2310 if (IS_ERR(handle)) {
2311 ret = PTR_ERR(handle);
2312 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2313 "%ld pages, ino %lu; err %d", __func__,
2314 wbc->nr_to_write, inode->i_ino, ret);
2315 blk_finish_plug(&plug);
2316 goto out_writepages;
2320 * Now call write_cache_pages_da() to find the next
2321 * contiguous region of logical blocks that need
2322 * blocks to be allocated by ext4 and submit them.
2324 ret = write_cache_pages_da(handle, mapping,
2325 wbc, &mpd, &done_index);
2327 * If we have a contiguous extent of pages and we
2328 * haven't done the I/O yet, map the blocks and submit
2331 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2332 mpage_da_map_and_submit(&mpd);
2333 ret = MPAGE_DA_EXTENT_TAIL;
2335 trace_ext4_da_write_pages(inode, &mpd);
2336 wbc->nr_to_write -= mpd.pages_written;
2338 ext4_journal_stop(handle);
2340 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2341 /* commit the transaction which would
2342 * free blocks released in the transaction
2345 jbd2_journal_force_commit_nested(sbi->s_journal);
2347 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2349 * Got one extent now try with rest of the pages.
2350 * If mpd.retval is set -EIO, journal is aborted.
2351 * So we don't need to write any more.
2353 pages_written += mpd.pages_written;
2356 } else if (wbc->nr_to_write)
2358 * There is no more writeout needed
2359 * or we requested for a noblocking writeout
2360 * and we found the device congested
2364 blk_finish_plug(&plug);
2365 if (!io_done && !cycled) {
2368 wbc->range_start = index << PAGE_CACHE_SHIFT;
2369 wbc->range_end = mapping->writeback_index - 1;
2374 wbc->range_cyclic = range_cyclic;
2375 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2377 * set the writeback_index so that range_cyclic
2378 * mode will write it back later
2380 mapping->writeback_index = done_index;
2383 wbc->nr_to_write -= nr_to_writebump;
2384 wbc->range_start = range_start;
2385 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2389 static int ext4_nonda_switch(struct super_block *sb)
2391 s64 free_blocks, dirty_blocks;
2392 struct ext4_sb_info *sbi = EXT4_SB(sb);
2395 * switch to non delalloc mode if we are running low
2396 * on free block. The free block accounting via percpu
2397 * counters can get slightly wrong with percpu_counter_batch getting
2398 * accumulated on each CPU without updating global counters
2399 * Delalloc need an accurate free block accounting. So switch
2400 * to non delalloc when we are near to error range.
2402 free_blocks = EXT4_C2B(sbi,
2403 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2404 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2406 * Start pushing delalloc when 1/2 of free blocks are dirty.
2408 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2409 !writeback_in_progress(sb->s_bdi) &&
2410 down_read_trylock(&sb->s_umount)) {
2411 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2412 up_read(&sb->s_umount);
2415 if (2 * free_blocks < 3 * dirty_blocks ||
2416 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2418 * free block count is less than 150% of dirty blocks
2419 * or free blocks is less than watermark
2426 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2427 loff_t pos, unsigned len, unsigned flags,
2428 struct page **pagep, void **fsdata)
2430 int ret, retries = 0;
2433 struct inode *inode = mapping->host;
2436 index = pos >> PAGE_CACHE_SHIFT;
2438 if (ext4_nonda_switch(inode->i_sb)) {
2439 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2440 return ext4_write_begin(file, mapping, pos,
2441 len, flags, pagep, fsdata);
2443 *fsdata = (void *)0;
2444 trace_ext4_da_write_begin(inode, pos, len, flags);
2446 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2447 ret = ext4_da_write_inline_data_begin(mapping, inode,
2460 * With delayed allocation, we don't log the i_disksize update
2461 * if there is delayed block allocation. But we still need
2462 * to journalling the i_disksize update if writes to the end
2463 * of file which has an already mapped buffer.
2465 handle = ext4_journal_start(inode, 1);
2466 if (IS_ERR(handle)) {
2467 ret = PTR_ERR(handle);
2470 /* We cannot recurse into the filesystem as the transaction is already
2472 flags |= AOP_FLAG_NOFS;
2474 page = grab_cache_page_write_begin(mapping, index, flags);
2476 ext4_journal_stop(handle);
2482 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2485 ext4_journal_stop(handle);
2486 page_cache_release(page);
2488 * block_write_begin may have instantiated a few blocks
2489 * outside i_size. Trim these off again. Don't need
2490 * i_size_read because we hold i_mutex.
2492 if (pos + len > inode->i_size)
2493 ext4_truncate_failed_write(inode);
2496 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2503 * Check if we should update i_disksize
2504 * when write to the end of file but not require block allocation
2506 static int ext4_da_should_update_i_disksize(struct page *page,
2507 unsigned long offset)
2509 struct buffer_head *bh;
2510 struct inode *inode = page->mapping->host;
2514 bh = page_buffers(page);
2515 idx = offset >> inode->i_blkbits;
2517 for (i = 0; i < idx; i++)
2518 bh = bh->b_this_page;
2520 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2525 static int ext4_da_write_end(struct file *file,
2526 struct address_space *mapping,
2527 loff_t pos, unsigned len, unsigned copied,
2528 struct page *page, void *fsdata)
2530 struct inode *inode = mapping->host;
2532 handle_t *handle = ext4_journal_current_handle();
2534 unsigned long start, end;
2535 int write_mode = (int)(unsigned long)fsdata;
2537 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2538 switch (ext4_inode_journal_mode(inode)) {
2539 case EXT4_INODE_ORDERED_DATA_MODE:
2540 return ext4_ordered_write_end(file, mapping, pos,
2541 len, copied, page, fsdata);
2542 case EXT4_INODE_WRITEBACK_DATA_MODE:
2543 return ext4_writeback_write_end(file, mapping, pos,
2544 len, copied, page, fsdata);
2550 trace_ext4_da_write_end(inode, pos, len, copied);
2551 start = pos & (PAGE_CACHE_SIZE - 1);
2552 end = start + copied - 1;
2555 * generic_write_end() will run mark_inode_dirty() if i_size
2556 * changes. So let's piggyback the i_disksize mark_inode_dirty
2559 new_i_size = pos + copied;
2560 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2561 if (ext4_has_inline_data(inode) ||
2562 ext4_da_should_update_i_disksize(page, end)) {
2563 down_write(&EXT4_I(inode)->i_data_sem);
2564 if (new_i_size > EXT4_I(inode)->i_disksize)
2565 EXT4_I(inode)->i_disksize = new_i_size;
2566 up_write(&EXT4_I(inode)->i_data_sem);
2567 /* We need to mark inode dirty even if
2568 * new_i_size is less that inode->i_size
2569 * bu greater than i_disksize.(hint delalloc)
2571 ext4_mark_inode_dirty(handle, inode);
2575 if (write_mode != CONVERT_INLINE_DATA &&
2576 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2577 ext4_has_inline_data(inode))
2578 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2581 ret2 = generic_write_end(file, mapping, pos, len, copied,
2587 ret2 = ext4_journal_stop(handle);
2591 return ret ? ret : copied;
2594 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2597 * Drop reserved blocks
2599 BUG_ON(!PageLocked(page));
2600 if (!page_has_buffers(page))
2603 ext4_da_page_release_reservation(page, offset);
2606 ext4_invalidatepage(page, offset);
2612 * Force all delayed allocation blocks to be allocated for a given inode.
2614 int ext4_alloc_da_blocks(struct inode *inode)
2616 trace_ext4_alloc_da_blocks(inode);
2618 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2619 !EXT4_I(inode)->i_reserved_meta_blocks)
2623 * We do something simple for now. The filemap_flush() will
2624 * also start triggering a write of the data blocks, which is
2625 * not strictly speaking necessary (and for users of
2626 * laptop_mode, not even desirable). However, to do otherwise
2627 * would require replicating code paths in:
2629 * ext4_da_writepages() ->
2630 * write_cache_pages() ---> (via passed in callback function)
2631 * __mpage_da_writepage() -->
2632 * mpage_add_bh_to_extent()
2633 * mpage_da_map_blocks()
2635 * The problem is that write_cache_pages(), located in
2636 * mm/page-writeback.c, marks pages clean in preparation for
2637 * doing I/O, which is not desirable if we're not planning on
2640 * We could call write_cache_pages(), and then redirty all of
2641 * the pages by calling redirty_page_for_writepage() but that
2642 * would be ugly in the extreme. So instead we would need to
2643 * replicate parts of the code in the above functions,
2644 * simplifying them because we wouldn't actually intend to
2645 * write out the pages, but rather only collect contiguous
2646 * logical block extents, call the multi-block allocator, and
2647 * then update the buffer heads with the block allocations.
2649 * For now, though, we'll cheat by calling filemap_flush(),
2650 * which will map the blocks, and start the I/O, but not
2651 * actually wait for the I/O to complete.
2653 return filemap_flush(inode->i_mapping);
2657 * bmap() is special. It gets used by applications such as lilo and by
2658 * the swapper to find the on-disk block of a specific piece of data.
2660 * Naturally, this is dangerous if the block concerned is still in the
2661 * journal. If somebody makes a swapfile on an ext4 data-journaling
2662 * filesystem and enables swap, then they may get a nasty shock when the
2663 * data getting swapped to that swapfile suddenly gets overwritten by
2664 * the original zero's written out previously to the journal and
2665 * awaiting writeback in the kernel's buffer cache.
2667 * So, if we see any bmap calls here on a modified, data-journaled file,
2668 * take extra steps to flush any blocks which might be in the cache.
2670 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2672 struct inode *inode = mapping->host;
2677 * We can get here for an inline file via the FIBMAP ioctl
2679 if (ext4_has_inline_data(inode))
2682 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2683 test_opt(inode->i_sb, DELALLOC)) {
2685 * With delalloc we want to sync the file
2686 * so that we can make sure we allocate
2689 filemap_write_and_wait(mapping);
2692 if (EXT4_JOURNAL(inode) &&
2693 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2695 * This is a REALLY heavyweight approach, but the use of
2696 * bmap on dirty files is expected to be extremely rare:
2697 * only if we run lilo or swapon on a freshly made file
2698 * do we expect this to happen.
2700 * (bmap requires CAP_SYS_RAWIO so this does not
2701 * represent an unprivileged user DOS attack --- we'd be
2702 * in trouble if mortal users could trigger this path at
2705 * NB. EXT4_STATE_JDATA is not set on files other than
2706 * regular files. If somebody wants to bmap a directory
2707 * or symlink and gets confused because the buffer
2708 * hasn't yet been flushed to disk, they deserve
2709 * everything they get.
2712 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2713 journal = EXT4_JOURNAL(inode);
2714 jbd2_journal_lock_updates(journal);
2715 err = jbd2_journal_flush(journal);
2716 jbd2_journal_unlock_updates(journal);
2722 return generic_block_bmap(mapping, block, ext4_get_block);
2725 static int ext4_readpage(struct file *file, struct page *page)
2728 struct inode *inode = page->mapping->host;
2730 trace_ext4_readpage(page);
2732 if (ext4_has_inline_data(inode))
2733 ret = ext4_readpage_inline(inode, page);
2736 return mpage_readpage(page, ext4_get_block);
2742 ext4_readpages(struct file *file, struct address_space *mapping,
2743 struct list_head *pages, unsigned nr_pages)
2745 struct inode *inode = mapping->host;
2747 /* If the file has inline data, no need to do readpages. */
2748 if (ext4_has_inline_data(inode))
2751 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2754 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2756 trace_ext4_invalidatepage(page, offset);
2758 /* No journalling happens on data buffers when this function is used */
2759 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2761 block_invalidatepage(page, offset);
2764 static int __ext4_journalled_invalidatepage(struct page *page,
2765 unsigned long offset)
2767 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2769 trace_ext4_journalled_invalidatepage(page, offset);
2772 * If it's a full truncate we just forget about the pending dirtying
2775 ClearPageChecked(page);
2777 return jbd2_journal_invalidatepage(journal, page, offset);
2780 /* Wrapper for aops... */
2781 static void ext4_journalled_invalidatepage(struct page *page,
2782 unsigned long offset)
2784 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
2787 static int ext4_releasepage(struct page *page, gfp_t wait)
2789 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2791 trace_ext4_releasepage(page);
2793 WARN_ON(PageChecked(page));
2794 if (!page_has_buffers(page))
2797 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2799 return try_to_free_buffers(page);
2803 * ext4_get_block used when preparing for a DIO write or buffer write.
2804 * We allocate an uinitialized extent if blocks haven't been allocated.
2805 * The extent will be converted to initialized after the IO is complete.
2807 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2808 struct buffer_head *bh_result, int create)
2810 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2811 inode->i_ino, create);
2812 return _ext4_get_block(inode, iblock, bh_result,
2813 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2816 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2817 struct buffer_head *bh_result, int create)
2819 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2820 inode->i_ino, create);
2821 return _ext4_get_block(inode, iblock, bh_result,
2822 EXT4_GET_BLOCKS_NO_LOCK);
2825 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2826 ssize_t size, void *private, int ret,
2829 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2830 ext4_io_end_t *io_end = iocb->private;
2832 /* if not async direct IO or dio with 0 bytes write, just return */
2833 if (!io_end || !size)
2836 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2837 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2838 iocb->private, io_end->inode->i_ino, iocb, offset,
2841 iocb->private = NULL;
2843 /* if not aio dio with unwritten extents, just free io and return */
2844 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2845 ext4_free_io_end(io_end);
2848 aio_complete(iocb, ret, 0);
2849 inode_dio_done(inode);
2853 io_end->offset = offset;
2854 io_end->size = size;
2856 io_end->iocb = iocb;
2857 io_end->result = ret;
2860 ext4_add_complete_io(io_end);
2864 * For ext4 extent files, ext4 will do direct-io write to holes,
2865 * preallocated extents, and those write extend the file, no need to
2866 * fall back to buffered IO.
2868 * For holes, we fallocate those blocks, mark them as uninitialized
2869 * If those blocks were preallocated, we mark sure they are split, but
2870 * still keep the range to write as uninitialized.
2872 * The unwritten extents will be converted to written when DIO is completed.
2873 * For async direct IO, since the IO may still pending when return, we
2874 * set up an end_io call back function, which will do the conversion
2875 * when async direct IO completed.
2877 * If the O_DIRECT write will extend the file then add this inode to the
2878 * orphan list. So recovery will truncate it back to the original size
2879 * if the machine crashes during the write.
2882 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2883 const struct iovec *iov, loff_t offset,
2884 unsigned long nr_segs)
2886 struct file *file = iocb->ki_filp;
2887 struct inode *inode = file->f_mapping->host;
2889 size_t count = iov_length(iov, nr_segs);
2891 get_block_t *get_block_func = NULL;
2893 loff_t final_size = offset + count;
2895 /* Use the old path for reads and writes beyond i_size. */
2896 if (rw != WRITE || final_size > inode->i_size)
2897 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2899 BUG_ON(iocb->private == NULL);
2901 /* If we do a overwrite dio, i_mutex locking can be released */
2902 overwrite = *((int *)iocb->private);
2905 atomic_inc(&inode->i_dio_count);
2906 down_read(&EXT4_I(inode)->i_data_sem);
2907 mutex_unlock(&inode->i_mutex);
2911 * We could direct write to holes and fallocate.
2913 * Allocated blocks to fill the hole are marked as
2914 * uninitialized to prevent parallel buffered read to expose
2915 * the stale data before DIO complete the data IO.
2917 * As to previously fallocated extents, ext4 get_block will
2918 * just simply mark the buffer mapped but still keep the
2919 * extents uninitialized.
2921 * For non AIO case, we will convert those unwritten extents
2922 * to written after return back from blockdev_direct_IO.
2924 * For async DIO, the conversion needs to be deferred when the
2925 * IO is completed. The ext4 end_io callback function will be
2926 * called to take care of the conversion work. Here for async
2927 * case, we allocate an io_end structure to hook to the iocb.
2929 iocb->private = NULL;
2930 ext4_inode_aio_set(inode, NULL);
2931 if (!is_sync_kiocb(iocb)) {
2932 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
2937 io_end->flag |= EXT4_IO_END_DIRECT;
2938 iocb->private = io_end;
2940 * we save the io structure for current async direct
2941 * IO, so that later ext4_map_blocks() could flag the
2942 * io structure whether there is a unwritten extents
2943 * needs to be converted when IO is completed.
2945 ext4_inode_aio_set(inode, io_end);
2949 get_block_func = ext4_get_block_write_nolock;
2951 get_block_func = ext4_get_block_write;
2952 dio_flags = DIO_LOCKING;
2954 ret = __blockdev_direct_IO(rw, iocb, inode,
2955 inode->i_sb->s_bdev, iov,
2963 ext4_inode_aio_set(inode, NULL);
2965 * The io_end structure takes a reference to the inode, that
2966 * structure needs to be destroyed and the reference to the
2967 * inode need to be dropped, when IO is complete, even with 0
2968 * byte write, or failed.
2970 * In the successful AIO DIO case, the io_end structure will
2971 * be destroyed and the reference to the inode will be dropped
2972 * after the end_io call back function is called.
2974 * In the case there is 0 byte write, or error case, since VFS
2975 * direct IO won't invoke the end_io call back function, we
2976 * need to free the end_io structure here.
2978 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
2979 ext4_free_io_end(iocb->private);
2980 iocb->private = NULL;
2981 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
2982 EXT4_STATE_DIO_UNWRITTEN)) {
2985 * for non AIO case, since the IO is already
2986 * completed, we could do the conversion right here
2988 err = ext4_convert_unwritten_extents(inode,
2992 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
2996 /* take i_mutex locking again if we do a ovewrite dio */
2998 inode_dio_done(inode);
2999 up_read(&EXT4_I(inode)->i_data_sem);
3000 mutex_lock(&inode->i_mutex);
3006 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3007 const struct iovec *iov, loff_t offset,
3008 unsigned long nr_segs)
3010 struct file *file = iocb->ki_filp;
3011 struct inode *inode = file->f_mapping->host;
3015 * If we are doing data journalling we don't support O_DIRECT
3017 if (ext4_should_journal_data(inode))
3020 /* Let buffer I/O handle the inline data case. */
3021 if (ext4_has_inline_data(inode))
3024 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3025 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3026 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3028 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3029 trace_ext4_direct_IO_exit(inode, offset,
3030 iov_length(iov, nr_segs), rw, ret);
3035 * Pages can be marked dirty completely asynchronously from ext4's journalling
3036 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3037 * much here because ->set_page_dirty is called under VFS locks. The page is
3038 * not necessarily locked.
3040 * We cannot just dirty the page and leave attached buffers clean, because the
3041 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3042 * or jbddirty because all the journalling code will explode.
3044 * So what we do is to mark the page "pending dirty" and next time writepage
3045 * is called, propagate that into the buffers appropriately.
3047 static int ext4_journalled_set_page_dirty(struct page *page)
3049 SetPageChecked(page);
3050 return __set_page_dirty_nobuffers(page);
3053 static const struct address_space_operations ext4_ordered_aops = {
3054 .readpage = ext4_readpage,
3055 .readpages = ext4_readpages,
3056 .writepage = ext4_writepage,
3057 .write_begin = ext4_write_begin,
3058 .write_end = ext4_ordered_write_end,
3060 .invalidatepage = ext4_invalidatepage,
3061 .releasepage = ext4_releasepage,
3062 .direct_IO = ext4_direct_IO,
3063 .migratepage = buffer_migrate_page,
3064 .is_partially_uptodate = block_is_partially_uptodate,
3065 .error_remove_page = generic_error_remove_page,
3068 static const struct address_space_operations ext4_writeback_aops = {
3069 .readpage = ext4_readpage,
3070 .readpages = ext4_readpages,
3071 .writepage = ext4_writepage,
3072 .write_begin = ext4_write_begin,
3073 .write_end = ext4_writeback_write_end,
3075 .invalidatepage = ext4_invalidatepage,
3076 .releasepage = ext4_releasepage,
3077 .direct_IO = ext4_direct_IO,
3078 .migratepage = buffer_migrate_page,
3079 .is_partially_uptodate = block_is_partially_uptodate,
3080 .error_remove_page = generic_error_remove_page,
3083 static const struct address_space_operations ext4_journalled_aops = {
3084 .readpage = ext4_readpage,
3085 .readpages = ext4_readpages,
3086 .writepage = ext4_writepage,
3087 .write_begin = ext4_write_begin,
3088 .write_end = ext4_journalled_write_end,
3089 .set_page_dirty = ext4_journalled_set_page_dirty,
3091 .invalidatepage = ext4_journalled_invalidatepage,
3092 .releasepage = ext4_releasepage,
3093 .direct_IO = ext4_direct_IO,
3094 .is_partially_uptodate = block_is_partially_uptodate,
3095 .error_remove_page = generic_error_remove_page,
3098 static const struct address_space_operations ext4_da_aops = {
3099 .readpage = ext4_readpage,
3100 .readpages = ext4_readpages,
3101 .writepage = ext4_writepage,
3102 .writepages = ext4_da_writepages,
3103 .write_begin = ext4_da_write_begin,
3104 .write_end = ext4_da_write_end,
3106 .invalidatepage = ext4_da_invalidatepage,
3107 .releasepage = ext4_releasepage,
3108 .direct_IO = ext4_direct_IO,
3109 .migratepage = buffer_migrate_page,
3110 .is_partially_uptodate = block_is_partially_uptodate,
3111 .error_remove_page = generic_error_remove_page,
3114 void ext4_set_aops(struct inode *inode)
3116 switch (ext4_inode_journal_mode(inode)) {
3117 case EXT4_INODE_ORDERED_DATA_MODE:
3118 if (test_opt(inode->i_sb, DELALLOC))
3119 inode->i_mapping->a_ops = &ext4_da_aops;
3121 inode->i_mapping->a_ops = &ext4_ordered_aops;
3123 case EXT4_INODE_WRITEBACK_DATA_MODE:
3124 if (test_opt(inode->i_sb, DELALLOC))
3125 inode->i_mapping->a_ops = &ext4_da_aops;
3127 inode->i_mapping->a_ops = &ext4_writeback_aops;
3129 case EXT4_INODE_JOURNAL_DATA_MODE:
3130 inode->i_mapping->a_ops = &ext4_journalled_aops;
3139 * ext4_discard_partial_page_buffers()
3140 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3141 * This function finds and locks the page containing the offset
3142 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3143 * Calling functions that already have the page locked should call
3144 * ext4_discard_partial_page_buffers_no_lock directly.
3146 int ext4_discard_partial_page_buffers(handle_t *handle,
3147 struct address_space *mapping, loff_t from,
3148 loff_t length, int flags)
3150 struct inode *inode = mapping->host;
3154 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3155 mapping_gfp_mask(mapping) & ~__GFP_FS);
3159 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3160 from, length, flags);
3163 page_cache_release(page);
3168 * ext4_discard_partial_page_buffers_no_lock()
3169 * Zeros a page range of length 'length' starting from offset 'from'.
3170 * Buffer heads that correspond to the block aligned regions of the
3171 * zeroed range will be unmapped. Unblock aligned regions
3172 * will have the corresponding buffer head mapped if needed so that
3173 * that region of the page can be updated with the partial zero out.
3175 * This function assumes that the page has already been locked. The
3176 * The range to be discarded must be contained with in the given page.
3177 * If the specified range exceeds the end of the page it will be shortened
3178 * to the end of the page that corresponds to 'from'. This function is
3179 * appropriate for updating a page and it buffer heads to be unmapped and
3180 * zeroed for blocks that have been either released, or are going to be
3183 * handle: The journal handle
3184 * inode: The files inode
3185 * page: A locked page that contains the offset "from"
3186 * from: The starting byte offset (from the beginning of the file)
3187 * to begin discarding
3188 * len: The length of bytes to discard
3189 * flags: Optional flags that may be used:
3191 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3192 * Only zero the regions of the page whose buffer heads
3193 * have already been unmapped. This flag is appropriate
3194 * for updating the contents of a page whose blocks may
3195 * have already been released, and we only want to zero
3196 * out the regions that correspond to those released blocks.
3198 * Returns zero on success or negative on failure.
3200 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3201 struct inode *inode, struct page *page, loff_t from,
3202 loff_t length, int flags)
3204 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3205 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3206 unsigned int blocksize, max, pos;
3208 struct buffer_head *bh;
3211 blocksize = inode->i_sb->s_blocksize;
3212 max = PAGE_CACHE_SIZE - offset;
3214 if (index != page->index)
3218 * correct length if it does not fall between
3219 * 'from' and the end of the page
3221 if (length > max || length < 0)
3224 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3226 if (!page_has_buffers(page))
3227 create_empty_buffers(page, blocksize, 0);
3229 /* Find the buffer that contains "offset" */
3230 bh = page_buffers(page);
3232 while (offset >= pos) {
3233 bh = bh->b_this_page;
3239 while (pos < offset + length) {
3240 unsigned int end_of_block, range_to_discard;
3244 /* The length of space left to zero and unmap */
3245 range_to_discard = offset + length - pos;
3247 /* The length of space until the end of the block */
3248 end_of_block = blocksize - (pos & (blocksize-1));
3251 * Do not unmap or zero past end of block
3252 * for this buffer head
3254 if (range_to_discard > end_of_block)
3255 range_to_discard = end_of_block;
3259 * Skip this buffer head if we are only zeroing unampped
3260 * regions of the page
3262 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3266 /* If the range is block aligned, unmap */
3267 if (range_to_discard == blocksize) {
3268 clear_buffer_dirty(bh);
3270 clear_buffer_mapped(bh);
3271 clear_buffer_req(bh);
3272 clear_buffer_new(bh);
3273 clear_buffer_delay(bh);
3274 clear_buffer_unwritten(bh);
3275 clear_buffer_uptodate(bh);
3276 zero_user(page, pos, range_to_discard);
3277 BUFFER_TRACE(bh, "Buffer discarded");
3282 * If this block is not completely contained in the range
3283 * to be discarded, then it is not going to be released. Because
3284 * we need to keep this block, we need to make sure this part
3285 * of the page is uptodate before we modify it by writeing
3286 * partial zeros on it.
3288 if (!buffer_mapped(bh)) {
3290 * Buffer head must be mapped before we can read
3293 BUFFER_TRACE(bh, "unmapped");
3294 ext4_get_block(inode, iblock, bh, 0);
3295 /* unmapped? It's a hole - nothing to do */
3296 if (!buffer_mapped(bh)) {
3297 BUFFER_TRACE(bh, "still unmapped");
3302 /* Ok, it's mapped. Make sure it's up-to-date */
3303 if (PageUptodate(page))
3304 set_buffer_uptodate(bh);
3306 if (!buffer_uptodate(bh)) {
3308 ll_rw_block(READ, 1, &bh);
3310 /* Uhhuh. Read error. Complain and punt.*/
3311 if (!buffer_uptodate(bh))
3315 if (ext4_should_journal_data(inode)) {
3316 BUFFER_TRACE(bh, "get write access");
3317 err = ext4_journal_get_write_access(handle, bh);
3322 zero_user(page, pos, range_to_discard);
3325 if (ext4_should_journal_data(inode)) {
3326 err = ext4_handle_dirty_metadata(handle, inode, bh);
3328 mark_buffer_dirty(bh);
3330 BUFFER_TRACE(bh, "Partial buffer zeroed");
3332 bh = bh->b_this_page;
3334 pos += range_to_discard;
3340 int ext4_can_truncate(struct inode *inode)
3342 if (S_ISREG(inode->i_mode))
3344 if (S_ISDIR(inode->i_mode))
3346 if (S_ISLNK(inode->i_mode))
3347 return !ext4_inode_is_fast_symlink(inode);
3352 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3353 * associated with the given offset and length
3355 * @inode: File inode
3356 * @offset: The offset where the hole will begin
3357 * @len: The length of the hole
3359 * Returns: 0 on success or negative on failure
3362 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3364 struct inode *inode = file->f_path.dentry->d_inode;
3365 if (!S_ISREG(inode->i_mode))
3368 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3369 return ext4_ind_punch_hole(file, offset, length);
3371 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3372 /* TODO: Add support for bigalloc file systems */
3376 trace_ext4_punch_hole(inode, offset, length);
3378 return ext4_ext_punch_hole(file, offset, length);
3384 * We block out ext4_get_block() block instantiations across the entire
3385 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3386 * simultaneously on behalf of the same inode.
3388 * As we work through the truncate and commit bits of it to the journal there
3389 * is one core, guiding principle: the file's tree must always be consistent on
3390 * disk. We must be able to restart the truncate after a crash.
3392 * The file's tree may be transiently inconsistent in memory (although it
3393 * probably isn't), but whenever we close off and commit a journal transaction,
3394 * the contents of (the filesystem + the journal) must be consistent and
3395 * restartable. It's pretty simple, really: bottom up, right to left (although
3396 * left-to-right works OK too).
3398 * Note that at recovery time, journal replay occurs *before* the restart of
3399 * truncate against the orphan inode list.
3401 * The committed inode has the new, desired i_size (which is the same as
3402 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3403 * that this inode's truncate did not complete and it will again call
3404 * ext4_truncate() to have another go. So there will be instantiated blocks
3405 * to the right of the truncation point in a crashed ext4 filesystem. But
3406 * that's fine - as long as they are linked from the inode, the post-crash
3407 * ext4_truncate() run will find them and release them.
3409 void ext4_truncate(struct inode *inode)
3411 trace_ext4_truncate_enter(inode);
3413 if (!ext4_can_truncate(inode))
3416 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3418 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3419 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3421 if (ext4_has_inline_data(inode)) {
3424 ext4_inline_data_truncate(inode, &has_inline);
3429 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3430 ext4_ext_truncate(inode);
3432 ext4_ind_truncate(inode);
3434 trace_ext4_truncate_exit(inode);
3438 * ext4_get_inode_loc returns with an extra refcount against the inode's
3439 * underlying buffer_head on success. If 'in_mem' is true, we have all
3440 * data in memory that is needed to recreate the on-disk version of this
3443 static int __ext4_get_inode_loc(struct inode *inode,
3444 struct ext4_iloc *iloc, int in_mem)
3446 struct ext4_group_desc *gdp;
3447 struct buffer_head *bh;
3448 struct super_block *sb = inode->i_sb;
3450 int inodes_per_block, inode_offset;
3453 if (!ext4_valid_inum(sb, inode->i_ino))
3456 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3457 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3462 * Figure out the offset within the block group inode table
3464 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3465 inode_offset = ((inode->i_ino - 1) %
3466 EXT4_INODES_PER_GROUP(sb));
3467 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3468 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3470 bh = sb_getblk(sb, block);
3473 if (!buffer_uptodate(bh)) {
3477 * If the buffer has the write error flag, we have failed
3478 * to write out another inode in the same block. In this
3479 * case, we don't have to read the block because we may
3480 * read the old inode data successfully.
3482 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3483 set_buffer_uptodate(bh);
3485 if (buffer_uptodate(bh)) {
3486 /* someone brought it uptodate while we waited */
3492 * If we have all information of the inode in memory and this
3493 * is the only valid inode in the block, we need not read the
3497 struct buffer_head *bitmap_bh;
3500 start = inode_offset & ~(inodes_per_block - 1);
3502 /* Is the inode bitmap in cache? */
3503 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3504 if (unlikely(!bitmap_bh))
3508 * If the inode bitmap isn't in cache then the
3509 * optimisation may end up performing two reads instead
3510 * of one, so skip it.
3512 if (!buffer_uptodate(bitmap_bh)) {
3516 for (i = start; i < start + inodes_per_block; i++) {
3517 if (i == inode_offset)
3519 if (ext4_test_bit(i, bitmap_bh->b_data))
3523 if (i == start + inodes_per_block) {
3524 /* all other inodes are free, so skip I/O */
3525 memset(bh->b_data, 0, bh->b_size);
3526 set_buffer_uptodate(bh);
3534 * If we need to do any I/O, try to pre-readahead extra
3535 * blocks from the inode table.
3537 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3538 ext4_fsblk_t b, end, table;
3541 table = ext4_inode_table(sb, gdp);
3542 /* s_inode_readahead_blks is always a power of 2 */
3543 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3546 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3547 num = EXT4_INODES_PER_GROUP(sb);
3548 if (ext4_has_group_desc_csum(sb))
3549 num -= ext4_itable_unused_count(sb, gdp);
3550 table += num / inodes_per_block;
3554 sb_breadahead(sb, b++);
3558 * There are other valid inodes in the buffer, this inode
3559 * has in-inode xattrs, or we don't have this inode in memory.
3560 * Read the block from disk.
3562 trace_ext4_load_inode(inode);
3564 bh->b_end_io = end_buffer_read_sync;
3565 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3567 if (!buffer_uptodate(bh)) {
3568 EXT4_ERROR_INODE_BLOCK(inode, block,
3569 "unable to read itable block");
3579 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3581 /* We have all inode data except xattrs in memory here. */
3582 return __ext4_get_inode_loc(inode, iloc,
3583 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3586 void ext4_set_inode_flags(struct inode *inode)
3588 unsigned int flags = EXT4_I(inode)->i_flags;
3590 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3591 if (flags & EXT4_SYNC_FL)
3592 inode->i_flags |= S_SYNC;
3593 if (flags & EXT4_APPEND_FL)
3594 inode->i_flags |= S_APPEND;
3595 if (flags & EXT4_IMMUTABLE_FL)
3596 inode->i_flags |= S_IMMUTABLE;
3597 if (flags & EXT4_NOATIME_FL)
3598 inode->i_flags |= S_NOATIME;
3599 if (flags & EXT4_DIRSYNC_FL)
3600 inode->i_flags |= S_DIRSYNC;
3603 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3604 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3606 unsigned int vfs_fl;
3607 unsigned long old_fl, new_fl;
3610 vfs_fl = ei->vfs_inode.i_flags;
3611 old_fl = ei->i_flags;
3612 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3613 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3615 if (vfs_fl & S_SYNC)
3616 new_fl |= EXT4_SYNC_FL;
3617 if (vfs_fl & S_APPEND)
3618 new_fl |= EXT4_APPEND_FL;
3619 if (vfs_fl & S_IMMUTABLE)
3620 new_fl |= EXT4_IMMUTABLE_FL;
3621 if (vfs_fl & S_NOATIME)
3622 new_fl |= EXT4_NOATIME_FL;
3623 if (vfs_fl & S_DIRSYNC)
3624 new_fl |= EXT4_DIRSYNC_FL;
3625 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3628 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3629 struct ext4_inode_info *ei)
3632 struct inode *inode = &(ei->vfs_inode);
3633 struct super_block *sb = inode->i_sb;
3635 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3636 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3637 /* we are using combined 48 bit field */
3638 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3639 le32_to_cpu(raw_inode->i_blocks_lo);
3640 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3641 /* i_blocks represent file system block size */
3642 return i_blocks << (inode->i_blkbits - 9);
3647 return le32_to_cpu(raw_inode->i_blocks_lo);
3651 static inline void ext4_iget_extra_inode(struct inode *inode,
3652 struct ext4_inode *raw_inode,
3653 struct ext4_inode_info *ei)
3655 __le32 *magic = (void *)raw_inode +
3656 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3657 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3658 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3659 ext4_find_inline_data_nolock(inode);
3661 EXT4_I(inode)->i_inline_off = 0;
3664 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3666 struct ext4_iloc iloc;
3667 struct ext4_inode *raw_inode;
3668 struct ext4_inode_info *ei;
3669 struct inode *inode;
3670 journal_t *journal = EXT4_SB(sb)->s_journal;
3676 inode = iget_locked(sb, ino);
3678 return ERR_PTR(-ENOMEM);
3679 if (!(inode->i_state & I_NEW))
3685 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3688 raw_inode = ext4_raw_inode(&iloc);
3690 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3691 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3692 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3693 EXT4_INODE_SIZE(inode->i_sb)) {
3694 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3695 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3696 EXT4_INODE_SIZE(inode->i_sb));
3701 ei->i_extra_isize = 0;
3703 /* Precompute checksum seed for inode metadata */
3704 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3705 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3706 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3708 __le32 inum = cpu_to_le32(inode->i_ino);
3709 __le32 gen = raw_inode->i_generation;
3710 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3712 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3716 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3717 EXT4_ERROR_INODE(inode, "checksum invalid");
3722 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3723 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3724 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3725 if (!(test_opt(inode->i_sb, NO_UID32))) {
3726 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3727 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3729 i_uid_write(inode, i_uid);
3730 i_gid_write(inode, i_gid);
3731 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3733 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3734 ei->i_inline_off = 0;
3735 ei->i_dir_start_lookup = 0;
3736 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3737 /* We now have enough fields to check if the inode was active or not.
3738 * This is needed because nfsd might try to access dead inodes
3739 * the test is that same one that e2fsck uses
3740 * NeilBrown 1999oct15
3742 if (inode->i_nlink == 0) {
3743 if (inode->i_mode == 0 ||
3744 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3745 /* this inode is deleted */
3749 /* The only unlinked inodes we let through here have
3750 * valid i_mode and are being read by the orphan
3751 * recovery code: that's fine, we're about to complete
3752 * the process of deleting those. */
3754 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3755 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3756 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3757 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3759 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3760 inode->i_size = ext4_isize(raw_inode);
3761 ei->i_disksize = inode->i_size;
3763 ei->i_reserved_quota = 0;
3765 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3766 ei->i_block_group = iloc.block_group;
3767 ei->i_last_alloc_group = ~0;
3769 * NOTE! The in-memory inode i_data array is in little-endian order
3770 * even on big-endian machines: we do NOT byteswap the block numbers!
3772 for (block = 0; block < EXT4_N_BLOCKS; block++)
3773 ei->i_data[block] = raw_inode->i_block[block];
3774 INIT_LIST_HEAD(&ei->i_orphan);
3777 * Set transaction id's of transactions that have to be committed
3778 * to finish f[data]sync. We set them to currently running transaction
3779 * as we cannot be sure that the inode or some of its metadata isn't
3780 * part of the transaction - the inode could have been reclaimed and
3781 * now it is reread from disk.
3784 transaction_t *transaction;
3787 read_lock(&journal->j_state_lock);
3788 if (journal->j_running_transaction)
3789 transaction = journal->j_running_transaction;
3791 transaction = journal->j_committing_transaction;
3793 tid = transaction->t_tid;
3795 tid = journal->j_commit_sequence;
3796 read_unlock(&journal->j_state_lock);
3797 ei->i_sync_tid = tid;
3798 ei->i_datasync_tid = tid;
3801 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3802 if (ei->i_extra_isize == 0) {
3803 /* The extra space is currently unused. Use it. */
3804 ei->i_extra_isize = sizeof(struct ext4_inode) -
3805 EXT4_GOOD_OLD_INODE_SIZE;
3807 ext4_iget_extra_inode(inode, raw_inode, ei);
3811 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3812 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3813 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3814 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3816 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3817 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3818 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3820 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3824 if (ei->i_file_acl &&
3825 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3826 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3830 } else if (!ext4_has_inline_data(inode)) {
3831 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3832 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3833 (S_ISLNK(inode->i_mode) &&
3834 !ext4_inode_is_fast_symlink(inode))))
3835 /* Validate extent which is part of inode */
3836 ret = ext4_ext_check_inode(inode);
3837 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3838 (S_ISLNK(inode->i_mode) &&
3839 !ext4_inode_is_fast_symlink(inode))) {
3840 /* Validate block references which are part of inode */
3841 ret = ext4_ind_check_inode(inode);
3847 if (S_ISREG(inode->i_mode)) {
3848 inode->i_op = &ext4_file_inode_operations;
3849 inode->i_fop = &ext4_file_operations;
3850 ext4_set_aops(inode);
3851 } else if (S_ISDIR(inode->i_mode)) {
3852 inode->i_op = &ext4_dir_inode_operations;
3853 inode->i_fop = &ext4_dir_operations;
3854 } else if (S_ISLNK(inode->i_mode)) {
3855 if (ext4_inode_is_fast_symlink(inode)) {
3856 inode->i_op = &ext4_fast_symlink_inode_operations;
3857 nd_terminate_link(ei->i_data, inode->i_size,
3858 sizeof(ei->i_data) - 1);
3860 inode->i_op = &ext4_symlink_inode_operations;
3861 ext4_set_aops(inode);
3863 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3864 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3865 inode->i_op = &ext4_special_inode_operations;
3866 if (raw_inode->i_block[0])
3867 init_special_inode(inode, inode->i_mode,
3868 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3870 init_special_inode(inode, inode->i_mode,
3871 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3874 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3878 ext4_set_inode_flags(inode);
3879 unlock_new_inode(inode);
3885 return ERR_PTR(ret);
3888 static int ext4_inode_blocks_set(handle_t *handle,
3889 struct ext4_inode *raw_inode,
3890 struct ext4_inode_info *ei)
3892 struct inode *inode = &(ei->vfs_inode);
3893 u64 i_blocks = inode->i_blocks;
3894 struct super_block *sb = inode->i_sb;
3896 if (i_blocks <= ~0U) {
3898 * i_blocks can be represented in a 32 bit variable
3899 * as multiple of 512 bytes
3901 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3902 raw_inode->i_blocks_high = 0;
3903 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3906 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3909 if (i_blocks <= 0xffffffffffffULL) {
3911 * i_blocks can be represented in a 48 bit variable
3912 * as multiple of 512 bytes
3914 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3915 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3916 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3918 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3919 /* i_block is stored in file system block size */
3920 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3921 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3922 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3928 * Post the struct inode info into an on-disk inode location in the
3929 * buffer-cache. This gobbles the caller's reference to the
3930 * buffer_head in the inode location struct.
3932 * The caller must have write access to iloc->bh.
3934 static int ext4_do_update_inode(handle_t *handle,
3935 struct inode *inode,
3936 struct ext4_iloc *iloc)
3938 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3939 struct ext4_inode_info *ei = EXT4_I(inode);
3940 struct buffer_head *bh = iloc->bh;
3941 int err = 0, rc, block;
3942 int need_datasync = 0;
3946 /* For fields not not tracking in the in-memory inode,
3947 * initialise them to zero for new inodes. */
3948 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
3949 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3951 ext4_get_inode_flags(ei);
3952 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3953 i_uid = i_uid_read(inode);
3954 i_gid = i_gid_read(inode);
3955 if (!(test_opt(inode->i_sb, NO_UID32))) {
3956 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3957 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
3959 * Fix up interoperability with old kernels. Otherwise, old inodes get
3960 * re-used with the upper 16 bits of the uid/gid intact
3963 raw_inode->i_uid_high =
3964 cpu_to_le16(high_16_bits(i_uid));
3965 raw_inode->i_gid_high =
3966 cpu_to_le16(high_16_bits(i_gid));
3968 raw_inode->i_uid_high = 0;
3969 raw_inode->i_gid_high = 0;
3972 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
3973 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
3974 raw_inode->i_uid_high = 0;
3975 raw_inode->i_gid_high = 0;
3977 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3979 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
3980 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
3981 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
3982 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
3984 if (ext4_inode_blocks_set(handle, raw_inode, ei))
3986 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3987 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
3988 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3989 cpu_to_le32(EXT4_OS_HURD))
3990 raw_inode->i_file_acl_high =
3991 cpu_to_le16(ei->i_file_acl >> 32);
3992 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
3993 if (ei->i_disksize != ext4_isize(raw_inode)) {
3994 ext4_isize_set(raw_inode, ei->i_disksize);
3997 if (ei->i_disksize > 0x7fffffffULL) {
3998 struct super_block *sb = inode->i_sb;
3999 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4000 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4001 EXT4_SB(sb)->s_es->s_rev_level ==
4002 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4003 /* If this is the first large file
4004 * created, add a flag to the superblock.
4006 err = ext4_journal_get_write_access(handle,
4007 EXT4_SB(sb)->s_sbh);
4010 ext4_update_dynamic_rev(sb);
4011 EXT4_SET_RO_COMPAT_FEATURE(sb,
4012 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4013 ext4_handle_sync(handle);
4014 err = ext4_handle_dirty_super(handle, sb);
4017 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4018 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4019 if (old_valid_dev(inode->i_rdev)) {
4020 raw_inode->i_block[0] =
4021 cpu_to_le32(old_encode_dev(inode->i_rdev));
4022 raw_inode->i_block[1] = 0;
4024 raw_inode->i_block[0] = 0;
4025 raw_inode->i_block[1] =
4026 cpu_to_le32(new_encode_dev(inode->i_rdev));
4027 raw_inode->i_block[2] = 0;
4029 } else if (!ext4_has_inline_data(inode)) {
4030 for (block = 0; block < EXT4_N_BLOCKS; block++)
4031 raw_inode->i_block[block] = ei->i_data[block];
4034 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4035 if (ei->i_extra_isize) {
4036 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4037 raw_inode->i_version_hi =
4038 cpu_to_le32(inode->i_version >> 32);
4039 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4042 ext4_inode_csum_set(inode, raw_inode, ei);
4044 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4045 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4048 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4050 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4053 ext4_std_error(inode->i_sb, err);
4058 * ext4_write_inode()
4060 * We are called from a few places:
4062 * - Within generic_file_write() for O_SYNC files.
4063 * Here, there will be no transaction running. We wait for any running
4064 * transaction to commit.
4066 * - Within sys_sync(), kupdate and such.
4067 * We wait on commit, if tol to.
4069 * - Within prune_icache() (PF_MEMALLOC == true)
4070 * Here we simply return. We can't afford to block kswapd on the
4073 * In all cases it is actually safe for us to return without doing anything,
4074 * because the inode has been copied into a raw inode buffer in
4075 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4078 * Note that we are absolutely dependent upon all inode dirtiers doing the
4079 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4080 * which we are interested.
4082 * It would be a bug for them to not do this. The code:
4084 * mark_inode_dirty(inode)
4086 * inode->i_size = expr;
4088 * is in error because a kswapd-driven write_inode() could occur while
4089 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4090 * will no longer be on the superblock's dirty inode list.
4092 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4096 if (current->flags & PF_MEMALLOC)
4099 if (EXT4_SB(inode->i_sb)->s_journal) {
4100 if (ext4_journal_current_handle()) {
4101 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4106 if (wbc->sync_mode != WB_SYNC_ALL)
4109 err = ext4_force_commit(inode->i_sb);
4111 struct ext4_iloc iloc;
4113 err = __ext4_get_inode_loc(inode, &iloc, 0);
4116 if (wbc->sync_mode == WB_SYNC_ALL)
4117 sync_dirty_buffer(iloc.bh);
4118 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4119 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4120 "IO error syncing inode");
4129 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4130 * buffers that are attached to a page stradding i_size and are undergoing
4131 * commit. In that case we have to wait for commit to finish and try again.
4133 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4137 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4138 tid_t commit_tid = 0;
4141 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4143 * All buffers in the last page remain valid? Then there's nothing to
4144 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4147 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4150 page = find_lock_page(inode->i_mapping,
4151 inode->i_size >> PAGE_CACHE_SHIFT);
4154 ret = __ext4_journalled_invalidatepage(page, offset);
4156 page_cache_release(page);
4160 read_lock(&journal->j_state_lock);
4161 if (journal->j_committing_transaction)
4162 commit_tid = journal->j_committing_transaction->t_tid;
4163 read_unlock(&journal->j_state_lock);
4165 jbd2_log_wait_commit(journal, commit_tid);
4172 * Called from notify_change.
4174 * We want to trap VFS attempts to truncate the file as soon as
4175 * possible. In particular, we want to make sure that when the VFS
4176 * shrinks i_size, we put the inode on the orphan list and modify
4177 * i_disksize immediately, so that during the subsequent flushing of
4178 * dirty pages and freeing of disk blocks, we can guarantee that any
4179 * commit will leave the blocks being flushed in an unused state on
4180 * disk. (On recovery, the inode will get truncated and the blocks will
4181 * be freed, so we have a strong guarantee that no future commit will
4182 * leave these blocks visible to the user.)
4184 * Another thing we have to assure is that if we are in ordered mode
4185 * and inode is still attached to the committing transaction, we must
4186 * we start writeout of all the dirty pages which are being truncated.
4187 * This way we are sure that all the data written in the previous
4188 * transaction are already on disk (truncate waits for pages under
4191 * Called with inode->i_mutex down.
4193 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4195 struct inode *inode = dentry->d_inode;
4198 const unsigned int ia_valid = attr->ia_valid;
4200 error = inode_change_ok(inode, attr);
4204 if (is_quota_modification(inode, attr))
4205 dquot_initialize(inode);
4206 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4207 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4210 /* (user+group)*(old+new) structure, inode write (sb,
4211 * inode block, ? - but truncate inode update has it) */
4212 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4213 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4214 if (IS_ERR(handle)) {
4215 error = PTR_ERR(handle);
4218 error = dquot_transfer(inode, attr);
4220 ext4_journal_stop(handle);
4223 /* Update corresponding info in inode so that everything is in
4224 * one transaction */
4225 if (attr->ia_valid & ATTR_UID)
4226 inode->i_uid = attr->ia_uid;
4227 if (attr->ia_valid & ATTR_GID)
4228 inode->i_gid = attr->ia_gid;
4229 error = ext4_mark_inode_dirty(handle, inode);
4230 ext4_journal_stop(handle);
4233 if (attr->ia_valid & ATTR_SIZE) {
4235 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4236 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4238 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4243 if (S_ISREG(inode->i_mode) &&
4244 attr->ia_valid & ATTR_SIZE &&
4245 (attr->ia_size < inode->i_size)) {
4248 handle = ext4_journal_start(inode, 3);
4249 if (IS_ERR(handle)) {
4250 error = PTR_ERR(handle);
4253 if (ext4_handle_valid(handle)) {
4254 error = ext4_orphan_add(handle, inode);
4257 EXT4_I(inode)->i_disksize = attr->ia_size;
4258 rc = ext4_mark_inode_dirty(handle, inode);
4261 ext4_journal_stop(handle);
4263 if (ext4_should_order_data(inode)) {
4264 error = ext4_begin_ordered_truncate(inode,
4267 /* Do as much error cleanup as possible */
4268 handle = ext4_journal_start(inode, 3);
4269 if (IS_ERR(handle)) {
4270 ext4_orphan_del(NULL, inode);
4273 ext4_orphan_del(handle, inode);
4275 ext4_journal_stop(handle);
4281 if (attr->ia_valid & ATTR_SIZE) {
4282 if (attr->ia_size != inode->i_size) {
4283 loff_t oldsize = inode->i_size;
4285 i_size_write(inode, attr->ia_size);
4287 * Blocks are going to be removed from the inode. Wait
4288 * for dio in flight. Temporarily disable
4289 * dioread_nolock to prevent livelock.
4292 if (!ext4_should_journal_data(inode)) {
4293 ext4_inode_block_unlocked_dio(inode);
4294 inode_dio_wait(inode);
4295 ext4_inode_resume_unlocked_dio(inode);
4297 ext4_wait_for_tail_page_commit(inode);
4300 * Truncate pagecache after we've waited for commit
4301 * in data=journal mode to make pages freeable.
4303 truncate_pagecache(inode, oldsize, inode->i_size);
4305 ext4_truncate(inode);
4309 setattr_copy(inode, attr);
4310 mark_inode_dirty(inode);
4314 * If the call to ext4_truncate failed to get a transaction handle at
4315 * all, we need to clean up the in-core orphan list manually.
4317 if (orphan && inode->i_nlink)
4318 ext4_orphan_del(NULL, inode);
4320 if (!rc && (ia_valid & ATTR_MODE))
4321 rc = ext4_acl_chmod(inode);
4324 ext4_std_error(inode->i_sb, error);
4330 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4333 struct inode *inode;
4334 unsigned long delalloc_blocks;
4336 inode = dentry->d_inode;
4337 generic_fillattr(inode, stat);
4340 * We can't update i_blocks if the block allocation is delayed
4341 * otherwise in the case of system crash before the real block
4342 * allocation is done, we will have i_blocks inconsistent with
4343 * on-disk file blocks.
4344 * We always keep i_blocks updated together with real
4345 * allocation. But to not confuse with user, stat
4346 * will return the blocks that include the delayed allocation
4347 * blocks for this file.
4349 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4350 EXT4_I(inode)->i_reserved_data_blocks);
4352 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4356 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4358 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4359 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4360 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4364 * Account for index blocks, block groups bitmaps and block group
4365 * descriptor blocks if modify datablocks and index blocks
4366 * worse case, the indexs blocks spread over different block groups
4368 * If datablocks are discontiguous, they are possible to spread over
4369 * different block groups too. If they are contiguous, with flexbg,
4370 * they could still across block group boundary.
4372 * Also account for superblock, inode, quota and xattr blocks
4374 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4376 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4382 * How many index blocks need to touch to modify nrblocks?
4383 * The "Chunk" flag indicating whether the nrblocks is
4384 * physically contiguous on disk
4386 * For Direct IO and fallocate, they calls get_block to allocate
4387 * one single extent at a time, so they could set the "Chunk" flag
4389 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4394 * Now let's see how many group bitmaps and group descriptors need
4404 if (groups > ngroups)
4406 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4407 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4409 /* bitmaps and block group descriptor blocks */
4410 ret += groups + gdpblocks;
4412 /* Blocks for super block, inode, quota and xattr blocks */
4413 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4419 * Calculate the total number of credits to reserve to fit
4420 * the modification of a single pages into a single transaction,
4421 * which may include multiple chunks of block allocations.
4423 * This could be called via ext4_write_begin()
4425 * We need to consider the worse case, when
4426 * one new block per extent.
4428 int ext4_writepage_trans_blocks(struct inode *inode)
4430 int bpp = ext4_journal_blocks_per_page(inode);
4433 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4435 /* Account for data blocks for journalled mode */
4436 if (ext4_should_journal_data(inode))
4442 * Calculate the journal credits for a chunk of data modification.
4444 * This is called from DIO, fallocate or whoever calling
4445 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4447 * journal buffers for data blocks are not included here, as DIO
4448 * and fallocate do no need to journal data buffers.
4450 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4452 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4456 * The caller must have previously called ext4_reserve_inode_write().
4457 * Give this, we know that the caller already has write access to iloc->bh.
4459 int ext4_mark_iloc_dirty(handle_t *handle,
4460 struct inode *inode, struct ext4_iloc *iloc)
4464 if (IS_I_VERSION(inode))
4465 inode_inc_iversion(inode);
4467 /* the do_update_inode consumes one bh->b_count */
4470 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4471 err = ext4_do_update_inode(handle, inode, iloc);
4477 * On success, We end up with an outstanding reference count against
4478 * iloc->bh. This _must_ be cleaned up later.
4482 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4483 struct ext4_iloc *iloc)
4487 err = ext4_get_inode_loc(inode, iloc);
4489 BUFFER_TRACE(iloc->bh, "get_write_access");
4490 err = ext4_journal_get_write_access(handle, iloc->bh);
4496 ext4_std_error(inode->i_sb, err);
4501 * Expand an inode by new_extra_isize bytes.
4502 * Returns 0 on success or negative error number on failure.
4504 static int ext4_expand_extra_isize(struct inode *inode,
4505 unsigned int new_extra_isize,
4506 struct ext4_iloc iloc,
4509 struct ext4_inode *raw_inode;
4510 struct ext4_xattr_ibody_header *header;
4512 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4515 raw_inode = ext4_raw_inode(&iloc);
4517 header = IHDR(inode, raw_inode);
4519 /* No extended attributes present */
4520 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4521 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4522 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4524 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4528 /* try to expand with EAs present */
4529 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4534 * What we do here is to mark the in-core inode as clean with respect to inode
4535 * dirtiness (it may still be data-dirty).
4536 * This means that the in-core inode may be reaped by prune_icache
4537 * without having to perform any I/O. This is a very good thing,
4538 * because *any* task may call prune_icache - even ones which
4539 * have a transaction open against a different journal.
4541 * Is this cheating? Not really. Sure, we haven't written the
4542 * inode out, but prune_icache isn't a user-visible syncing function.
4543 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4544 * we start and wait on commits.
4546 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4548 struct ext4_iloc iloc;
4549 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4550 static unsigned int mnt_count;
4554 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4555 err = ext4_reserve_inode_write(handle, inode, &iloc);
4556 if (ext4_handle_valid(handle) &&
4557 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4558 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4560 * We need extra buffer credits since we may write into EA block
4561 * with this same handle. If journal_extend fails, then it will
4562 * only result in a minor loss of functionality for that inode.
4563 * If this is felt to be critical, then e2fsck should be run to
4564 * force a large enough s_min_extra_isize.
4566 if ((jbd2_journal_extend(handle,
4567 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4568 ret = ext4_expand_extra_isize(inode,
4569 sbi->s_want_extra_isize,
4572 ext4_set_inode_state(inode,
4573 EXT4_STATE_NO_EXPAND);
4575 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4576 ext4_warning(inode->i_sb,
4577 "Unable to expand inode %lu. Delete"
4578 " some EAs or run e2fsck.",
4581 le16_to_cpu(sbi->s_es->s_mnt_count);
4587 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4592 * ext4_dirty_inode() is called from __mark_inode_dirty()
4594 * We're really interested in the case where a file is being extended.
4595 * i_size has been changed by generic_commit_write() and we thus need
4596 * to include the updated inode in the current transaction.
4598 * Also, dquot_alloc_block() will always dirty the inode when blocks
4599 * are allocated to the file.
4601 * If the inode is marked synchronous, we don't honour that here - doing
4602 * so would cause a commit on atime updates, which we don't bother doing.
4603 * We handle synchronous inodes at the highest possible level.
4605 void ext4_dirty_inode(struct inode *inode, int flags)
4609 handle = ext4_journal_start(inode, 2);
4613 ext4_mark_inode_dirty(handle, inode);
4615 ext4_journal_stop(handle);
4622 * Bind an inode's backing buffer_head into this transaction, to prevent
4623 * it from being flushed to disk early. Unlike
4624 * ext4_reserve_inode_write, this leaves behind no bh reference and
4625 * returns no iloc structure, so the caller needs to repeat the iloc
4626 * lookup to mark the inode dirty later.
4628 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4630 struct ext4_iloc iloc;
4634 err = ext4_get_inode_loc(inode, &iloc);
4636 BUFFER_TRACE(iloc.bh, "get_write_access");
4637 err = jbd2_journal_get_write_access(handle, iloc.bh);
4639 err = ext4_handle_dirty_metadata(handle,
4645 ext4_std_error(inode->i_sb, err);
4650 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4657 * We have to be very careful here: changing a data block's
4658 * journaling status dynamically is dangerous. If we write a
4659 * data block to the journal, change the status and then delete
4660 * that block, we risk forgetting to revoke the old log record
4661 * from the journal and so a subsequent replay can corrupt data.
4662 * So, first we make sure that the journal is empty and that
4663 * nobody is changing anything.
4666 journal = EXT4_JOURNAL(inode);
4669 if (is_journal_aborted(journal))
4671 /* We have to allocate physical blocks for delalloc blocks
4672 * before flushing journal. otherwise delalloc blocks can not
4673 * be allocated any more. even more truncate on delalloc blocks
4674 * could trigger BUG by flushing delalloc blocks in journal.
4675 * There is no delalloc block in non-journal data mode.
4677 if (val && test_opt(inode->i_sb, DELALLOC)) {
4678 err = ext4_alloc_da_blocks(inode);
4683 /* Wait for all existing dio workers */
4684 ext4_inode_block_unlocked_dio(inode);
4685 inode_dio_wait(inode);
4687 jbd2_journal_lock_updates(journal);
4690 * OK, there are no updates running now, and all cached data is
4691 * synced to disk. We are now in a completely consistent state
4692 * which doesn't have anything in the journal, and we know that
4693 * no filesystem updates are running, so it is safe to modify
4694 * the inode's in-core data-journaling state flag now.
4698 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4700 jbd2_journal_flush(journal);
4701 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4703 ext4_set_aops(inode);
4705 jbd2_journal_unlock_updates(journal);
4706 ext4_inode_resume_unlocked_dio(inode);
4708 /* Finally we can mark the inode as dirty. */
4710 handle = ext4_journal_start(inode, 1);
4712 return PTR_ERR(handle);
4714 err = ext4_mark_inode_dirty(handle, inode);
4715 ext4_handle_sync(handle);
4716 ext4_journal_stop(handle);
4717 ext4_std_error(inode->i_sb, err);
4722 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4724 return !buffer_mapped(bh);
4727 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4729 struct page *page = vmf->page;
4733 struct file *file = vma->vm_file;
4734 struct inode *inode = file->f_path.dentry->d_inode;
4735 struct address_space *mapping = inode->i_mapping;
4737 get_block_t *get_block;
4740 sb_start_pagefault(inode->i_sb);
4741 file_update_time(vma->vm_file);
4742 /* Delalloc case is easy... */
4743 if (test_opt(inode->i_sb, DELALLOC) &&
4744 !ext4_should_journal_data(inode) &&
4745 !ext4_nonda_switch(inode->i_sb)) {
4747 ret = __block_page_mkwrite(vma, vmf,
4748 ext4_da_get_block_prep);
4749 } while (ret == -ENOSPC &&
4750 ext4_should_retry_alloc(inode->i_sb, &retries));
4755 size = i_size_read(inode);
4756 /* Page got truncated from under us? */
4757 if (page->mapping != mapping || page_offset(page) > size) {
4759 ret = VM_FAULT_NOPAGE;
4763 if (page->index == size >> PAGE_CACHE_SHIFT)
4764 len = size & ~PAGE_CACHE_MASK;
4766 len = PAGE_CACHE_SIZE;
4768 * Return if we have all the buffers mapped. This avoids the need to do
4769 * journal_start/journal_stop which can block and take a long time
4771 if (page_has_buffers(page)) {
4772 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
4774 ext4_bh_unmapped)) {
4775 /* Wait so that we don't change page under IO */
4776 wait_on_page_writeback(page);
4777 ret = VM_FAULT_LOCKED;
4782 /* OK, we need to fill the hole... */
4783 if (ext4_should_dioread_nolock(inode))
4784 get_block = ext4_get_block_write;
4786 get_block = ext4_get_block;
4788 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4789 if (IS_ERR(handle)) {
4790 ret = VM_FAULT_SIGBUS;
4793 ret = __block_page_mkwrite(vma, vmf, get_block);
4794 if (!ret && ext4_should_journal_data(inode)) {
4795 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
4796 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4798 ret = VM_FAULT_SIGBUS;
4799 ext4_journal_stop(handle);
4802 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4804 ext4_journal_stop(handle);
4805 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4808 ret = block_page_mkwrite_return(ret);
4810 sb_end_pagefault(inode->i_sb);