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 * @bh - bh of the block (used to access block's state)
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,
1655 sector_t logical, size_t b_size,
1656 unsigned long b_state)
1659 int nrblocks = mpd->b_size >> mpd->inode->i_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/mpd->inode->i_sb->s_blocksize)
1670 /* check if thereserved 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.
1680 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1681 EXT4_MAX_TRANS_DATA) {
1683 * Adding the new buffer_head would make it cross the
1684 * allowed limit for which we have journal credit
1685 * reserved. So limit the new bh->b_size
1687 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1688 mpd->inode->i_blkbits;
1689 /* we will do mpage_da_submit_io in the next loop */
1693 * First block in the extent
1695 if (mpd->b_size == 0) {
1696 mpd->b_blocknr = logical;
1697 mpd->b_size = b_size;
1698 mpd->b_state = b_state & BH_FLAGS;
1702 next = mpd->b_blocknr + nrblocks;
1704 * Can we merge the block to our big extent?
1706 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1707 mpd->b_size += b_size;
1713 * We couldn't merge the block to our extent, so we
1714 * need to flush current extent and start new one
1716 mpage_da_map_and_submit(mpd);
1720 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1722 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1726 * This function is grabs code from the very beginning of
1727 * ext4_map_blocks, but assumes that the caller is from delayed write
1728 * time. This function looks up the requested blocks and sets the
1729 * buffer delay bit under the protection of i_data_sem.
1731 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1732 struct ext4_map_blocks *map,
1733 struct buffer_head *bh)
1736 sector_t invalid_block = ~((sector_t) 0xffff);
1738 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1742 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1743 "logical block %lu\n", inode->i_ino, map->m_len,
1744 (unsigned long) map->m_lblk);
1746 * Try to see if we can get the block without requesting a new
1747 * file system block.
1749 down_read((&EXT4_I(inode)->i_data_sem));
1750 if (ext4_has_inline_data(inode)) {
1752 * We will soon create blocks for this page, and let
1753 * us pretend as if the blocks aren't allocated yet.
1754 * In case of clusters, we have to handle the work
1755 * of mapping from cluster so that the reserved space
1756 * is calculated properly.
1758 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1759 ext4_find_delalloc_cluster(inode, map->m_lblk))
1760 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1762 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1763 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1765 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1769 * XXX: __block_prepare_write() unmaps passed block,
1772 /* If the block was allocated from previously allocated cluster,
1773 * then we dont need to reserve it again. */
1774 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1775 retval = ext4_da_reserve_space(inode, iblock);
1777 /* not enough space to reserve */
1781 retval = ext4_es_insert_extent(inode, map->m_lblk, map->m_len);
1785 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1786 * and it should not appear on the bh->b_state.
1788 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1790 map_bh(bh, inode->i_sb, invalid_block);
1792 set_buffer_delay(bh);
1796 up_read((&EXT4_I(inode)->i_data_sem));
1802 * This is a special get_blocks_t callback which is used by
1803 * ext4_da_write_begin(). It will either return mapped block or
1804 * reserve space for a single block.
1806 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1807 * We also have b_blocknr = -1 and b_bdev initialized properly
1809 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1810 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1811 * initialized properly.
1813 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1814 struct buffer_head *bh, int create)
1816 struct ext4_map_blocks map;
1819 BUG_ON(create == 0);
1820 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1822 map.m_lblk = iblock;
1826 * first, we need to know whether the block is allocated already
1827 * preallocated blocks are unmapped but should treated
1828 * the same as allocated blocks.
1830 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1834 map_bh(bh, inode->i_sb, map.m_pblk);
1835 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1837 if (buffer_unwritten(bh)) {
1838 /* A delayed write to unwritten bh should be marked
1839 * new and mapped. Mapped ensures that we don't do
1840 * get_block multiple times when we write to the same
1841 * offset and new ensures that we do proper zero out
1842 * for partial write.
1845 set_buffer_mapped(bh);
1850 static int bget_one(handle_t *handle, struct buffer_head *bh)
1856 static int bput_one(handle_t *handle, struct buffer_head *bh)
1862 static int __ext4_journalled_writepage(struct page *page,
1865 struct address_space *mapping = page->mapping;
1866 struct inode *inode = mapping->host;
1867 struct buffer_head *page_bufs = NULL;
1868 handle_t *handle = NULL;
1869 int ret = 0, err = 0;
1870 int inline_data = ext4_has_inline_data(inode);
1871 struct buffer_head *inode_bh = NULL;
1873 ClearPageChecked(page);
1876 BUG_ON(page->index != 0);
1877 BUG_ON(len > ext4_get_max_inline_size(inode));
1878 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1879 if (inode_bh == NULL)
1882 page_bufs = page_buffers(page);
1887 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1890 /* As soon as we unlock the page, it can go away, but we have
1891 * references to buffers so we are safe */
1894 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1895 if (IS_ERR(handle)) {
1896 ret = PTR_ERR(handle);
1900 BUG_ON(!ext4_handle_valid(handle));
1903 ret = ext4_journal_get_write_access(handle, inode_bh);
1905 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1908 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1909 do_journal_get_write_access);
1911 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1916 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1917 err = ext4_journal_stop(handle);
1921 if (!ext4_has_inline_data(inode))
1922 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1924 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1931 * Note that we don't need to start a transaction unless we're journaling data
1932 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1933 * need to file the inode to the transaction's list in ordered mode because if
1934 * we are writing back data added by write(), the inode is already there and if
1935 * we are writing back data modified via mmap(), no one guarantees in which
1936 * transaction the data will hit the disk. In case we are journaling data, we
1937 * cannot start transaction directly because transaction start ranks above page
1938 * lock so we have to do some magic.
1940 * This function can get called via...
1941 * - ext4_da_writepages after taking page lock (have journal handle)
1942 * - journal_submit_inode_data_buffers (no journal handle)
1943 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1944 * - grab_page_cache when doing write_begin (have journal handle)
1946 * We don't do any block allocation in this function. If we have page with
1947 * multiple blocks we need to write those buffer_heads that are mapped. This
1948 * is important for mmaped based write. So if we do with blocksize 1K
1949 * truncate(f, 1024);
1950 * a = mmap(f, 0, 4096);
1952 * truncate(f, 4096);
1953 * we have in the page first buffer_head mapped via page_mkwrite call back
1954 * but other buffer_heads would be unmapped but dirty (dirty done via the
1955 * do_wp_page). So writepage should write the first block. If we modify
1956 * the mmap area beyond 1024 we will again get a page_fault and the
1957 * page_mkwrite callback will do the block allocation and mark the
1958 * buffer_heads mapped.
1960 * We redirty the page if we have any buffer_heads that is either delay or
1961 * unwritten in the page.
1963 * We can get recursively called as show below.
1965 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1968 * But since we don't do any block allocation we should not deadlock.
1969 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1971 static int ext4_writepage(struct page *page,
1972 struct writeback_control *wbc)
1977 struct buffer_head *page_bufs = NULL;
1978 struct inode *inode = page->mapping->host;
1979 struct ext4_io_submit io_submit;
1981 trace_ext4_writepage(page);
1982 size = i_size_read(inode);
1983 if (page->index == size >> PAGE_CACHE_SHIFT)
1984 len = size & ~PAGE_CACHE_MASK;
1986 len = PAGE_CACHE_SIZE;
1988 page_bufs = page_buffers(page);
1989 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1990 ext4_bh_delay_or_unwritten)) {
1992 * We don't want to do block allocation, so redirty
1993 * the page and return. We may reach here when we do
1994 * a journal commit via journal_submit_inode_data_buffers.
1995 * We can also reach here via shrink_page_list but it
1996 * should never be for direct reclaim so warn if that
1999 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2001 redirty_page_for_writepage(wbc, page);
2006 if (PageChecked(page) && ext4_should_journal_data(inode))
2008 * It's mmapped pagecache. Add buffers and journal it. There
2009 * doesn't seem much point in redirtying the page here.
2011 return __ext4_journalled_writepage(page, len);
2013 memset(&io_submit, 0, sizeof(io_submit));
2014 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2015 ext4_io_submit(&io_submit);
2020 * This is called via ext4_da_writepages() to
2021 * calculate the total number of credits to reserve to fit
2022 * a single extent allocation into a single transaction,
2023 * ext4_da_writpeages() will loop calling this before
2024 * the block allocation.
2027 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2029 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2032 * With non-extent format the journal credit needed to
2033 * insert nrblocks contiguous block is dependent on
2034 * number of contiguous block. So we will limit
2035 * number of contiguous block to a sane value
2037 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2038 (max_blocks > EXT4_MAX_TRANS_DATA))
2039 max_blocks = EXT4_MAX_TRANS_DATA;
2041 return ext4_chunk_trans_blocks(inode, max_blocks);
2045 * write_cache_pages_da - walk the list of dirty pages of the given
2046 * address space and accumulate pages that need writing, and call
2047 * mpage_da_map_and_submit to map a single contiguous memory region
2048 * and then write them.
2050 static int write_cache_pages_da(handle_t *handle,
2051 struct address_space *mapping,
2052 struct writeback_control *wbc,
2053 struct mpage_da_data *mpd,
2054 pgoff_t *done_index)
2056 struct buffer_head *bh, *head;
2057 struct inode *inode = mapping->host;
2058 struct pagevec pvec;
2059 unsigned int nr_pages;
2062 long nr_to_write = wbc->nr_to_write;
2063 int i, tag, ret = 0;
2065 memset(mpd, 0, sizeof(struct mpage_da_data));
2068 pagevec_init(&pvec, 0);
2069 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2070 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2072 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2073 tag = PAGECACHE_TAG_TOWRITE;
2075 tag = PAGECACHE_TAG_DIRTY;
2077 *done_index = index;
2078 while (index <= end) {
2079 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2080 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2084 for (i = 0; i < nr_pages; i++) {
2085 struct page *page = pvec.pages[i];
2088 * At this point, the page may be truncated or
2089 * invalidated (changing page->mapping to NULL), or
2090 * even swizzled back from swapper_space to tmpfs file
2091 * mapping. However, page->index will not change
2092 * because we have a reference on the page.
2094 if (page->index > end)
2097 *done_index = page->index + 1;
2100 * If we can't merge this page, and we have
2101 * accumulated an contiguous region, write it
2103 if ((mpd->next_page != page->index) &&
2104 (mpd->next_page != mpd->first_page)) {
2105 mpage_da_map_and_submit(mpd);
2106 goto ret_extent_tail;
2112 * If the page is no longer dirty, or its
2113 * mapping no longer corresponds to inode we
2114 * are writing (which means it has been
2115 * truncated or invalidated), or the page is
2116 * already under writeback and we are not
2117 * doing a data integrity writeback, skip the page
2119 if (!PageDirty(page) ||
2120 (PageWriteback(page) &&
2121 (wbc->sync_mode == WB_SYNC_NONE)) ||
2122 unlikely(page->mapping != mapping)) {
2127 wait_on_page_writeback(page);
2128 BUG_ON(PageWriteback(page));
2131 * If we have inline data and arrive here, it means that
2132 * we will soon create the block for the 1st page, so
2133 * we'd better clear the inline data here.
2135 if (ext4_has_inline_data(inode)) {
2136 BUG_ON(ext4_test_inode_state(inode,
2137 EXT4_STATE_MAY_INLINE_DATA));
2138 ext4_destroy_inline_data(handle, inode);
2141 if (mpd->next_page != page->index)
2142 mpd->first_page = page->index;
2143 mpd->next_page = page->index + 1;
2144 logical = (sector_t) page->index <<
2145 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2147 /* Add all dirty buffers to mpd */
2148 head = page_buffers(page);
2151 BUG_ON(buffer_locked(bh));
2153 * We need to try to allocate unmapped blocks
2154 * in the same page. Otherwise we won't make
2155 * progress with the page in ext4_writepage
2157 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2158 mpage_add_bh_to_extent(mpd, logical,
2162 goto ret_extent_tail;
2163 } else if (buffer_dirty(bh) &&
2164 buffer_mapped(bh)) {
2166 * mapped dirty buffer. We need to
2167 * update the b_state because we look
2168 * at b_state in mpage_da_map_blocks.
2169 * We don't update b_size because if we
2170 * find an unmapped buffer_head later
2171 * we need to use the b_state flag of
2174 if (mpd->b_size == 0)
2176 bh->b_state & BH_FLAGS;
2179 } while ((bh = bh->b_this_page) != head);
2181 if (nr_to_write > 0) {
2183 if (nr_to_write == 0 &&
2184 wbc->sync_mode == WB_SYNC_NONE)
2186 * We stop writing back only if we are
2187 * not doing integrity sync. In case of
2188 * integrity sync we have to keep going
2189 * because someone may be concurrently
2190 * dirtying pages, and we might have
2191 * synced a lot of newly appeared dirty
2192 * pages, but have not synced all of the
2198 pagevec_release(&pvec);
2203 ret = MPAGE_DA_EXTENT_TAIL;
2205 pagevec_release(&pvec);
2211 static int ext4_da_writepages(struct address_space *mapping,
2212 struct writeback_control *wbc)
2215 int range_whole = 0;
2216 handle_t *handle = NULL;
2217 struct mpage_da_data mpd;
2218 struct inode *inode = mapping->host;
2219 int pages_written = 0;
2220 unsigned int max_pages;
2221 int range_cyclic, cycled = 1, io_done = 0;
2222 int needed_blocks, ret = 0;
2223 long desired_nr_to_write, nr_to_writebump = 0;
2224 loff_t range_start = wbc->range_start;
2225 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2226 pgoff_t done_index = 0;
2228 struct blk_plug plug;
2230 trace_ext4_da_writepages(inode, wbc);
2233 * No pages to write? This is mainly a kludge to avoid starting
2234 * a transaction for special inodes like journal inode on last iput()
2235 * because that could violate lock ordering on umount
2237 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2241 * If the filesystem has aborted, it is read-only, so return
2242 * right away instead of dumping stack traces later on that
2243 * will obscure the real source of the problem. We test
2244 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2245 * the latter could be true if the filesystem is mounted
2246 * read-only, and in that case, ext4_da_writepages should
2247 * *never* be called, so if that ever happens, we would want
2250 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2253 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2256 range_cyclic = wbc->range_cyclic;
2257 if (wbc->range_cyclic) {
2258 index = mapping->writeback_index;
2261 wbc->range_start = index << PAGE_CACHE_SHIFT;
2262 wbc->range_end = LLONG_MAX;
2263 wbc->range_cyclic = 0;
2266 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2267 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2271 * This works around two forms of stupidity. The first is in
2272 * the writeback code, which caps the maximum number of pages
2273 * written to be 1024 pages. This is wrong on multiple
2274 * levels; different architectues have a different page size,
2275 * which changes the maximum amount of data which gets
2276 * written. Secondly, 4 megabytes is way too small. XFS
2277 * forces this value to be 16 megabytes by multiplying
2278 * nr_to_write parameter by four, and then relies on its
2279 * allocator to allocate larger extents to make them
2280 * contiguous. Unfortunately this brings us to the second
2281 * stupidity, which is that ext4's mballoc code only allocates
2282 * at most 2048 blocks. So we force contiguous writes up to
2283 * the number of dirty blocks in the inode, or
2284 * sbi->max_writeback_mb_bump whichever is smaller.
2286 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2287 if (!range_cyclic && range_whole) {
2288 if (wbc->nr_to_write == LONG_MAX)
2289 desired_nr_to_write = wbc->nr_to_write;
2291 desired_nr_to_write = wbc->nr_to_write * 8;
2293 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2295 if (desired_nr_to_write > max_pages)
2296 desired_nr_to_write = max_pages;
2298 if (wbc->nr_to_write < desired_nr_to_write) {
2299 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2300 wbc->nr_to_write = desired_nr_to_write;
2304 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2305 tag_pages_for_writeback(mapping, index, end);
2307 blk_start_plug(&plug);
2308 while (!ret && wbc->nr_to_write > 0) {
2311 * we insert one extent at a time. So we need
2312 * credit needed for single extent allocation.
2313 * journalled mode is currently not supported
2316 BUG_ON(ext4_should_journal_data(inode));
2317 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2319 /* start a new transaction*/
2320 handle = ext4_journal_start(inode, needed_blocks);
2321 if (IS_ERR(handle)) {
2322 ret = PTR_ERR(handle);
2323 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2324 "%ld pages, ino %lu; err %d", __func__,
2325 wbc->nr_to_write, inode->i_ino, ret);
2326 blk_finish_plug(&plug);
2327 goto out_writepages;
2331 * Now call write_cache_pages_da() to find the next
2332 * contiguous region of logical blocks that need
2333 * blocks to be allocated by ext4 and submit them.
2335 ret = write_cache_pages_da(handle, mapping,
2336 wbc, &mpd, &done_index);
2338 * If we have a contiguous extent of pages and we
2339 * haven't done the I/O yet, map the blocks and submit
2342 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2343 mpage_da_map_and_submit(&mpd);
2344 ret = MPAGE_DA_EXTENT_TAIL;
2346 trace_ext4_da_write_pages(inode, &mpd);
2347 wbc->nr_to_write -= mpd.pages_written;
2349 ext4_journal_stop(handle);
2351 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2352 /* commit the transaction which would
2353 * free blocks released in the transaction
2356 jbd2_journal_force_commit_nested(sbi->s_journal);
2358 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2360 * Got one extent now try with rest of the pages.
2361 * If mpd.retval is set -EIO, journal is aborted.
2362 * So we don't need to write any more.
2364 pages_written += mpd.pages_written;
2367 } else if (wbc->nr_to_write)
2369 * There is no more writeout needed
2370 * or we requested for a noblocking writeout
2371 * and we found the device congested
2375 blk_finish_plug(&plug);
2376 if (!io_done && !cycled) {
2379 wbc->range_start = index << PAGE_CACHE_SHIFT;
2380 wbc->range_end = mapping->writeback_index - 1;
2385 wbc->range_cyclic = range_cyclic;
2386 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2388 * set the writeback_index so that range_cyclic
2389 * mode will write it back later
2391 mapping->writeback_index = done_index;
2394 wbc->nr_to_write -= nr_to_writebump;
2395 wbc->range_start = range_start;
2396 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2400 static int ext4_nonda_switch(struct super_block *sb)
2402 s64 free_blocks, dirty_blocks;
2403 struct ext4_sb_info *sbi = EXT4_SB(sb);
2406 * switch to non delalloc mode if we are running low
2407 * on free block. The free block accounting via percpu
2408 * counters can get slightly wrong with percpu_counter_batch getting
2409 * accumulated on each CPU without updating global counters
2410 * Delalloc need an accurate free block accounting. So switch
2411 * to non delalloc when we are near to error range.
2413 free_blocks = EXT4_C2B(sbi,
2414 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2415 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2417 * Start pushing delalloc when 1/2 of free blocks are dirty.
2419 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2420 !writeback_in_progress(sb->s_bdi) &&
2421 down_read_trylock(&sb->s_umount)) {
2422 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2423 up_read(&sb->s_umount);
2426 if (2 * free_blocks < 3 * dirty_blocks ||
2427 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2429 * free block count is less than 150% of dirty blocks
2430 * or free blocks is less than watermark
2437 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2438 loff_t pos, unsigned len, unsigned flags,
2439 struct page **pagep, void **fsdata)
2441 int ret, retries = 0;
2444 struct inode *inode = mapping->host;
2447 index = pos >> PAGE_CACHE_SHIFT;
2449 if (ext4_nonda_switch(inode->i_sb)) {
2450 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2451 return ext4_write_begin(file, mapping, pos,
2452 len, flags, pagep, fsdata);
2454 *fsdata = (void *)0;
2455 trace_ext4_da_write_begin(inode, pos, len, flags);
2457 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2458 ret = ext4_da_write_inline_data_begin(mapping, inode,
2471 * With delayed allocation, we don't log the i_disksize update
2472 * if there is delayed block allocation. But we still need
2473 * to journalling the i_disksize update if writes to the end
2474 * of file which has an already mapped buffer.
2476 handle = ext4_journal_start(inode, 1);
2477 if (IS_ERR(handle)) {
2478 ret = PTR_ERR(handle);
2481 /* We cannot recurse into the filesystem as the transaction is already
2483 flags |= AOP_FLAG_NOFS;
2485 page = grab_cache_page_write_begin(mapping, index, flags);
2487 ext4_journal_stop(handle);
2493 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2496 ext4_journal_stop(handle);
2497 page_cache_release(page);
2499 * block_write_begin may have instantiated a few blocks
2500 * outside i_size. Trim these off again. Don't need
2501 * i_size_read because we hold i_mutex.
2503 if (pos + len > inode->i_size)
2504 ext4_truncate_failed_write(inode);
2507 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2514 * Check if we should update i_disksize
2515 * when write to the end of file but not require block allocation
2517 static int ext4_da_should_update_i_disksize(struct page *page,
2518 unsigned long offset)
2520 struct buffer_head *bh;
2521 struct inode *inode = page->mapping->host;
2525 bh = page_buffers(page);
2526 idx = offset >> inode->i_blkbits;
2528 for (i = 0; i < idx; i++)
2529 bh = bh->b_this_page;
2531 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2536 static int ext4_da_write_end(struct file *file,
2537 struct address_space *mapping,
2538 loff_t pos, unsigned len, unsigned copied,
2539 struct page *page, void *fsdata)
2541 struct inode *inode = mapping->host;
2543 handle_t *handle = ext4_journal_current_handle();
2545 unsigned long start, end;
2546 int write_mode = (int)(unsigned long)fsdata;
2548 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2549 switch (ext4_inode_journal_mode(inode)) {
2550 case EXT4_INODE_ORDERED_DATA_MODE:
2551 return ext4_ordered_write_end(file, mapping, pos,
2552 len, copied, page, fsdata);
2553 case EXT4_INODE_WRITEBACK_DATA_MODE:
2554 return ext4_writeback_write_end(file, mapping, pos,
2555 len, copied, page, fsdata);
2561 trace_ext4_da_write_end(inode, pos, len, copied);
2562 start = pos & (PAGE_CACHE_SIZE - 1);
2563 end = start + copied - 1;
2566 * generic_write_end() will run mark_inode_dirty() if i_size
2567 * changes. So let's piggyback the i_disksize mark_inode_dirty
2570 new_i_size = pos + copied;
2571 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2572 if (ext4_has_inline_data(inode) ||
2573 ext4_da_should_update_i_disksize(page, end)) {
2574 down_write(&EXT4_I(inode)->i_data_sem);
2575 if (new_i_size > EXT4_I(inode)->i_disksize)
2576 EXT4_I(inode)->i_disksize = new_i_size;
2577 up_write(&EXT4_I(inode)->i_data_sem);
2578 /* We need to mark inode dirty even if
2579 * new_i_size is less that inode->i_size
2580 * bu greater than i_disksize.(hint delalloc)
2582 ext4_mark_inode_dirty(handle, inode);
2586 if (write_mode != CONVERT_INLINE_DATA &&
2587 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2588 ext4_has_inline_data(inode))
2589 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2592 ret2 = generic_write_end(file, mapping, pos, len, copied,
2598 ret2 = ext4_journal_stop(handle);
2602 return ret ? ret : copied;
2605 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2608 * Drop reserved blocks
2610 BUG_ON(!PageLocked(page));
2611 if (!page_has_buffers(page))
2614 ext4_da_page_release_reservation(page, offset);
2617 ext4_invalidatepage(page, offset);
2623 * Force all delayed allocation blocks to be allocated for a given inode.
2625 int ext4_alloc_da_blocks(struct inode *inode)
2627 trace_ext4_alloc_da_blocks(inode);
2629 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2630 !EXT4_I(inode)->i_reserved_meta_blocks)
2634 * We do something simple for now. The filemap_flush() will
2635 * also start triggering a write of the data blocks, which is
2636 * not strictly speaking necessary (and for users of
2637 * laptop_mode, not even desirable). However, to do otherwise
2638 * would require replicating code paths in:
2640 * ext4_da_writepages() ->
2641 * write_cache_pages() ---> (via passed in callback function)
2642 * __mpage_da_writepage() -->
2643 * mpage_add_bh_to_extent()
2644 * mpage_da_map_blocks()
2646 * The problem is that write_cache_pages(), located in
2647 * mm/page-writeback.c, marks pages clean in preparation for
2648 * doing I/O, which is not desirable if we're not planning on
2651 * We could call write_cache_pages(), and then redirty all of
2652 * the pages by calling redirty_page_for_writepage() but that
2653 * would be ugly in the extreme. So instead we would need to
2654 * replicate parts of the code in the above functions,
2655 * simplifying them because we wouldn't actually intend to
2656 * write out the pages, but rather only collect contiguous
2657 * logical block extents, call the multi-block allocator, and
2658 * then update the buffer heads with the block allocations.
2660 * For now, though, we'll cheat by calling filemap_flush(),
2661 * which will map the blocks, and start the I/O, but not
2662 * actually wait for the I/O to complete.
2664 return filemap_flush(inode->i_mapping);
2668 * bmap() is special. It gets used by applications such as lilo and by
2669 * the swapper to find the on-disk block of a specific piece of data.
2671 * Naturally, this is dangerous if the block concerned is still in the
2672 * journal. If somebody makes a swapfile on an ext4 data-journaling
2673 * filesystem and enables swap, then they may get a nasty shock when the
2674 * data getting swapped to that swapfile suddenly gets overwritten by
2675 * the original zero's written out previously to the journal and
2676 * awaiting writeback in the kernel's buffer cache.
2678 * So, if we see any bmap calls here on a modified, data-journaled file,
2679 * take extra steps to flush any blocks which might be in the cache.
2681 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2683 struct inode *inode = mapping->host;
2688 * We can get here for an inline file via the FIBMAP ioctl
2690 if (ext4_has_inline_data(inode))
2693 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2694 test_opt(inode->i_sb, DELALLOC)) {
2696 * With delalloc we want to sync the file
2697 * so that we can make sure we allocate
2700 filemap_write_and_wait(mapping);
2703 if (EXT4_JOURNAL(inode) &&
2704 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2706 * This is a REALLY heavyweight approach, but the use of
2707 * bmap on dirty files is expected to be extremely rare:
2708 * only if we run lilo or swapon on a freshly made file
2709 * do we expect this to happen.
2711 * (bmap requires CAP_SYS_RAWIO so this does not
2712 * represent an unprivileged user DOS attack --- we'd be
2713 * in trouble if mortal users could trigger this path at
2716 * NB. EXT4_STATE_JDATA is not set on files other than
2717 * regular files. If somebody wants to bmap a directory
2718 * or symlink and gets confused because the buffer
2719 * hasn't yet been flushed to disk, they deserve
2720 * everything they get.
2723 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2724 journal = EXT4_JOURNAL(inode);
2725 jbd2_journal_lock_updates(journal);
2726 err = jbd2_journal_flush(journal);
2727 jbd2_journal_unlock_updates(journal);
2733 return generic_block_bmap(mapping, block, ext4_get_block);
2736 static int ext4_readpage(struct file *file, struct page *page)
2739 struct inode *inode = page->mapping->host;
2741 trace_ext4_readpage(page);
2743 if (ext4_has_inline_data(inode))
2744 ret = ext4_readpage_inline(inode, page);
2747 return mpage_readpage(page, ext4_get_block);
2753 ext4_readpages(struct file *file, struct address_space *mapping,
2754 struct list_head *pages, unsigned nr_pages)
2756 struct inode *inode = mapping->host;
2758 /* If the file has inline data, no need to do readpages. */
2759 if (ext4_has_inline_data(inode))
2762 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2765 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2767 trace_ext4_invalidatepage(page, offset);
2769 /* No journalling happens on data buffers when this function is used */
2770 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2772 block_invalidatepage(page, offset);
2775 static int __ext4_journalled_invalidatepage(struct page *page,
2776 unsigned long offset)
2778 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2780 trace_ext4_journalled_invalidatepage(page, offset);
2783 * If it's a full truncate we just forget about the pending dirtying
2786 ClearPageChecked(page);
2788 return jbd2_journal_invalidatepage(journal, page, offset);
2791 /* Wrapper for aops... */
2792 static void ext4_journalled_invalidatepage(struct page *page,
2793 unsigned long offset)
2795 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
2798 static int ext4_releasepage(struct page *page, gfp_t wait)
2800 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2802 trace_ext4_releasepage(page);
2804 WARN_ON(PageChecked(page));
2805 if (!page_has_buffers(page))
2808 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2810 return try_to_free_buffers(page);
2814 * ext4_get_block used when preparing for a DIO write or buffer write.
2815 * We allocate an uinitialized extent if blocks haven't been allocated.
2816 * The extent will be converted to initialized after the IO is complete.
2818 int ext4_get_block_write(struct inode *inode, sector_t iblock,
2819 struct buffer_head *bh_result, int create)
2821 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2822 inode->i_ino, create);
2823 return _ext4_get_block(inode, iblock, bh_result,
2824 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2827 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2828 struct buffer_head *bh_result, int create)
2830 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2831 inode->i_ino, create);
2832 return _ext4_get_block(inode, iblock, bh_result,
2833 EXT4_GET_BLOCKS_NO_LOCK);
2836 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2837 ssize_t size, void *private, int ret,
2840 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2841 ext4_io_end_t *io_end = iocb->private;
2843 /* if not async direct IO or dio with 0 bytes write, just return */
2844 if (!io_end || !size)
2847 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2848 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2849 iocb->private, io_end->inode->i_ino, iocb, offset,
2852 iocb->private = NULL;
2854 /* if not aio dio with unwritten extents, just free io and return */
2855 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2856 ext4_free_io_end(io_end);
2859 aio_complete(iocb, ret, 0);
2860 inode_dio_done(inode);
2864 io_end->offset = offset;
2865 io_end->size = size;
2867 io_end->iocb = iocb;
2868 io_end->result = ret;
2871 ext4_add_complete_io(io_end);
2875 * For ext4 extent files, ext4 will do direct-io write to holes,
2876 * preallocated extents, and those write extend the file, no need to
2877 * fall back to buffered IO.
2879 * For holes, we fallocate those blocks, mark them as uninitialized
2880 * If those blocks were preallocated, we mark sure they are split, but
2881 * still keep the range to write as uninitialized.
2883 * The unwritten extents will be converted to written when DIO is completed.
2884 * For async direct IO, since the IO may still pending when return, we
2885 * set up an end_io call back function, which will do the conversion
2886 * when async direct IO completed.
2888 * If the O_DIRECT write will extend the file then add this inode to the
2889 * orphan list. So recovery will truncate it back to the original size
2890 * if the machine crashes during the write.
2893 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2894 const struct iovec *iov, loff_t offset,
2895 unsigned long nr_segs)
2897 struct file *file = iocb->ki_filp;
2898 struct inode *inode = file->f_mapping->host;
2900 size_t count = iov_length(iov, nr_segs);
2902 get_block_t *get_block_func = NULL;
2904 loff_t final_size = offset + count;
2906 /* Use the old path for reads and writes beyond i_size. */
2907 if (rw != WRITE || final_size > inode->i_size)
2908 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2910 BUG_ON(iocb->private == NULL);
2912 /* If we do a overwrite dio, i_mutex locking can be released */
2913 overwrite = *((int *)iocb->private);
2916 atomic_inc(&inode->i_dio_count);
2917 down_read(&EXT4_I(inode)->i_data_sem);
2918 mutex_unlock(&inode->i_mutex);
2922 * We could direct write to holes and fallocate.
2924 * Allocated blocks to fill the hole are marked as
2925 * uninitialized to prevent parallel buffered read to expose
2926 * the stale data before DIO complete the data IO.
2928 * As to previously fallocated extents, ext4 get_block will
2929 * just simply mark the buffer mapped but still keep the
2930 * extents uninitialized.
2932 * For non AIO case, we will convert those unwritten extents
2933 * to written after return back from blockdev_direct_IO.
2935 * For async DIO, the conversion needs to be deferred when the
2936 * IO is completed. The ext4 end_io callback function will be
2937 * called to take care of the conversion work. Here for async
2938 * case, we allocate an io_end structure to hook to the iocb.
2940 iocb->private = NULL;
2941 ext4_inode_aio_set(inode, NULL);
2942 if (!is_sync_kiocb(iocb)) {
2943 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
2948 io_end->flag |= EXT4_IO_END_DIRECT;
2949 iocb->private = io_end;
2951 * we save the io structure for current async direct
2952 * IO, so that later ext4_map_blocks() could flag the
2953 * io structure whether there is a unwritten extents
2954 * needs to be converted when IO is completed.
2956 ext4_inode_aio_set(inode, io_end);
2960 get_block_func = ext4_get_block_write_nolock;
2962 get_block_func = ext4_get_block_write;
2963 dio_flags = DIO_LOCKING;
2965 ret = __blockdev_direct_IO(rw, iocb, inode,
2966 inode->i_sb->s_bdev, iov,
2974 ext4_inode_aio_set(inode, NULL);
2976 * The io_end structure takes a reference to the inode, that
2977 * structure needs to be destroyed and the reference to the
2978 * inode need to be dropped, when IO is complete, even with 0
2979 * byte write, or failed.
2981 * In the successful AIO DIO case, the io_end structure will
2982 * be destroyed and the reference to the inode will be dropped
2983 * after the end_io call back function is called.
2985 * In the case there is 0 byte write, or error case, since VFS
2986 * direct IO won't invoke the end_io call back function, we
2987 * need to free the end_io structure here.
2989 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
2990 ext4_free_io_end(iocb->private);
2991 iocb->private = NULL;
2992 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
2993 EXT4_STATE_DIO_UNWRITTEN)) {
2996 * for non AIO case, since the IO is already
2997 * completed, we could do the conversion right here
2999 err = ext4_convert_unwritten_extents(inode,
3003 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3007 /* take i_mutex locking again if we do a ovewrite dio */
3009 inode_dio_done(inode);
3010 up_read(&EXT4_I(inode)->i_data_sem);
3011 mutex_lock(&inode->i_mutex);
3017 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3018 const struct iovec *iov, loff_t offset,
3019 unsigned long nr_segs)
3021 struct file *file = iocb->ki_filp;
3022 struct inode *inode = file->f_mapping->host;
3026 * If we are doing data journalling we don't support O_DIRECT
3028 if (ext4_should_journal_data(inode))
3031 /* Let buffer I/O handle the inline data case. */
3032 if (ext4_has_inline_data(inode))
3035 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3036 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3037 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3039 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3040 trace_ext4_direct_IO_exit(inode, offset,
3041 iov_length(iov, nr_segs), rw, ret);
3046 * Pages can be marked dirty completely asynchronously from ext4's journalling
3047 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3048 * much here because ->set_page_dirty is called under VFS locks. The page is
3049 * not necessarily locked.
3051 * We cannot just dirty the page and leave attached buffers clean, because the
3052 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3053 * or jbddirty because all the journalling code will explode.
3055 * So what we do is to mark the page "pending dirty" and next time writepage
3056 * is called, propagate that into the buffers appropriately.
3058 static int ext4_journalled_set_page_dirty(struct page *page)
3060 SetPageChecked(page);
3061 return __set_page_dirty_nobuffers(page);
3064 static const struct address_space_operations ext4_ordered_aops = {
3065 .readpage = ext4_readpage,
3066 .readpages = ext4_readpages,
3067 .writepage = ext4_writepage,
3068 .write_begin = ext4_write_begin,
3069 .write_end = ext4_ordered_write_end,
3071 .invalidatepage = ext4_invalidatepage,
3072 .releasepage = ext4_releasepage,
3073 .direct_IO = ext4_direct_IO,
3074 .migratepage = buffer_migrate_page,
3075 .is_partially_uptodate = block_is_partially_uptodate,
3076 .error_remove_page = generic_error_remove_page,
3079 static const struct address_space_operations ext4_writeback_aops = {
3080 .readpage = ext4_readpage,
3081 .readpages = ext4_readpages,
3082 .writepage = ext4_writepage,
3083 .write_begin = ext4_write_begin,
3084 .write_end = ext4_writeback_write_end,
3086 .invalidatepage = ext4_invalidatepage,
3087 .releasepage = ext4_releasepage,
3088 .direct_IO = ext4_direct_IO,
3089 .migratepage = buffer_migrate_page,
3090 .is_partially_uptodate = block_is_partially_uptodate,
3091 .error_remove_page = generic_error_remove_page,
3094 static const struct address_space_operations ext4_journalled_aops = {
3095 .readpage = ext4_readpage,
3096 .readpages = ext4_readpages,
3097 .writepage = ext4_writepage,
3098 .write_begin = ext4_write_begin,
3099 .write_end = ext4_journalled_write_end,
3100 .set_page_dirty = ext4_journalled_set_page_dirty,
3102 .invalidatepage = ext4_journalled_invalidatepage,
3103 .releasepage = ext4_releasepage,
3104 .direct_IO = ext4_direct_IO,
3105 .is_partially_uptodate = block_is_partially_uptodate,
3106 .error_remove_page = generic_error_remove_page,
3109 static const struct address_space_operations ext4_da_aops = {
3110 .readpage = ext4_readpage,
3111 .readpages = ext4_readpages,
3112 .writepage = ext4_writepage,
3113 .writepages = ext4_da_writepages,
3114 .write_begin = ext4_da_write_begin,
3115 .write_end = ext4_da_write_end,
3117 .invalidatepage = ext4_da_invalidatepage,
3118 .releasepage = ext4_releasepage,
3119 .direct_IO = ext4_direct_IO,
3120 .migratepage = buffer_migrate_page,
3121 .is_partially_uptodate = block_is_partially_uptodate,
3122 .error_remove_page = generic_error_remove_page,
3125 void ext4_set_aops(struct inode *inode)
3127 switch (ext4_inode_journal_mode(inode)) {
3128 case EXT4_INODE_ORDERED_DATA_MODE:
3129 if (test_opt(inode->i_sb, DELALLOC))
3130 inode->i_mapping->a_ops = &ext4_da_aops;
3132 inode->i_mapping->a_ops = &ext4_ordered_aops;
3134 case EXT4_INODE_WRITEBACK_DATA_MODE:
3135 if (test_opt(inode->i_sb, DELALLOC))
3136 inode->i_mapping->a_ops = &ext4_da_aops;
3138 inode->i_mapping->a_ops = &ext4_writeback_aops;
3140 case EXT4_INODE_JOURNAL_DATA_MODE:
3141 inode->i_mapping->a_ops = &ext4_journalled_aops;
3150 * ext4_discard_partial_page_buffers()
3151 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3152 * This function finds and locks the page containing the offset
3153 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3154 * Calling functions that already have the page locked should call
3155 * ext4_discard_partial_page_buffers_no_lock directly.
3157 int ext4_discard_partial_page_buffers(handle_t *handle,
3158 struct address_space *mapping, loff_t from,
3159 loff_t length, int flags)
3161 struct inode *inode = mapping->host;
3165 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3166 mapping_gfp_mask(mapping) & ~__GFP_FS);
3170 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3171 from, length, flags);
3174 page_cache_release(page);
3179 * ext4_discard_partial_page_buffers_no_lock()
3180 * Zeros a page range of length 'length' starting from offset 'from'.
3181 * Buffer heads that correspond to the block aligned regions of the
3182 * zeroed range will be unmapped. Unblock aligned regions
3183 * will have the corresponding buffer head mapped if needed so that
3184 * that region of the page can be updated with the partial zero out.
3186 * This function assumes that the page has already been locked. The
3187 * The range to be discarded must be contained with in the given page.
3188 * If the specified range exceeds the end of the page it will be shortened
3189 * to the end of the page that corresponds to 'from'. This function is
3190 * appropriate for updating a page and it buffer heads to be unmapped and
3191 * zeroed for blocks that have been either released, or are going to be
3194 * handle: The journal handle
3195 * inode: The files inode
3196 * page: A locked page that contains the offset "from"
3197 * from: The starting byte offset (from the beginning of the file)
3198 * to begin discarding
3199 * len: The length of bytes to discard
3200 * flags: Optional flags that may be used:
3202 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3203 * Only zero the regions of the page whose buffer heads
3204 * have already been unmapped. This flag is appropriate
3205 * for updating the contents of a page whose blocks may
3206 * have already been released, and we only want to zero
3207 * out the regions that correspond to those released blocks.
3209 * Returns zero on success or negative on failure.
3211 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3212 struct inode *inode, struct page *page, loff_t from,
3213 loff_t length, int flags)
3215 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3216 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3217 unsigned int blocksize, max, pos;
3219 struct buffer_head *bh;
3222 blocksize = inode->i_sb->s_blocksize;
3223 max = PAGE_CACHE_SIZE - offset;
3225 if (index != page->index)
3229 * correct length if it does not fall between
3230 * 'from' and the end of the page
3232 if (length > max || length < 0)
3235 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3237 if (!page_has_buffers(page))
3238 create_empty_buffers(page, blocksize, 0);
3240 /* Find the buffer that contains "offset" */
3241 bh = page_buffers(page);
3243 while (offset >= pos) {
3244 bh = bh->b_this_page;
3250 while (pos < offset + length) {
3251 unsigned int end_of_block, range_to_discard;
3255 /* The length of space left to zero and unmap */
3256 range_to_discard = offset + length - pos;
3258 /* The length of space until the end of the block */
3259 end_of_block = blocksize - (pos & (blocksize-1));
3262 * Do not unmap or zero past end of block
3263 * for this buffer head
3265 if (range_to_discard > end_of_block)
3266 range_to_discard = end_of_block;
3270 * Skip this buffer head if we are only zeroing unampped
3271 * regions of the page
3273 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3277 /* If the range is block aligned, unmap */
3278 if (range_to_discard == blocksize) {
3279 clear_buffer_dirty(bh);
3281 clear_buffer_mapped(bh);
3282 clear_buffer_req(bh);
3283 clear_buffer_new(bh);
3284 clear_buffer_delay(bh);
3285 clear_buffer_unwritten(bh);
3286 clear_buffer_uptodate(bh);
3287 zero_user(page, pos, range_to_discard);
3288 BUFFER_TRACE(bh, "Buffer discarded");
3293 * If this block is not completely contained in the range
3294 * to be discarded, then it is not going to be released. Because
3295 * we need to keep this block, we need to make sure this part
3296 * of the page is uptodate before we modify it by writeing
3297 * partial zeros on it.
3299 if (!buffer_mapped(bh)) {
3301 * Buffer head must be mapped before we can read
3304 BUFFER_TRACE(bh, "unmapped");
3305 ext4_get_block(inode, iblock, bh, 0);
3306 /* unmapped? It's a hole - nothing to do */
3307 if (!buffer_mapped(bh)) {
3308 BUFFER_TRACE(bh, "still unmapped");
3313 /* Ok, it's mapped. Make sure it's up-to-date */
3314 if (PageUptodate(page))
3315 set_buffer_uptodate(bh);
3317 if (!buffer_uptodate(bh)) {
3319 ll_rw_block(READ, 1, &bh);
3321 /* Uhhuh. Read error. Complain and punt.*/
3322 if (!buffer_uptodate(bh))
3326 if (ext4_should_journal_data(inode)) {
3327 BUFFER_TRACE(bh, "get write access");
3328 err = ext4_journal_get_write_access(handle, bh);
3333 zero_user(page, pos, range_to_discard);
3336 if (ext4_should_journal_data(inode)) {
3337 err = ext4_handle_dirty_metadata(handle, inode, bh);
3339 mark_buffer_dirty(bh);
3341 BUFFER_TRACE(bh, "Partial buffer zeroed");
3343 bh = bh->b_this_page;
3345 pos += range_to_discard;
3351 int ext4_can_truncate(struct inode *inode)
3353 if (S_ISREG(inode->i_mode))
3355 if (S_ISDIR(inode->i_mode))
3357 if (S_ISLNK(inode->i_mode))
3358 return !ext4_inode_is_fast_symlink(inode);
3363 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3364 * associated with the given offset and length
3366 * @inode: File inode
3367 * @offset: The offset where the hole will begin
3368 * @len: The length of the hole
3370 * Returns: 0 on success or negative on failure
3373 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3375 struct inode *inode = file->f_path.dentry->d_inode;
3376 if (!S_ISREG(inode->i_mode))
3379 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3380 return ext4_ind_punch_hole(file, offset, length);
3382 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3383 /* TODO: Add support for bigalloc file systems */
3387 trace_ext4_punch_hole(inode, offset, length);
3389 return ext4_ext_punch_hole(file, offset, length);
3395 * We block out ext4_get_block() block instantiations across the entire
3396 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3397 * simultaneously on behalf of the same inode.
3399 * As we work through the truncate and commit bits of it to the journal there
3400 * is one core, guiding principle: the file's tree must always be consistent on
3401 * disk. We must be able to restart the truncate after a crash.
3403 * The file's tree may be transiently inconsistent in memory (although it
3404 * probably isn't), but whenever we close off and commit a journal transaction,
3405 * the contents of (the filesystem + the journal) must be consistent and
3406 * restartable. It's pretty simple, really: bottom up, right to left (although
3407 * left-to-right works OK too).
3409 * Note that at recovery time, journal replay occurs *before* the restart of
3410 * truncate against the orphan inode list.
3412 * The committed inode has the new, desired i_size (which is the same as
3413 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3414 * that this inode's truncate did not complete and it will again call
3415 * ext4_truncate() to have another go. So there will be instantiated blocks
3416 * to the right of the truncation point in a crashed ext4 filesystem. But
3417 * that's fine - as long as they are linked from the inode, the post-crash
3418 * ext4_truncate() run will find them and release them.
3420 void ext4_truncate(struct inode *inode)
3422 trace_ext4_truncate_enter(inode);
3424 if (!ext4_can_truncate(inode))
3427 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3429 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3430 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3432 if (ext4_has_inline_data(inode)) {
3435 ext4_inline_data_truncate(inode, &has_inline);
3440 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3441 ext4_ext_truncate(inode);
3443 ext4_ind_truncate(inode);
3445 trace_ext4_truncate_exit(inode);
3449 * ext4_get_inode_loc returns with an extra refcount against the inode's
3450 * underlying buffer_head on success. If 'in_mem' is true, we have all
3451 * data in memory that is needed to recreate the on-disk version of this
3454 static int __ext4_get_inode_loc(struct inode *inode,
3455 struct ext4_iloc *iloc, int in_mem)
3457 struct ext4_group_desc *gdp;
3458 struct buffer_head *bh;
3459 struct super_block *sb = inode->i_sb;
3461 int inodes_per_block, inode_offset;
3464 if (!ext4_valid_inum(sb, inode->i_ino))
3467 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3468 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3473 * Figure out the offset within the block group inode table
3475 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3476 inode_offset = ((inode->i_ino - 1) %
3477 EXT4_INODES_PER_GROUP(sb));
3478 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3479 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3481 bh = sb_getblk(sb, block);
3484 if (!buffer_uptodate(bh)) {
3488 * If the buffer has the write error flag, we have failed
3489 * to write out another inode in the same block. In this
3490 * case, we don't have to read the block because we may
3491 * read the old inode data successfully.
3493 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3494 set_buffer_uptodate(bh);
3496 if (buffer_uptodate(bh)) {
3497 /* someone brought it uptodate while we waited */
3503 * If we have all information of the inode in memory and this
3504 * is the only valid inode in the block, we need not read the
3508 struct buffer_head *bitmap_bh;
3511 start = inode_offset & ~(inodes_per_block - 1);
3513 /* Is the inode bitmap in cache? */
3514 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3515 if (unlikely(!bitmap_bh))
3519 * If the inode bitmap isn't in cache then the
3520 * optimisation may end up performing two reads instead
3521 * of one, so skip it.
3523 if (!buffer_uptodate(bitmap_bh)) {
3527 for (i = start; i < start + inodes_per_block; i++) {
3528 if (i == inode_offset)
3530 if (ext4_test_bit(i, bitmap_bh->b_data))
3534 if (i == start + inodes_per_block) {
3535 /* all other inodes are free, so skip I/O */
3536 memset(bh->b_data, 0, bh->b_size);
3537 set_buffer_uptodate(bh);
3545 * If we need to do any I/O, try to pre-readahead extra
3546 * blocks from the inode table.
3548 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3549 ext4_fsblk_t b, end, table;
3552 table = ext4_inode_table(sb, gdp);
3553 /* s_inode_readahead_blks is always a power of 2 */
3554 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3557 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3558 num = EXT4_INODES_PER_GROUP(sb);
3559 if (ext4_has_group_desc_csum(sb))
3560 num -= ext4_itable_unused_count(sb, gdp);
3561 table += num / inodes_per_block;
3565 sb_breadahead(sb, b++);
3569 * There are other valid inodes in the buffer, this inode
3570 * has in-inode xattrs, or we don't have this inode in memory.
3571 * Read the block from disk.
3573 trace_ext4_load_inode(inode);
3575 bh->b_end_io = end_buffer_read_sync;
3576 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3578 if (!buffer_uptodate(bh)) {
3579 EXT4_ERROR_INODE_BLOCK(inode, block,
3580 "unable to read itable block");
3590 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3592 /* We have all inode data except xattrs in memory here. */
3593 return __ext4_get_inode_loc(inode, iloc,
3594 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3597 void ext4_set_inode_flags(struct inode *inode)
3599 unsigned int flags = EXT4_I(inode)->i_flags;
3601 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3602 if (flags & EXT4_SYNC_FL)
3603 inode->i_flags |= S_SYNC;
3604 if (flags & EXT4_APPEND_FL)
3605 inode->i_flags |= S_APPEND;
3606 if (flags & EXT4_IMMUTABLE_FL)
3607 inode->i_flags |= S_IMMUTABLE;
3608 if (flags & EXT4_NOATIME_FL)
3609 inode->i_flags |= S_NOATIME;
3610 if (flags & EXT4_DIRSYNC_FL)
3611 inode->i_flags |= S_DIRSYNC;
3614 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3615 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3617 unsigned int vfs_fl;
3618 unsigned long old_fl, new_fl;
3621 vfs_fl = ei->vfs_inode.i_flags;
3622 old_fl = ei->i_flags;
3623 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3624 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3626 if (vfs_fl & S_SYNC)
3627 new_fl |= EXT4_SYNC_FL;
3628 if (vfs_fl & S_APPEND)
3629 new_fl |= EXT4_APPEND_FL;
3630 if (vfs_fl & S_IMMUTABLE)
3631 new_fl |= EXT4_IMMUTABLE_FL;
3632 if (vfs_fl & S_NOATIME)
3633 new_fl |= EXT4_NOATIME_FL;
3634 if (vfs_fl & S_DIRSYNC)
3635 new_fl |= EXT4_DIRSYNC_FL;
3636 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3639 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3640 struct ext4_inode_info *ei)
3643 struct inode *inode = &(ei->vfs_inode);
3644 struct super_block *sb = inode->i_sb;
3646 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3647 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3648 /* we are using combined 48 bit field */
3649 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3650 le32_to_cpu(raw_inode->i_blocks_lo);
3651 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3652 /* i_blocks represent file system block size */
3653 return i_blocks << (inode->i_blkbits - 9);
3658 return le32_to_cpu(raw_inode->i_blocks_lo);
3662 static inline void ext4_iget_extra_inode(struct inode *inode,
3663 struct ext4_inode *raw_inode,
3664 struct ext4_inode_info *ei)
3666 __le32 *magic = (void *)raw_inode +
3667 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3668 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3669 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3670 ext4_find_inline_data_nolock(inode);
3672 EXT4_I(inode)->i_inline_off = 0;
3675 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3677 struct ext4_iloc iloc;
3678 struct ext4_inode *raw_inode;
3679 struct ext4_inode_info *ei;
3680 struct inode *inode;
3681 journal_t *journal = EXT4_SB(sb)->s_journal;
3687 inode = iget_locked(sb, ino);
3689 return ERR_PTR(-ENOMEM);
3690 if (!(inode->i_state & I_NEW))
3696 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3699 raw_inode = ext4_raw_inode(&iloc);
3701 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3702 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3703 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3704 EXT4_INODE_SIZE(inode->i_sb)) {
3705 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3706 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3707 EXT4_INODE_SIZE(inode->i_sb));
3712 ei->i_extra_isize = 0;
3714 /* Precompute checksum seed for inode metadata */
3715 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3716 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3717 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3719 __le32 inum = cpu_to_le32(inode->i_ino);
3720 __le32 gen = raw_inode->i_generation;
3721 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3723 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3727 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3728 EXT4_ERROR_INODE(inode, "checksum invalid");
3733 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3734 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3735 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3736 if (!(test_opt(inode->i_sb, NO_UID32))) {
3737 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3738 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3740 i_uid_write(inode, i_uid);
3741 i_gid_write(inode, i_gid);
3742 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3744 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3745 ei->i_inline_off = 0;
3746 ei->i_dir_start_lookup = 0;
3747 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3748 /* We now have enough fields to check if the inode was active or not.
3749 * This is needed because nfsd might try to access dead inodes
3750 * the test is that same one that e2fsck uses
3751 * NeilBrown 1999oct15
3753 if (inode->i_nlink == 0) {
3754 if (inode->i_mode == 0 ||
3755 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3756 /* this inode is deleted */
3760 /* The only unlinked inodes we let through here have
3761 * valid i_mode and are being read by the orphan
3762 * recovery code: that's fine, we're about to complete
3763 * the process of deleting those. */
3765 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3766 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3767 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3768 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3770 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3771 inode->i_size = ext4_isize(raw_inode);
3772 ei->i_disksize = inode->i_size;
3774 ei->i_reserved_quota = 0;
3776 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3777 ei->i_block_group = iloc.block_group;
3778 ei->i_last_alloc_group = ~0;
3780 * NOTE! The in-memory inode i_data array is in little-endian order
3781 * even on big-endian machines: we do NOT byteswap the block numbers!
3783 for (block = 0; block < EXT4_N_BLOCKS; block++)
3784 ei->i_data[block] = raw_inode->i_block[block];
3785 INIT_LIST_HEAD(&ei->i_orphan);
3788 * Set transaction id's of transactions that have to be committed
3789 * to finish f[data]sync. We set them to currently running transaction
3790 * as we cannot be sure that the inode or some of its metadata isn't
3791 * part of the transaction - the inode could have been reclaimed and
3792 * now it is reread from disk.
3795 transaction_t *transaction;
3798 read_lock(&journal->j_state_lock);
3799 if (journal->j_running_transaction)
3800 transaction = journal->j_running_transaction;
3802 transaction = journal->j_committing_transaction;
3804 tid = transaction->t_tid;
3806 tid = journal->j_commit_sequence;
3807 read_unlock(&journal->j_state_lock);
3808 ei->i_sync_tid = tid;
3809 ei->i_datasync_tid = tid;
3812 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3813 if (ei->i_extra_isize == 0) {
3814 /* The extra space is currently unused. Use it. */
3815 ei->i_extra_isize = sizeof(struct ext4_inode) -
3816 EXT4_GOOD_OLD_INODE_SIZE;
3818 ext4_iget_extra_inode(inode, raw_inode, ei);
3822 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3823 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3824 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3825 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3827 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3828 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3829 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3831 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3835 if (ei->i_file_acl &&
3836 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3837 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3841 } else if (!ext4_has_inline_data(inode)) {
3842 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3843 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3844 (S_ISLNK(inode->i_mode) &&
3845 !ext4_inode_is_fast_symlink(inode))))
3846 /* Validate extent which is part of inode */
3847 ret = ext4_ext_check_inode(inode);
3848 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3849 (S_ISLNK(inode->i_mode) &&
3850 !ext4_inode_is_fast_symlink(inode))) {
3851 /* Validate block references which are part of inode */
3852 ret = ext4_ind_check_inode(inode);
3858 if (S_ISREG(inode->i_mode)) {
3859 inode->i_op = &ext4_file_inode_operations;
3860 inode->i_fop = &ext4_file_operations;
3861 ext4_set_aops(inode);
3862 } else if (S_ISDIR(inode->i_mode)) {
3863 inode->i_op = &ext4_dir_inode_operations;
3864 inode->i_fop = &ext4_dir_operations;
3865 } else if (S_ISLNK(inode->i_mode)) {
3866 if (ext4_inode_is_fast_symlink(inode)) {
3867 inode->i_op = &ext4_fast_symlink_inode_operations;
3868 nd_terminate_link(ei->i_data, inode->i_size,
3869 sizeof(ei->i_data) - 1);
3871 inode->i_op = &ext4_symlink_inode_operations;
3872 ext4_set_aops(inode);
3874 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3875 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3876 inode->i_op = &ext4_special_inode_operations;
3877 if (raw_inode->i_block[0])
3878 init_special_inode(inode, inode->i_mode,
3879 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3881 init_special_inode(inode, inode->i_mode,
3882 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3885 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3889 ext4_set_inode_flags(inode);
3890 unlock_new_inode(inode);
3896 return ERR_PTR(ret);
3899 static int ext4_inode_blocks_set(handle_t *handle,
3900 struct ext4_inode *raw_inode,
3901 struct ext4_inode_info *ei)
3903 struct inode *inode = &(ei->vfs_inode);
3904 u64 i_blocks = inode->i_blocks;
3905 struct super_block *sb = inode->i_sb;
3907 if (i_blocks <= ~0U) {
3909 * i_blocks can be represented in a 32 bit variable
3910 * as multiple of 512 bytes
3912 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3913 raw_inode->i_blocks_high = 0;
3914 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3917 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3920 if (i_blocks <= 0xffffffffffffULL) {
3922 * i_blocks can be represented in a 48 bit variable
3923 * as multiple of 512 bytes
3925 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3926 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3927 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3929 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3930 /* i_block is stored in file system block size */
3931 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3932 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3933 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3939 * Post the struct inode info into an on-disk inode location in the
3940 * buffer-cache. This gobbles the caller's reference to the
3941 * buffer_head in the inode location struct.
3943 * The caller must have write access to iloc->bh.
3945 static int ext4_do_update_inode(handle_t *handle,
3946 struct inode *inode,
3947 struct ext4_iloc *iloc)
3949 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3950 struct ext4_inode_info *ei = EXT4_I(inode);
3951 struct buffer_head *bh = iloc->bh;
3952 int err = 0, rc, block;
3953 int need_datasync = 0;
3957 /* For fields not not tracking in the in-memory inode,
3958 * initialise them to zero for new inodes. */
3959 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
3960 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3962 ext4_get_inode_flags(ei);
3963 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3964 i_uid = i_uid_read(inode);
3965 i_gid = i_gid_read(inode);
3966 if (!(test_opt(inode->i_sb, NO_UID32))) {
3967 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3968 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
3970 * Fix up interoperability with old kernels. Otherwise, old inodes get
3971 * re-used with the upper 16 bits of the uid/gid intact
3974 raw_inode->i_uid_high =
3975 cpu_to_le16(high_16_bits(i_uid));
3976 raw_inode->i_gid_high =
3977 cpu_to_le16(high_16_bits(i_gid));
3979 raw_inode->i_uid_high = 0;
3980 raw_inode->i_gid_high = 0;
3983 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
3984 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
3985 raw_inode->i_uid_high = 0;
3986 raw_inode->i_gid_high = 0;
3988 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3990 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
3991 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
3992 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
3993 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
3995 if (ext4_inode_blocks_set(handle, raw_inode, ei))
3997 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3998 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
3999 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4000 cpu_to_le32(EXT4_OS_HURD))
4001 raw_inode->i_file_acl_high =
4002 cpu_to_le16(ei->i_file_acl >> 32);
4003 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4004 if (ei->i_disksize != ext4_isize(raw_inode)) {
4005 ext4_isize_set(raw_inode, ei->i_disksize);
4008 if (ei->i_disksize > 0x7fffffffULL) {
4009 struct super_block *sb = inode->i_sb;
4010 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4011 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4012 EXT4_SB(sb)->s_es->s_rev_level ==
4013 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4014 /* If this is the first large file
4015 * created, add a flag to the superblock.
4017 err = ext4_journal_get_write_access(handle,
4018 EXT4_SB(sb)->s_sbh);
4021 ext4_update_dynamic_rev(sb);
4022 EXT4_SET_RO_COMPAT_FEATURE(sb,
4023 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4024 ext4_handle_sync(handle);
4025 err = ext4_handle_dirty_super(handle, sb);
4028 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4029 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4030 if (old_valid_dev(inode->i_rdev)) {
4031 raw_inode->i_block[0] =
4032 cpu_to_le32(old_encode_dev(inode->i_rdev));
4033 raw_inode->i_block[1] = 0;
4035 raw_inode->i_block[0] = 0;
4036 raw_inode->i_block[1] =
4037 cpu_to_le32(new_encode_dev(inode->i_rdev));
4038 raw_inode->i_block[2] = 0;
4040 } else if (!ext4_has_inline_data(inode)) {
4041 for (block = 0; block < EXT4_N_BLOCKS; block++)
4042 raw_inode->i_block[block] = ei->i_data[block];
4045 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4046 if (ei->i_extra_isize) {
4047 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4048 raw_inode->i_version_hi =
4049 cpu_to_le32(inode->i_version >> 32);
4050 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4053 ext4_inode_csum_set(inode, raw_inode, ei);
4055 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4056 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4059 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4061 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4064 ext4_std_error(inode->i_sb, err);
4069 * ext4_write_inode()
4071 * We are called from a few places:
4073 * - Within generic_file_write() for O_SYNC files.
4074 * Here, there will be no transaction running. We wait for any running
4075 * transaction to commit.
4077 * - Within sys_sync(), kupdate and such.
4078 * We wait on commit, if tol to.
4080 * - Within prune_icache() (PF_MEMALLOC == true)
4081 * Here we simply return. We can't afford to block kswapd on the
4084 * In all cases it is actually safe for us to return without doing anything,
4085 * because the inode has been copied into a raw inode buffer in
4086 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4089 * Note that we are absolutely dependent upon all inode dirtiers doing the
4090 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4091 * which we are interested.
4093 * It would be a bug for them to not do this. The code:
4095 * mark_inode_dirty(inode)
4097 * inode->i_size = expr;
4099 * is in error because a kswapd-driven write_inode() could occur while
4100 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4101 * will no longer be on the superblock's dirty inode list.
4103 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4107 if (current->flags & PF_MEMALLOC)
4110 if (EXT4_SB(inode->i_sb)->s_journal) {
4111 if (ext4_journal_current_handle()) {
4112 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4117 if (wbc->sync_mode != WB_SYNC_ALL)
4120 err = ext4_force_commit(inode->i_sb);
4122 struct ext4_iloc iloc;
4124 err = __ext4_get_inode_loc(inode, &iloc, 0);
4127 if (wbc->sync_mode == WB_SYNC_ALL)
4128 sync_dirty_buffer(iloc.bh);
4129 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4130 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4131 "IO error syncing inode");
4140 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4141 * buffers that are attached to a page stradding i_size and are undergoing
4142 * commit. In that case we have to wait for commit to finish and try again.
4144 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4148 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4149 tid_t commit_tid = 0;
4152 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4154 * All buffers in the last page remain valid? Then there's nothing to
4155 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4158 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4161 page = find_lock_page(inode->i_mapping,
4162 inode->i_size >> PAGE_CACHE_SHIFT);
4165 ret = __ext4_journalled_invalidatepage(page, offset);
4167 page_cache_release(page);
4171 read_lock(&journal->j_state_lock);
4172 if (journal->j_committing_transaction)
4173 commit_tid = journal->j_committing_transaction->t_tid;
4174 read_unlock(&journal->j_state_lock);
4176 jbd2_log_wait_commit(journal, commit_tid);
4183 * Called from notify_change.
4185 * We want to trap VFS attempts to truncate the file as soon as
4186 * possible. In particular, we want to make sure that when the VFS
4187 * shrinks i_size, we put the inode on the orphan list and modify
4188 * i_disksize immediately, so that during the subsequent flushing of
4189 * dirty pages and freeing of disk blocks, we can guarantee that any
4190 * commit will leave the blocks being flushed in an unused state on
4191 * disk. (On recovery, the inode will get truncated and the blocks will
4192 * be freed, so we have a strong guarantee that no future commit will
4193 * leave these blocks visible to the user.)
4195 * Another thing we have to assure is that if we are in ordered mode
4196 * and inode is still attached to the committing transaction, we must
4197 * we start writeout of all the dirty pages which are being truncated.
4198 * This way we are sure that all the data written in the previous
4199 * transaction are already on disk (truncate waits for pages under
4202 * Called with inode->i_mutex down.
4204 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4206 struct inode *inode = dentry->d_inode;
4209 const unsigned int ia_valid = attr->ia_valid;
4211 error = inode_change_ok(inode, attr);
4215 if (is_quota_modification(inode, attr))
4216 dquot_initialize(inode);
4217 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4218 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4221 /* (user+group)*(old+new) structure, inode write (sb,
4222 * inode block, ? - but truncate inode update has it) */
4223 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4224 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4225 if (IS_ERR(handle)) {
4226 error = PTR_ERR(handle);
4229 error = dquot_transfer(inode, attr);
4231 ext4_journal_stop(handle);
4234 /* Update corresponding info in inode so that everything is in
4235 * one transaction */
4236 if (attr->ia_valid & ATTR_UID)
4237 inode->i_uid = attr->ia_uid;
4238 if (attr->ia_valid & ATTR_GID)
4239 inode->i_gid = attr->ia_gid;
4240 error = ext4_mark_inode_dirty(handle, inode);
4241 ext4_journal_stop(handle);
4244 if (attr->ia_valid & ATTR_SIZE) {
4246 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4247 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4249 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4254 if (S_ISREG(inode->i_mode) &&
4255 attr->ia_valid & ATTR_SIZE &&
4256 (attr->ia_size < inode->i_size)) {
4259 handle = ext4_journal_start(inode, 3);
4260 if (IS_ERR(handle)) {
4261 error = PTR_ERR(handle);
4264 if (ext4_handle_valid(handle)) {
4265 error = ext4_orphan_add(handle, inode);
4268 EXT4_I(inode)->i_disksize = attr->ia_size;
4269 rc = ext4_mark_inode_dirty(handle, inode);
4272 ext4_journal_stop(handle);
4274 if (ext4_should_order_data(inode)) {
4275 error = ext4_begin_ordered_truncate(inode,
4278 /* Do as much error cleanup as possible */
4279 handle = ext4_journal_start(inode, 3);
4280 if (IS_ERR(handle)) {
4281 ext4_orphan_del(NULL, inode);
4284 ext4_orphan_del(handle, inode);
4286 ext4_journal_stop(handle);
4292 if (attr->ia_valid & ATTR_SIZE) {
4293 if (attr->ia_size != inode->i_size) {
4294 loff_t oldsize = inode->i_size;
4296 i_size_write(inode, attr->ia_size);
4298 * Blocks are going to be removed from the inode. Wait
4299 * for dio in flight. Temporarily disable
4300 * dioread_nolock to prevent livelock.
4303 if (!ext4_should_journal_data(inode)) {
4304 ext4_inode_block_unlocked_dio(inode);
4305 inode_dio_wait(inode);
4306 ext4_inode_resume_unlocked_dio(inode);
4308 ext4_wait_for_tail_page_commit(inode);
4311 * Truncate pagecache after we've waited for commit
4312 * in data=journal mode to make pages freeable.
4314 truncate_pagecache(inode, oldsize, inode->i_size);
4316 ext4_truncate(inode);
4320 setattr_copy(inode, attr);
4321 mark_inode_dirty(inode);
4325 * If the call to ext4_truncate failed to get a transaction handle at
4326 * all, we need to clean up the in-core orphan list manually.
4328 if (orphan && inode->i_nlink)
4329 ext4_orphan_del(NULL, inode);
4331 if (!rc && (ia_valid & ATTR_MODE))
4332 rc = ext4_acl_chmod(inode);
4335 ext4_std_error(inode->i_sb, error);
4341 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4344 struct inode *inode;
4345 unsigned long delalloc_blocks;
4347 inode = dentry->d_inode;
4348 generic_fillattr(inode, stat);
4351 * We can't update i_blocks if the block allocation is delayed
4352 * otherwise in the case of system crash before the real block
4353 * allocation is done, we will have i_blocks inconsistent with
4354 * on-disk file blocks.
4355 * We always keep i_blocks updated together with real
4356 * allocation. But to not confuse with user, stat
4357 * will return the blocks that include the delayed allocation
4358 * blocks for this file.
4360 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4361 EXT4_I(inode)->i_reserved_data_blocks);
4363 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4367 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4369 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4370 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4371 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4375 * Account for index blocks, block groups bitmaps and block group
4376 * descriptor blocks if modify datablocks and index blocks
4377 * worse case, the indexs blocks spread over different block groups
4379 * If datablocks are discontiguous, they are possible to spread over
4380 * different block groups too. If they are contiguous, with flexbg,
4381 * they could still across block group boundary.
4383 * Also account for superblock, inode, quota and xattr blocks
4385 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4387 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4393 * How many index blocks need to touch to modify nrblocks?
4394 * The "Chunk" flag indicating whether the nrblocks is
4395 * physically contiguous on disk
4397 * For Direct IO and fallocate, they calls get_block to allocate
4398 * one single extent at a time, so they could set the "Chunk" flag
4400 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4405 * Now let's see how many group bitmaps and group descriptors need
4415 if (groups > ngroups)
4417 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4418 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4420 /* bitmaps and block group descriptor blocks */
4421 ret += groups + gdpblocks;
4423 /* Blocks for super block, inode, quota and xattr blocks */
4424 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4430 * Calculate the total number of credits to reserve to fit
4431 * the modification of a single pages into a single transaction,
4432 * which may include multiple chunks of block allocations.
4434 * This could be called via ext4_write_begin()
4436 * We need to consider the worse case, when
4437 * one new block per extent.
4439 int ext4_writepage_trans_blocks(struct inode *inode)
4441 int bpp = ext4_journal_blocks_per_page(inode);
4444 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4446 /* Account for data blocks for journalled mode */
4447 if (ext4_should_journal_data(inode))
4453 * Calculate the journal credits for a chunk of data modification.
4455 * This is called from DIO, fallocate or whoever calling
4456 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4458 * journal buffers for data blocks are not included here, as DIO
4459 * and fallocate do no need to journal data buffers.
4461 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4463 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4467 * The caller must have previously called ext4_reserve_inode_write().
4468 * Give this, we know that the caller already has write access to iloc->bh.
4470 int ext4_mark_iloc_dirty(handle_t *handle,
4471 struct inode *inode, struct ext4_iloc *iloc)
4475 if (IS_I_VERSION(inode))
4476 inode_inc_iversion(inode);
4478 /* the do_update_inode consumes one bh->b_count */
4481 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4482 err = ext4_do_update_inode(handle, inode, iloc);
4488 * On success, We end up with an outstanding reference count against
4489 * iloc->bh. This _must_ be cleaned up later.
4493 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4494 struct ext4_iloc *iloc)
4498 err = ext4_get_inode_loc(inode, iloc);
4500 BUFFER_TRACE(iloc->bh, "get_write_access");
4501 err = ext4_journal_get_write_access(handle, iloc->bh);
4507 ext4_std_error(inode->i_sb, err);
4512 * Expand an inode by new_extra_isize bytes.
4513 * Returns 0 on success or negative error number on failure.
4515 static int ext4_expand_extra_isize(struct inode *inode,
4516 unsigned int new_extra_isize,
4517 struct ext4_iloc iloc,
4520 struct ext4_inode *raw_inode;
4521 struct ext4_xattr_ibody_header *header;
4523 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4526 raw_inode = ext4_raw_inode(&iloc);
4528 header = IHDR(inode, raw_inode);
4530 /* No extended attributes present */
4531 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4532 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4533 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4535 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4539 /* try to expand with EAs present */
4540 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4545 * What we do here is to mark the in-core inode as clean with respect to inode
4546 * dirtiness (it may still be data-dirty).
4547 * This means that the in-core inode may be reaped by prune_icache
4548 * without having to perform any I/O. This is a very good thing,
4549 * because *any* task may call prune_icache - even ones which
4550 * have a transaction open against a different journal.
4552 * Is this cheating? Not really. Sure, we haven't written the
4553 * inode out, but prune_icache isn't a user-visible syncing function.
4554 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4555 * we start and wait on commits.
4557 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4559 struct ext4_iloc iloc;
4560 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4561 static unsigned int mnt_count;
4565 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4566 err = ext4_reserve_inode_write(handle, inode, &iloc);
4567 if (ext4_handle_valid(handle) &&
4568 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4569 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4571 * We need extra buffer credits since we may write into EA block
4572 * with this same handle. If journal_extend fails, then it will
4573 * only result in a minor loss of functionality for that inode.
4574 * If this is felt to be critical, then e2fsck should be run to
4575 * force a large enough s_min_extra_isize.
4577 if ((jbd2_journal_extend(handle,
4578 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4579 ret = ext4_expand_extra_isize(inode,
4580 sbi->s_want_extra_isize,
4583 ext4_set_inode_state(inode,
4584 EXT4_STATE_NO_EXPAND);
4586 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4587 ext4_warning(inode->i_sb,
4588 "Unable to expand inode %lu. Delete"
4589 " some EAs or run e2fsck.",
4592 le16_to_cpu(sbi->s_es->s_mnt_count);
4598 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4603 * ext4_dirty_inode() is called from __mark_inode_dirty()
4605 * We're really interested in the case where a file is being extended.
4606 * i_size has been changed by generic_commit_write() and we thus need
4607 * to include the updated inode in the current transaction.
4609 * Also, dquot_alloc_block() will always dirty the inode when blocks
4610 * are allocated to the file.
4612 * If the inode is marked synchronous, we don't honour that here - doing
4613 * so would cause a commit on atime updates, which we don't bother doing.
4614 * We handle synchronous inodes at the highest possible level.
4616 void ext4_dirty_inode(struct inode *inode, int flags)
4620 handle = ext4_journal_start(inode, 2);
4624 ext4_mark_inode_dirty(handle, inode);
4626 ext4_journal_stop(handle);
4633 * Bind an inode's backing buffer_head into this transaction, to prevent
4634 * it from being flushed to disk early. Unlike
4635 * ext4_reserve_inode_write, this leaves behind no bh reference and
4636 * returns no iloc structure, so the caller needs to repeat the iloc
4637 * lookup to mark the inode dirty later.
4639 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4641 struct ext4_iloc iloc;
4645 err = ext4_get_inode_loc(inode, &iloc);
4647 BUFFER_TRACE(iloc.bh, "get_write_access");
4648 err = jbd2_journal_get_write_access(handle, iloc.bh);
4650 err = ext4_handle_dirty_metadata(handle,
4656 ext4_std_error(inode->i_sb, err);
4661 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4668 * We have to be very careful here: changing a data block's
4669 * journaling status dynamically is dangerous. If we write a
4670 * data block to the journal, change the status and then delete
4671 * that block, we risk forgetting to revoke the old log record
4672 * from the journal and so a subsequent replay can corrupt data.
4673 * So, first we make sure that the journal is empty and that
4674 * nobody is changing anything.
4677 journal = EXT4_JOURNAL(inode);
4680 if (is_journal_aborted(journal))
4682 /* We have to allocate physical blocks for delalloc blocks
4683 * before flushing journal. otherwise delalloc blocks can not
4684 * be allocated any more. even more truncate on delalloc blocks
4685 * could trigger BUG by flushing delalloc blocks in journal.
4686 * There is no delalloc block in non-journal data mode.
4688 if (val && test_opt(inode->i_sb, DELALLOC)) {
4689 err = ext4_alloc_da_blocks(inode);
4694 /* Wait for all existing dio workers */
4695 ext4_inode_block_unlocked_dio(inode);
4696 inode_dio_wait(inode);
4698 jbd2_journal_lock_updates(journal);
4701 * OK, there are no updates running now, and all cached data is
4702 * synced to disk. We are now in a completely consistent state
4703 * which doesn't have anything in the journal, and we know that
4704 * no filesystem updates are running, so it is safe to modify
4705 * the inode's in-core data-journaling state flag now.
4709 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4711 jbd2_journal_flush(journal);
4712 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4714 ext4_set_aops(inode);
4716 jbd2_journal_unlock_updates(journal);
4717 ext4_inode_resume_unlocked_dio(inode);
4719 /* Finally we can mark the inode as dirty. */
4721 handle = ext4_journal_start(inode, 1);
4723 return PTR_ERR(handle);
4725 err = ext4_mark_inode_dirty(handle, inode);
4726 ext4_handle_sync(handle);
4727 ext4_journal_stop(handle);
4728 ext4_std_error(inode->i_sb, err);
4733 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4735 return !buffer_mapped(bh);
4738 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4740 struct page *page = vmf->page;
4744 struct file *file = vma->vm_file;
4745 struct inode *inode = file->f_path.dentry->d_inode;
4746 struct address_space *mapping = inode->i_mapping;
4748 get_block_t *get_block;
4751 sb_start_pagefault(inode->i_sb);
4752 file_update_time(vma->vm_file);
4753 /* Delalloc case is easy... */
4754 if (test_opt(inode->i_sb, DELALLOC) &&
4755 !ext4_should_journal_data(inode) &&
4756 !ext4_nonda_switch(inode->i_sb)) {
4758 ret = __block_page_mkwrite(vma, vmf,
4759 ext4_da_get_block_prep);
4760 } while (ret == -ENOSPC &&
4761 ext4_should_retry_alloc(inode->i_sb, &retries));
4766 size = i_size_read(inode);
4767 /* Page got truncated from under us? */
4768 if (page->mapping != mapping || page_offset(page) > size) {
4770 ret = VM_FAULT_NOPAGE;
4774 if (page->index == size >> PAGE_CACHE_SHIFT)
4775 len = size & ~PAGE_CACHE_MASK;
4777 len = PAGE_CACHE_SIZE;
4779 * Return if we have all the buffers mapped. This avoids the need to do
4780 * journal_start/journal_stop which can block and take a long time
4782 if (page_has_buffers(page)) {
4783 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
4785 ext4_bh_unmapped)) {
4786 /* Wait so that we don't change page under IO */
4787 wait_on_page_writeback(page);
4788 ret = VM_FAULT_LOCKED;
4793 /* OK, we need to fill the hole... */
4794 if (ext4_should_dioread_nolock(inode))
4795 get_block = ext4_get_block_write;
4797 get_block = ext4_get_block;
4799 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4800 if (IS_ERR(handle)) {
4801 ret = VM_FAULT_SIGBUS;
4804 ret = __block_page_mkwrite(vma, vmf, get_block);
4805 if (!ret && ext4_should_journal_data(inode)) {
4806 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
4807 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4809 ret = VM_FAULT_SIGBUS;
4810 ext4_journal_stop(handle);
4813 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4815 ext4_journal_stop(handle);
4816 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4819 ret = block_page_mkwrite_return(ret);
4821 sb_end_pagefault(inode->i_sb);