2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
96 return provided == calculated;
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 trace_ext4_begin_ordered_truncate(inode, new_size);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode)->jinode)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
134 static void ext4_invalidatepage(struct page *page, unsigned long offset);
135 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
136 struct buffer_head *bh_result, int create);
137 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
138 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
139 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
140 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
141 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
142 struct inode *inode, struct page *page, loff_t from,
143 loff_t length, int flags);
146 * Test whether an inode is a fast symlink.
148 static int ext4_inode_is_fast_symlink(struct inode *inode)
150 int ea_blocks = EXT4_I(inode)->i_file_acl ?
151 (inode->i_sb->s_blocksize >> 9) : 0;
153 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
157 * Restart the transaction associated with *handle. This does a commit,
158 * so before we call here everything must be consistently dirtied against
161 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
167 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
168 * moment, get_block can be called only for blocks inside i_size since
169 * page cache has been already dropped and writes are blocked by
170 * i_mutex. So we can safely drop the i_data_sem here.
172 BUG_ON(EXT4_JOURNAL(inode) == NULL);
173 jbd_debug(2, "restarting handle %p\n", handle);
174 up_write(&EXT4_I(inode)->i_data_sem);
175 ret = ext4_journal_restart(handle, nblocks);
176 down_write(&EXT4_I(inode)->i_data_sem);
177 ext4_discard_preallocations(inode);
183 * Called at the last iput() if i_nlink is zero.
185 void ext4_evict_inode(struct inode *inode)
190 trace_ext4_evict_inode(inode);
192 ext4_ioend_wait(inode);
194 if (inode->i_nlink) {
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
210 * Note that directories do not have this problem because they
211 * don't use page cache.
213 if (ext4_should_journal_data(inode) &&
214 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
218 jbd2_log_start_commit(journal, commit_tid);
219 jbd2_log_wait_commit(journal, commit_tid);
220 filemap_write_and_wait(&inode->i_data);
222 truncate_inode_pages(&inode->i_data, 0);
226 if (!is_bad_inode(inode))
227 dquot_initialize(inode);
229 if (ext4_should_order_data(inode))
230 ext4_begin_ordered_truncate(inode, 0);
231 truncate_inode_pages(&inode->i_data, 0);
233 if (is_bad_inode(inode))
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
240 sb_start_intwrite(inode->i_sb);
241 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
242 if (IS_ERR(handle)) {
243 ext4_std_error(inode->i_sb, PTR_ERR(handle));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL, inode);
250 sb_end_intwrite(inode->i_sb);
255 ext4_handle_sync(handle);
257 err = ext4_mark_inode_dirty(handle, inode);
259 ext4_warning(inode->i_sb,
260 "couldn't mark inode dirty (err %d)", err);
264 ext4_truncate(inode);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle, 3)) {
273 err = ext4_journal_extend(handle, 3);
275 err = ext4_journal_restart(handle, 3);
277 ext4_warning(inode->i_sb,
278 "couldn't extend journal (err %d)", err);
280 ext4_journal_stop(handle);
281 ext4_orphan_del(NULL, inode);
282 sb_end_intwrite(inode->i_sb);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle, inode);
296 EXT4_I(inode)->i_dtime = get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle, inode))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode);
309 ext4_free_inode(handle, inode);
310 ext4_journal_stop(handle);
311 sb_end_intwrite(inode->i_sb);
314 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
318 qsize_t *ext4_get_reserved_space(struct inode *inode)
320 return &EXT4_I(inode)->i_reserved_quota;
325 * Calculate the number of metadata blocks need to reserve
326 * to allocate a block located at @lblock
328 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
330 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
331 return ext4_ext_calc_metadata_amount(inode, lblock);
333 return ext4_ind_calc_metadata_amount(inode, lblock);
337 * Called with i_data_sem down, which is important since we can call
338 * ext4_discard_preallocations() from here.
340 void ext4_da_update_reserve_space(struct inode *inode,
341 int used, int quota_claim)
343 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
344 struct ext4_inode_info *ei = EXT4_I(inode);
346 spin_lock(&ei->i_block_reservation_lock);
347 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
348 if (unlikely(used > ei->i_reserved_data_blocks)) {
349 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
350 "with only %d reserved data blocks",
351 __func__, inode->i_ino, used,
352 ei->i_reserved_data_blocks);
354 used = ei->i_reserved_data_blocks;
357 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
358 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
359 "with only %d reserved metadata blocks\n", __func__,
360 inode->i_ino, ei->i_allocated_meta_blocks,
361 ei->i_reserved_meta_blocks);
363 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
366 /* Update per-inode reservations */
367 ei->i_reserved_data_blocks -= used;
368 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
369 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
370 used + ei->i_allocated_meta_blocks);
371 ei->i_allocated_meta_blocks = 0;
373 if (ei->i_reserved_data_blocks == 0) {
375 * We can release all of the reserved metadata blocks
376 * only when we have written all of the delayed
379 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
380 ei->i_reserved_meta_blocks);
381 ei->i_reserved_meta_blocks = 0;
382 ei->i_da_metadata_calc_len = 0;
384 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
386 /* Update quota subsystem for data blocks */
388 dquot_claim_block(inode, EXT4_C2B(sbi, used));
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
393 * not re-claim the quota for fallocated blocks.
395 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
403 if ((ei->i_reserved_data_blocks == 0) &&
404 (atomic_read(&inode->i_writecount) == 0))
405 ext4_discard_preallocations(inode);
408 static int __check_block_validity(struct inode *inode, const char *func,
410 struct ext4_map_blocks *map)
412 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
414 ext4_error_inode(inode, func, line, map->m_pblk,
415 "lblock %lu mapped to illegal pblock "
416 "(length %d)", (unsigned long) map->m_lblk,
423 #define check_block_validity(inode, map) \
424 __check_block_validity((inode), __func__, __LINE__, (map))
427 * Return the number of contiguous dirty pages in a given inode
428 * starting at page frame idx.
430 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
431 unsigned int max_pages)
433 struct address_space *mapping = inode->i_mapping;
437 int i, nr_pages, done = 0;
441 pagevec_init(&pvec, 0);
444 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
446 (pgoff_t)PAGEVEC_SIZE);
449 for (i = 0; i < nr_pages; i++) {
450 struct page *page = pvec.pages[i];
451 struct buffer_head *bh, *head;
454 if (unlikely(page->mapping != mapping) ||
456 PageWriteback(page) ||
457 page->index != idx) {
462 if (page_has_buffers(page)) {
463 bh = head = page_buffers(page);
465 if (!buffer_delay(bh) &&
466 !buffer_unwritten(bh))
468 bh = bh->b_this_page;
469 } while (!done && (bh != head));
476 if (num >= max_pages) {
481 pagevec_release(&pvec);
487 * The ext4_map_blocks() function tries to look up the requested blocks,
488 * and returns if the blocks are already mapped.
490 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
491 * and store the allocated blocks in the result buffer head and mark it
494 * If file type is extents based, it will call ext4_ext_map_blocks(),
495 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
498 * On success, it returns the number of blocks being mapped or allocate.
499 * if create==0 and the blocks are pre-allocated and uninitialized block,
500 * the result buffer head is unmapped. If the create ==1, it will make sure
501 * the buffer head is mapped.
503 * It returns 0 if plain look up failed (blocks have not been allocated), in
504 * that case, buffer head is unmapped
506 * It returns the error in case of allocation failure.
508 int ext4_map_blocks(handle_t *handle, struct inode *inode,
509 struct ext4_map_blocks *map, int flags)
514 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
515 "logical block %lu\n", inode->i_ino, flags, map->m_len,
516 (unsigned long) map->m_lblk);
518 * Try to see if we can get the block without requesting a new
521 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
522 down_read((&EXT4_I(inode)->i_data_sem));
523 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
524 retval = ext4_ext_map_blocks(handle, inode, map, flags &
525 EXT4_GET_BLOCKS_KEEP_SIZE);
527 retval = ext4_ind_map_blocks(handle, inode, map, flags &
528 EXT4_GET_BLOCKS_KEEP_SIZE);
530 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
531 up_read((&EXT4_I(inode)->i_data_sem));
533 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
535 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
536 /* delayed alloc may be allocated by fallocate and
537 * coverted to initialized by directIO.
538 * we need to handle delayed extent here.
540 down_write((&EXT4_I(inode)->i_data_sem));
543 ret = check_block_validity(inode, map);
548 /* If it is only a block(s) look up */
549 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
553 * Returns if the blocks have already allocated
555 * Note that if blocks have been preallocated
556 * ext4_ext_get_block() returns the create = 0
557 * with buffer head unmapped.
559 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
563 * When we call get_blocks without the create flag, the
564 * BH_Unwritten flag could have gotten set if the blocks
565 * requested were part of a uninitialized extent. We need to
566 * clear this flag now that we are committed to convert all or
567 * part of the uninitialized extent to be an initialized
568 * extent. This is because we need to avoid the combination
569 * of BH_Unwritten and BH_Mapped flags being simultaneously
570 * set on the buffer_head.
572 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
575 * New blocks allocate and/or writing to uninitialized extent
576 * will possibly result in updating i_data, so we take
577 * the write lock of i_data_sem, and call get_blocks()
578 * with create == 1 flag.
580 down_write((&EXT4_I(inode)->i_data_sem));
583 * if the caller is from delayed allocation writeout path
584 * we have already reserved fs blocks for allocation
585 * let the underlying get_block() function know to
586 * avoid double accounting
588 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
589 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
591 * We need to check for EXT4 here because migrate
592 * could have changed the inode type in between
594 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
595 retval = ext4_ext_map_blocks(handle, inode, map, flags);
597 retval = ext4_ind_map_blocks(handle, inode, map, flags);
599 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
601 * We allocated new blocks which will result in
602 * i_data's format changing. Force the migrate
603 * to fail by clearing migrate flags
605 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
609 * Update reserved blocks/metadata blocks after successful
610 * block allocation which had been deferred till now. We don't
611 * support fallocate for non extent files. So we can update
612 * reserve space here.
615 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
616 ext4_da_update_reserve_space(inode, retval, 1);
618 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
619 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
621 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
624 /* delayed allocation blocks has been allocated */
625 ret = ext4_es_remove_extent(inode, map->m_lblk,
632 up_write((&EXT4_I(inode)->i_data_sem));
633 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
634 int ret = check_block_validity(inode, map);
641 /* Maximum number of blocks we map for direct IO at once. */
642 #define DIO_MAX_BLOCKS 4096
644 static int _ext4_get_block(struct inode *inode, sector_t iblock,
645 struct buffer_head *bh, int flags)
647 handle_t *handle = ext4_journal_current_handle();
648 struct ext4_map_blocks map;
649 int ret = 0, started = 0;
652 if (ext4_has_inline_data(inode))
656 map.m_len = bh->b_size >> inode->i_blkbits;
658 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
659 /* Direct IO write... */
660 if (map.m_len > DIO_MAX_BLOCKS)
661 map.m_len = DIO_MAX_BLOCKS;
662 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
663 handle = ext4_journal_start(inode, dio_credits);
664 if (IS_ERR(handle)) {
665 ret = PTR_ERR(handle);
671 ret = ext4_map_blocks(handle, inode, &map, flags);
673 map_bh(bh, inode->i_sb, map.m_pblk);
674 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
675 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
679 ext4_journal_stop(handle);
683 int ext4_get_block(struct inode *inode, sector_t iblock,
684 struct buffer_head *bh, int create)
686 return _ext4_get_block(inode, iblock, bh,
687 create ? EXT4_GET_BLOCKS_CREATE : 0);
691 * `handle' can be NULL if create is zero
693 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
694 ext4_lblk_t block, int create, int *errp)
696 struct ext4_map_blocks map;
697 struct buffer_head *bh;
700 J_ASSERT(handle != NULL || create == 0);
704 err = ext4_map_blocks(handle, inode, &map,
705 create ? EXT4_GET_BLOCKS_CREATE : 0);
707 /* ensure we send some value back into *errp */
715 bh = sb_getblk(inode->i_sb, map.m_pblk);
720 if (map.m_flags & EXT4_MAP_NEW) {
721 J_ASSERT(create != 0);
722 J_ASSERT(handle != NULL);
725 * Now that we do not always journal data, we should
726 * keep in mind whether this should always journal the
727 * new buffer as metadata. For now, regular file
728 * writes use ext4_get_block instead, so it's not a
732 BUFFER_TRACE(bh, "call get_create_access");
733 fatal = ext4_journal_get_create_access(handle, bh);
734 if (!fatal && !buffer_uptodate(bh)) {
735 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
736 set_buffer_uptodate(bh);
739 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
740 err = ext4_handle_dirty_metadata(handle, inode, bh);
744 BUFFER_TRACE(bh, "not a new buffer");
754 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
755 ext4_lblk_t block, int create, int *err)
757 struct buffer_head *bh;
759 bh = ext4_getblk(handle, inode, block, create, err);
762 if (buffer_uptodate(bh))
764 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
766 if (buffer_uptodate(bh))
773 static int walk_page_buffers(handle_t *handle,
774 struct buffer_head *head,
778 int (*fn)(handle_t *handle,
779 struct buffer_head *bh))
781 struct buffer_head *bh;
782 unsigned block_start, block_end;
783 unsigned blocksize = head->b_size;
785 struct buffer_head *next;
787 for (bh = head, block_start = 0;
788 ret == 0 && (bh != head || !block_start);
789 block_start = block_end, bh = next) {
790 next = bh->b_this_page;
791 block_end = block_start + blocksize;
792 if (block_end <= from || block_start >= to) {
793 if (partial && !buffer_uptodate(bh))
797 err = (*fn)(handle, bh);
805 * To preserve ordering, it is essential that the hole instantiation and
806 * the data write be encapsulated in a single transaction. We cannot
807 * close off a transaction and start a new one between the ext4_get_block()
808 * and the commit_write(). So doing the jbd2_journal_start at the start of
809 * prepare_write() is the right place.
811 * Also, this function can nest inside ext4_writepage() ->
812 * block_write_full_page(). In that case, we *know* that ext4_writepage()
813 * has generated enough buffer credits to do the whole page. So we won't
814 * block on the journal in that case, which is good, because the caller may
817 * By accident, ext4 can be reentered when a transaction is open via
818 * quota file writes. If we were to commit the transaction while thus
819 * reentered, there can be a deadlock - we would be holding a quota
820 * lock, and the commit would never complete if another thread had a
821 * transaction open and was blocking on the quota lock - a ranking
824 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
825 * will _not_ run commit under these circumstances because handle->h_ref
826 * is elevated. We'll still have enough credits for the tiny quotafile
829 static int do_journal_get_write_access(handle_t *handle,
830 struct buffer_head *bh)
832 int dirty = buffer_dirty(bh);
835 if (!buffer_mapped(bh) || buffer_freed(bh))
838 * __block_write_begin() could have dirtied some buffers. Clean
839 * the dirty bit as jbd2_journal_get_write_access() could complain
840 * otherwise about fs integrity issues. Setting of the dirty bit
841 * by __block_write_begin() isn't a real problem here as we clear
842 * the bit before releasing a page lock and thus writeback cannot
843 * ever write the buffer.
846 clear_buffer_dirty(bh);
847 ret = ext4_journal_get_write_access(handle, bh);
849 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
853 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
854 struct buffer_head *bh_result, int create);
855 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
856 struct buffer_head *bh_result, int create);
857 static int ext4_write_begin(struct file *file, struct address_space *mapping,
858 loff_t pos, unsigned len, unsigned flags,
859 struct page **pagep, void **fsdata)
861 struct inode *inode = mapping->host;
862 int ret, needed_blocks;
869 trace_ext4_write_begin(inode, pos, len, flags);
871 * Reserve one block more for addition to orphan list in case
872 * we allocate blocks but write fails for some reason
874 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
875 index = pos >> PAGE_CACHE_SHIFT;
876 from = pos & (PAGE_CACHE_SIZE - 1);
880 handle = ext4_journal_start(inode, needed_blocks);
881 if (IS_ERR(handle)) {
882 ret = PTR_ERR(handle);
886 /* We cannot recurse into the filesystem as the transaction is already
888 flags |= AOP_FLAG_NOFS;
890 page = grab_cache_page_write_begin(mapping, index, flags);
892 ext4_journal_stop(handle);
898 if (ext4_should_dioread_nolock(inode))
899 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
901 ret = __block_write_begin(page, pos, len, ext4_get_block);
903 if (!ret && ext4_should_journal_data(inode)) {
904 ret = walk_page_buffers(handle, page_buffers(page),
905 from, to, NULL, do_journal_get_write_access);
910 page_cache_release(page);
912 * __block_write_begin may have instantiated a few blocks
913 * outside i_size. Trim these off again. Don't need
914 * i_size_read because we hold i_mutex.
916 * Add inode to orphan list in case we crash before
919 if (pos + len > inode->i_size && ext4_can_truncate(inode))
920 ext4_orphan_add(handle, inode);
922 ext4_journal_stop(handle);
923 if (pos + len > inode->i_size) {
924 ext4_truncate_failed_write(inode);
926 * If truncate failed early the inode might
927 * still be on the orphan list; we need to
928 * make sure the inode is removed from the
929 * orphan list in that case.
932 ext4_orphan_del(NULL, inode);
936 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
942 /* For write_end() in data=journal mode */
943 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
945 if (!buffer_mapped(bh) || buffer_freed(bh))
947 set_buffer_uptodate(bh);
948 return ext4_handle_dirty_metadata(handle, NULL, bh);
951 static int ext4_generic_write_end(struct file *file,
952 struct address_space *mapping,
953 loff_t pos, unsigned len, unsigned copied,
954 struct page *page, void *fsdata)
956 int i_size_changed = 0;
957 struct inode *inode = mapping->host;
958 handle_t *handle = ext4_journal_current_handle();
960 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
963 * No need to use i_size_read() here, the i_size
964 * cannot change under us because we hold i_mutex.
966 * But it's important to update i_size while still holding page lock:
967 * page writeout could otherwise come in and zero beyond i_size.
969 if (pos + copied > inode->i_size) {
970 i_size_write(inode, pos + copied);
974 if (pos + copied > EXT4_I(inode)->i_disksize) {
975 /* We need to mark inode dirty even if
976 * new_i_size is less that inode->i_size
977 * bu greater than i_disksize.(hint delalloc)
979 ext4_update_i_disksize(inode, (pos + copied));
983 page_cache_release(page);
986 * Don't mark the inode dirty under page lock. First, it unnecessarily
987 * makes the holding time of page lock longer. Second, it forces lock
988 * ordering of page lock and transaction start for journaling
992 ext4_mark_inode_dirty(handle, inode);
998 * We need to pick up the new inode size which generic_commit_write gave us
999 * `file' can be NULL - eg, when called from page_symlink().
1001 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1002 * buffers are managed internally.
1004 static int ext4_ordered_write_end(struct file *file,
1005 struct address_space *mapping,
1006 loff_t pos, unsigned len, unsigned copied,
1007 struct page *page, void *fsdata)
1009 handle_t *handle = ext4_journal_current_handle();
1010 struct inode *inode = mapping->host;
1013 trace_ext4_ordered_write_end(inode, pos, len, copied);
1014 ret = ext4_jbd2_file_inode(handle, inode);
1017 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1020 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1021 /* if we have allocated more blocks and copied
1022 * less. We will have blocks allocated outside
1023 * inode->i_size. So truncate them
1025 ext4_orphan_add(handle, inode);
1030 page_cache_release(page);
1033 ret2 = ext4_journal_stop(handle);
1037 if (pos + len > inode->i_size) {
1038 ext4_truncate_failed_write(inode);
1040 * If truncate failed early the inode might still be
1041 * on the orphan list; we need to make sure the inode
1042 * is removed from the orphan list in that case.
1045 ext4_orphan_del(NULL, inode);
1049 return ret ? ret : copied;
1052 static int ext4_writeback_write_end(struct file *file,
1053 struct address_space *mapping,
1054 loff_t pos, unsigned len, unsigned copied,
1055 struct page *page, void *fsdata)
1057 handle_t *handle = ext4_journal_current_handle();
1058 struct inode *inode = mapping->host;
1061 trace_ext4_writeback_write_end(inode, pos, len, copied);
1062 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1065 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1066 /* if we have allocated more blocks and copied
1067 * less. We will have blocks allocated outside
1068 * inode->i_size. So truncate them
1070 ext4_orphan_add(handle, inode);
1075 ret2 = ext4_journal_stop(handle);
1079 if (pos + len > inode->i_size) {
1080 ext4_truncate_failed_write(inode);
1082 * If truncate failed early the inode might still be
1083 * on the orphan list; we need to make sure the inode
1084 * is removed from the orphan list in that case.
1087 ext4_orphan_del(NULL, inode);
1090 return ret ? ret : copied;
1093 static int ext4_journalled_write_end(struct file *file,
1094 struct address_space *mapping,
1095 loff_t pos, unsigned len, unsigned copied,
1096 struct page *page, void *fsdata)
1098 handle_t *handle = ext4_journal_current_handle();
1099 struct inode *inode = mapping->host;
1105 trace_ext4_journalled_write_end(inode, pos, len, copied);
1106 from = pos & (PAGE_CACHE_SIZE - 1);
1109 BUG_ON(!ext4_handle_valid(handle));
1112 if (!PageUptodate(page))
1114 page_zero_new_buffers(page, from+copied, to);
1117 ret = walk_page_buffers(handle, page_buffers(page), from,
1118 to, &partial, write_end_fn);
1120 SetPageUptodate(page);
1121 new_i_size = pos + copied;
1122 if (new_i_size > inode->i_size)
1123 i_size_write(inode, pos+copied);
1124 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1125 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1126 if (new_i_size > EXT4_I(inode)->i_disksize) {
1127 ext4_update_i_disksize(inode, new_i_size);
1128 ret2 = ext4_mark_inode_dirty(handle, inode);
1134 page_cache_release(page);
1135 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1136 /* if we have allocated more blocks and copied
1137 * less. We will have blocks allocated outside
1138 * inode->i_size. So truncate them
1140 ext4_orphan_add(handle, inode);
1142 ret2 = ext4_journal_stop(handle);
1145 if (pos + len > inode->i_size) {
1146 ext4_truncate_failed_write(inode);
1148 * If truncate failed early the inode might still be
1149 * on the orphan list; we need to make sure the inode
1150 * is removed from the orphan list in that case.
1153 ext4_orphan_del(NULL, inode);
1156 return ret ? ret : copied;
1160 * Reserve a single cluster located at lblock
1162 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1165 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1166 struct ext4_inode_info *ei = EXT4_I(inode);
1167 unsigned int md_needed;
1169 ext4_lblk_t save_last_lblock;
1173 * We will charge metadata quota at writeout time; this saves
1174 * us from metadata over-estimation, though we may go over by
1175 * a small amount in the end. Here we just reserve for data.
1177 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1182 * recalculate the amount of metadata blocks to reserve
1183 * in order to allocate nrblocks
1184 * worse case is one extent per block
1187 spin_lock(&ei->i_block_reservation_lock);
1189 * ext4_calc_metadata_amount() has side effects, which we have
1190 * to be prepared undo if we fail to claim space.
1192 save_len = ei->i_da_metadata_calc_len;
1193 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1194 md_needed = EXT4_NUM_B2C(sbi,
1195 ext4_calc_metadata_amount(inode, lblock));
1196 trace_ext4_da_reserve_space(inode, md_needed);
1199 * We do still charge estimated metadata to the sb though;
1200 * we cannot afford to run out of free blocks.
1202 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1203 ei->i_da_metadata_calc_len = save_len;
1204 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1205 spin_unlock(&ei->i_block_reservation_lock);
1206 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1210 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1213 ei->i_reserved_data_blocks++;
1214 ei->i_reserved_meta_blocks += md_needed;
1215 spin_unlock(&ei->i_block_reservation_lock);
1217 return 0; /* success */
1220 static void ext4_da_release_space(struct inode *inode, int to_free)
1222 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1223 struct ext4_inode_info *ei = EXT4_I(inode);
1226 return; /* Nothing to release, exit */
1228 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1230 trace_ext4_da_release_space(inode, to_free);
1231 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1233 * if there aren't enough reserved blocks, then the
1234 * counter is messed up somewhere. Since this
1235 * function is called from invalidate page, it's
1236 * harmless to return without any action.
1238 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1239 "ino %lu, to_free %d with only %d reserved "
1240 "data blocks", inode->i_ino, to_free,
1241 ei->i_reserved_data_blocks);
1243 to_free = ei->i_reserved_data_blocks;
1245 ei->i_reserved_data_blocks -= to_free;
1247 if (ei->i_reserved_data_blocks == 0) {
1249 * We can release all of the reserved metadata blocks
1250 * only when we have written all of the delayed
1251 * allocation blocks.
1252 * Note that in case of bigalloc, i_reserved_meta_blocks,
1253 * i_reserved_data_blocks, etc. refer to number of clusters.
1255 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1256 ei->i_reserved_meta_blocks);
1257 ei->i_reserved_meta_blocks = 0;
1258 ei->i_da_metadata_calc_len = 0;
1261 /* update fs dirty data blocks counter */
1262 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1264 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1266 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1269 static void ext4_da_page_release_reservation(struct page *page,
1270 unsigned long offset)
1273 struct buffer_head *head, *bh;
1274 unsigned int curr_off = 0;
1275 struct inode *inode = page->mapping->host;
1276 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1280 head = page_buffers(page);
1283 unsigned int next_off = curr_off + bh->b_size;
1285 if ((offset <= curr_off) && (buffer_delay(bh))) {
1287 clear_buffer_delay(bh);
1289 curr_off = next_off;
1290 } while ((bh = bh->b_this_page) != head);
1293 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1294 ext4_es_remove_extent(inode, lblk, to_release);
1297 /* If we have released all the blocks belonging to a cluster, then we
1298 * need to release the reserved space for that cluster. */
1299 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1300 while (num_clusters > 0) {
1301 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1302 ((num_clusters - 1) << sbi->s_cluster_bits);
1303 if (sbi->s_cluster_ratio == 1 ||
1304 !ext4_find_delalloc_cluster(inode, lblk))
1305 ext4_da_release_space(inode, 1);
1312 * Delayed allocation stuff
1316 * mpage_da_submit_io - walks through extent of pages and try to write
1317 * them with writepage() call back
1319 * @mpd->inode: inode
1320 * @mpd->first_page: first page of the extent
1321 * @mpd->next_page: page after the last page of the extent
1323 * By the time mpage_da_submit_io() is called we expect all blocks
1324 * to be allocated. this may be wrong if allocation failed.
1326 * As pages are already locked by write_cache_pages(), we can't use it
1328 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1329 struct ext4_map_blocks *map)
1331 struct pagevec pvec;
1332 unsigned long index, end;
1333 int ret = 0, err, nr_pages, i;
1334 struct inode *inode = mpd->inode;
1335 struct address_space *mapping = inode->i_mapping;
1336 loff_t size = i_size_read(inode);
1337 unsigned int len, block_start;
1338 struct buffer_head *bh, *page_bufs = NULL;
1339 int journal_data = ext4_should_journal_data(inode);
1340 sector_t pblock = 0, cur_logical = 0;
1341 struct ext4_io_submit io_submit;
1343 BUG_ON(mpd->next_page <= mpd->first_page);
1344 memset(&io_submit, 0, sizeof(io_submit));
1346 * We need to start from the first_page to the next_page - 1
1347 * to make sure we also write the mapped dirty buffer_heads.
1348 * If we look at mpd->b_blocknr we would only be looking
1349 * at the currently mapped buffer_heads.
1351 index = mpd->first_page;
1352 end = mpd->next_page - 1;
1354 pagevec_init(&pvec, 0);
1355 while (index <= end) {
1356 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1359 for (i = 0; i < nr_pages; i++) {
1360 int commit_write = 0, skip_page = 0;
1361 struct page *page = pvec.pages[i];
1363 index = page->index;
1367 if (index == size >> PAGE_CACHE_SHIFT)
1368 len = size & ~PAGE_CACHE_MASK;
1370 len = PAGE_CACHE_SIZE;
1372 cur_logical = index << (PAGE_CACHE_SHIFT -
1374 pblock = map->m_pblk + (cur_logical -
1379 BUG_ON(!PageLocked(page));
1380 BUG_ON(PageWriteback(page));
1383 * If the page does not have buffers (for
1384 * whatever reason), try to create them using
1385 * __block_write_begin. If this fails,
1386 * skip the page and move on.
1388 if (!page_has_buffers(page)) {
1389 if (__block_write_begin(page, 0, len,
1390 noalloc_get_block_write)) {
1398 bh = page_bufs = page_buffers(page);
1403 if (map && (cur_logical >= map->m_lblk) &&
1404 (cur_logical <= (map->m_lblk +
1405 (map->m_len - 1)))) {
1406 if (buffer_delay(bh)) {
1407 clear_buffer_delay(bh);
1408 bh->b_blocknr = pblock;
1410 if (buffer_unwritten(bh) ||
1412 BUG_ON(bh->b_blocknr != pblock);
1413 if (map->m_flags & EXT4_MAP_UNINIT)
1414 set_buffer_uninit(bh);
1415 clear_buffer_unwritten(bh);
1419 * skip page if block allocation undone and
1422 if (ext4_bh_delay_or_unwritten(NULL, bh))
1424 bh = bh->b_this_page;
1425 block_start += bh->b_size;
1428 } while (bh != page_bufs);
1434 /* mark the buffer_heads as dirty & uptodate */
1435 block_commit_write(page, 0, len);
1437 clear_page_dirty_for_io(page);
1439 * Delalloc doesn't support data journalling,
1440 * but eventually maybe we'll lift this
1443 if (unlikely(journal_data && PageChecked(page)))
1444 err = __ext4_journalled_writepage(page, len);
1445 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1446 err = ext4_bio_write_page(&io_submit, page,
1448 else if (buffer_uninit(page_bufs)) {
1449 ext4_set_bh_endio(page_bufs, inode);
1450 err = block_write_full_page_endio(page,
1451 noalloc_get_block_write,
1452 mpd->wbc, ext4_end_io_buffer_write);
1454 err = block_write_full_page(page,
1455 noalloc_get_block_write, mpd->wbc);
1458 mpd->pages_written++;
1460 * In error case, we have to continue because
1461 * remaining pages are still locked
1466 pagevec_release(&pvec);
1468 ext4_io_submit(&io_submit);
1472 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1476 struct pagevec pvec;
1477 struct inode *inode = mpd->inode;
1478 struct address_space *mapping = inode->i_mapping;
1479 ext4_lblk_t start, last;
1481 index = mpd->first_page;
1482 end = mpd->next_page - 1;
1484 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1485 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1486 ext4_es_remove_extent(inode, start, last - start + 1);
1488 pagevec_init(&pvec, 0);
1489 while (index <= end) {
1490 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1493 for (i = 0; i < nr_pages; i++) {
1494 struct page *page = pvec.pages[i];
1495 if (page->index > end)
1497 BUG_ON(!PageLocked(page));
1498 BUG_ON(PageWriteback(page));
1499 block_invalidatepage(page, 0);
1500 ClearPageUptodate(page);
1503 index = pvec.pages[nr_pages - 1]->index + 1;
1504 pagevec_release(&pvec);
1509 static void ext4_print_free_blocks(struct inode *inode)
1511 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1512 struct super_block *sb = inode->i_sb;
1514 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1515 EXT4_C2B(EXT4_SB(inode->i_sb),
1516 ext4_count_free_clusters(inode->i_sb)));
1517 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1518 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1519 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1520 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1521 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1522 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1523 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1524 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1525 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1526 EXT4_I(inode)->i_reserved_data_blocks);
1527 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1528 EXT4_I(inode)->i_reserved_meta_blocks);
1533 * mpage_da_map_and_submit - go through given space, map them
1534 * if necessary, and then submit them for I/O
1536 * @mpd - bh describing space
1538 * The function skips space we know is already mapped to disk blocks.
1541 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1543 int err, blks, get_blocks_flags;
1544 struct ext4_map_blocks map, *mapp = NULL;
1545 sector_t next = mpd->b_blocknr;
1546 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1547 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1548 handle_t *handle = NULL;
1551 * If the blocks are mapped already, or we couldn't accumulate
1552 * any blocks, then proceed immediately to the submission stage.
1554 if ((mpd->b_size == 0) ||
1555 ((mpd->b_state & (1 << BH_Mapped)) &&
1556 !(mpd->b_state & (1 << BH_Delay)) &&
1557 !(mpd->b_state & (1 << BH_Unwritten))))
1560 handle = ext4_journal_current_handle();
1564 * Call ext4_map_blocks() to allocate any delayed allocation
1565 * blocks, or to convert an uninitialized extent to be
1566 * initialized (in the case where we have written into
1567 * one or more preallocated blocks).
1569 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1570 * indicate that we are on the delayed allocation path. This
1571 * affects functions in many different parts of the allocation
1572 * call path. This flag exists primarily because we don't
1573 * want to change *many* call functions, so ext4_map_blocks()
1574 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1575 * inode's allocation semaphore is taken.
1577 * If the blocks in questions were delalloc blocks, set
1578 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1579 * variables are updated after the blocks have been allocated.
1582 map.m_len = max_blocks;
1583 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1584 if (ext4_should_dioread_nolock(mpd->inode))
1585 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1586 if (mpd->b_state & (1 << BH_Delay))
1587 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1589 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1591 struct super_block *sb = mpd->inode->i_sb;
1595 * If get block returns EAGAIN or ENOSPC and there
1596 * appears to be free blocks we will just let
1597 * mpage_da_submit_io() unlock all of the pages.
1602 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1608 * get block failure will cause us to loop in
1609 * writepages, because a_ops->writepage won't be able
1610 * to make progress. The page will be redirtied by
1611 * writepage and writepages will again try to write
1614 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1615 ext4_msg(sb, KERN_CRIT,
1616 "delayed block allocation failed for inode %lu "
1617 "at logical offset %llu with max blocks %zd "
1618 "with error %d", mpd->inode->i_ino,
1619 (unsigned long long) next,
1620 mpd->b_size >> mpd->inode->i_blkbits, err);
1621 ext4_msg(sb, KERN_CRIT,
1622 "This should not happen!! Data will be lost\n");
1624 ext4_print_free_blocks(mpd->inode);
1626 /* invalidate all the pages */
1627 ext4_da_block_invalidatepages(mpd);
1629 /* Mark this page range as having been completed */
1636 if (map.m_flags & EXT4_MAP_NEW) {
1637 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1640 for (i = 0; i < map.m_len; i++)
1641 unmap_underlying_metadata(bdev, map.m_pblk + i);
1645 * Update on-disk size along with block allocation.
1647 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1648 if (disksize > i_size_read(mpd->inode))
1649 disksize = i_size_read(mpd->inode);
1650 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1651 ext4_update_i_disksize(mpd->inode, disksize);
1652 err = ext4_mark_inode_dirty(handle, mpd->inode);
1654 ext4_error(mpd->inode->i_sb,
1655 "Failed to mark inode %lu dirty",
1660 mpage_da_submit_io(mpd, mapp);
1664 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1665 (1 << BH_Delay) | (1 << BH_Unwritten))
1668 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1670 * @mpd->lbh - extent of blocks
1671 * @logical - logical number of the block in the file
1672 * @bh - bh of the block (used to access block's state)
1674 * the function is used to collect contig. blocks in same state
1676 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1677 sector_t logical, size_t b_size,
1678 unsigned long b_state)
1681 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1684 * XXX Don't go larger than mballoc is willing to allocate
1685 * This is a stopgap solution. We eventually need to fold
1686 * mpage_da_submit_io() into this function and then call
1687 * ext4_map_blocks() multiple times in a loop
1689 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1692 /* check if thereserved journal credits might overflow */
1693 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1694 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1696 * With non-extent format we are limited by the journal
1697 * credit available. Total credit needed to insert
1698 * nrblocks contiguous blocks is dependent on the
1699 * nrblocks. So limit nrblocks.
1702 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1703 EXT4_MAX_TRANS_DATA) {
1705 * Adding the new buffer_head would make it cross the
1706 * allowed limit for which we have journal credit
1707 * reserved. So limit the new bh->b_size
1709 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1710 mpd->inode->i_blkbits;
1711 /* we will do mpage_da_submit_io in the next loop */
1715 * First block in the extent
1717 if (mpd->b_size == 0) {
1718 mpd->b_blocknr = logical;
1719 mpd->b_size = b_size;
1720 mpd->b_state = b_state & BH_FLAGS;
1724 next = mpd->b_blocknr + nrblocks;
1726 * Can we merge the block to our big extent?
1728 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1729 mpd->b_size += b_size;
1735 * We couldn't merge the block to our extent, so we
1736 * need to flush current extent and start new one
1738 mpage_da_map_and_submit(mpd);
1742 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1744 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1748 * This function is grabs code from the very beginning of
1749 * ext4_map_blocks, but assumes that the caller is from delayed write
1750 * time. This function looks up the requested blocks and sets the
1751 * buffer delay bit under the protection of i_data_sem.
1753 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1754 struct ext4_map_blocks *map,
1755 struct buffer_head *bh)
1758 sector_t invalid_block = ~((sector_t) 0xffff);
1760 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1764 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1765 "logical block %lu\n", inode->i_ino, map->m_len,
1766 (unsigned long) map->m_lblk);
1768 * Try to see if we can get the block without requesting a new
1769 * file system block.
1771 down_read((&EXT4_I(inode)->i_data_sem));
1772 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1773 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1775 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1779 * XXX: __block_prepare_write() unmaps passed block,
1782 /* If the block was allocated from previously allocated cluster,
1783 * then we dont need to reserve it again. */
1784 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1785 retval = ext4_da_reserve_space(inode, iblock);
1787 /* not enough space to reserve */
1791 retval = ext4_es_insert_extent(inode, map->m_lblk, map->m_len);
1795 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1796 * and it should not appear on the bh->b_state.
1798 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1800 map_bh(bh, inode->i_sb, invalid_block);
1802 set_buffer_delay(bh);
1806 up_read((&EXT4_I(inode)->i_data_sem));
1812 * This is a special get_blocks_t callback which is used by
1813 * ext4_da_write_begin(). It will either return mapped block or
1814 * reserve space for a single block.
1816 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1817 * We also have b_blocknr = -1 and b_bdev initialized properly
1819 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1820 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1821 * initialized properly.
1823 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1824 struct buffer_head *bh, int create)
1826 struct ext4_map_blocks map;
1829 BUG_ON(create == 0);
1830 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1832 map.m_lblk = iblock;
1836 * first, we need to know whether the block is allocated already
1837 * preallocated blocks are unmapped but should treated
1838 * the same as allocated blocks.
1840 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1844 map_bh(bh, inode->i_sb, map.m_pblk);
1845 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1847 if (buffer_unwritten(bh)) {
1848 /* A delayed write to unwritten bh should be marked
1849 * new and mapped. Mapped ensures that we don't do
1850 * get_block multiple times when we write to the same
1851 * offset and new ensures that we do proper zero out
1852 * for partial write.
1855 set_buffer_mapped(bh);
1861 * This function is used as a standard get_block_t calback function
1862 * when there is no desire to allocate any blocks. It is used as a
1863 * callback function for block_write_begin() and block_write_full_page().
1864 * These functions should only try to map a single block at a time.
1866 * Since this function doesn't do block allocations even if the caller
1867 * requests it by passing in create=1, it is critically important that
1868 * any caller checks to make sure that any buffer heads are returned
1869 * by this function are either all already mapped or marked for
1870 * delayed allocation before calling block_write_full_page(). Otherwise,
1871 * b_blocknr could be left unitialized, and the page write functions will
1872 * be taken by surprise.
1874 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1875 struct buffer_head *bh_result, int create)
1877 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1878 return _ext4_get_block(inode, iblock, bh_result, 0);
1881 static int bget_one(handle_t *handle, struct buffer_head *bh)
1887 static int bput_one(handle_t *handle, struct buffer_head *bh)
1893 static int __ext4_journalled_writepage(struct page *page,
1896 struct address_space *mapping = page->mapping;
1897 struct inode *inode = mapping->host;
1898 struct buffer_head *page_bufs;
1899 handle_t *handle = NULL;
1903 ClearPageChecked(page);
1904 page_bufs = page_buffers(page);
1906 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1907 /* As soon as we unlock the page, it can go away, but we have
1908 * references to buffers so we are safe */
1911 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1912 if (IS_ERR(handle)) {
1913 ret = PTR_ERR(handle);
1917 BUG_ON(!ext4_handle_valid(handle));
1919 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1920 do_journal_get_write_access);
1922 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1926 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1927 err = ext4_journal_stop(handle);
1931 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1932 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1938 * Note that we don't need to start a transaction unless we're journaling data
1939 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1940 * need to file the inode to the transaction's list in ordered mode because if
1941 * we are writing back data added by write(), the inode is already there and if
1942 * we are writing back data modified via mmap(), no one guarantees in which
1943 * transaction the data will hit the disk. In case we are journaling data, we
1944 * cannot start transaction directly because transaction start ranks above page
1945 * lock so we have to do some magic.
1947 * This function can get called via...
1948 * - ext4_da_writepages after taking page lock (have journal handle)
1949 * - journal_submit_inode_data_buffers (no journal handle)
1950 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1951 * - grab_page_cache when doing write_begin (have journal handle)
1953 * We don't do any block allocation in this function. If we have page with
1954 * multiple blocks we need to write those buffer_heads that are mapped. This
1955 * is important for mmaped based write. So if we do with blocksize 1K
1956 * truncate(f, 1024);
1957 * a = mmap(f, 0, 4096);
1959 * truncate(f, 4096);
1960 * we have in the page first buffer_head mapped via page_mkwrite call back
1961 * but other buffer_heads would be unmapped but dirty (dirty done via the
1962 * do_wp_page). So writepage should write the first block. If we modify
1963 * the mmap area beyond 1024 we will again get a page_fault and the
1964 * page_mkwrite callback will do the block allocation and mark the
1965 * buffer_heads mapped.
1967 * We redirty the page if we have any buffer_heads that is either delay or
1968 * unwritten in the page.
1970 * We can get recursively called as show below.
1972 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1975 * But since we don't do any block allocation we should not deadlock.
1976 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1978 static int ext4_writepage(struct page *page,
1979 struct writeback_control *wbc)
1981 int ret = 0, commit_write = 0;
1984 struct buffer_head *page_bufs = NULL;
1985 struct inode *inode = page->mapping->host;
1987 trace_ext4_writepage(page);
1988 size = i_size_read(inode);
1989 if (page->index == size >> PAGE_CACHE_SHIFT)
1990 len = size & ~PAGE_CACHE_MASK;
1992 len = PAGE_CACHE_SIZE;
1995 * If the page does not have buffers (for whatever reason),
1996 * try to create them using __block_write_begin. If this
1997 * fails, redirty the page and move on.
1999 if (!page_has_buffers(page)) {
2000 if (__block_write_begin(page, 0, len,
2001 noalloc_get_block_write)) {
2003 redirty_page_for_writepage(wbc, page);
2009 page_bufs = page_buffers(page);
2010 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2011 ext4_bh_delay_or_unwritten)) {
2013 * We don't want to do block allocation, so redirty
2014 * the page and return. We may reach here when we do
2015 * a journal commit via journal_submit_inode_data_buffers.
2016 * We can also reach here via shrink_page_list but it
2017 * should never be for direct reclaim so warn if that
2020 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2025 /* now mark the buffer_heads as dirty and uptodate */
2026 block_commit_write(page, 0, len);
2028 if (PageChecked(page) && ext4_should_journal_data(inode))
2030 * It's mmapped pagecache. Add buffers and journal it. There
2031 * doesn't seem much point in redirtying the page here.
2033 return __ext4_journalled_writepage(page, len);
2035 if (buffer_uninit(page_bufs)) {
2036 ext4_set_bh_endio(page_bufs, inode);
2037 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2038 wbc, ext4_end_io_buffer_write);
2040 ret = block_write_full_page(page, noalloc_get_block_write,
2047 * This is called via ext4_da_writepages() to
2048 * calculate the total number of credits to reserve to fit
2049 * a single extent allocation into a single transaction,
2050 * ext4_da_writpeages() will loop calling this before
2051 * the block allocation.
2054 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2056 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2059 * With non-extent format the journal credit needed to
2060 * insert nrblocks contiguous block is dependent on
2061 * number of contiguous block. So we will limit
2062 * number of contiguous block to a sane value
2064 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2065 (max_blocks > EXT4_MAX_TRANS_DATA))
2066 max_blocks = EXT4_MAX_TRANS_DATA;
2068 return ext4_chunk_trans_blocks(inode, max_blocks);
2072 * write_cache_pages_da - walk the list of dirty pages of the given
2073 * address space and accumulate pages that need writing, and call
2074 * mpage_da_map_and_submit to map a single contiguous memory region
2075 * and then write them.
2077 static int write_cache_pages_da(struct address_space *mapping,
2078 struct writeback_control *wbc,
2079 struct mpage_da_data *mpd,
2080 pgoff_t *done_index)
2082 struct buffer_head *bh, *head;
2083 struct inode *inode = mapping->host;
2084 struct pagevec pvec;
2085 unsigned int nr_pages;
2088 long nr_to_write = wbc->nr_to_write;
2089 int i, tag, ret = 0;
2091 memset(mpd, 0, sizeof(struct mpage_da_data));
2094 pagevec_init(&pvec, 0);
2095 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2096 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2098 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2099 tag = PAGECACHE_TAG_TOWRITE;
2101 tag = PAGECACHE_TAG_DIRTY;
2103 *done_index = index;
2104 while (index <= end) {
2105 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2106 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2110 for (i = 0; i < nr_pages; i++) {
2111 struct page *page = pvec.pages[i];
2114 * At this point, the page may be truncated or
2115 * invalidated (changing page->mapping to NULL), or
2116 * even swizzled back from swapper_space to tmpfs file
2117 * mapping. However, page->index will not change
2118 * because we have a reference on the page.
2120 if (page->index > end)
2123 *done_index = page->index + 1;
2126 * If we can't merge this page, and we have
2127 * accumulated an contiguous region, write it
2129 if ((mpd->next_page != page->index) &&
2130 (mpd->next_page != mpd->first_page)) {
2131 mpage_da_map_and_submit(mpd);
2132 goto ret_extent_tail;
2138 * If the page is no longer dirty, or its
2139 * mapping no longer corresponds to inode we
2140 * are writing (which means it has been
2141 * truncated or invalidated), or the page is
2142 * already under writeback and we are not
2143 * doing a data integrity writeback, skip the page
2145 if (!PageDirty(page) ||
2146 (PageWriteback(page) &&
2147 (wbc->sync_mode == WB_SYNC_NONE)) ||
2148 unlikely(page->mapping != mapping)) {
2153 wait_on_page_writeback(page);
2154 BUG_ON(PageWriteback(page));
2156 if (mpd->next_page != page->index)
2157 mpd->first_page = page->index;
2158 mpd->next_page = page->index + 1;
2159 logical = (sector_t) page->index <<
2160 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2162 if (!page_has_buffers(page)) {
2163 mpage_add_bh_to_extent(mpd, logical,
2165 (1 << BH_Dirty) | (1 << BH_Uptodate));
2167 goto ret_extent_tail;
2170 * Page with regular buffer heads,
2171 * just add all dirty ones
2173 head = page_buffers(page);
2176 BUG_ON(buffer_locked(bh));
2178 * We need to try to allocate
2179 * unmapped blocks in the same page.
2180 * Otherwise we won't make progress
2181 * with the page in ext4_writepage
2183 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2184 mpage_add_bh_to_extent(mpd, logical,
2188 goto ret_extent_tail;
2189 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2191 * mapped dirty buffer. We need
2192 * to update the b_state
2193 * because we look at b_state
2194 * in mpage_da_map_blocks. We
2195 * don't update b_size because
2196 * if we find an unmapped
2197 * buffer_head later we need to
2198 * use the b_state flag of that
2201 if (mpd->b_size == 0)
2202 mpd->b_state = bh->b_state & BH_FLAGS;
2205 } while ((bh = bh->b_this_page) != head);
2208 if (nr_to_write > 0) {
2210 if (nr_to_write == 0 &&
2211 wbc->sync_mode == WB_SYNC_NONE)
2213 * We stop writing back only if we are
2214 * not doing integrity sync. In case of
2215 * integrity sync we have to keep going
2216 * because someone may be concurrently
2217 * dirtying pages, and we might have
2218 * synced a lot of newly appeared dirty
2219 * pages, but have not synced all of the
2225 pagevec_release(&pvec);
2230 ret = MPAGE_DA_EXTENT_TAIL;
2232 pagevec_release(&pvec);
2238 static int ext4_da_writepages(struct address_space *mapping,
2239 struct writeback_control *wbc)
2242 int range_whole = 0;
2243 handle_t *handle = NULL;
2244 struct mpage_da_data mpd;
2245 struct inode *inode = mapping->host;
2246 int pages_written = 0;
2247 unsigned int max_pages;
2248 int range_cyclic, cycled = 1, io_done = 0;
2249 int needed_blocks, ret = 0;
2250 long desired_nr_to_write, nr_to_writebump = 0;
2251 loff_t range_start = wbc->range_start;
2252 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2253 pgoff_t done_index = 0;
2255 struct blk_plug plug;
2257 trace_ext4_da_writepages(inode, wbc);
2260 * No pages to write? This is mainly a kludge to avoid starting
2261 * a transaction for special inodes like journal inode on last iput()
2262 * because that could violate lock ordering on umount
2264 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2268 * If the filesystem has aborted, it is read-only, so return
2269 * right away instead of dumping stack traces later on that
2270 * will obscure the real source of the problem. We test
2271 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2272 * the latter could be true if the filesystem is mounted
2273 * read-only, and in that case, ext4_da_writepages should
2274 * *never* be called, so if that ever happens, we would want
2277 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2280 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2283 range_cyclic = wbc->range_cyclic;
2284 if (wbc->range_cyclic) {
2285 index = mapping->writeback_index;
2288 wbc->range_start = index << PAGE_CACHE_SHIFT;
2289 wbc->range_end = LLONG_MAX;
2290 wbc->range_cyclic = 0;
2293 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2294 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2298 * This works around two forms of stupidity. The first is in
2299 * the writeback code, which caps the maximum number of pages
2300 * written to be 1024 pages. This is wrong on multiple
2301 * levels; different architectues have a different page size,
2302 * which changes the maximum amount of data which gets
2303 * written. Secondly, 4 megabytes is way too small. XFS
2304 * forces this value to be 16 megabytes by multiplying
2305 * nr_to_write parameter by four, and then relies on its
2306 * allocator to allocate larger extents to make them
2307 * contiguous. Unfortunately this brings us to the second
2308 * stupidity, which is that ext4's mballoc code only allocates
2309 * at most 2048 blocks. So we force contiguous writes up to
2310 * the number of dirty blocks in the inode, or
2311 * sbi->max_writeback_mb_bump whichever is smaller.
2313 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2314 if (!range_cyclic && range_whole) {
2315 if (wbc->nr_to_write == LONG_MAX)
2316 desired_nr_to_write = wbc->nr_to_write;
2318 desired_nr_to_write = wbc->nr_to_write * 8;
2320 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2322 if (desired_nr_to_write > max_pages)
2323 desired_nr_to_write = max_pages;
2325 if (wbc->nr_to_write < desired_nr_to_write) {
2326 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2327 wbc->nr_to_write = desired_nr_to_write;
2331 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2332 tag_pages_for_writeback(mapping, index, end);
2334 blk_start_plug(&plug);
2335 while (!ret && wbc->nr_to_write > 0) {
2338 * we insert one extent at a time. So we need
2339 * credit needed for single extent allocation.
2340 * journalled mode is currently not supported
2343 BUG_ON(ext4_should_journal_data(inode));
2344 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2346 /* start a new transaction*/
2347 handle = ext4_journal_start(inode, needed_blocks);
2348 if (IS_ERR(handle)) {
2349 ret = PTR_ERR(handle);
2350 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2351 "%ld pages, ino %lu; err %d", __func__,
2352 wbc->nr_to_write, inode->i_ino, ret);
2353 blk_finish_plug(&plug);
2354 goto out_writepages;
2358 * Now call write_cache_pages_da() to find the next
2359 * contiguous region of logical blocks that need
2360 * blocks to be allocated by ext4 and submit them.
2362 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2364 * If we have a contiguous extent of pages and we
2365 * haven't done the I/O yet, map the blocks and submit
2368 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2369 mpage_da_map_and_submit(&mpd);
2370 ret = MPAGE_DA_EXTENT_TAIL;
2372 trace_ext4_da_write_pages(inode, &mpd);
2373 wbc->nr_to_write -= mpd.pages_written;
2375 ext4_journal_stop(handle);
2377 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2378 /* commit the transaction which would
2379 * free blocks released in the transaction
2382 jbd2_journal_force_commit_nested(sbi->s_journal);
2384 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2386 * Got one extent now try with rest of the pages.
2387 * If mpd.retval is set -EIO, journal is aborted.
2388 * So we don't need to write any more.
2390 pages_written += mpd.pages_written;
2393 } else if (wbc->nr_to_write)
2395 * There is no more writeout needed
2396 * or we requested for a noblocking writeout
2397 * and we found the device congested
2401 blk_finish_plug(&plug);
2402 if (!io_done && !cycled) {
2405 wbc->range_start = index << PAGE_CACHE_SHIFT;
2406 wbc->range_end = mapping->writeback_index - 1;
2411 wbc->range_cyclic = range_cyclic;
2412 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2414 * set the writeback_index so that range_cyclic
2415 * mode will write it back later
2417 mapping->writeback_index = done_index;
2420 wbc->nr_to_write -= nr_to_writebump;
2421 wbc->range_start = range_start;
2422 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2426 #define FALL_BACK_TO_NONDELALLOC 1
2427 static int ext4_nonda_switch(struct super_block *sb)
2429 s64 free_blocks, dirty_blocks;
2430 struct ext4_sb_info *sbi = EXT4_SB(sb);
2433 * switch to non delalloc mode if we are running low
2434 * on free block. The free block accounting via percpu
2435 * counters can get slightly wrong with percpu_counter_batch getting
2436 * accumulated on each CPU without updating global counters
2437 * Delalloc need an accurate free block accounting. So switch
2438 * to non delalloc when we are near to error range.
2440 free_blocks = EXT4_C2B(sbi,
2441 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2442 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2444 * Start pushing delalloc when 1/2 of free blocks are dirty.
2446 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2447 !writeback_in_progress(sb->s_bdi) &&
2448 down_read_trylock(&sb->s_umount)) {
2449 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2450 up_read(&sb->s_umount);
2453 if (2 * free_blocks < 3 * dirty_blocks ||
2454 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2456 * free block count is less than 150% of dirty blocks
2457 * or free blocks is less than watermark
2464 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2465 loff_t pos, unsigned len, unsigned flags,
2466 struct page **pagep, void **fsdata)
2468 int ret, retries = 0;
2471 struct inode *inode = mapping->host;
2474 index = pos >> PAGE_CACHE_SHIFT;
2476 if (ext4_nonda_switch(inode->i_sb)) {
2477 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2478 return ext4_write_begin(file, mapping, pos,
2479 len, flags, pagep, fsdata);
2481 *fsdata = (void *)0;
2482 trace_ext4_da_write_begin(inode, pos, len, flags);
2485 * With delayed allocation, we don't log the i_disksize update
2486 * if there is delayed block allocation. But we still need
2487 * to journalling the i_disksize update if writes to the end
2488 * of file which has an already mapped buffer.
2490 handle = ext4_journal_start(inode, 1);
2491 if (IS_ERR(handle)) {
2492 ret = PTR_ERR(handle);
2495 /* We cannot recurse into the filesystem as the transaction is already
2497 flags |= AOP_FLAG_NOFS;
2499 page = grab_cache_page_write_begin(mapping, index, flags);
2501 ext4_journal_stop(handle);
2507 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2510 ext4_journal_stop(handle);
2511 page_cache_release(page);
2513 * block_write_begin may have instantiated a few blocks
2514 * outside i_size. Trim these off again. Don't need
2515 * i_size_read because we hold i_mutex.
2517 if (pos + len > inode->i_size)
2518 ext4_truncate_failed_write(inode);
2521 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2528 * Check if we should update i_disksize
2529 * when write to the end of file but not require block allocation
2531 static int ext4_da_should_update_i_disksize(struct page *page,
2532 unsigned long offset)
2534 struct buffer_head *bh;
2535 struct inode *inode = page->mapping->host;
2539 bh = page_buffers(page);
2540 idx = offset >> inode->i_blkbits;
2542 for (i = 0; i < idx; i++)
2543 bh = bh->b_this_page;
2545 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2550 static int ext4_da_write_end(struct file *file,
2551 struct address_space *mapping,
2552 loff_t pos, unsigned len, unsigned copied,
2553 struct page *page, void *fsdata)
2555 struct inode *inode = mapping->host;
2557 handle_t *handle = ext4_journal_current_handle();
2559 unsigned long start, end;
2560 int write_mode = (int)(unsigned long)fsdata;
2562 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2563 switch (ext4_inode_journal_mode(inode)) {
2564 case EXT4_INODE_ORDERED_DATA_MODE:
2565 return ext4_ordered_write_end(file, mapping, pos,
2566 len, copied, page, fsdata);
2567 case EXT4_INODE_WRITEBACK_DATA_MODE:
2568 return ext4_writeback_write_end(file, mapping, pos,
2569 len, copied, page, fsdata);
2575 trace_ext4_da_write_end(inode, pos, len, copied);
2576 start = pos & (PAGE_CACHE_SIZE - 1);
2577 end = start + copied - 1;
2580 * generic_write_end() will run mark_inode_dirty() if i_size
2581 * changes. So let's piggyback the i_disksize mark_inode_dirty
2585 new_i_size = pos + copied;
2586 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2587 if (ext4_da_should_update_i_disksize(page, end)) {
2588 down_write(&EXT4_I(inode)->i_data_sem);
2589 if (new_i_size > EXT4_I(inode)->i_disksize)
2590 EXT4_I(inode)->i_disksize = new_i_size;
2591 up_write(&EXT4_I(inode)->i_data_sem);
2592 /* We need to mark inode dirty even if
2593 * new_i_size is less that inode->i_size
2594 * bu greater than i_disksize.(hint delalloc)
2596 ext4_mark_inode_dirty(handle, inode);
2599 ret2 = generic_write_end(file, mapping, pos, len, copied,
2604 ret2 = ext4_journal_stop(handle);
2608 return ret ? ret : copied;
2611 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2614 * Drop reserved blocks
2616 BUG_ON(!PageLocked(page));
2617 if (!page_has_buffers(page))
2620 ext4_da_page_release_reservation(page, offset);
2623 ext4_invalidatepage(page, offset);
2629 * Force all delayed allocation blocks to be allocated for a given inode.
2631 int ext4_alloc_da_blocks(struct inode *inode)
2633 trace_ext4_alloc_da_blocks(inode);
2635 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2636 !EXT4_I(inode)->i_reserved_meta_blocks)
2640 * We do something simple for now. The filemap_flush() will
2641 * also start triggering a write of the data blocks, which is
2642 * not strictly speaking necessary (and for users of
2643 * laptop_mode, not even desirable). However, to do otherwise
2644 * would require replicating code paths in:
2646 * ext4_da_writepages() ->
2647 * write_cache_pages() ---> (via passed in callback function)
2648 * __mpage_da_writepage() -->
2649 * mpage_add_bh_to_extent()
2650 * mpage_da_map_blocks()
2652 * The problem is that write_cache_pages(), located in
2653 * mm/page-writeback.c, marks pages clean in preparation for
2654 * doing I/O, which is not desirable if we're not planning on
2657 * We could call write_cache_pages(), and then redirty all of
2658 * the pages by calling redirty_page_for_writepage() but that
2659 * would be ugly in the extreme. So instead we would need to
2660 * replicate parts of the code in the above functions,
2661 * simplifying them because we wouldn't actually intend to
2662 * write out the pages, but rather only collect contiguous
2663 * logical block extents, call the multi-block allocator, and
2664 * then update the buffer heads with the block allocations.
2666 * For now, though, we'll cheat by calling filemap_flush(),
2667 * which will map the blocks, and start the I/O, but not
2668 * actually wait for the I/O to complete.
2670 return filemap_flush(inode->i_mapping);
2674 * bmap() is special. It gets used by applications such as lilo and by
2675 * the swapper to find the on-disk block of a specific piece of data.
2677 * Naturally, this is dangerous if the block concerned is still in the
2678 * journal. If somebody makes a swapfile on an ext4 data-journaling
2679 * filesystem and enables swap, then they may get a nasty shock when the
2680 * data getting swapped to that swapfile suddenly gets overwritten by
2681 * the original zero's written out previously to the journal and
2682 * awaiting writeback in the kernel's buffer cache.
2684 * So, if we see any bmap calls here on a modified, data-journaled file,
2685 * take extra steps to flush any blocks which might be in the cache.
2687 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2689 struct inode *inode = mapping->host;
2694 * We can get here for an inline file via the FIBMAP ioctl
2696 if (ext4_has_inline_data(inode))
2699 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2700 test_opt(inode->i_sb, DELALLOC)) {
2702 * With delalloc we want to sync the file
2703 * so that we can make sure we allocate
2706 filemap_write_and_wait(mapping);
2709 if (EXT4_JOURNAL(inode) &&
2710 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2712 * This is a REALLY heavyweight approach, but the use of
2713 * bmap on dirty files is expected to be extremely rare:
2714 * only if we run lilo or swapon on a freshly made file
2715 * do we expect this to happen.
2717 * (bmap requires CAP_SYS_RAWIO so this does not
2718 * represent an unprivileged user DOS attack --- we'd be
2719 * in trouble if mortal users could trigger this path at
2722 * NB. EXT4_STATE_JDATA is not set on files other than
2723 * regular files. If somebody wants to bmap a directory
2724 * or symlink and gets confused because the buffer
2725 * hasn't yet been flushed to disk, they deserve
2726 * everything they get.
2729 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2730 journal = EXT4_JOURNAL(inode);
2731 jbd2_journal_lock_updates(journal);
2732 err = jbd2_journal_flush(journal);
2733 jbd2_journal_unlock_updates(journal);
2739 return generic_block_bmap(mapping, block, ext4_get_block);
2742 static int ext4_readpage(struct file *file, struct page *page)
2745 struct inode *inode = page->mapping->host;
2747 trace_ext4_readpage(page);
2749 if (ext4_has_inline_data(inode))
2750 ret = ext4_readpage_inline(inode, page);
2753 return mpage_readpage(page, ext4_get_block);
2759 ext4_readpages(struct file *file, struct address_space *mapping,
2760 struct list_head *pages, unsigned nr_pages)
2762 struct inode *inode = mapping->host;
2764 /* If the file has inline data, no need to do readpages. */
2765 if (ext4_has_inline_data(inode))
2768 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2771 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2773 struct buffer_head *head, *bh;
2774 unsigned int curr_off = 0;
2776 if (!page_has_buffers(page))
2778 head = bh = page_buffers(page);
2780 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2782 ext4_free_io_end(bh->b_private);
2783 bh->b_private = NULL;
2784 bh->b_end_io = NULL;
2786 curr_off = curr_off + bh->b_size;
2787 bh = bh->b_this_page;
2788 } while (bh != head);
2791 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2793 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2795 trace_ext4_invalidatepage(page, offset);
2798 * free any io_end structure allocated for buffers to be discarded
2800 if (ext4_should_dioread_nolock(page->mapping->host))
2801 ext4_invalidatepage_free_endio(page, offset);
2803 * If it's a full truncate we just forget about the pending dirtying
2806 ClearPageChecked(page);
2809 jbd2_journal_invalidatepage(journal, page, offset);
2811 block_invalidatepage(page, offset);
2814 static int ext4_releasepage(struct page *page, gfp_t wait)
2816 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2818 trace_ext4_releasepage(page);
2820 WARN_ON(PageChecked(page));
2821 if (!page_has_buffers(page))
2824 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2826 return try_to_free_buffers(page);
2830 * ext4_get_block used when preparing for a DIO write or buffer write.
2831 * We allocate an uinitialized extent if blocks haven't been allocated.
2832 * The extent will be converted to initialized after the IO is complete.
2834 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2835 struct buffer_head *bh_result, int create)
2837 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2838 inode->i_ino, create);
2839 return _ext4_get_block(inode, iblock, bh_result,
2840 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2843 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2844 struct buffer_head *bh_result, int create)
2846 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2847 inode->i_ino, create);
2848 return _ext4_get_block(inode, iblock, bh_result,
2849 EXT4_GET_BLOCKS_NO_LOCK);
2852 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2853 ssize_t size, void *private, int ret,
2856 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2857 ext4_io_end_t *io_end = iocb->private;
2859 /* if not async direct IO or dio with 0 bytes write, just return */
2860 if (!io_end || !size)
2863 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2864 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2865 iocb->private, io_end->inode->i_ino, iocb, offset,
2868 iocb->private = NULL;
2870 /* if not aio dio with unwritten extents, just free io and return */
2871 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2872 ext4_free_io_end(io_end);
2875 aio_complete(iocb, ret, 0);
2876 inode_dio_done(inode);
2880 io_end->offset = offset;
2881 io_end->size = size;
2883 io_end->iocb = iocb;
2884 io_end->result = ret;
2887 ext4_add_complete_io(io_end);
2890 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2892 ext4_io_end_t *io_end = bh->b_private;
2893 struct inode *inode;
2895 if (!test_clear_buffer_uninit(bh) || !io_end)
2898 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2899 ext4_msg(io_end->inode->i_sb, KERN_INFO,
2900 "sb umounted, discard end_io request for inode %lu",
2901 io_end->inode->i_ino);
2902 ext4_free_io_end(io_end);
2907 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2908 * but being more careful is always safe for the future change.
2910 inode = io_end->inode;
2911 ext4_set_io_unwritten_flag(inode, io_end);
2912 ext4_add_complete_io(io_end);
2914 bh->b_private = NULL;
2915 bh->b_end_io = NULL;
2916 clear_buffer_uninit(bh);
2917 end_buffer_async_write(bh, uptodate);
2920 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2922 ext4_io_end_t *io_end;
2923 struct page *page = bh->b_page;
2924 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2925 size_t size = bh->b_size;
2928 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2930 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2934 io_end->offset = offset;
2935 io_end->size = size;
2937 * We need to hold a reference to the page to make sure it
2938 * doesn't get evicted before ext4_end_io_work() has a chance
2939 * to convert the extent from written to unwritten.
2941 io_end->page = page;
2942 get_page(io_end->page);
2944 bh->b_private = io_end;
2945 bh->b_end_io = ext4_end_io_buffer_write;
2950 * For ext4 extent files, ext4 will do direct-io write to holes,
2951 * preallocated extents, and those write extend the file, no need to
2952 * fall back to buffered IO.
2954 * For holes, we fallocate those blocks, mark them as uninitialized
2955 * If those blocks were preallocated, we mark sure they are split, but
2956 * still keep the range to write as uninitialized.
2958 * The unwritten extents will be converted to written when DIO is completed.
2959 * For async direct IO, since the IO may still pending when return, we
2960 * set up an end_io call back function, which will do the conversion
2961 * when async direct IO completed.
2963 * If the O_DIRECT write will extend the file then add this inode to the
2964 * orphan list. So recovery will truncate it back to the original size
2965 * if the machine crashes during the write.
2968 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2969 const struct iovec *iov, loff_t offset,
2970 unsigned long nr_segs)
2972 struct file *file = iocb->ki_filp;
2973 struct inode *inode = file->f_mapping->host;
2975 size_t count = iov_length(iov, nr_segs);
2977 get_block_t *get_block_func = NULL;
2979 loff_t final_size = offset + count;
2981 /* Use the old path for reads and writes beyond i_size. */
2982 if (rw != WRITE || final_size > inode->i_size)
2983 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2985 BUG_ON(iocb->private == NULL);
2987 /* If we do a overwrite dio, i_mutex locking can be released */
2988 overwrite = *((int *)iocb->private);
2991 atomic_inc(&inode->i_dio_count);
2992 down_read(&EXT4_I(inode)->i_data_sem);
2993 mutex_unlock(&inode->i_mutex);
2997 * We could direct write to holes and fallocate.
2999 * Allocated blocks to fill the hole are marked as
3000 * uninitialized to prevent parallel buffered read to expose
3001 * the stale data before DIO complete the data IO.
3003 * As to previously fallocated extents, ext4 get_block will
3004 * just simply mark the buffer mapped but still keep the
3005 * extents uninitialized.
3007 * For non AIO case, we will convert those unwritten extents
3008 * to written after return back from blockdev_direct_IO.
3010 * For async DIO, the conversion needs to be deferred when the
3011 * IO is completed. The ext4 end_io callback function will be
3012 * called to take care of the conversion work. Here for async
3013 * case, we allocate an io_end structure to hook to the iocb.
3015 iocb->private = NULL;
3016 ext4_inode_aio_set(inode, NULL);
3017 if (!is_sync_kiocb(iocb)) {
3018 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3023 io_end->flag |= EXT4_IO_END_DIRECT;
3024 iocb->private = io_end;
3026 * we save the io structure for current async direct
3027 * IO, so that later ext4_map_blocks() could flag the
3028 * io structure whether there is a unwritten extents
3029 * needs to be converted when IO is completed.
3031 ext4_inode_aio_set(inode, io_end);
3035 get_block_func = ext4_get_block_write_nolock;
3037 get_block_func = ext4_get_block_write;
3038 dio_flags = DIO_LOCKING;
3040 ret = __blockdev_direct_IO(rw, iocb, inode,
3041 inode->i_sb->s_bdev, iov,
3049 ext4_inode_aio_set(inode, NULL);
3051 * The io_end structure takes a reference to the inode, that
3052 * structure needs to be destroyed and the reference to the
3053 * inode need to be dropped, when IO is complete, even with 0
3054 * byte write, or failed.
3056 * In the successful AIO DIO case, the io_end structure will
3057 * be destroyed and the reference to the inode will be dropped
3058 * after the end_io call back function is called.
3060 * In the case there is 0 byte write, or error case, since VFS
3061 * direct IO won't invoke the end_io call back function, we
3062 * need to free the end_io structure here.
3064 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3065 ext4_free_io_end(iocb->private);
3066 iocb->private = NULL;
3067 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3068 EXT4_STATE_DIO_UNWRITTEN)) {
3071 * for non AIO case, since the IO is already
3072 * completed, we could do the conversion right here
3074 err = ext4_convert_unwritten_extents(inode,
3078 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3082 /* take i_mutex locking again if we do a ovewrite dio */
3084 inode_dio_done(inode);
3085 up_read(&EXT4_I(inode)->i_data_sem);
3086 mutex_lock(&inode->i_mutex);
3092 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3093 const struct iovec *iov, loff_t offset,
3094 unsigned long nr_segs)
3096 struct file *file = iocb->ki_filp;
3097 struct inode *inode = file->f_mapping->host;
3101 * If we are doing data journalling we don't support O_DIRECT
3103 if (ext4_should_journal_data(inode))
3106 /* Let buffer I/O handle the inline data case. */
3107 if (ext4_has_inline_data(inode))
3110 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3111 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3112 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3114 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3115 trace_ext4_direct_IO_exit(inode, offset,
3116 iov_length(iov, nr_segs), rw, ret);
3121 * Pages can be marked dirty completely asynchronously from ext4's journalling
3122 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3123 * much here because ->set_page_dirty is called under VFS locks. The page is
3124 * not necessarily locked.
3126 * We cannot just dirty the page and leave attached buffers clean, because the
3127 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3128 * or jbddirty because all the journalling code will explode.
3130 * So what we do is to mark the page "pending dirty" and next time writepage
3131 * is called, propagate that into the buffers appropriately.
3133 static int ext4_journalled_set_page_dirty(struct page *page)
3135 SetPageChecked(page);
3136 return __set_page_dirty_nobuffers(page);
3139 static const struct address_space_operations ext4_ordered_aops = {
3140 .readpage = ext4_readpage,
3141 .readpages = ext4_readpages,
3142 .writepage = ext4_writepage,
3143 .write_begin = ext4_write_begin,
3144 .write_end = ext4_ordered_write_end,
3146 .invalidatepage = ext4_invalidatepage,
3147 .releasepage = ext4_releasepage,
3148 .direct_IO = ext4_direct_IO,
3149 .migratepage = buffer_migrate_page,
3150 .is_partially_uptodate = block_is_partially_uptodate,
3151 .error_remove_page = generic_error_remove_page,
3154 static const struct address_space_operations ext4_writeback_aops = {
3155 .readpage = ext4_readpage,
3156 .readpages = ext4_readpages,
3157 .writepage = ext4_writepage,
3158 .write_begin = ext4_write_begin,
3159 .write_end = ext4_writeback_write_end,
3161 .invalidatepage = ext4_invalidatepage,
3162 .releasepage = ext4_releasepage,
3163 .direct_IO = ext4_direct_IO,
3164 .migratepage = buffer_migrate_page,
3165 .is_partially_uptodate = block_is_partially_uptodate,
3166 .error_remove_page = generic_error_remove_page,
3169 static const struct address_space_operations ext4_journalled_aops = {
3170 .readpage = ext4_readpage,
3171 .readpages = ext4_readpages,
3172 .writepage = ext4_writepage,
3173 .write_begin = ext4_write_begin,
3174 .write_end = ext4_journalled_write_end,
3175 .set_page_dirty = ext4_journalled_set_page_dirty,
3177 .invalidatepage = ext4_invalidatepage,
3178 .releasepage = ext4_releasepage,
3179 .direct_IO = ext4_direct_IO,
3180 .is_partially_uptodate = block_is_partially_uptodate,
3181 .error_remove_page = generic_error_remove_page,
3184 static const struct address_space_operations ext4_da_aops = {
3185 .readpage = ext4_readpage,
3186 .readpages = ext4_readpages,
3187 .writepage = ext4_writepage,
3188 .writepages = ext4_da_writepages,
3189 .write_begin = ext4_da_write_begin,
3190 .write_end = ext4_da_write_end,
3192 .invalidatepage = ext4_da_invalidatepage,
3193 .releasepage = ext4_releasepage,
3194 .direct_IO = ext4_direct_IO,
3195 .migratepage = buffer_migrate_page,
3196 .is_partially_uptodate = block_is_partially_uptodate,
3197 .error_remove_page = generic_error_remove_page,
3200 void ext4_set_aops(struct inode *inode)
3202 switch (ext4_inode_journal_mode(inode)) {
3203 case EXT4_INODE_ORDERED_DATA_MODE:
3204 if (test_opt(inode->i_sb, DELALLOC))
3205 inode->i_mapping->a_ops = &ext4_da_aops;
3207 inode->i_mapping->a_ops = &ext4_ordered_aops;
3209 case EXT4_INODE_WRITEBACK_DATA_MODE:
3210 if (test_opt(inode->i_sb, DELALLOC))
3211 inode->i_mapping->a_ops = &ext4_da_aops;
3213 inode->i_mapping->a_ops = &ext4_writeback_aops;
3215 case EXT4_INODE_JOURNAL_DATA_MODE:
3216 inode->i_mapping->a_ops = &ext4_journalled_aops;
3225 * ext4_discard_partial_page_buffers()
3226 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3227 * This function finds and locks the page containing the offset
3228 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3229 * Calling functions that already have the page locked should call
3230 * ext4_discard_partial_page_buffers_no_lock directly.
3232 int ext4_discard_partial_page_buffers(handle_t *handle,
3233 struct address_space *mapping, loff_t from,
3234 loff_t length, int flags)
3236 struct inode *inode = mapping->host;
3240 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3241 mapping_gfp_mask(mapping) & ~__GFP_FS);
3245 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3246 from, length, flags);
3249 page_cache_release(page);
3254 * ext4_discard_partial_page_buffers_no_lock()
3255 * Zeros a page range of length 'length' starting from offset 'from'.
3256 * Buffer heads that correspond to the block aligned regions of the
3257 * zeroed range will be unmapped. Unblock aligned regions
3258 * will have the corresponding buffer head mapped if needed so that
3259 * that region of the page can be updated with the partial zero out.
3261 * This function assumes that the page has already been locked. The
3262 * The range to be discarded must be contained with in the given page.
3263 * If the specified range exceeds the end of the page it will be shortened
3264 * to the end of the page that corresponds to 'from'. This function is
3265 * appropriate for updating a page and it buffer heads to be unmapped and
3266 * zeroed for blocks that have been either released, or are going to be
3269 * handle: The journal handle
3270 * inode: The files inode
3271 * page: A locked page that contains the offset "from"
3272 * from: The starting byte offset (from the beginning of the file)
3273 * to begin discarding
3274 * len: The length of bytes to discard
3275 * flags: Optional flags that may be used:
3277 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3278 * Only zero the regions of the page whose buffer heads
3279 * have already been unmapped. This flag is appropriate
3280 * for updating the contents of a page whose blocks may
3281 * have already been released, and we only want to zero
3282 * out the regions that correspond to those released blocks.
3284 * Returns zero on success or negative on failure.
3286 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3287 struct inode *inode, struct page *page, loff_t from,
3288 loff_t length, int flags)
3290 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3291 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3292 unsigned int blocksize, max, pos;
3294 struct buffer_head *bh;
3297 blocksize = inode->i_sb->s_blocksize;
3298 max = PAGE_CACHE_SIZE - offset;
3300 if (index != page->index)
3304 * correct length if it does not fall between
3305 * 'from' and the end of the page
3307 if (length > max || length < 0)
3310 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3312 if (!page_has_buffers(page))
3313 create_empty_buffers(page, blocksize, 0);
3315 /* Find the buffer that contains "offset" */
3316 bh = page_buffers(page);
3318 while (offset >= pos) {
3319 bh = bh->b_this_page;
3325 while (pos < offset + length) {
3326 unsigned int end_of_block, range_to_discard;
3330 /* The length of space left to zero and unmap */
3331 range_to_discard = offset + length - pos;
3333 /* The length of space until the end of the block */
3334 end_of_block = blocksize - (pos & (blocksize-1));
3337 * Do not unmap or zero past end of block
3338 * for this buffer head
3340 if (range_to_discard > end_of_block)
3341 range_to_discard = end_of_block;
3345 * Skip this buffer head if we are only zeroing unampped
3346 * regions of the page
3348 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3352 /* If the range is block aligned, unmap */
3353 if (range_to_discard == blocksize) {
3354 clear_buffer_dirty(bh);
3356 clear_buffer_mapped(bh);
3357 clear_buffer_req(bh);
3358 clear_buffer_new(bh);
3359 clear_buffer_delay(bh);
3360 clear_buffer_unwritten(bh);
3361 clear_buffer_uptodate(bh);
3362 zero_user(page, pos, range_to_discard);
3363 BUFFER_TRACE(bh, "Buffer discarded");
3368 * If this block is not completely contained in the range
3369 * to be discarded, then it is not going to be released. Because
3370 * we need to keep this block, we need to make sure this part
3371 * of the page is uptodate before we modify it by writeing
3372 * partial zeros on it.
3374 if (!buffer_mapped(bh)) {
3376 * Buffer head must be mapped before we can read
3379 BUFFER_TRACE(bh, "unmapped");
3380 ext4_get_block(inode, iblock, bh, 0);
3381 /* unmapped? It's a hole - nothing to do */
3382 if (!buffer_mapped(bh)) {
3383 BUFFER_TRACE(bh, "still unmapped");
3388 /* Ok, it's mapped. Make sure it's up-to-date */
3389 if (PageUptodate(page))
3390 set_buffer_uptodate(bh);
3392 if (!buffer_uptodate(bh)) {
3394 ll_rw_block(READ, 1, &bh);
3396 /* Uhhuh. Read error. Complain and punt.*/
3397 if (!buffer_uptodate(bh))
3401 if (ext4_should_journal_data(inode)) {
3402 BUFFER_TRACE(bh, "get write access");
3403 err = ext4_journal_get_write_access(handle, bh);
3408 zero_user(page, pos, range_to_discard);
3411 if (ext4_should_journal_data(inode)) {
3412 err = ext4_handle_dirty_metadata(handle, inode, bh);
3414 mark_buffer_dirty(bh);
3416 BUFFER_TRACE(bh, "Partial buffer zeroed");
3418 bh = bh->b_this_page;
3420 pos += range_to_discard;
3426 int ext4_can_truncate(struct inode *inode)
3428 if (S_ISREG(inode->i_mode))
3430 if (S_ISDIR(inode->i_mode))
3432 if (S_ISLNK(inode->i_mode))
3433 return !ext4_inode_is_fast_symlink(inode);
3438 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3439 * associated with the given offset and length
3441 * @inode: File inode
3442 * @offset: The offset where the hole will begin
3443 * @len: The length of the hole
3445 * Returns: 0 on success or negative on failure
3448 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3450 struct inode *inode = file->f_path.dentry->d_inode;
3451 if (!S_ISREG(inode->i_mode))
3454 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3455 /* TODO: Add support for non extent hole punching */
3459 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3460 /* TODO: Add support for bigalloc file systems */
3464 return ext4_ext_punch_hole(file, offset, length);
3470 * We block out ext4_get_block() block instantiations across the entire
3471 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3472 * simultaneously on behalf of the same inode.
3474 * As we work through the truncate and commit bits of it to the journal there
3475 * is one core, guiding principle: the file's tree must always be consistent on
3476 * disk. We must be able to restart the truncate after a crash.
3478 * The file's tree may be transiently inconsistent in memory (although it
3479 * probably isn't), but whenever we close off and commit a journal transaction,
3480 * the contents of (the filesystem + the journal) must be consistent and
3481 * restartable. It's pretty simple, really: bottom up, right to left (although
3482 * left-to-right works OK too).
3484 * Note that at recovery time, journal replay occurs *before* the restart of
3485 * truncate against the orphan inode list.
3487 * The committed inode has the new, desired i_size (which is the same as
3488 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3489 * that this inode's truncate did not complete and it will again call
3490 * ext4_truncate() to have another go. So there will be instantiated blocks
3491 * to the right of the truncation point in a crashed ext4 filesystem. But
3492 * that's fine - as long as they are linked from the inode, the post-crash
3493 * ext4_truncate() run will find them and release them.
3495 void ext4_truncate(struct inode *inode)
3497 trace_ext4_truncate_enter(inode);
3499 if (!ext4_can_truncate(inode))
3502 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3504 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3505 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3507 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3508 ext4_ext_truncate(inode);
3510 ext4_ind_truncate(inode);
3512 trace_ext4_truncate_exit(inode);
3516 * ext4_get_inode_loc returns with an extra refcount against the inode's
3517 * underlying buffer_head on success. If 'in_mem' is true, we have all
3518 * data in memory that is needed to recreate the on-disk version of this
3521 static int __ext4_get_inode_loc(struct inode *inode,
3522 struct ext4_iloc *iloc, int in_mem)
3524 struct ext4_group_desc *gdp;
3525 struct buffer_head *bh;
3526 struct super_block *sb = inode->i_sb;
3528 int inodes_per_block, inode_offset;
3531 if (!ext4_valid_inum(sb, inode->i_ino))
3534 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3535 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3540 * Figure out the offset within the block group inode table
3542 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3543 inode_offset = ((inode->i_ino - 1) %
3544 EXT4_INODES_PER_GROUP(sb));
3545 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3546 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3548 bh = sb_getblk(sb, block);
3550 EXT4_ERROR_INODE_BLOCK(inode, block,
3551 "unable to read itable block");
3554 if (!buffer_uptodate(bh)) {
3558 * If the buffer has the write error flag, we have failed
3559 * to write out another inode in the same block. In this
3560 * case, we don't have to read the block because we may
3561 * read the old inode data successfully.
3563 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3564 set_buffer_uptodate(bh);
3566 if (buffer_uptodate(bh)) {
3567 /* someone brought it uptodate while we waited */
3573 * If we have all information of the inode in memory and this
3574 * is the only valid inode in the block, we need not read the
3578 struct buffer_head *bitmap_bh;
3581 start = inode_offset & ~(inodes_per_block - 1);
3583 /* Is the inode bitmap in cache? */
3584 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3589 * If the inode bitmap isn't in cache then the
3590 * optimisation may end up performing two reads instead
3591 * of one, so skip it.
3593 if (!buffer_uptodate(bitmap_bh)) {
3597 for (i = start; i < start + inodes_per_block; i++) {
3598 if (i == inode_offset)
3600 if (ext4_test_bit(i, bitmap_bh->b_data))
3604 if (i == start + inodes_per_block) {
3605 /* all other inodes are free, so skip I/O */
3606 memset(bh->b_data, 0, bh->b_size);
3607 set_buffer_uptodate(bh);
3615 * If we need to do any I/O, try to pre-readahead extra
3616 * blocks from the inode table.
3618 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3619 ext4_fsblk_t b, end, table;
3622 table = ext4_inode_table(sb, gdp);
3623 /* s_inode_readahead_blks is always a power of 2 */
3624 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3627 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3628 num = EXT4_INODES_PER_GROUP(sb);
3629 if (ext4_has_group_desc_csum(sb))
3630 num -= ext4_itable_unused_count(sb, gdp);
3631 table += num / inodes_per_block;
3635 sb_breadahead(sb, b++);
3639 * There are other valid inodes in the buffer, this inode
3640 * has in-inode xattrs, or we don't have this inode in memory.
3641 * Read the block from disk.
3643 trace_ext4_load_inode(inode);
3645 bh->b_end_io = end_buffer_read_sync;
3646 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3648 if (!buffer_uptodate(bh)) {
3649 EXT4_ERROR_INODE_BLOCK(inode, block,
3650 "unable to read itable block");
3660 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3662 /* We have all inode data except xattrs in memory here. */
3663 return __ext4_get_inode_loc(inode, iloc,
3664 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3667 void ext4_set_inode_flags(struct inode *inode)
3669 unsigned int flags = EXT4_I(inode)->i_flags;
3671 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3672 if (flags & EXT4_SYNC_FL)
3673 inode->i_flags |= S_SYNC;
3674 if (flags & EXT4_APPEND_FL)
3675 inode->i_flags |= S_APPEND;
3676 if (flags & EXT4_IMMUTABLE_FL)
3677 inode->i_flags |= S_IMMUTABLE;
3678 if (flags & EXT4_NOATIME_FL)
3679 inode->i_flags |= S_NOATIME;
3680 if (flags & EXT4_DIRSYNC_FL)
3681 inode->i_flags |= S_DIRSYNC;
3684 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3685 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3687 unsigned int vfs_fl;
3688 unsigned long old_fl, new_fl;
3691 vfs_fl = ei->vfs_inode.i_flags;
3692 old_fl = ei->i_flags;
3693 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3694 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3696 if (vfs_fl & S_SYNC)
3697 new_fl |= EXT4_SYNC_FL;
3698 if (vfs_fl & S_APPEND)
3699 new_fl |= EXT4_APPEND_FL;
3700 if (vfs_fl & S_IMMUTABLE)
3701 new_fl |= EXT4_IMMUTABLE_FL;
3702 if (vfs_fl & S_NOATIME)
3703 new_fl |= EXT4_NOATIME_FL;
3704 if (vfs_fl & S_DIRSYNC)
3705 new_fl |= EXT4_DIRSYNC_FL;
3706 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3709 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3710 struct ext4_inode_info *ei)
3713 struct inode *inode = &(ei->vfs_inode);
3714 struct super_block *sb = inode->i_sb;
3716 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3717 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3718 /* we are using combined 48 bit field */
3719 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3720 le32_to_cpu(raw_inode->i_blocks_lo);
3721 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3722 /* i_blocks represent file system block size */
3723 return i_blocks << (inode->i_blkbits - 9);
3728 return le32_to_cpu(raw_inode->i_blocks_lo);
3732 static inline void ext4_iget_extra_inode(struct inode *inode,
3733 struct ext4_inode *raw_inode,
3734 struct ext4_inode_info *ei)
3736 __le32 *magic = (void *)raw_inode +
3737 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3738 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3739 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3740 ext4_find_inline_data_nolock(inode);
3744 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3746 struct ext4_iloc iloc;
3747 struct ext4_inode *raw_inode;
3748 struct ext4_inode_info *ei;
3749 struct inode *inode;
3750 journal_t *journal = EXT4_SB(sb)->s_journal;
3756 inode = iget_locked(sb, ino);
3758 return ERR_PTR(-ENOMEM);
3759 if (!(inode->i_state & I_NEW))
3765 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3768 raw_inode = ext4_raw_inode(&iloc);
3770 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3771 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3772 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3773 EXT4_INODE_SIZE(inode->i_sb)) {
3774 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3775 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3776 EXT4_INODE_SIZE(inode->i_sb));
3781 ei->i_extra_isize = 0;
3783 /* Precompute checksum seed for inode metadata */
3784 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3785 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3786 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3788 __le32 inum = cpu_to_le32(inode->i_ino);
3789 __le32 gen = raw_inode->i_generation;
3790 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3792 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3796 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3797 EXT4_ERROR_INODE(inode, "checksum invalid");
3802 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3803 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3804 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3805 if (!(test_opt(inode->i_sb, NO_UID32))) {
3806 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3807 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3809 i_uid_write(inode, i_uid);
3810 i_gid_write(inode, i_gid);
3811 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3813 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3814 ei->i_inline_off = 0;
3815 ei->i_dir_start_lookup = 0;
3816 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3817 /* We now have enough fields to check if the inode was active or not.
3818 * This is needed because nfsd might try to access dead inodes
3819 * the test is that same one that e2fsck uses
3820 * NeilBrown 1999oct15
3822 if (inode->i_nlink == 0) {
3823 if (inode->i_mode == 0 ||
3824 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3825 /* this inode is deleted */
3829 /* The only unlinked inodes we let through here have
3830 * valid i_mode and are being read by the orphan
3831 * recovery code: that's fine, we're about to complete
3832 * the process of deleting those. */
3834 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3835 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3836 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3837 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3839 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3840 inode->i_size = ext4_isize(raw_inode);
3841 ei->i_disksize = inode->i_size;
3843 ei->i_reserved_quota = 0;
3845 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3846 ei->i_block_group = iloc.block_group;
3847 ei->i_last_alloc_group = ~0;
3849 * NOTE! The in-memory inode i_data array is in little-endian order
3850 * even on big-endian machines: we do NOT byteswap the block numbers!
3852 for (block = 0; block < EXT4_N_BLOCKS; block++)
3853 ei->i_data[block] = raw_inode->i_block[block];
3854 INIT_LIST_HEAD(&ei->i_orphan);
3857 * Set transaction id's of transactions that have to be committed
3858 * to finish f[data]sync. We set them to currently running transaction
3859 * as we cannot be sure that the inode or some of its metadata isn't
3860 * part of the transaction - the inode could have been reclaimed and
3861 * now it is reread from disk.
3864 transaction_t *transaction;
3867 read_lock(&journal->j_state_lock);
3868 if (journal->j_running_transaction)
3869 transaction = journal->j_running_transaction;
3871 transaction = journal->j_committing_transaction;
3873 tid = transaction->t_tid;
3875 tid = journal->j_commit_sequence;
3876 read_unlock(&journal->j_state_lock);
3877 ei->i_sync_tid = tid;
3878 ei->i_datasync_tid = tid;
3881 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3882 if (ei->i_extra_isize == 0) {
3883 /* The extra space is currently unused. Use it. */
3884 ei->i_extra_isize = sizeof(struct ext4_inode) -
3885 EXT4_GOOD_OLD_INODE_SIZE;
3887 ext4_iget_extra_inode(inode, raw_inode, ei);
3891 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3892 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3893 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3894 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3896 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3897 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3898 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3900 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3904 if (ei->i_file_acl &&
3905 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3906 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3910 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3911 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3912 (S_ISLNK(inode->i_mode) &&
3913 !ext4_inode_is_fast_symlink(inode)))
3914 /* Validate extent which is part of inode */
3915 ret = ext4_ext_check_inode(inode);
3916 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3917 (S_ISLNK(inode->i_mode) &&
3918 !ext4_inode_is_fast_symlink(inode))) {
3919 /* Validate block references which are part of inode */
3920 ret = ext4_ind_check_inode(inode);
3925 if (S_ISREG(inode->i_mode)) {
3926 inode->i_op = &ext4_file_inode_operations;
3927 inode->i_fop = &ext4_file_operations;
3928 ext4_set_aops(inode);
3929 } else if (S_ISDIR(inode->i_mode)) {
3930 inode->i_op = &ext4_dir_inode_operations;
3931 inode->i_fop = &ext4_dir_operations;
3932 } else if (S_ISLNK(inode->i_mode)) {
3933 if (ext4_inode_is_fast_symlink(inode)) {
3934 inode->i_op = &ext4_fast_symlink_inode_operations;
3935 nd_terminate_link(ei->i_data, inode->i_size,
3936 sizeof(ei->i_data) - 1);
3938 inode->i_op = &ext4_symlink_inode_operations;
3939 ext4_set_aops(inode);
3941 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3942 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3943 inode->i_op = &ext4_special_inode_operations;
3944 if (raw_inode->i_block[0])
3945 init_special_inode(inode, inode->i_mode,
3946 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3948 init_special_inode(inode, inode->i_mode,
3949 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3952 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3956 ext4_set_inode_flags(inode);
3957 unlock_new_inode(inode);
3963 return ERR_PTR(ret);
3966 static int ext4_inode_blocks_set(handle_t *handle,
3967 struct ext4_inode *raw_inode,
3968 struct ext4_inode_info *ei)
3970 struct inode *inode = &(ei->vfs_inode);
3971 u64 i_blocks = inode->i_blocks;
3972 struct super_block *sb = inode->i_sb;
3974 if (i_blocks <= ~0U) {
3976 * i_blocks can be represented in a 32 bit variable
3977 * as multiple of 512 bytes
3979 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3980 raw_inode->i_blocks_high = 0;
3981 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3984 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3987 if (i_blocks <= 0xffffffffffffULL) {
3989 * i_blocks can be represented in a 48 bit variable
3990 * as multiple of 512 bytes
3992 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3993 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3994 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3996 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3997 /* i_block is stored in file system block size */
3998 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3999 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4000 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4006 * Post the struct inode info into an on-disk inode location in the
4007 * buffer-cache. This gobbles the caller's reference to the
4008 * buffer_head in the inode location struct.
4010 * The caller must have write access to iloc->bh.
4012 static int ext4_do_update_inode(handle_t *handle,
4013 struct inode *inode,
4014 struct ext4_iloc *iloc)
4016 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4017 struct ext4_inode_info *ei = EXT4_I(inode);
4018 struct buffer_head *bh = iloc->bh;
4019 int err = 0, rc, block;
4020 int need_datasync = 0;
4024 /* For fields not not tracking in the in-memory inode,
4025 * initialise them to zero for new inodes. */
4026 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4027 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4029 ext4_get_inode_flags(ei);
4030 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4031 i_uid = i_uid_read(inode);
4032 i_gid = i_gid_read(inode);
4033 if (!(test_opt(inode->i_sb, NO_UID32))) {
4034 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4035 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4037 * Fix up interoperability with old kernels. Otherwise, old inodes get
4038 * re-used with the upper 16 bits of the uid/gid intact
4041 raw_inode->i_uid_high =
4042 cpu_to_le16(high_16_bits(i_uid));
4043 raw_inode->i_gid_high =
4044 cpu_to_le16(high_16_bits(i_gid));
4046 raw_inode->i_uid_high = 0;
4047 raw_inode->i_gid_high = 0;
4050 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4051 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4052 raw_inode->i_uid_high = 0;
4053 raw_inode->i_gid_high = 0;
4055 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4057 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4058 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4059 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4060 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4062 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4064 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4065 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4066 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4067 cpu_to_le32(EXT4_OS_HURD))
4068 raw_inode->i_file_acl_high =
4069 cpu_to_le16(ei->i_file_acl >> 32);
4070 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4071 if (ei->i_disksize != ext4_isize(raw_inode)) {
4072 ext4_isize_set(raw_inode, ei->i_disksize);
4075 if (ei->i_disksize > 0x7fffffffULL) {
4076 struct super_block *sb = inode->i_sb;
4077 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4078 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4079 EXT4_SB(sb)->s_es->s_rev_level ==
4080 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4081 /* If this is the first large file
4082 * created, add a flag to the superblock.
4084 err = ext4_journal_get_write_access(handle,
4085 EXT4_SB(sb)->s_sbh);
4088 ext4_update_dynamic_rev(sb);
4089 EXT4_SET_RO_COMPAT_FEATURE(sb,
4090 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4091 ext4_handle_sync(handle);
4092 err = ext4_handle_dirty_super(handle, sb);
4095 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4096 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4097 if (old_valid_dev(inode->i_rdev)) {
4098 raw_inode->i_block[0] =
4099 cpu_to_le32(old_encode_dev(inode->i_rdev));
4100 raw_inode->i_block[1] = 0;
4102 raw_inode->i_block[0] = 0;
4103 raw_inode->i_block[1] =
4104 cpu_to_le32(new_encode_dev(inode->i_rdev));
4105 raw_inode->i_block[2] = 0;
4108 for (block = 0; block < EXT4_N_BLOCKS; block++)
4109 raw_inode->i_block[block] = ei->i_data[block];
4111 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4112 if (ei->i_extra_isize) {
4113 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4114 raw_inode->i_version_hi =
4115 cpu_to_le32(inode->i_version >> 32);
4116 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4119 ext4_inode_csum_set(inode, raw_inode, ei);
4121 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4122 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4125 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4127 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4130 ext4_std_error(inode->i_sb, err);
4135 * ext4_write_inode()
4137 * We are called from a few places:
4139 * - Within generic_file_write() for O_SYNC files.
4140 * Here, there will be no transaction running. We wait for any running
4141 * transaction to commit.
4143 * - Within sys_sync(), kupdate and such.
4144 * We wait on commit, if tol to.
4146 * - Within prune_icache() (PF_MEMALLOC == true)
4147 * Here we simply return. We can't afford to block kswapd on the
4150 * In all cases it is actually safe for us to return without doing anything,
4151 * because the inode has been copied into a raw inode buffer in
4152 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4155 * Note that we are absolutely dependent upon all inode dirtiers doing the
4156 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4157 * which we are interested.
4159 * It would be a bug for them to not do this. The code:
4161 * mark_inode_dirty(inode)
4163 * inode->i_size = expr;
4165 * is in error because a kswapd-driven write_inode() could occur while
4166 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4167 * will no longer be on the superblock's dirty inode list.
4169 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4173 if (current->flags & PF_MEMALLOC)
4176 if (EXT4_SB(inode->i_sb)->s_journal) {
4177 if (ext4_journal_current_handle()) {
4178 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4183 if (wbc->sync_mode != WB_SYNC_ALL)
4186 err = ext4_force_commit(inode->i_sb);
4188 struct ext4_iloc iloc;
4190 err = __ext4_get_inode_loc(inode, &iloc, 0);
4193 if (wbc->sync_mode == WB_SYNC_ALL)
4194 sync_dirty_buffer(iloc.bh);
4195 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4196 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4197 "IO error syncing inode");
4208 * Called from notify_change.
4210 * We want to trap VFS attempts to truncate the file as soon as
4211 * possible. In particular, we want to make sure that when the VFS
4212 * shrinks i_size, we put the inode on the orphan list and modify
4213 * i_disksize immediately, so that during the subsequent flushing of
4214 * dirty pages and freeing of disk blocks, we can guarantee that any
4215 * commit will leave the blocks being flushed in an unused state on
4216 * disk. (On recovery, the inode will get truncated and the blocks will
4217 * be freed, so we have a strong guarantee that no future commit will
4218 * leave these blocks visible to the user.)
4220 * Another thing we have to assure is that if we are in ordered mode
4221 * and inode is still attached to the committing transaction, we must
4222 * we start writeout of all the dirty pages which are being truncated.
4223 * This way we are sure that all the data written in the previous
4224 * transaction are already on disk (truncate waits for pages under
4227 * Called with inode->i_mutex down.
4229 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4231 struct inode *inode = dentry->d_inode;
4234 const unsigned int ia_valid = attr->ia_valid;
4236 error = inode_change_ok(inode, attr);
4240 if (is_quota_modification(inode, attr))
4241 dquot_initialize(inode);
4242 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4243 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4246 /* (user+group)*(old+new) structure, inode write (sb,
4247 * inode block, ? - but truncate inode update has it) */
4248 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4249 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4250 if (IS_ERR(handle)) {
4251 error = PTR_ERR(handle);
4254 error = dquot_transfer(inode, attr);
4256 ext4_journal_stop(handle);
4259 /* Update corresponding info in inode so that everything is in
4260 * one transaction */
4261 if (attr->ia_valid & ATTR_UID)
4262 inode->i_uid = attr->ia_uid;
4263 if (attr->ia_valid & ATTR_GID)
4264 inode->i_gid = attr->ia_gid;
4265 error = ext4_mark_inode_dirty(handle, inode);
4266 ext4_journal_stop(handle);
4269 if (attr->ia_valid & ATTR_SIZE) {
4271 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4272 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4274 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4279 if (S_ISREG(inode->i_mode) &&
4280 attr->ia_valid & ATTR_SIZE &&
4281 (attr->ia_size < inode->i_size)) {
4284 handle = ext4_journal_start(inode, 3);
4285 if (IS_ERR(handle)) {
4286 error = PTR_ERR(handle);
4289 if (ext4_handle_valid(handle)) {
4290 error = ext4_orphan_add(handle, inode);
4293 EXT4_I(inode)->i_disksize = attr->ia_size;
4294 rc = ext4_mark_inode_dirty(handle, inode);
4297 ext4_journal_stop(handle);
4299 if (ext4_should_order_data(inode)) {
4300 error = ext4_begin_ordered_truncate(inode,
4303 /* Do as much error cleanup as possible */
4304 handle = ext4_journal_start(inode, 3);
4305 if (IS_ERR(handle)) {
4306 ext4_orphan_del(NULL, inode);
4309 ext4_orphan_del(handle, inode);
4311 ext4_journal_stop(handle);
4317 if (attr->ia_valid & ATTR_SIZE) {
4318 if (attr->ia_size != i_size_read(inode)) {
4319 truncate_setsize(inode, attr->ia_size);
4320 /* Inode size will be reduced, wait for dio in flight.
4321 * Temporarily disable dioread_nolock to prevent
4324 ext4_inode_block_unlocked_dio(inode);
4325 inode_dio_wait(inode);
4326 ext4_inode_resume_unlocked_dio(inode);
4329 ext4_truncate(inode);
4333 setattr_copy(inode, attr);
4334 mark_inode_dirty(inode);
4338 * If the call to ext4_truncate failed to get a transaction handle at
4339 * all, we need to clean up the in-core orphan list manually.
4341 if (orphan && inode->i_nlink)
4342 ext4_orphan_del(NULL, inode);
4344 if (!rc && (ia_valid & ATTR_MODE))
4345 rc = ext4_acl_chmod(inode);
4348 ext4_std_error(inode->i_sb, error);
4354 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4357 struct inode *inode;
4358 unsigned long delalloc_blocks;
4360 inode = dentry->d_inode;
4361 generic_fillattr(inode, stat);
4364 * We can't update i_blocks if the block allocation is delayed
4365 * otherwise in the case of system crash before the real block
4366 * allocation is done, we will have i_blocks inconsistent with
4367 * on-disk file blocks.
4368 * We always keep i_blocks updated together with real
4369 * allocation. But to not confuse with user, stat
4370 * will return the blocks that include the delayed allocation
4371 * blocks for this file.
4373 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4374 EXT4_I(inode)->i_reserved_data_blocks);
4376 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4380 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4382 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4383 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4384 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4388 * Account for index blocks, block groups bitmaps and block group
4389 * descriptor blocks if modify datablocks and index blocks
4390 * worse case, the indexs blocks spread over different block groups
4392 * If datablocks are discontiguous, they are possible to spread over
4393 * different block groups too. If they are contiguous, with flexbg,
4394 * they could still across block group boundary.
4396 * Also account for superblock, inode, quota and xattr blocks
4398 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4400 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4406 * How many index blocks need to touch to modify nrblocks?
4407 * The "Chunk" flag indicating whether the nrblocks is
4408 * physically contiguous on disk
4410 * For Direct IO and fallocate, they calls get_block to allocate
4411 * one single extent at a time, so they could set the "Chunk" flag
4413 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4418 * Now let's see how many group bitmaps and group descriptors need
4428 if (groups > ngroups)
4430 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4431 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4433 /* bitmaps and block group descriptor blocks */
4434 ret += groups + gdpblocks;
4436 /* Blocks for super block, inode, quota and xattr blocks */
4437 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4443 * Calculate the total number of credits to reserve to fit
4444 * the modification of a single pages into a single transaction,
4445 * which may include multiple chunks of block allocations.
4447 * This could be called via ext4_write_begin()
4449 * We need to consider the worse case, when
4450 * one new block per extent.
4452 int ext4_writepage_trans_blocks(struct inode *inode)
4454 int bpp = ext4_journal_blocks_per_page(inode);
4457 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4459 /* Account for data blocks for journalled mode */
4460 if (ext4_should_journal_data(inode))
4466 * Calculate the journal credits for a chunk of data modification.
4468 * This is called from DIO, fallocate or whoever calling
4469 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4471 * journal buffers for data blocks are not included here, as DIO
4472 * and fallocate do no need to journal data buffers.
4474 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4476 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4480 * The caller must have previously called ext4_reserve_inode_write().
4481 * Give this, we know that the caller already has write access to iloc->bh.
4483 int ext4_mark_iloc_dirty(handle_t *handle,
4484 struct inode *inode, struct ext4_iloc *iloc)
4488 if (IS_I_VERSION(inode))
4489 inode_inc_iversion(inode);
4491 /* the do_update_inode consumes one bh->b_count */
4494 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4495 err = ext4_do_update_inode(handle, inode, iloc);
4501 * On success, We end up with an outstanding reference count against
4502 * iloc->bh. This _must_ be cleaned up later.
4506 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4507 struct ext4_iloc *iloc)
4511 err = ext4_get_inode_loc(inode, iloc);
4513 BUFFER_TRACE(iloc->bh, "get_write_access");
4514 err = ext4_journal_get_write_access(handle, iloc->bh);
4520 ext4_std_error(inode->i_sb, err);
4525 * Expand an inode by new_extra_isize bytes.
4526 * Returns 0 on success or negative error number on failure.
4528 static int ext4_expand_extra_isize(struct inode *inode,
4529 unsigned int new_extra_isize,
4530 struct ext4_iloc iloc,
4533 struct ext4_inode *raw_inode;
4534 struct ext4_xattr_ibody_header *header;
4536 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4539 raw_inode = ext4_raw_inode(&iloc);
4541 header = IHDR(inode, raw_inode);
4543 /* No extended attributes present */
4544 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4545 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4546 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4548 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4552 /* try to expand with EAs present */
4553 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4558 * What we do here is to mark the in-core inode as clean with respect to inode
4559 * dirtiness (it may still be data-dirty).
4560 * This means that the in-core inode may be reaped by prune_icache
4561 * without having to perform any I/O. This is a very good thing,
4562 * because *any* task may call prune_icache - even ones which
4563 * have a transaction open against a different journal.
4565 * Is this cheating? Not really. Sure, we haven't written the
4566 * inode out, but prune_icache isn't a user-visible syncing function.
4567 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4568 * we start and wait on commits.
4570 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4572 struct ext4_iloc iloc;
4573 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4574 static unsigned int mnt_count;
4578 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4579 err = ext4_reserve_inode_write(handle, inode, &iloc);
4580 if (ext4_handle_valid(handle) &&
4581 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4582 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4584 * We need extra buffer credits since we may write into EA block
4585 * with this same handle. If journal_extend fails, then it will
4586 * only result in a minor loss of functionality for that inode.
4587 * If this is felt to be critical, then e2fsck should be run to
4588 * force a large enough s_min_extra_isize.
4590 if ((jbd2_journal_extend(handle,
4591 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4592 ret = ext4_expand_extra_isize(inode,
4593 sbi->s_want_extra_isize,
4596 ext4_set_inode_state(inode,
4597 EXT4_STATE_NO_EXPAND);
4599 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4600 ext4_warning(inode->i_sb,
4601 "Unable to expand inode %lu. Delete"
4602 " some EAs or run e2fsck.",
4605 le16_to_cpu(sbi->s_es->s_mnt_count);
4611 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4616 * ext4_dirty_inode() is called from __mark_inode_dirty()
4618 * We're really interested in the case where a file is being extended.
4619 * i_size has been changed by generic_commit_write() and we thus need
4620 * to include the updated inode in the current transaction.
4622 * Also, dquot_alloc_block() will always dirty the inode when blocks
4623 * are allocated to the file.
4625 * If the inode is marked synchronous, we don't honour that here - doing
4626 * so would cause a commit on atime updates, which we don't bother doing.
4627 * We handle synchronous inodes at the highest possible level.
4629 void ext4_dirty_inode(struct inode *inode, int flags)
4633 handle = ext4_journal_start(inode, 2);
4637 ext4_mark_inode_dirty(handle, inode);
4639 ext4_journal_stop(handle);
4646 * Bind an inode's backing buffer_head into this transaction, to prevent
4647 * it from being flushed to disk early. Unlike
4648 * ext4_reserve_inode_write, this leaves behind no bh reference and
4649 * returns no iloc structure, so the caller needs to repeat the iloc
4650 * lookup to mark the inode dirty later.
4652 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4654 struct ext4_iloc iloc;
4658 err = ext4_get_inode_loc(inode, &iloc);
4660 BUFFER_TRACE(iloc.bh, "get_write_access");
4661 err = jbd2_journal_get_write_access(handle, iloc.bh);
4663 err = ext4_handle_dirty_metadata(handle,
4669 ext4_std_error(inode->i_sb, err);
4674 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4681 * We have to be very careful here: changing a data block's
4682 * journaling status dynamically is dangerous. If we write a
4683 * data block to the journal, change the status and then delete
4684 * that block, we risk forgetting to revoke the old log record
4685 * from the journal and so a subsequent replay can corrupt data.
4686 * So, first we make sure that the journal is empty and that
4687 * nobody is changing anything.
4690 journal = EXT4_JOURNAL(inode);
4693 if (is_journal_aborted(journal))
4695 /* We have to allocate physical blocks for delalloc blocks
4696 * before flushing journal. otherwise delalloc blocks can not
4697 * be allocated any more. even more truncate on delalloc blocks
4698 * could trigger BUG by flushing delalloc blocks in journal.
4699 * There is no delalloc block in non-journal data mode.
4701 if (val && test_opt(inode->i_sb, DELALLOC)) {
4702 err = ext4_alloc_da_blocks(inode);
4707 /* Wait for all existing dio workers */
4708 ext4_inode_block_unlocked_dio(inode);
4709 inode_dio_wait(inode);
4711 jbd2_journal_lock_updates(journal);
4714 * OK, there are no updates running now, and all cached data is
4715 * synced to disk. We are now in a completely consistent state
4716 * which doesn't have anything in the journal, and we know that
4717 * no filesystem updates are running, so it is safe to modify
4718 * the inode's in-core data-journaling state flag now.
4722 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4724 jbd2_journal_flush(journal);
4725 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4727 ext4_set_aops(inode);
4729 jbd2_journal_unlock_updates(journal);
4730 ext4_inode_resume_unlocked_dio(inode);
4732 /* Finally we can mark the inode as dirty. */
4734 handle = ext4_journal_start(inode, 1);
4736 return PTR_ERR(handle);
4738 err = ext4_mark_inode_dirty(handle, inode);
4739 ext4_handle_sync(handle);
4740 ext4_journal_stop(handle);
4741 ext4_std_error(inode->i_sb, err);
4746 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4748 return !buffer_mapped(bh);
4751 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4753 struct page *page = vmf->page;
4757 struct file *file = vma->vm_file;
4758 struct inode *inode = file->f_path.dentry->d_inode;
4759 struct address_space *mapping = inode->i_mapping;
4761 get_block_t *get_block;
4764 sb_start_pagefault(inode->i_sb);
4765 file_update_time(vma->vm_file);
4766 /* Delalloc case is easy... */
4767 if (test_opt(inode->i_sb, DELALLOC) &&
4768 !ext4_should_journal_data(inode) &&
4769 !ext4_nonda_switch(inode->i_sb)) {
4771 ret = __block_page_mkwrite(vma, vmf,
4772 ext4_da_get_block_prep);
4773 } while (ret == -ENOSPC &&
4774 ext4_should_retry_alloc(inode->i_sb, &retries));
4779 size = i_size_read(inode);
4780 /* Page got truncated from under us? */
4781 if (page->mapping != mapping || page_offset(page) > size) {
4783 ret = VM_FAULT_NOPAGE;
4787 if (page->index == size >> PAGE_CACHE_SHIFT)
4788 len = size & ~PAGE_CACHE_MASK;
4790 len = PAGE_CACHE_SIZE;
4792 * Return if we have all the buffers mapped. This avoids the need to do
4793 * journal_start/journal_stop which can block and take a long time
4795 if (page_has_buffers(page)) {
4796 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4797 ext4_bh_unmapped)) {
4798 /* Wait so that we don't change page under IO */
4799 wait_on_page_writeback(page);
4800 ret = VM_FAULT_LOCKED;
4805 /* OK, we need to fill the hole... */
4806 if (ext4_should_dioread_nolock(inode))
4807 get_block = ext4_get_block_write;
4809 get_block = ext4_get_block;
4811 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4812 if (IS_ERR(handle)) {
4813 ret = VM_FAULT_SIGBUS;
4816 ret = __block_page_mkwrite(vma, vmf, get_block);
4817 if (!ret && ext4_should_journal_data(inode)) {
4818 if (walk_page_buffers(handle, page_buffers(page), 0,
4819 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4821 ret = VM_FAULT_SIGBUS;
4822 ext4_journal_stop(handle);
4825 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4827 ext4_journal_stop(handle);
4828 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4831 ret = block_page_mkwrite_return(ret);
4833 sb_end_pagefault(inode->i_sb);