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 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
489 static void set_buffers_da_mapped(struct inode *inode,
490 struct ext4_map_blocks *map)
492 struct address_space *mapping = inode->i_mapping;
497 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
498 end = (map->m_lblk + map->m_len - 1) >>
499 (PAGE_CACHE_SHIFT - inode->i_blkbits);
501 pagevec_init(&pvec, 0);
502 while (index <= end) {
503 nr_pages = pagevec_lookup(&pvec, mapping, index,
505 (pgoff_t)PAGEVEC_SIZE));
508 for (i = 0; i < nr_pages; i++) {
509 struct page *page = pvec.pages[i];
510 struct buffer_head *bh, *head;
512 if (unlikely(page->mapping != mapping) ||
516 if (page_has_buffers(page)) {
517 bh = head = page_buffers(page);
519 set_buffer_da_mapped(bh);
520 bh = bh->b_this_page;
521 } while (bh != head);
525 pagevec_release(&pvec);
530 * The ext4_map_blocks() function tries to look up the requested blocks,
531 * and returns if the blocks are already mapped.
533 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
534 * and store the allocated blocks in the result buffer head and mark it
537 * If file type is extents based, it will call ext4_ext_map_blocks(),
538 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
541 * On success, it returns the number of blocks being mapped or allocate.
542 * if create==0 and the blocks are pre-allocated and uninitialized block,
543 * the result buffer head is unmapped. If the create ==1, it will make sure
544 * the buffer head is mapped.
546 * It returns 0 if plain look up failed (blocks have not been allocated), in
547 * that case, buffer head is unmapped
549 * It returns the error in case of allocation failure.
551 int ext4_map_blocks(handle_t *handle, struct inode *inode,
552 struct ext4_map_blocks *map, int flags)
557 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
558 "logical block %lu\n", inode->i_ino, flags, map->m_len,
559 (unsigned long) map->m_lblk);
561 * Try to see if we can get the block without requesting a new
564 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
565 down_read((&EXT4_I(inode)->i_data_sem));
566 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
567 retval = ext4_ext_map_blocks(handle, inode, map, flags &
568 EXT4_GET_BLOCKS_KEEP_SIZE);
570 retval = ext4_ind_map_blocks(handle, inode, map, flags &
571 EXT4_GET_BLOCKS_KEEP_SIZE);
573 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
574 up_read((&EXT4_I(inode)->i_data_sem));
576 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
577 int ret = check_block_validity(inode, map);
582 /* If it is only a block(s) look up */
583 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
587 * Returns if the blocks have already allocated
589 * Note that if blocks have been preallocated
590 * ext4_ext_get_block() returns the create = 0
591 * with buffer head unmapped.
593 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
597 * When we call get_blocks without the create flag, the
598 * BH_Unwritten flag could have gotten set if the blocks
599 * requested were part of a uninitialized extent. We need to
600 * clear this flag now that we are committed to convert all or
601 * part of the uninitialized extent to be an initialized
602 * extent. This is because we need to avoid the combination
603 * of BH_Unwritten and BH_Mapped flags being simultaneously
604 * set on the buffer_head.
606 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
609 * New blocks allocate and/or writing to uninitialized extent
610 * will possibly result in updating i_data, so we take
611 * the write lock of i_data_sem, and call get_blocks()
612 * with create == 1 flag.
614 down_write((&EXT4_I(inode)->i_data_sem));
617 * if the caller is from delayed allocation writeout path
618 * we have already reserved fs blocks for allocation
619 * let the underlying get_block() function know to
620 * avoid double accounting
622 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
623 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
625 * We need to check for EXT4 here because migrate
626 * could have changed the inode type in between
628 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
629 retval = ext4_ext_map_blocks(handle, inode, map, flags);
631 retval = ext4_ind_map_blocks(handle, inode, map, flags);
633 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
635 * We allocated new blocks which will result in
636 * i_data's format changing. Force the migrate
637 * to fail by clearing migrate flags
639 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
643 * Update reserved blocks/metadata blocks after successful
644 * block allocation which had been deferred till now. We don't
645 * support fallocate for non extent files. So we can update
646 * reserve space here.
649 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
650 ext4_da_update_reserve_space(inode, retval, 1);
652 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
653 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
655 /* If we have successfully mapped the delayed allocated blocks,
656 * set the BH_Da_Mapped bit on them. Its important to do this
657 * under the protection of i_data_sem.
659 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
660 set_buffers_da_mapped(inode, map);
663 up_write((&EXT4_I(inode)->i_data_sem));
664 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
665 int ret = check_block_validity(inode, map);
672 /* Maximum number of blocks we map for direct IO at once. */
673 #define DIO_MAX_BLOCKS 4096
675 static int _ext4_get_block(struct inode *inode, sector_t iblock,
676 struct buffer_head *bh, int flags)
678 handle_t *handle = ext4_journal_current_handle();
679 struct ext4_map_blocks map;
680 int ret = 0, started = 0;
684 map.m_len = bh->b_size >> inode->i_blkbits;
686 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
687 /* Direct IO write... */
688 if (map.m_len > DIO_MAX_BLOCKS)
689 map.m_len = DIO_MAX_BLOCKS;
690 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
691 handle = ext4_journal_start(inode, dio_credits);
692 if (IS_ERR(handle)) {
693 ret = PTR_ERR(handle);
699 ret = ext4_map_blocks(handle, inode, &map, flags);
701 map_bh(bh, inode->i_sb, map.m_pblk);
702 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
703 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
707 ext4_journal_stop(handle);
711 int ext4_get_block(struct inode *inode, sector_t iblock,
712 struct buffer_head *bh, int create)
714 return _ext4_get_block(inode, iblock, bh,
715 create ? EXT4_GET_BLOCKS_CREATE : 0);
719 * `handle' can be NULL if create is zero
721 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
722 ext4_lblk_t block, int create, int *errp)
724 struct ext4_map_blocks map;
725 struct buffer_head *bh;
728 J_ASSERT(handle != NULL || create == 0);
732 err = ext4_map_blocks(handle, inode, &map,
733 create ? EXT4_GET_BLOCKS_CREATE : 0);
735 /* ensure we send some value back into *errp */
743 bh = sb_getblk(inode->i_sb, map.m_pblk);
748 if (map.m_flags & EXT4_MAP_NEW) {
749 J_ASSERT(create != 0);
750 J_ASSERT(handle != NULL);
753 * Now that we do not always journal data, we should
754 * keep in mind whether this should always journal the
755 * new buffer as metadata. For now, regular file
756 * writes use ext4_get_block instead, so it's not a
760 BUFFER_TRACE(bh, "call get_create_access");
761 fatal = ext4_journal_get_create_access(handle, bh);
762 if (!fatal && !buffer_uptodate(bh)) {
763 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
764 set_buffer_uptodate(bh);
767 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
768 err = ext4_handle_dirty_metadata(handle, inode, bh);
772 BUFFER_TRACE(bh, "not a new buffer");
782 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
783 ext4_lblk_t block, int create, int *err)
785 struct buffer_head *bh;
787 bh = ext4_getblk(handle, inode, block, create, err);
790 if (buffer_uptodate(bh))
792 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
794 if (buffer_uptodate(bh))
801 static int walk_page_buffers(handle_t *handle,
802 struct buffer_head *head,
806 int (*fn)(handle_t *handle,
807 struct buffer_head *bh))
809 struct buffer_head *bh;
810 unsigned block_start, block_end;
811 unsigned blocksize = head->b_size;
813 struct buffer_head *next;
815 for (bh = head, block_start = 0;
816 ret == 0 && (bh != head || !block_start);
817 block_start = block_end, bh = next) {
818 next = bh->b_this_page;
819 block_end = block_start + blocksize;
820 if (block_end <= from || block_start >= to) {
821 if (partial && !buffer_uptodate(bh))
825 err = (*fn)(handle, bh);
833 * To preserve ordering, it is essential that the hole instantiation and
834 * the data write be encapsulated in a single transaction. We cannot
835 * close off a transaction and start a new one between the ext4_get_block()
836 * and the commit_write(). So doing the jbd2_journal_start at the start of
837 * prepare_write() is the right place.
839 * Also, this function can nest inside ext4_writepage() ->
840 * block_write_full_page(). In that case, we *know* that ext4_writepage()
841 * has generated enough buffer credits to do the whole page. So we won't
842 * block on the journal in that case, which is good, because the caller may
845 * By accident, ext4 can be reentered when a transaction is open via
846 * quota file writes. If we were to commit the transaction while thus
847 * reentered, there can be a deadlock - we would be holding a quota
848 * lock, and the commit would never complete if another thread had a
849 * transaction open and was blocking on the quota lock - a ranking
852 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
853 * will _not_ run commit under these circumstances because handle->h_ref
854 * is elevated. We'll still have enough credits for the tiny quotafile
857 static int do_journal_get_write_access(handle_t *handle,
858 struct buffer_head *bh)
860 int dirty = buffer_dirty(bh);
863 if (!buffer_mapped(bh) || buffer_freed(bh))
866 * __block_write_begin() could have dirtied some buffers. Clean
867 * the dirty bit as jbd2_journal_get_write_access() could complain
868 * otherwise about fs integrity issues. Setting of the dirty bit
869 * by __block_write_begin() isn't a real problem here as we clear
870 * the bit before releasing a page lock and thus writeback cannot
871 * ever write the buffer.
874 clear_buffer_dirty(bh);
875 ret = ext4_journal_get_write_access(handle, bh);
877 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
881 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
882 struct buffer_head *bh_result, int create);
883 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
884 struct buffer_head *bh_result, int create);
885 static int ext4_write_begin(struct file *file, struct address_space *mapping,
886 loff_t pos, unsigned len, unsigned flags,
887 struct page **pagep, void **fsdata)
889 struct inode *inode = mapping->host;
890 int ret, needed_blocks;
897 trace_ext4_write_begin(inode, pos, len, flags);
899 * Reserve one block more for addition to orphan list in case
900 * we allocate blocks but write fails for some reason
902 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
903 index = pos >> PAGE_CACHE_SHIFT;
904 from = pos & (PAGE_CACHE_SIZE - 1);
908 handle = ext4_journal_start(inode, needed_blocks);
909 if (IS_ERR(handle)) {
910 ret = PTR_ERR(handle);
914 /* We cannot recurse into the filesystem as the transaction is already
916 flags |= AOP_FLAG_NOFS;
918 page = grab_cache_page_write_begin(mapping, index, flags);
920 ext4_journal_stop(handle);
926 if (ext4_should_dioread_nolock(inode))
927 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
929 ret = __block_write_begin(page, pos, len, ext4_get_block);
931 if (!ret && ext4_should_journal_data(inode)) {
932 ret = walk_page_buffers(handle, page_buffers(page),
933 from, to, NULL, do_journal_get_write_access);
938 page_cache_release(page);
940 * __block_write_begin may have instantiated a few blocks
941 * outside i_size. Trim these off again. Don't need
942 * i_size_read because we hold i_mutex.
944 * Add inode to orphan list in case we crash before
947 if (pos + len > inode->i_size && ext4_can_truncate(inode))
948 ext4_orphan_add(handle, inode);
950 ext4_journal_stop(handle);
951 if (pos + len > inode->i_size) {
952 ext4_truncate_failed_write(inode);
954 * If truncate failed early the inode might
955 * still be on the orphan list; we need to
956 * make sure the inode is removed from the
957 * orphan list in that case.
960 ext4_orphan_del(NULL, inode);
964 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
970 /* For write_end() in data=journal mode */
971 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
973 if (!buffer_mapped(bh) || buffer_freed(bh))
975 set_buffer_uptodate(bh);
976 return ext4_handle_dirty_metadata(handle, NULL, bh);
979 static int ext4_generic_write_end(struct file *file,
980 struct address_space *mapping,
981 loff_t pos, unsigned len, unsigned copied,
982 struct page *page, void *fsdata)
984 int i_size_changed = 0;
985 struct inode *inode = mapping->host;
986 handle_t *handle = ext4_journal_current_handle();
988 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
991 * No need to use i_size_read() here, the i_size
992 * cannot change under us because we hold i_mutex.
994 * But it's important to update i_size while still holding page lock:
995 * page writeout could otherwise come in and zero beyond i_size.
997 if (pos + copied > inode->i_size) {
998 i_size_write(inode, pos + copied);
1002 if (pos + copied > EXT4_I(inode)->i_disksize) {
1003 /* We need to mark inode dirty even if
1004 * new_i_size is less that inode->i_size
1005 * bu greater than i_disksize.(hint delalloc)
1007 ext4_update_i_disksize(inode, (pos + copied));
1011 page_cache_release(page);
1014 * Don't mark the inode dirty under page lock. First, it unnecessarily
1015 * makes the holding time of page lock longer. Second, it forces lock
1016 * ordering of page lock and transaction start for journaling
1020 ext4_mark_inode_dirty(handle, inode);
1026 * We need to pick up the new inode size which generic_commit_write gave us
1027 * `file' can be NULL - eg, when called from page_symlink().
1029 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1030 * buffers are managed internally.
1032 static int ext4_ordered_write_end(struct file *file,
1033 struct address_space *mapping,
1034 loff_t pos, unsigned len, unsigned copied,
1035 struct page *page, void *fsdata)
1037 handle_t *handle = ext4_journal_current_handle();
1038 struct inode *inode = mapping->host;
1041 trace_ext4_ordered_write_end(inode, pos, len, copied);
1042 ret = ext4_jbd2_file_inode(handle, inode);
1045 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1048 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1049 /* if we have allocated more blocks and copied
1050 * less. We will have blocks allocated outside
1051 * inode->i_size. So truncate them
1053 ext4_orphan_add(handle, inode);
1058 page_cache_release(page);
1061 ret2 = ext4_journal_stop(handle);
1065 if (pos + len > inode->i_size) {
1066 ext4_truncate_failed_write(inode);
1068 * If truncate failed early the inode might still be
1069 * on the orphan list; we need to make sure the inode
1070 * is removed from the orphan list in that case.
1073 ext4_orphan_del(NULL, inode);
1077 return ret ? ret : copied;
1080 static int ext4_writeback_write_end(struct file *file,
1081 struct address_space *mapping,
1082 loff_t pos, unsigned len, unsigned copied,
1083 struct page *page, void *fsdata)
1085 handle_t *handle = ext4_journal_current_handle();
1086 struct inode *inode = mapping->host;
1089 trace_ext4_writeback_write_end(inode, pos, len, copied);
1090 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1093 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1094 /* if we have allocated more blocks and copied
1095 * less. We will have blocks allocated outside
1096 * inode->i_size. So truncate them
1098 ext4_orphan_add(handle, inode);
1103 ret2 = ext4_journal_stop(handle);
1107 if (pos + len > inode->i_size) {
1108 ext4_truncate_failed_write(inode);
1110 * If truncate failed early the inode might still be
1111 * on the orphan list; we need to make sure the inode
1112 * is removed from the orphan list in that case.
1115 ext4_orphan_del(NULL, inode);
1118 return ret ? ret : copied;
1121 static int ext4_journalled_write_end(struct file *file,
1122 struct address_space *mapping,
1123 loff_t pos, unsigned len, unsigned copied,
1124 struct page *page, void *fsdata)
1126 handle_t *handle = ext4_journal_current_handle();
1127 struct inode *inode = mapping->host;
1133 trace_ext4_journalled_write_end(inode, pos, len, copied);
1134 from = pos & (PAGE_CACHE_SIZE - 1);
1137 BUG_ON(!ext4_handle_valid(handle));
1140 if (!PageUptodate(page))
1142 page_zero_new_buffers(page, from+copied, to);
1145 ret = walk_page_buffers(handle, page_buffers(page), from,
1146 to, &partial, write_end_fn);
1148 SetPageUptodate(page);
1149 new_i_size = pos + copied;
1150 if (new_i_size > inode->i_size)
1151 i_size_write(inode, pos+copied);
1152 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1153 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1154 if (new_i_size > EXT4_I(inode)->i_disksize) {
1155 ext4_update_i_disksize(inode, new_i_size);
1156 ret2 = ext4_mark_inode_dirty(handle, inode);
1162 page_cache_release(page);
1163 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1164 /* if we have allocated more blocks and copied
1165 * less. We will have blocks allocated outside
1166 * inode->i_size. So truncate them
1168 ext4_orphan_add(handle, inode);
1170 ret2 = ext4_journal_stop(handle);
1173 if (pos + len > inode->i_size) {
1174 ext4_truncate_failed_write(inode);
1176 * If truncate failed early the inode might still be
1177 * on the orphan list; we need to make sure the inode
1178 * is removed from the orphan list in that case.
1181 ext4_orphan_del(NULL, inode);
1184 return ret ? ret : copied;
1188 * Reserve a single cluster located at lblock
1190 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1193 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1194 struct ext4_inode_info *ei = EXT4_I(inode);
1195 unsigned int md_needed;
1197 ext4_lblk_t save_last_lblock;
1201 * We will charge metadata quota at writeout time; this saves
1202 * us from metadata over-estimation, though we may go over by
1203 * a small amount in the end. Here we just reserve for data.
1205 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1210 * recalculate the amount of metadata blocks to reserve
1211 * in order to allocate nrblocks
1212 * worse case is one extent per block
1215 spin_lock(&ei->i_block_reservation_lock);
1217 * ext4_calc_metadata_amount() has side effects, which we have
1218 * to be prepared undo if we fail to claim space.
1220 save_len = ei->i_da_metadata_calc_len;
1221 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1222 md_needed = EXT4_NUM_B2C(sbi,
1223 ext4_calc_metadata_amount(inode, lblock));
1224 trace_ext4_da_reserve_space(inode, md_needed);
1227 * We do still charge estimated metadata to the sb though;
1228 * we cannot afford to run out of free blocks.
1230 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1231 ei->i_da_metadata_calc_len = save_len;
1232 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1233 spin_unlock(&ei->i_block_reservation_lock);
1234 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1238 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1241 ei->i_reserved_data_blocks++;
1242 ei->i_reserved_meta_blocks += md_needed;
1243 spin_unlock(&ei->i_block_reservation_lock);
1245 return 0; /* success */
1248 static void ext4_da_release_space(struct inode *inode, int to_free)
1250 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1251 struct ext4_inode_info *ei = EXT4_I(inode);
1254 return; /* Nothing to release, exit */
1256 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1258 trace_ext4_da_release_space(inode, to_free);
1259 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1261 * if there aren't enough reserved blocks, then the
1262 * counter is messed up somewhere. Since this
1263 * function is called from invalidate page, it's
1264 * harmless to return without any action.
1266 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1267 "ino %lu, to_free %d with only %d reserved "
1268 "data blocks", inode->i_ino, to_free,
1269 ei->i_reserved_data_blocks);
1271 to_free = ei->i_reserved_data_blocks;
1273 ei->i_reserved_data_blocks -= to_free;
1275 if (ei->i_reserved_data_blocks == 0) {
1277 * We can release all of the reserved metadata blocks
1278 * only when we have written all of the delayed
1279 * allocation blocks.
1280 * Note that in case of bigalloc, i_reserved_meta_blocks,
1281 * i_reserved_data_blocks, etc. refer to number of clusters.
1283 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1284 ei->i_reserved_meta_blocks);
1285 ei->i_reserved_meta_blocks = 0;
1286 ei->i_da_metadata_calc_len = 0;
1289 /* update fs dirty data blocks counter */
1290 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1292 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1294 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1297 static void ext4_da_page_release_reservation(struct page *page,
1298 unsigned long offset)
1301 struct buffer_head *head, *bh;
1302 unsigned int curr_off = 0;
1303 struct inode *inode = page->mapping->host;
1304 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1307 head = page_buffers(page);
1310 unsigned int next_off = curr_off + bh->b_size;
1312 if ((offset <= curr_off) && (buffer_delay(bh))) {
1314 clear_buffer_delay(bh);
1315 clear_buffer_da_mapped(bh);
1317 curr_off = next_off;
1318 } while ((bh = bh->b_this_page) != head);
1320 /* If we have released all the blocks belonging to a cluster, then we
1321 * need to release the reserved space for that cluster. */
1322 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1323 while (num_clusters > 0) {
1325 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1326 ((num_clusters - 1) << sbi->s_cluster_bits);
1327 if (sbi->s_cluster_ratio == 1 ||
1328 !ext4_find_delalloc_cluster(inode, lblk, 1))
1329 ext4_da_release_space(inode, 1);
1336 * Delayed allocation stuff
1340 * mpage_da_submit_io - walks through extent of pages and try to write
1341 * them with writepage() call back
1343 * @mpd->inode: inode
1344 * @mpd->first_page: first page of the extent
1345 * @mpd->next_page: page after the last page of the extent
1347 * By the time mpage_da_submit_io() is called we expect all blocks
1348 * to be allocated. this may be wrong if allocation failed.
1350 * As pages are already locked by write_cache_pages(), we can't use it
1352 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1353 struct ext4_map_blocks *map)
1355 struct pagevec pvec;
1356 unsigned long index, end;
1357 int ret = 0, err, nr_pages, i;
1358 struct inode *inode = mpd->inode;
1359 struct address_space *mapping = inode->i_mapping;
1360 loff_t size = i_size_read(inode);
1361 unsigned int len, block_start;
1362 struct buffer_head *bh, *page_bufs = NULL;
1363 int journal_data = ext4_should_journal_data(inode);
1364 sector_t pblock = 0, cur_logical = 0;
1365 struct ext4_io_submit io_submit;
1367 BUG_ON(mpd->next_page <= mpd->first_page);
1368 memset(&io_submit, 0, sizeof(io_submit));
1370 * We need to start from the first_page to the next_page - 1
1371 * to make sure we also write the mapped dirty buffer_heads.
1372 * If we look at mpd->b_blocknr we would only be looking
1373 * at the currently mapped buffer_heads.
1375 index = mpd->first_page;
1376 end = mpd->next_page - 1;
1378 pagevec_init(&pvec, 0);
1379 while (index <= end) {
1380 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1383 for (i = 0; i < nr_pages; i++) {
1384 int commit_write = 0, skip_page = 0;
1385 struct page *page = pvec.pages[i];
1387 index = page->index;
1391 if (index == size >> PAGE_CACHE_SHIFT)
1392 len = size & ~PAGE_CACHE_MASK;
1394 len = PAGE_CACHE_SIZE;
1396 cur_logical = index << (PAGE_CACHE_SHIFT -
1398 pblock = map->m_pblk + (cur_logical -
1403 BUG_ON(!PageLocked(page));
1404 BUG_ON(PageWriteback(page));
1407 * If the page does not have buffers (for
1408 * whatever reason), try to create them using
1409 * __block_write_begin. If this fails,
1410 * skip the page and move on.
1412 if (!page_has_buffers(page)) {
1413 if (__block_write_begin(page, 0, len,
1414 noalloc_get_block_write)) {
1422 bh = page_bufs = page_buffers(page);
1427 if (map && (cur_logical >= map->m_lblk) &&
1428 (cur_logical <= (map->m_lblk +
1429 (map->m_len - 1)))) {
1430 if (buffer_delay(bh)) {
1431 clear_buffer_delay(bh);
1432 bh->b_blocknr = pblock;
1434 if (buffer_da_mapped(bh))
1435 clear_buffer_da_mapped(bh);
1436 if (buffer_unwritten(bh) ||
1438 BUG_ON(bh->b_blocknr != pblock);
1439 if (map->m_flags & EXT4_MAP_UNINIT)
1440 set_buffer_uninit(bh);
1441 clear_buffer_unwritten(bh);
1445 * skip page if block allocation undone and
1448 if (ext4_bh_delay_or_unwritten(NULL, bh))
1450 bh = bh->b_this_page;
1451 block_start += bh->b_size;
1454 } while (bh != page_bufs);
1460 /* mark the buffer_heads as dirty & uptodate */
1461 block_commit_write(page, 0, len);
1463 clear_page_dirty_for_io(page);
1465 * Delalloc doesn't support data journalling,
1466 * but eventually maybe we'll lift this
1469 if (unlikely(journal_data && PageChecked(page)))
1470 err = __ext4_journalled_writepage(page, len);
1471 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1472 err = ext4_bio_write_page(&io_submit, page,
1474 else if (buffer_uninit(page_bufs)) {
1475 ext4_set_bh_endio(page_bufs, inode);
1476 err = block_write_full_page_endio(page,
1477 noalloc_get_block_write,
1478 mpd->wbc, ext4_end_io_buffer_write);
1480 err = block_write_full_page(page,
1481 noalloc_get_block_write, mpd->wbc);
1484 mpd->pages_written++;
1486 * In error case, we have to continue because
1487 * remaining pages are still locked
1492 pagevec_release(&pvec);
1494 ext4_io_submit(&io_submit);
1498 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1502 struct pagevec pvec;
1503 struct inode *inode = mpd->inode;
1504 struct address_space *mapping = inode->i_mapping;
1506 index = mpd->first_page;
1507 end = mpd->next_page - 1;
1508 while (index <= end) {
1509 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1512 for (i = 0; i < nr_pages; i++) {
1513 struct page *page = pvec.pages[i];
1514 if (page->index > end)
1516 BUG_ON(!PageLocked(page));
1517 BUG_ON(PageWriteback(page));
1518 block_invalidatepage(page, 0);
1519 ClearPageUptodate(page);
1522 index = pvec.pages[nr_pages - 1]->index + 1;
1523 pagevec_release(&pvec);
1528 static void ext4_print_free_blocks(struct inode *inode)
1530 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1531 struct super_block *sb = inode->i_sb;
1533 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1534 EXT4_C2B(EXT4_SB(inode->i_sb),
1535 ext4_count_free_clusters(inode->i_sb)));
1536 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1537 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1538 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1539 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1540 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1541 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1542 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1543 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1544 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1545 EXT4_I(inode)->i_reserved_data_blocks);
1546 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1547 EXT4_I(inode)->i_reserved_meta_blocks);
1552 * mpage_da_map_and_submit - go through given space, map them
1553 * if necessary, and then submit them for I/O
1555 * @mpd - bh describing space
1557 * The function skips space we know is already mapped to disk blocks.
1560 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1562 int err, blks, get_blocks_flags;
1563 struct ext4_map_blocks map, *mapp = NULL;
1564 sector_t next = mpd->b_blocknr;
1565 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1566 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1567 handle_t *handle = NULL;
1570 * If the blocks are mapped already, or we couldn't accumulate
1571 * any blocks, then proceed immediately to the submission stage.
1573 if ((mpd->b_size == 0) ||
1574 ((mpd->b_state & (1 << BH_Mapped)) &&
1575 !(mpd->b_state & (1 << BH_Delay)) &&
1576 !(mpd->b_state & (1 << BH_Unwritten))))
1579 handle = ext4_journal_current_handle();
1583 * Call ext4_map_blocks() to allocate any delayed allocation
1584 * blocks, or to convert an uninitialized extent to be
1585 * initialized (in the case where we have written into
1586 * one or more preallocated blocks).
1588 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1589 * indicate that we are on the delayed allocation path. This
1590 * affects functions in many different parts of the allocation
1591 * call path. This flag exists primarily because we don't
1592 * want to change *many* call functions, so ext4_map_blocks()
1593 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1594 * inode's allocation semaphore is taken.
1596 * If the blocks in questions were delalloc blocks, set
1597 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1598 * variables are updated after the blocks have been allocated.
1601 map.m_len = max_blocks;
1602 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1603 if (ext4_should_dioread_nolock(mpd->inode))
1604 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1605 if (mpd->b_state & (1 << BH_Delay))
1606 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1608 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1610 struct super_block *sb = mpd->inode->i_sb;
1614 * If get block returns EAGAIN or ENOSPC and there
1615 * appears to be free blocks we will just let
1616 * mpage_da_submit_io() unlock all of the pages.
1621 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1627 * get block failure will cause us to loop in
1628 * writepages, because a_ops->writepage won't be able
1629 * to make progress. The page will be redirtied by
1630 * writepage and writepages will again try to write
1633 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1634 ext4_msg(sb, KERN_CRIT,
1635 "delayed block allocation failed for inode %lu "
1636 "at logical offset %llu with max blocks %zd "
1637 "with error %d", mpd->inode->i_ino,
1638 (unsigned long long) next,
1639 mpd->b_size >> mpd->inode->i_blkbits, err);
1640 ext4_msg(sb, KERN_CRIT,
1641 "This should not happen!! Data will be lost\n");
1643 ext4_print_free_blocks(mpd->inode);
1645 /* invalidate all the pages */
1646 ext4_da_block_invalidatepages(mpd);
1648 /* Mark this page range as having been completed */
1655 if (map.m_flags & EXT4_MAP_NEW) {
1656 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1659 for (i = 0; i < map.m_len; i++)
1660 unmap_underlying_metadata(bdev, map.m_pblk + i);
1662 if (ext4_should_order_data(mpd->inode)) {
1663 err = ext4_jbd2_file_inode(handle, mpd->inode);
1665 /* Only if the journal is aborted */
1673 * Update on-disk size along with block allocation.
1675 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1676 if (disksize > i_size_read(mpd->inode))
1677 disksize = i_size_read(mpd->inode);
1678 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1679 ext4_update_i_disksize(mpd->inode, disksize);
1680 err = ext4_mark_inode_dirty(handle, mpd->inode);
1682 ext4_error(mpd->inode->i_sb,
1683 "Failed to mark inode %lu dirty",
1688 mpage_da_submit_io(mpd, mapp);
1692 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1693 (1 << BH_Delay) | (1 << BH_Unwritten))
1696 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1698 * @mpd->lbh - extent of blocks
1699 * @logical - logical number of the block in the file
1700 * @bh - bh of the block (used to access block's state)
1702 * the function is used to collect contig. blocks in same state
1704 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1705 sector_t logical, size_t b_size,
1706 unsigned long b_state)
1709 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1712 * XXX Don't go larger than mballoc is willing to allocate
1713 * This is a stopgap solution. We eventually need to fold
1714 * mpage_da_submit_io() into this function and then call
1715 * ext4_map_blocks() multiple times in a loop
1717 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1720 /* check if thereserved journal credits might overflow */
1721 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1722 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1724 * With non-extent format we are limited by the journal
1725 * credit available. Total credit needed to insert
1726 * nrblocks contiguous blocks is dependent on the
1727 * nrblocks. So limit nrblocks.
1730 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1731 EXT4_MAX_TRANS_DATA) {
1733 * Adding the new buffer_head would make it cross the
1734 * allowed limit for which we have journal credit
1735 * reserved. So limit the new bh->b_size
1737 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1738 mpd->inode->i_blkbits;
1739 /* we will do mpage_da_submit_io in the next loop */
1743 * First block in the extent
1745 if (mpd->b_size == 0) {
1746 mpd->b_blocknr = logical;
1747 mpd->b_size = b_size;
1748 mpd->b_state = b_state & BH_FLAGS;
1752 next = mpd->b_blocknr + nrblocks;
1754 * Can we merge the block to our big extent?
1756 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1757 mpd->b_size += b_size;
1763 * We couldn't merge the block to our extent, so we
1764 * need to flush current extent and start new one
1766 mpage_da_map_and_submit(mpd);
1770 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1772 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1776 * This function is grabs code from the very beginning of
1777 * ext4_map_blocks, but assumes that the caller is from delayed write
1778 * time. This function looks up the requested blocks and sets the
1779 * buffer delay bit under the protection of i_data_sem.
1781 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1782 struct ext4_map_blocks *map,
1783 struct buffer_head *bh)
1786 sector_t invalid_block = ~((sector_t) 0xffff);
1788 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1792 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1793 "logical block %lu\n", inode->i_ino, map->m_len,
1794 (unsigned long) map->m_lblk);
1796 * Try to see if we can get the block without requesting a new
1797 * file system block.
1799 down_read((&EXT4_I(inode)->i_data_sem));
1800 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1801 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1803 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1807 * XXX: __block_prepare_write() unmaps passed block,
1810 /* If the block was allocated from previously allocated cluster,
1811 * then we dont need to reserve it again. */
1812 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1813 retval = ext4_da_reserve_space(inode, iblock);
1815 /* not enough space to reserve */
1819 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1820 * and it should not appear on the bh->b_state.
1822 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1824 map_bh(bh, inode->i_sb, invalid_block);
1826 set_buffer_delay(bh);
1830 up_read((&EXT4_I(inode)->i_data_sem));
1836 * This is a special get_blocks_t callback which is used by
1837 * ext4_da_write_begin(). It will either return mapped block or
1838 * reserve space for a single block.
1840 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1841 * We also have b_blocknr = -1 and b_bdev initialized properly
1843 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1844 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1845 * initialized properly.
1847 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1848 struct buffer_head *bh, int create)
1850 struct ext4_map_blocks map;
1853 BUG_ON(create == 0);
1854 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1856 map.m_lblk = iblock;
1860 * first, we need to know whether the block is allocated already
1861 * preallocated blocks are unmapped but should treated
1862 * the same as allocated blocks.
1864 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1868 map_bh(bh, inode->i_sb, map.m_pblk);
1869 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1871 if (buffer_unwritten(bh)) {
1872 /* A delayed write to unwritten bh should be marked
1873 * new and mapped. Mapped ensures that we don't do
1874 * get_block multiple times when we write to the same
1875 * offset and new ensures that we do proper zero out
1876 * for partial write.
1879 set_buffer_mapped(bh);
1885 * This function is used as a standard get_block_t calback function
1886 * when there is no desire to allocate any blocks. It is used as a
1887 * callback function for block_write_begin() and block_write_full_page().
1888 * These functions should only try to map a single block at a time.
1890 * Since this function doesn't do block allocations even if the caller
1891 * requests it by passing in create=1, it is critically important that
1892 * any caller checks to make sure that any buffer heads are returned
1893 * by this function are either all already mapped or marked for
1894 * delayed allocation before calling block_write_full_page(). Otherwise,
1895 * b_blocknr could be left unitialized, and the page write functions will
1896 * be taken by surprise.
1898 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1899 struct buffer_head *bh_result, int create)
1901 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1902 return _ext4_get_block(inode, iblock, bh_result, 0);
1905 static int bget_one(handle_t *handle, struct buffer_head *bh)
1911 static int bput_one(handle_t *handle, struct buffer_head *bh)
1917 static int __ext4_journalled_writepage(struct page *page,
1920 struct address_space *mapping = page->mapping;
1921 struct inode *inode = mapping->host;
1922 struct buffer_head *page_bufs;
1923 handle_t *handle = NULL;
1927 ClearPageChecked(page);
1928 page_bufs = page_buffers(page);
1930 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1931 /* As soon as we unlock the page, it can go away, but we have
1932 * references to buffers so we are safe */
1935 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1936 if (IS_ERR(handle)) {
1937 ret = PTR_ERR(handle);
1941 BUG_ON(!ext4_handle_valid(handle));
1943 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1944 do_journal_get_write_access);
1946 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1950 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1951 err = ext4_journal_stop(handle);
1955 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1956 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1962 * Note that we don't need to start a transaction unless we're journaling data
1963 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1964 * need to file the inode to the transaction's list in ordered mode because if
1965 * we are writing back data added by write(), the inode is already there and if
1966 * we are writing back data modified via mmap(), no one guarantees in which
1967 * transaction the data will hit the disk. In case we are journaling data, we
1968 * cannot start transaction directly because transaction start ranks above page
1969 * lock so we have to do some magic.
1971 * This function can get called via...
1972 * - ext4_da_writepages after taking page lock (have journal handle)
1973 * - journal_submit_inode_data_buffers (no journal handle)
1974 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1975 * - grab_page_cache when doing write_begin (have journal handle)
1977 * We don't do any block allocation in this function. If we have page with
1978 * multiple blocks we need to write those buffer_heads that are mapped. This
1979 * is important for mmaped based write. So if we do with blocksize 1K
1980 * truncate(f, 1024);
1981 * a = mmap(f, 0, 4096);
1983 * truncate(f, 4096);
1984 * we have in the page first buffer_head mapped via page_mkwrite call back
1985 * but other buffer_heads would be unmapped but dirty (dirty done via the
1986 * do_wp_page). So writepage should write the first block. If we modify
1987 * the mmap area beyond 1024 we will again get a page_fault and the
1988 * page_mkwrite callback will do the block allocation and mark the
1989 * buffer_heads mapped.
1991 * We redirty the page if we have any buffer_heads that is either delay or
1992 * unwritten in the page.
1994 * We can get recursively called as show below.
1996 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1999 * But since we don't do any block allocation we should not deadlock.
2000 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2002 static int ext4_writepage(struct page *page,
2003 struct writeback_control *wbc)
2005 int ret = 0, commit_write = 0;
2008 struct buffer_head *page_bufs = NULL;
2009 struct inode *inode = page->mapping->host;
2011 trace_ext4_writepage(page);
2012 size = i_size_read(inode);
2013 if (page->index == size >> PAGE_CACHE_SHIFT)
2014 len = size & ~PAGE_CACHE_MASK;
2016 len = PAGE_CACHE_SIZE;
2019 * If the page does not have buffers (for whatever reason),
2020 * try to create them using __block_write_begin. If this
2021 * fails, redirty the page and move on.
2023 if (!page_has_buffers(page)) {
2024 if (__block_write_begin(page, 0, len,
2025 noalloc_get_block_write)) {
2027 redirty_page_for_writepage(wbc, page);
2033 page_bufs = page_buffers(page);
2034 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2035 ext4_bh_delay_or_unwritten)) {
2037 * We don't want to do block allocation, so redirty
2038 * the page and return. We may reach here when we do
2039 * a journal commit via journal_submit_inode_data_buffers.
2040 * We can also reach here via shrink_page_list but it
2041 * should never be for direct reclaim so warn if that
2044 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2049 /* now mark the buffer_heads as dirty and uptodate */
2050 block_commit_write(page, 0, len);
2052 if (PageChecked(page) && ext4_should_journal_data(inode))
2054 * It's mmapped pagecache. Add buffers and journal it. There
2055 * doesn't seem much point in redirtying the page here.
2057 return __ext4_journalled_writepage(page, len);
2059 if (buffer_uninit(page_bufs)) {
2060 ext4_set_bh_endio(page_bufs, inode);
2061 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2062 wbc, ext4_end_io_buffer_write);
2064 ret = block_write_full_page(page, noalloc_get_block_write,
2071 * This is called via ext4_da_writepages() to
2072 * calculate the total number of credits to reserve to fit
2073 * a single extent allocation into a single transaction,
2074 * ext4_da_writpeages() will loop calling this before
2075 * the block allocation.
2078 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2080 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2083 * With non-extent format the journal credit needed to
2084 * insert nrblocks contiguous block is dependent on
2085 * number of contiguous block. So we will limit
2086 * number of contiguous block to a sane value
2088 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2089 (max_blocks > EXT4_MAX_TRANS_DATA))
2090 max_blocks = EXT4_MAX_TRANS_DATA;
2092 return ext4_chunk_trans_blocks(inode, max_blocks);
2096 * write_cache_pages_da - walk the list of dirty pages of the given
2097 * address space and accumulate pages that need writing, and call
2098 * mpage_da_map_and_submit to map a single contiguous memory region
2099 * and then write them.
2101 static int write_cache_pages_da(struct address_space *mapping,
2102 struct writeback_control *wbc,
2103 struct mpage_da_data *mpd,
2104 pgoff_t *done_index)
2106 struct buffer_head *bh, *head;
2107 struct inode *inode = mapping->host;
2108 struct pagevec pvec;
2109 unsigned int nr_pages;
2112 long nr_to_write = wbc->nr_to_write;
2113 int i, tag, ret = 0;
2115 memset(mpd, 0, sizeof(struct mpage_da_data));
2118 pagevec_init(&pvec, 0);
2119 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2120 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2122 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2123 tag = PAGECACHE_TAG_TOWRITE;
2125 tag = PAGECACHE_TAG_DIRTY;
2127 *done_index = index;
2128 while (index <= end) {
2129 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2130 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2134 for (i = 0; i < nr_pages; i++) {
2135 struct page *page = pvec.pages[i];
2138 * At this point, the page may be truncated or
2139 * invalidated (changing page->mapping to NULL), or
2140 * even swizzled back from swapper_space to tmpfs file
2141 * mapping. However, page->index will not change
2142 * because we have a reference on the page.
2144 if (page->index > end)
2147 *done_index = page->index + 1;
2150 * If we can't merge this page, and we have
2151 * accumulated an contiguous region, write it
2153 if ((mpd->next_page != page->index) &&
2154 (mpd->next_page != mpd->first_page)) {
2155 mpage_da_map_and_submit(mpd);
2156 goto ret_extent_tail;
2162 * If the page is no longer dirty, or its
2163 * mapping no longer corresponds to inode we
2164 * are writing (which means it has been
2165 * truncated or invalidated), or the page is
2166 * already under writeback and we are not
2167 * doing a data integrity writeback, skip the page
2169 if (!PageDirty(page) ||
2170 (PageWriteback(page) &&
2171 (wbc->sync_mode == WB_SYNC_NONE)) ||
2172 unlikely(page->mapping != mapping)) {
2177 wait_on_page_writeback(page);
2178 BUG_ON(PageWriteback(page));
2180 if (mpd->next_page != page->index)
2181 mpd->first_page = page->index;
2182 mpd->next_page = page->index + 1;
2183 logical = (sector_t) page->index <<
2184 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2186 if (!page_has_buffers(page)) {
2187 mpage_add_bh_to_extent(mpd, logical,
2189 (1 << BH_Dirty) | (1 << BH_Uptodate));
2191 goto ret_extent_tail;
2194 * Page with regular buffer heads,
2195 * just add all dirty ones
2197 head = page_buffers(page);
2200 BUG_ON(buffer_locked(bh));
2202 * We need to try to allocate
2203 * unmapped blocks in the same page.
2204 * Otherwise we won't make progress
2205 * with the page in ext4_writepage
2207 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2208 mpage_add_bh_to_extent(mpd, logical,
2212 goto ret_extent_tail;
2213 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2215 * mapped dirty buffer. We need
2216 * to update the b_state
2217 * because we look at b_state
2218 * in mpage_da_map_blocks. We
2219 * don't update b_size because
2220 * if we find an unmapped
2221 * buffer_head later we need to
2222 * use the b_state flag of that
2225 if (mpd->b_size == 0)
2226 mpd->b_state = bh->b_state & BH_FLAGS;
2229 } while ((bh = bh->b_this_page) != head);
2232 if (nr_to_write > 0) {
2234 if (nr_to_write == 0 &&
2235 wbc->sync_mode == WB_SYNC_NONE)
2237 * We stop writing back only if we are
2238 * not doing integrity sync. In case of
2239 * integrity sync we have to keep going
2240 * because someone may be concurrently
2241 * dirtying pages, and we might have
2242 * synced a lot of newly appeared dirty
2243 * pages, but have not synced all of the
2249 pagevec_release(&pvec);
2254 ret = MPAGE_DA_EXTENT_TAIL;
2256 pagevec_release(&pvec);
2262 static int ext4_da_writepages(struct address_space *mapping,
2263 struct writeback_control *wbc)
2266 int range_whole = 0;
2267 handle_t *handle = NULL;
2268 struct mpage_da_data mpd;
2269 struct inode *inode = mapping->host;
2270 int pages_written = 0;
2271 unsigned int max_pages;
2272 int range_cyclic, cycled = 1, io_done = 0;
2273 int needed_blocks, ret = 0;
2274 long desired_nr_to_write, nr_to_writebump = 0;
2275 loff_t range_start = wbc->range_start;
2276 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2277 pgoff_t done_index = 0;
2279 struct blk_plug plug;
2281 trace_ext4_da_writepages(inode, wbc);
2284 * No pages to write? This is mainly a kludge to avoid starting
2285 * a transaction for special inodes like journal inode on last iput()
2286 * because that could violate lock ordering on umount
2288 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2292 * If the filesystem has aborted, it is read-only, so return
2293 * right away instead of dumping stack traces later on that
2294 * will obscure the real source of the problem. We test
2295 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2296 * the latter could be true if the filesystem is mounted
2297 * read-only, and in that case, ext4_da_writepages should
2298 * *never* be called, so if that ever happens, we would want
2301 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2304 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2307 range_cyclic = wbc->range_cyclic;
2308 if (wbc->range_cyclic) {
2309 index = mapping->writeback_index;
2312 wbc->range_start = index << PAGE_CACHE_SHIFT;
2313 wbc->range_end = LLONG_MAX;
2314 wbc->range_cyclic = 0;
2317 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2318 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2322 * This works around two forms of stupidity. The first is in
2323 * the writeback code, which caps the maximum number of pages
2324 * written to be 1024 pages. This is wrong on multiple
2325 * levels; different architectues have a different page size,
2326 * which changes the maximum amount of data which gets
2327 * written. Secondly, 4 megabytes is way too small. XFS
2328 * forces this value to be 16 megabytes by multiplying
2329 * nr_to_write parameter by four, and then relies on its
2330 * allocator to allocate larger extents to make them
2331 * contiguous. Unfortunately this brings us to the second
2332 * stupidity, which is that ext4's mballoc code only allocates
2333 * at most 2048 blocks. So we force contiguous writes up to
2334 * the number of dirty blocks in the inode, or
2335 * sbi->max_writeback_mb_bump whichever is smaller.
2337 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2338 if (!range_cyclic && range_whole) {
2339 if (wbc->nr_to_write == LONG_MAX)
2340 desired_nr_to_write = wbc->nr_to_write;
2342 desired_nr_to_write = wbc->nr_to_write * 8;
2344 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2346 if (desired_nr_to_write > max_pages)
2347 desired_nr_to_write = max_pages;
2349 if (wbc->nr_to_write < desired_nr_to_write) {
2350 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2351 wbc->nr_to_write = desired_nr_to_write;
2355 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2356 tag_pages_for_writeback(mapping, index, end);
2358 blk_start_plug(&plug);
2359 while (!ret && wbc->nr_to_write > 0) {
2362 * we insert one extent at a time. So we need
2363 * credit needed for single extent allocation.
2364 * journalled mode is currently not supported
2367 BUG_ON(ext4_should_journal_data(inode));
2368 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2370 /* start a new transaction*/
2371 handle = ext4_journal_start(inode, needed_blocks);
2372 if (IS_ERR(handle)) {
2373 ret = PTR_ERR(handle);
2374 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2375 "%ld pages, ino %lu; err %d", __func__,
2376 wbc->nr_to_write, inode->i_ino, ret);
2377 blk_finish_plug(&plug);
2378 goto out_writepages;
2382 * Now call write_cache_pages_da() to find the next
2383 * contiguous region of logical blocks that need
2384 * blocks to be allocated by ext4 and submit them.
2386 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2388 * If we have a contiguous extent of pages and we
2389 * haven't done the I/O yet, map the blocks and submit
2392 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2393 mpage_da_map_and_submit(&mpd);
2394 ret = MPAGE_DA_EXTENT_TAIL;
2396 trace_ext4_da_write_pages(inode, &mpd);
2397 wbc->nr_to_write -= mpd.pages_written;
2399 ext4_journal_stop(handle);
2401 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2402 /* commit the transaction which would
2403 * free blocks released in the transaction
2406 jbd2_journal_force_commit_nested(sbi->s_journal);
2408 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2410 * Got one extent now try with rest of the pages.
2411 * If mpd.retval is set -EIO, journal is aborted.
2412 * So we don't need to write any more.
2414 pages_written += mpd.pages_written;
2417 } else if (wbc->nr_to_write)
2419 * There is no more writeout needed
2420 * or we requested for a noblocking writeout
2421 * and we found the device congested
2425 blk_finish_plug(&plug);
2426 if (!io_done && !cycled) {
2429 wbc->range_start = index << PAGE_CACHE_SHIFT;
2430 wbc->range_end = mapping->writeback_index - 1;
2435 wbc->range_cyclic = range_cyclic;
2436 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2438 * set the writeback_index so that range_cyclic
2439 * mode will write it back later
2441 mapping->writeback_index = done_index;
2444 wbc->nr_to_write -= nr_to_writebump;
2445 wbc->range_start = range_start;
2446 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2450 #define FALL_BACK_TO_NONDELALLOC 1
2451 static int ext4_nonda_switch(struct super_block *sb)
2453 s64 free_blocks, dirty_blocks;
2454 struct ext4_sb_info *sbi = EXT4_SB(sb);
2457 * switch to non delalloc mode if we are running low
2458 * on free block. The free block accounting via percpu
2459 * counters can get slightly wrong with percpu_counter_batch getting
2460 * accumulated on each CPU without updating global counters
2461 * Delalloc need an accurate free block accounting. So switch
2462 * to non delalloc when we are near to error range.
2464 free_blocks = EXT4_C2B(sbi,
2465 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2466 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2468 * Start pushing delalloc when 1/2 of free blocks are dirty.
2470 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2471 !writeback_in_progress(sb->s_bdi) &&
2472 down_read_trylock(&sb->s_umount)) {
2473 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2474 up_read(&sb->s_umount);
2477 if (2 * free_blocks < 3 * dirty_blocks ||
2478 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2480 * free block count is less than 150% of dirty blocks
2481 * or free blocks is less than watermark
2488 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2489 loff_t pos, unsigned len, unsigned flags,
2490 struct page **pagep, void **fsdata)
2492 int ret, retries = 0;
2495 struct inode *inode = mapping->host;
2498 index = pos >> PAGE_CACHE_SHIFT;
2500 if (ext4_nonda_switch(inode->i_sb)) {
2501 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2502 return ext4_write_begin(file, mapping, pos,
2503 len, flags, pagep, fsdata);
2505 *fsdata = (void *)0;
2506 trace_ext4_da_write_begin(inode, pos, len, flags);
2509 * With delayed allocation, we don't log the i_disksize update
2510 * if there is delayed block allocation. But we still need
2511 * to journalling the i_disksize update if writes to the end
2512 * of file which has an already mapped buffer.
2514 handle = ext4_journal_start(inode, 1);
2515 if (IS_ERR(handle)) {
2516 ret = PTR_ERR(handle);
2519 /* We cannot recurse into the filesystem as the transaction is already
2521 flags |= AOP_FLAG_NOFS;
2523 page = grab_cache_page_write_begin(mapping, index, flags);
2525 ext4_journal_stop(handle);
2531 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2534 ext4_journal_stop(handle);
2535 page_cache_release(page);
2537 * block_write_begin may have instantiated a few blocks
2538 * outside i_size. Trim these off again. Don't need
2539 * i_size_read because we hold i_mutex.
2541 if (pos + len > inode->i_size)
2542 ext4_truncate_failed_write(inode);
2545 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2552 * Check if we should update i_disksize
2553 * when write to the end of file but not require block allocation
2555 static int ext4_da_should_update_i_disksize(struct page *page,
2556 unsigned long offset)
2558 struct buffer_head *bh;
2559 struct inode *inode = page->mapping->host;
2563 bh = page_buffers(page);
2564 idx = offset >> inode->i_blkbits;
2566 for (i = 0; i < idx; i++)
2567 bh = bh->b_this_page;
2569 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2574 static int ext4_da_write_end(struct file *file,
2575 struct address_space *mapping,
2576 loff_t pos, unsigned len, unsigned copied,
2577 struct page *page, void *fsdata)
2579 struct inode *inode = mapping->host;
2581 handle_t *handle = ext4_journal_current_handle();
2583 unsigned long start, end;
2584 int write_mode = (int)(unsigned long)fsdata;
2586 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2587 switch (ext4_inode_journal_mode(inode)) {
2588 case EXT4_INODE_ORDERED_DATA_MODE:
2589 return ext4_ordered_write_end(file, mapping, pos,
2590 len, copied, page, fsdata);
2591 case EXT4_INODE_WRITEBACK_DATA_MODE:
2592 return ext4_writeback_write_end(file, mapping, pos,
2593 len, copied, page, fsdata);
2599 trace_ext4_da_write_end(inode, pos, len, copied);
2600 start = pos & (PAGE_CACHE_SIZE - 1);
2601 end = start + copied - 1;
2604 * generic_write_end() will run mark_inode_dirty() if i_size
2605 * changes. So let's piggyback the i_disksize mark_inode_dirty
2609 new_i_size = pos + copied;
2610 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2611 if (ext4_da_should_update_i_disksize(page, end)) {
2612 down_write(&EXT4_I(inode)->i_data_sem);
2613 if (new_i_size > EXT4_I(inode)->i_disksize) {
2615 * Updating i_disksize when extending file
2616 * without needing block allocation
2618 if (ext4_should_order_data(inode))
2619 ret = ext4_jbd2_file_inode(handle,
2622 EXT4_I(inode)->i_disksize = new_i_size;
2624 up_write(&EXT4_I(inode)->i_data_sem);
2625 /* We need to mark inode dirty even if
2626 * new_i_size is less that inode->i_size
2627 * bu greater than i_disksize.(hint delalloc)
2629 ext4_mark_inode_dirty(handle, inode);
2632 ret2 = generic_write_end(file, mapping, pos, len, copied,
2637 ret2 = ext4_journal_stop(handle);
2641 return ret ? ret : copied;
2644 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2647 * Drop reserved blocks
2649 BUG_ON(!PageLocked(page));
2650 if (!page_has_buffers(page))
2653 ext4_da_page_release_reservation(page, offset);
2656 ext4_invalidatepage(page, offset);
2662 * Force all delayed allocation blocks to be allocated for a given inode.
2664 int ext4_alloc_da_blocks(struct inode *inode)
2666 trace_ext4_alloc_da_blocks(inode);
2668 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2669 !EXT4_I(inode)->i_reserved_meta_blocks)
2673 * We do something simple for now. The filemap_flush() will
2674 * also start triggering a write of the data blocks, which is
2675 * not strictly speaking necessary (and for users of
2676 * laptop_mode, not even desirable). However, to do otherwise
2677 * would require replicating code paths in:
2679 * ext4_da_writepages() ->
2680 * write_cache_pages() ---> (via passed in callback function)
2681 * __mpage_da_writepage() -->
2682 * mpage_add_bh_to_extent()
2683 * mpage_da_map_blocks()
2685 * The problem is that write_cache_pages(), located in
2686 * mm/page-writeback.c, marks pages clean in preparation for
2687 * doing I/O, which is not desirable if we're not planning on
2690 * We could call write_cache_pages(), and then redirty all of
2691 * the pages by calling redirty_page_for_writepage() but that
2692 * would be ugly in the extreme. So instead we would need to
2693 * replicate parts of the code in the above functions,
2694 * simplifying them because we wouldn't actually intend to
2695 * write out the pages, but rather only collect contiguous
2696 * logical block extents, call the multi-block allocator, and
2697 * then update the buffer heads with the block allocations.
2699 * For now, though, we'll cheat by calling filemap_flush(),
2700 * which will map the blocks, and start the I/O, but not
2701 * actually wait for the I/O to complete.
2703 return filemap_flush(inode->i_mapping);
2707 * bmap() is special. It gets used by applications such as lilo and by
2708 * the swapper to find the on-disk block of a specific piece of data.
2710 * Naturally, this is dangerous if the block concerned is still in the
2711 * journal. If somebody makes a swapfile on an ext4 data-journaling
2712 * filesystem and enables swap, then they may get a nasty shock when the
2713 * data getting swapped to that swapfile suddenly gets overwritten by
2714 * the original zero's written out previously to the journal and
2715 * awaiting writeback in the kernel's buffer cache.
2717 * So, if we see any bmap calls here on a modified, data-journaled file,
2718 * take extra steps to flush any blocks which might be in the cache.
2720 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2722 struct inode *inode = mapping->host;
2726 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2727 test_opt(inode->i_sb, DELALLOC)) {
2729 * With delalloc we want to sync the file
2730 * so that we can make sure we allocate
2733 filemap_write_and_wait(mapping);
2736 if (EXT4_JOURNAL(inode) &&
2737 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2739 * This is a REALLY heavyweight approach, but the use of
2740 * bmap on dirty files is expected to be extremely rare:
2741 * only if we run lilo or swapon on a freshly made file
2742 * do we expect this to happen.
2744 * (bmap requires CAP_SYS_RAWIO so this does not
2745 * represent an unprivileged user DOS attack --- we'd be
2746 * in trouble if mortal users could trigger this path at
2749 * NB. EXT4_STATE_JDATA is not set on files other than
2750 * regular files. If somebody wants to bmap a directory
2751 * or symlink and gets confused because the buffer
2752 * hasn't yet been flushed to disk, they deserve
2753 * everything they get.
2756 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2757 journal = EXT4_JOURNAL(inode);
2758 jbd2_journal_lock_updates(journal);
2759 err = jbd2_journal_flush(journal);
2760 jbd2_journal_unlock_updates(journal);
2766 return generic_block_bmap(mapping, block, ext4_get_block);
2769 static int ext4_readpage(struct file *file, struct page *page)
2771 trace_ext4_readpage(page);
2772 return mpage_readpage(page, ext4_get_block);
2776 ext4_readpages(struct file *file, struct address_space *mapping,
2777 struct list_head *pages, unsigned nr_pages)
2779 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2782 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2784 struct buffer_head *head, *bh;
2785 unsigned int curr_off = 0;
2787 if (!page_has_buffers(page))
2789 head = bh = page_buffers(page);
2791 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2793 ext4_free_io_end(bh->b_private);
2794 bh->b_private = NULL;
2795 bh->b_end_io = NULL;
2797 curr_off = curr_off + bh->b_size;
2798 bh = bh->b_this_page;
2799 } while (bh != head);
2802 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2804 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2806 trace_ext4_invalidatepage(page, offset);
2809 * free any io_end structure allocated for buffers to be discarded
2811 if (ext4_should_dioread_nolock(page->mapping->host))
2812 ext4_invalidatepage_free_endio(page, offset);
2814 * If it's a full truncate we just forget about the pending dirtying
2817 ClearPageChecked(page);
2820 jbd2_journal_invalidatepage(journal, page, offset);
2822 block_invalidatepage(page, offset);
2825 static int ext4_releasepage(struct page *page, gfp_t wait)
2827 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2829 trace_ext4_releasepage(page);
2831 WARN_ON(PageChecked(page));
2832 if (!page_has_buffers(page))
2835 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2837 return try_to_free_buffers(page);
2841 * ext4_get_block used when preparing for a DIO write or buffer write.
2842 * We allocate an uinitialized extent if blocks haven't been allocated.
2843 * The extent will be converted to initialized after the IO is complete.
2845 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2846 struct buffer_head *bh_result, int create)
2848 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2849 inode->i_ino, create);
2850 return _ext4_get_block(inode, iblock, bh_result,
2851 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2854 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2855 struct buffer_head *bh_result, int create)
2857 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2858 inode->i_ino, create);
2859 return _ext4_get_block(inode, iblock, bh_result,
2860 EXT4_GET_BLOCKS_NO_LOCK);
2863 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2864 ssize_t size, void *private, int ret,
2867 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2868 ext4_io_end_t *io_end = iocb->private;
2870 /* if not async direct IO or dio with 0 bytes write, just return */
2871 if (!io_end || !size)
2874 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2875 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2876 iocb->private, io_end->inode->i_ino, iocb, offset,
2879 iocb->private = NULL;
2881 /* if not aio dio with unwritten extents, just free io and return */
2882 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2883 ext4_free_io_end(io_end);
2886 aio_complete(iocb, ret, 0);
2887 inode_dio_done(inode);
2891 io_end->offset = offset;
2892 io_end->size = size;
2894 io_end->iocb = iocb;
2895 io_end->result = ret;
2898 ext4_add_complete_io(io_end);
2901 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2903 ext4_io_end_t *io_end = bh->b_private;
2904 struct inode *inode;
2906 if (!test_clear_buffer_uninit(bh) || !io_end)
2909 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2910 ext4_msg(io_end->inode->i_sb, KERN_INFO,
2911 "sb umounted, discard end_io request for inode %lu",
2912 io_end->inode->i_ino);
2913 ext4_free_io_end(io_end);
2918 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2919 * but being more careful is always safe for the future change.
2921 inode = io_end->inode;
2922 ext4_set_io_unwritten_flag(inode, io_end);
2923 ext4_add_complete_io(io_end);
2925 bh->b_private = NULL;
2926 bh->b_end_io = NULL;
2927 clear_buffer_uninit(bh);
2928 end_buffer_async_write(bh, uptodate);
2931 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2933 ext4_io_end_t *io_end;
2934 struct page *page = bh->b_page;
2935 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2936 size_t size = bh->b_size;
2939 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2941 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2945 io_end->offset = offset;
2946 io_end->size = size;
2948 * We need to hold a reference to the page to make sure it
2949 * doesn't get evicted before ext4_end_io_work() has a chance
2950 * to convert the extent from written to unwritten.
2952 io_end->page = page;
2953 get_page(io_end->page);
2955 bh->b_private = io_end;
2956 bh->b_end_io = ext4_end_io_buffer_write;
2961 * For ext4 extent files, ext4 will do direct-io write to holes,
2962 * preallocated extents, and those write extend the file, no need to
2963 * fall back to buffered IO.
2965 * For holes, we fallocate those blocks, mark them as uninitialized
2966 * If those blocks were preallocated, we mark sure they are splited, but
2967 * still keep the range to write as uninitialized.
2969 * The unwrritten extents will be converted to written when DIO is completed.
2970 * For async direct IO, since the IO may still pending when return, we
2971 * set up an end_io call back function, which will do the conversion
2972 * when async direct IO completed.
2974 * If the O_DIRECT write will extend the file then add this inode to the
2975 * orphan list. So recovery will truncate it back to the original size
2976 * if the machine crashes during the write.
2979 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2980 const struct iovec *iov, loff_t offset,
2981 unsigned long nr_segs)
2983 struct file *file = iocb->ki_filp;
2984 struct inode *inode = file->f_mapping->host;
2986 size_t count = iov_length(iov, nr_segs);
2988 loff_t final_size = offset + count;
2989 if (rw == WRITE && final_size <= inode->i_size) {
2991 get_block_t *get_block_func = NULL;
2994 BUG_ON(iocb->private == NULL);
2996 /* If we do a overwrite dio, i_mutex locking can be released */
2997 overwrite = *((int *)iocb->private);
3000 atomic_inc(&inode->i_dio_count);
3001 down_read(&EXT4_I(inode)->i_data_sem);
3002 mutex_unlock(&inode->i_mutex);
3006 * We could direct write to holes and fallocate.
3008 * Allocated blocks to fill the hole are marked as uninitialized
3009 * to prevent parallel buffered read to expose the stale data
3010 * before DIO complete the data IO.
3012 * As to previously fallocated extents, ext4 get_block
3013 * will just simply mark the buffer mapped but still
3014 * keep the extents uninitialized.
3016 * for non AIO case, we will convert those unwritten extents
3017 * to written after return back from blockdev_direct_IO.
3019 * for async DIO, the conversion needs to be defered when
3020 * the IO is completed. The ext4 end_io callback function
3021 * will be called to take care of the conversion work.
3022 * Here for async case, we allocate an io_end structure to
3025 iocb->private = NULL;
3026 ext4_inode_aio_set(inode, NULL);
3027 if (!is_sync_kiocb(iocb)) {
3028 ext4_io_end_t *io_end =
3029 ext4_init_io_end(inode, GFP_NOFS);
3034 io_end->flag |= EXT4_IO_END_DIRECT;
3035 iocb->private = io_end;
3037 * we save the io structure for current async
3038 * direct IO, so that later ext4_map_blocks()
3039 * could flag the io structure whether there
3040 * is a unwritten extents needs to be converted
3041 * when IO is completed.
3043 ext4_inode_aio_set(inode, io_end);
3047 get_block_func = ext4_get_block_write_nolock;
3049 get_block_func = ext4_get_block_write;
3050 dio_flags = DIO_LOCKING;
3052 ret = __blockdev_direct_IO(rw, iocb, inode,
3053 inode->i_sb->s_bdev, iov,
3061 ext4_inode_aio_set(inode, NULL);
3063 * The io_end structure takes a reference to the inode,
3064 * that structure needs to be destroyed and the
3065 * reference to the inode need to be dropped, when IO is
3066 * complete, even with 0 byte write, or failed.
3068 * In the successful AIO DIO case, the io_end structure will be
3069 * desctroyed and the reference to the inode will be dropped
3070 * after the end_io call back function is called.
3072 * In the case there is 0 byte write, or error case, since
3073 * VFS direct IO won't invoke the end_io call back function,
3074 * we need to free the end_io structure here.
3076 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3077 ext4_free_io_end(iocb->private);
3078 iocb->private = NULL;
3079 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3080 EXT4_STATE_DIO_UNWRITTEN)) {
3083 * for non AIO case, since the IO is already
3084 * completed, we could do the conversion right here
3086 err = ext4_convert_unwritten_extents(inode,
3090 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3094 /* take i_mutex locking again if we do a ovewrite dio */
3096 inode_dio_done(inode);
3097 up_read(&EXT4_I(inode)->i_data_sem);
3098 mutex_lock(&inode->i_mutex);
3104 /* for write the the end of file case, we fall back to old way */
3105 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3108 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3109 const struct iovec *iov, loff_t offset,
3110 unsigned long nr_segs)
3112 struct file *file = iocb->ki_filp;
3113 struct inode *inode = file->f_mapping->host;
3117 * If we are doing data journalling we don't support O_DIRECT
3119 if (ext4_should_journal_data(inode))
3122 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3123 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3124 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3126 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3127 trace_ext4_direct_IO_exit(inode, offset,
3128 iov_length(iov, nr_segs), rw, ret);
3133 * Pages can be marked dirty completely asynchronously from ext4's journalling
3134 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3135 * much here because ->set_page_dirty is called under VFS locks. The page is
3136 * not necessarily locked.
3138 * We cannot just dirty the page and leave attached buffers clean, because the
3139 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3140 * or jbddirty because all the journalling code will explode.
3142 * So what we do is to mark the page "pending dirty" and next time writepage
3143 * is called, propagate that into the buffers appropriately.
3145 static int ext4_journalled_set_page_dirty(struct page *page)
3147 SetPageChecked(page);
3148 return __set_page_dirty_nobuffers(page);
3151 static const struct address_space_operations ext4_ordered_aops = {
3152 .readpage = ext4_readpage,
3153 .readpages = ext4_readpages,
3154 .writepage = ext4_writepage,
3155 .write_begin = ext4_write_begin,
3156 .write_end = ext4_ordered_write_end,
3158 .invalidatepage = ext4_invalidatepage,
3159 .releasepage = ext4_releasepage,
3160 .direct_IO = ext4_direct_IO,
3161 .migratepage = buffer_migrate_page,
3162 .is_partially_uptodate = block_is_partially_uptodate,
3163 .error_remove_page = generic_error_remove_page,
3166 static const struct address_space_operations ext4_writeback_aops = {
3167 .readpage = ext4_readpage,
3168 .readpages = ext4_readpages,
3169 .writepage = ext4_writepage,
3170 .write_begin = ext4_write_begin,
3171 .write_end = ext4_writeback_write_end,
3173 .invalidatepage = ext4_invalidatepage,
3174 .releasepage = ext4_releasepage,
3175 .direct_IO = ext4_direct_IO,
3176 .migratepage = buffer_migrate_page,
3177 .is_partially_uptodate = block_is_partially_uptodate,
3178 .error_remove_page = generic_error_remove_page,
3181 static const struct address_space_operations ext4_journalled_aops = {
3182 .readpage = ext4_readpage,
3183 .readpages = ext4_readpages,
3184 .writepage = ext4_writepage,
3185 .write_begin = ext4_write_begin,
3186 .write_end = ext4_journalled_write_end,
3187 .set_page_dirty = ext4_journalled_set_page_dirty,
3189 .invalidatepage = ext4_invalidatepage,
3190 .releasepage = ext4_releasepage,
3191 .direct_IO = ext4_direct_IO,
3192 .is_partially_uptodate = block_is_partially_uptodate,
3193 .error_remove_page = generic_error_remove_page,
3196 static const struct address_space_operations ext4_da_aops = {
3197 .readpage = ext4_readpage,
3198 .readpages = ext4_readpages,
3199 .writepage = ext4_writepage,
3200 .writepages = ext4_da_writepages,
3201 .write_begin = ext4_da_write_begin,
3202 .write_end = ext4_da_write_end,
3204 .invalidatepage = ext4_da_invalidatepage,
3205 .releasepage = ext4_releasepage,
3206 .direct_IO = ext4_direct_IO,
3207 .migratepage = buffer_migrate_page,
3208 .is_partially_uptodate = block_is_partially_uptodate,
3209 .error_remove_page = generic_error_remove_page,
3212 void ext4_set_aops(struct inode *inode)
3214 switch (ext4_inode_journal_mode(inode)) {
3215 case EXT4_INODE_ORDERED_DATA_MODE:
3216 if (test_opt(inode->i_sb, DELALLOC))
3217 inode->i_mapping->a_ops = &ext4_da_aops;
3219 inode->i_mapping->a_ops = &ext4_ordered_aops;
3221 case EXT4_INODE_WRITEBACK_DATA_MODE:
3222 if (test_opt(inode->i_sb, DELALLOC))
3223 inode->i_mapping->a_ops = &ext4_da_aops;
3225 inode->i_mapping->a_ops = &ext4_writeback_aops;
3227 case EXT4_INODE_JOURNAL_DATA_MODE:
3228 inode->i_mapping->a_ops = &ext4_journalled_aops;
3237 * ext4_discard_partial_page_buffers()
3238 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3239 * This function finds and locks the page containing the offset
3240 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3241 * Calling functions that already have the page locked should call
3242 * ext4_discard_partial_page_buffers_no_lock directly.
3244 int ext4_discard_partial_page_buffers(handle_t *handle,
3245 struct address_space *mapping, loff_t from,
3246 loff_t length, int flags)
3248 struct inode *inode = mapping->host;
3252 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3253 mapping_gfp_mask(mapping) & ~__GFP_FS);
3257 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3258 from, length, flags);
3261 page_cache_release(page);
3266 * ext4_discard_partial_page_buffers_no_lock()
3267 * Zeros a page range of length 'length' starting from offset 'from'.
3268 * Buffer heads that correspond to the block aligned regions of the
3269 * zeroed range will be unmapped. Unblock aligned regions
3270 * will have the corresponding buffer head mapped if needed so that
3271 * that region of the page can be updated with the partial zero out.
3273 * This function assumes that the page has already been locked. The
3274 * The range to be discarded must be contained with in the given page.
3275 * If the specified range exceeds the end of the page it will be shortened
3276 * to the end of the page that corresponds to 'from'. This function is
3277 * appropriate for updating a page and it buffer heads to be unmapped and
3278 * zeroed for blocks that have been either released, or are going to be
3281 * handle: The journal handle
3282 * inode: The files inode
3283 * page: A locked page that contains the offset "from"
3284 * from: The starting byte offset (from the beginning of the file)
3285 * to begin discarding
3286 * len: The length of bytes to discard
3287 * flags: Optional flags that may be used:
3289 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3290 * Only zero the regions of the page whose buffer heads
3291 * have already been unmapped. This flag is appropriate
3292 * for updating the contents of a page whose blocks may
3293 * have already been released, and we only want to zero
3294 * out the regions that correspond to those released blocks.
3296 * Returns zero on success or negative on failure.
3298 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3299 struct inode *inode, struct page *page, loff_t from,
3300 loff_t length, int flags)
3302 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3303 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3304 unsigned int blocksize, max, pos;
3306 struct buffer_head *bh;
3309 blocksize = inode->i_sb->s_blocksize;
3310 max = PAGE_CACHE_SIZE - offset;
3312 if (index != page->index)
3316 * correct length if it does not fall between
3317 * 'from' and the end of the page
3319 if (length > max || length < 0)
3322 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3324 if (!page_has_buffers(page))
3325 create_empty_buffers(page, blocksize, 0);
3327 /* Find the buffer that contains "offset" */
3328 bh = page_buffers(page);
3330 while (offset >= pos) {
3331 bh = bh->b_this_page;
3337 while (pos < offset + length) {
3338 unsigned int end_of_block, range_to_discard;
3342 /* The length of space left to zero and unmap */
3343 range_to_discard = offset + length - pos;
3345 /* The length of space until the end of the block */
3346 end_of_block = blocksize - (pos & (blocksize-1));
3349 * Do not unmap or zero past end of block
3350 * for this buffer head
3352 if (range_to_discard > end_of_block)
3353 range_to_discard = end_of_block;
3357 * Skip this buffer head if we are only zeroing unampped
3358 * regions of the page
3360 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3364 /* If the range is block aligned, unmap */
3365 if (range_to_discard == blocksize) {
3366 clear_buffer_dirty(bh);
3368 clear_buffer_mapped(bh);
3369 clear_buffer_req(bh);
3370 clear_buffer_new(bh);
3371 clear_buffer_delay(bh);
3372 clear_buffer_unwritten(bh);
3373 clear_buffer_uptodate(bh);
3374 zero_user(page, pos, range_to_discard);
3375 BUFFER_TRACE(bh, "Buffer discarded");
3380 * If this block is not completely contained in the range
3381 * to be discarded, then it is not going to be released. Because
3382 * we need to keep this block, we need to make sure this part
3383 * of the page is uptodate before we modify it by writeing
3384 * partial zeros on it.
3386 if (!buffer_mapped(bh)) {
3388 * Buffer head must be mapped before we can read
3391 BUFFER_TRACE(bh, "unmapped");
3392 ext4_get_block(inode, iblock, bh, 0);
3393 /* unmapped? It's a hole - nothing to do */
3394 if (!buffer_mapped(bh)) {
3395 BUFFER_TRACE(bh, "still unmapped");
3400 /* Ok, it's mapped. Make sure it's up-to-date */
3401 if (PageUptodate(page))
3402 set_buffer_uptodate(bh);
3404 if (!buffer_uptodate(bh)) {
3406 ll_rw_block(READ, 1, &bh);
3408 /* Uhhuh. Read error. Complain and punt.*/
3409 if (!buffer_uptodate(bh))
3413 if (ext4_should_journal_data(inode)) {
3414 BUFFER_TRACE(bh, "get write access");
3415 err = ext4_journal_get_write_access(handle, bh);
3420 zero_user(page, pos, range_to_discard);
3423 if (ext4_should_journal_data(inode)) {
3424 err = ext4_handle_dirty_metadata(handle, inode, bh);
3426 mark_buffer_dirty(bh);
3428 BUFFER_TRACE(bh, "Partial buffer zeroed");
3430 bh = bh->b_this_page;
3432 pos += range_to_discard;
3438 int ext4_can_truncate(struct inode *inode)
3440 if (S_ISREG(inode->i_mode))
3442 if (S_ISDIR(inode->i_mode))
3444 if (S_ISLNK(inode->i_mode))
3445 return !ext4_inode_is_fast_symlink(inode);
3450 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3451 * associated with the given offset and length
3453 * @inode: File inode
3454 * @offset: The offset where the hole will begin
3455 * @len: The length of the hole
3457 * Returns: 0 on success or negative on failure
3460 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3462 struct inode *inode = file->f_path.dentry->d_inode;
3463 if (!S_ISREG(inode->i_mode))
3466 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3467 /* TODO: Add support for non extent hole punching */
3471 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3472 /* TODO: Add support for bigalloc file systems */
3476 return ext4_ext_punch_hole(file, offset, length);
3482 * We block out ext4_get_block() block instantiations across the entire
3483 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3484 * simultaneously on behalf of the same inode.
3486 * As we work through the truncate and commit bits of it to the journal there
3487 * is one core, guiding principle: the file's tree must always be consistent on
3488 * disk. We must be able to restart the truncate after a crash.
3490 * The file's tree may be transiently inconsistent in memory (although it
3491 * probably isn't), but whenever we close off and commit a journal transaction,
3492 * the contents of (the filesystem + the journal) must be consistent and
3493 * restartable. It's pretty simple, really: bottom up, right to left (although
3494 * left-to-right works OK too).
3496 * Note that at recovery time, journal replay occurs *before* the restart of
3497 * truncate against the orphan inode list.
3499 * The committed inode has the new, desired i_size (which is the same as
3500 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3501 * that this inode's truncate did not complete and it will again call
3502 * ext4_truncate() to have another go. So there will be instantiated blocks
3503 * to the right of the truncation point in a crashed ext4 filesystem. But
3504 * that's fine - as long as they are linked from the inode, the post-crash
3505 * ext4_truncate() run will find them and release them.
3507 void ext4_truncate(struct inode *inode)
3509 trace_ext4_truncate_enter(inode);
3511 if (!ext4_can_truncate(inode))
3514 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3516 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3517 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3519 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3520 ext4_ext_truncate(inode);
3522 ext4_ind_truncate(inode);
3524 trace_ext4_truncate_exit(inode);
3528 * ext4_get_inode_loc returns with an extra refcount against the inode's
3529 * underlying buffer_head on success. If 'in_mem' is true, we have all
3530 * data in memory that is needed to recreate the on-disk version of this
3533 static int __ext4_get_inode_loc(struct inode *inode,
3534 struct ext4_iloc *iloc, int in_mem)
3536 struct ext4_group_desc *gdp;
3537 struct buffer_head *bh;
3538 struct super_block *sb = inode->i_sb;
3540 int inodes_per_block, inode_offset;
3543 if (!ext4_valid_inum(sb, inode->i_ino))
3546 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3547 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3552 * Figure out the offset within the block group inode table
3554 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3555 inode_offset = ((inode->i_ino - 1) %
3556 EXT4_INODES_PER_GROUP(sb));
3557 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3558 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3560 bh = sb_getblk(sb, block);
3562 EXT4_ERROR_INODE_BLOCK(inode, block,
3563 "unable to read itable block");
3566 if (!buffer_uptodate(bh)) {
3570 * If the buffer has the write error flag, we have failed
3571 * to write out another inode in the same block. In this
3572 * case, we don't have to read the block because we may
3573 * read the old inode data successfully.
3575 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3576 set_buffer_uptodate(bh);
3578 if (buffer_uptodate(bh)) {
3579 /* someone brought it uptodate while we waited */
3585 * If we have all information of the inode in memory and this
3586 * is the only valid inode in the block, we need not read the
3590 struct buffer_head *bitmap_bh;
3593 start = inode_offset & ~(inodes_per_block - 1);
3595 /* Is the inode bitmap in cache? */
3596 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3601 * If the inode bitmap isn't in cache then the
3602 * optimisation may end up performing two reads instead
3603 * of one, so skip it.
3605 if (!buffer_uptodate(bitmap_bh)) {
3609 for (i = start; i < start + inodes_per_block; i++) {
3610 if (i == inode_offset)
3612 if (ext4_test_bit(i, bitmap_bh->b_data))
3616 if (i == start + inodes_per_block) {
3617 /* all other inodes are free, so skip I/O */
3618 memset(bh->b_data, 0, bh->b_size);
3619 set_buffer_uptodate(bh);
3627 * If we need to do any I/O, try to pre-readahead extra
3628 * blocks from the inode table.
3630 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3631 ext4_fsblk_t b, end, table;
3634 table = ext4_inode_table(sb, gdp);
3635 /* s_inode_readahead_blks is always a power of 2 */
3636 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3639 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3640 num = EXT4_INODES_PER_GROUP(sb);
3641 if (ext4_has_group_desc_csum(sb))
3642 num -= ext4_itable_unused_count(sb, gdp);
3643 table += num / inodes_per_block;
3647 sb_breadahead(sb, b++);
3651 * There are other valid inodes in the buffer, this inode
3652 * has in-inode xattrs, or we don't have this inode in memory.
3653 * Read the block from disk.
3655 trace_ext4_load_inode(inode);
3657 bh->b_end_io = end_buffer_read_sync;
3658 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3660 if (!buffer_uptodate(bh)) {
3661 EXT4_ERROR_INODE_BLOCK(inode, block,
3662 "unable to read itable block");
3672 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3674 /* We have all inode data except xattrs in memory here. */
3675 return __ext4_get_inode_loc(inode, iloc,
3676 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3679 void ext4_set_inode_flags(struct inode *inode)
3681 unsigned int flags = EXT4_I(inode)->i_flags;
3683 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3684 if (flags & EXT4_SYNC_FL)
3685 inode->i_flags |= S_SYNC;
3686 if (flags & EXT4_APPEND_FL)
3687 inode->i_flags |= S_APPEND;
3688 if (flags & EXT4_IMMUTABLE_FL)
3689 inode->i_flags |= S_IMMUTABLE;
3690 if (flags & EXT4_NOATIME_FL)
3691 inode->i_flags |= S_NOATIME;
3692 if (flags & EXT4_DIRSYNC_FL)
3693 inode->i_flags |= S_DIRSYNC;
3696 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3697 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3699 unsigned int vfs_fl;
3700 unsigned long old_fl, new_fl;
3703 vfs_fl = ei->vfs_inode.i_flags;
3704 old_fl = ei->i_flags;
3705 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3706 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3708 if (vfs_fl & S_SYNC)
3709 new_fl |= EXT4_SYNC_FL;
3710 if (vfs_fl & S_APPEND)
3711 new_fl |= EXT4_APPEND_FL;
3712 if (vfs_fl & S_IMMUTABLE)
3713 new_fl |= EXT4_IMMUTABLE_FL;
3714 if (vfs_fl & S_NOATIME)
3715 new_fl |= EXT4_NOATIME_FL;
3716 if (vfs_fl & S_DIRSYNC)
3717 new_fl |= EXT4_DIRSYNC_FL;
3718 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3721 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3722 struct ext4_inode_info *ei)
3725 struct inode *inode = &(ei->vfs_inode);
3726 struct super_block *sb = inode->i_sb;
3728 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3729 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3730 /* we are using combined 48 bit field */
3731 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3732 le32_to_cpu(raw_inode->i_blocks_lo);
3733 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3734 /* i_blocks represent file system block size */
3735 return i_blocks << (inode->i_blkbits - 9);
3740 return le32_to_cpu(raw_inode->i_blocks_lo);
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_dir_start_lookup = 0;
3815 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3816 /* We now have enough fields to check if the inode was active or not.
3817 * This is needed because nfsd might try to access dead inodes
3818 * the test is that same one that e2fsck uses
3819 * NeilBrown 1999oct15
3821 if (inode->i_nlink == 0) {
3822 if (inode->i_mode == 0 ||
3823 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3824 /* this inode is deleted */
3828 /* The only unlinked inodes we let through here have
3829 * valid i_mode and are being read by the orphan
3830 * recovery code: that's fine, we're about to complete
3831 * the process of deleting those. */
3833 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3834 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3835 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3836 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3838 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3839 inode->i_size = ext4_isize(raw_inode);
3840 ei->i_disksize = inode->i_size;
3842 ei->i_reserved_quota = 0;
3844 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3845 ei->i_block_group = iloc.block_group;
3846 ei->i_last_alloc_group = ~0;
3848 * NOTE! The in-memory inode i_data array is in little-endian order
3849 * even on big-endian machines: we do NOT byteswap the block numbers!
3851 for (block = 0; block < EXT4_N_BLOCKS; block++)
3852 ei->i_data[block] = raw_inode->i_block[block];
3853 INIT_LIST_HEAD(&ei->i_orphan);
3856 * Set transaction id's of transactions that have to be committed
3857 * to finish f[data]sync. We set them to currently running transaction
3858 * as we cannot be sure that the inode or some of its metadata isn't
3859 * part of the transaction - the inode could have been reclaimed and
3860 * now it is reread from disk.
3863 transaction_t *transaction;
3866 read_lock(&journal->j_state_lock);
3867 if (journal->j_running_transaction)
3868 transaction = journal->j_running_transaction;
3870 transaction = journal->j_committing_transaction;
3872 tid = transaction->t_tid;
3874 tid = journal->j_commit_sequence;
3875 read_unlock(&journal->j_state_lock);
3876 ei->i_sync_tid = tid;
3877 ei->i_datasync_tid = tid;
3880 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3881 if (ei->i_extra_isize == 0) {
3882 /* The extra space is currently unused. Use it. */
3883 ei->i_extra_isize = sizeof(struct ext4_inode) -
3884 EXT4_GOOD_OLD_INODE_SIZE;
3886 __le32 *magic = (void *)raw_inode +
3887 EXT4_GOOD_OLD_INODE_SIZE +
3889 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3890 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3894 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3895 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3896 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3897 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3899 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3900 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3901 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3903 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3907 if (ei->i_file_acl &&
3908 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3909 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3913 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3914 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3915 (S_ISLNK(inode->i_mode) &&
3916 !ext4_inode_is_fast_symlink(inode)))
3917 /* Validate extent which is part of inode */
3918 ret = ext4_ext_check_inode(inode);
3919 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3920 (S_ISLNK(inode->i_mode) &&
3921 !ext4_inode_is_fast_symlink(inode))) {
3922 /* Validate block references which are part of inode */
3923 ret = ext4_ind_check_inode(inode);
3928 if (S_ISREG(inode->i_mode)) {
3929 inode->i_op = &ext4_file_inode_operations;
3930 inode->i_fop = &ext4_file_operations;
3931 ext4_set_aops(inode);
3932 } else if (S_ISDIR(inode->i_mode)) {
3933 inode->i_op = &ext4_dir_inode_operations;
3934 inode->i_fop = &ext4_dir_operations;
3935 } else if (S_ISLNK(inode->i_mode)) {
3936 if (ext4_inode_is_fast_symlink(inode)) {
3937 inode->i_op = &ext4_fast_symlink_inode_operations;
3938 nd_terminate_link(ei->i_data, inode->i_size,
3939 sizeof(ei->i_data) - 1);
3941 inode->i_op = &ext4_symlink_inode_operations;
3942 ext4_set_aops(inode);
3944 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3945 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3946 inode->i_op = &ext4_special_inode_operations;
3947 if (raw_inode->i_block[0])
3948 init_special_inode(inode, inode->i_mode,
3949 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3951 init_special_inode(inode, inode->i_mode,
3952 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3955 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3959 ext4_set_inode_flags(inode);
3960 unlock_new_inode(inode);
3966 return ERR_PTR(ret);
3969 static int ext4_inode_blocks_set(handle_t *handle,
3970 struct ext4_inode *raw_inode,
3971 struct ext4_inode_info *ei)
3973 struct inode *inode = &(ei->vfs_inode);
3974 u64 i_blocks = inode->i_blocks;
3975 struct super_block *sb = inode->i_sb;
3977 if (i_blocks <= ~0U) {
3979 * i_blocks can be represented in a 32 bit variable
3980 * as multiple of 512 bytes
3982 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3983 raw_inode->i_blocks_high = 0;
3984 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3987 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3990 if (i_blocks <= 0xffffffffffffULL) {
3992 * i_blocks can be represented in a 48 bit variable
3993 * as multiple of 512 bytes
3995 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3996 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3997 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3999 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4000 /* i_block is stored in file system block size */
4001 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4002 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4003 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4009 * Post the struct inode info into an on-disk inode location in the
4010 * buffer-cache. This gobbles the caller's reference to the
4011 * buffer_head in the inode location struct.
4013 * The caller must have write access to iloc->bh.
4015 static int ext4_do_update_inode(handle_t *handle,
4016 struct inode *inode,
4017 struct ext4_iloc *iloc)
4019 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4020 struct ext4_inode_info *ei = EXT4_I(inode);
4021 struct buffer_head *bh = iloc->bh;
4022 int err = 0, rc, block;
4023 int need_datasync = 0;
4027 /* For fields not not tracking in the in-memory inode,
4028 * initialise them to zero for new inodes. */
4029 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4030 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4032 ext4_get_inode_flags(ei);
4033 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4034 i_uid = i_uid_read(inode);
4035 i_gid = i_gid_read(inode);
4036 if (!(test_opt(inode->i_sb, NO_UID32))) {
4037 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4038 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4040 * Fix up interoperability with old kernels. Otherwise, old inodes get
4041 * re-used with the upper 16 bits of the uid/gid intact
4044 raw_inode->i_uid_high =
4045 cpu_to_le16(high_16_bits(i_uid));
4046 raw_inode->i_gid_high =
4047 cpu_to_le16(high_16_bits(i_gid));
4049 raw_inode->i_uid_high = 0;
4050 raw_inode->i_gid_high = 0;
4053 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4054 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4055 raw_inode->i_uid_high = 0;
4056 raw_inode->i_gid_high = 0;
4058 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4060 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4061 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4062 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4063 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4065 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4067 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4068 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4069 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4070 cpu_to_le32(EXT4_OS_HURD))
4071 raw_inode->i_file_acl_high =
4072 cpu_to_le16(ei->i_file_acl >> 32);
4073 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4074 if (ei->i_disksize != ext4_isize(raw_inode)) {
4075 ext4_isize_set(raw_inode, ei->i_disksize);
4078 if (ei->i_disksize > 0x7fffffffULL) {
4079 struct super_block *sb = inode->i_sb;
4080 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4081 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4082 EXT4_SB(sb)->s_es->s_rev_level ==
4083 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4084 /* If this is the first large file
4085 * created, add a flag to the superblock.
4087 err = ext4_journal_get_write_access(handle,
4088 EXT4_SB(sb)->s_sbh);
4091 ext4_update_dynamic_rev(sb);
4092 EXT4_SET_RO_COMPAT_FEATURE(sb,
4093 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4094 ext4_handle_sync(handle);
4095 err = ext4_handle_dirty_super(handle, sb);
4098 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4099 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4100 if (old_valid_dev(inode->i_rdev)) {
4101 raw_inode->i_block[0] =
4102 cpu_to_le32(old_encode_dev(inode->i_rdev));
4103 raw_inode->i_block[1] = 0;
4105 raw_inode->i_block[0] = 0;
4106 raw_inode->i_block[1] =
4107 cpu_to_le32(new_encode_dev(inode->i_rdev));
4108 raw_inode->i_block[2] = 0;
4111 for (block = 0; block < EXT4_N_BLOCKS; block++)
4112 raw_inode->i_block[block] = ei->i_data[block];
4114 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4115 if (ei->i_extra_isize) {
4116 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4117 raw_inode->i_version_hi =
4118 cpu_to_le32(inode->i_version >> 32);
4119 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4122 ext4_inode_csum_set(inode, raw_inode, ei);
4124 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4125 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4128 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4130 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4133 ext4_std_error(inode->i_sb, err);
4138 * ext4_write_inode()
4140 * We are called from a few places:
4142 * - Within generic_file_write() for O_SYNC files.
4143 * Here, there will be no transaction running. We wait for any running
4144 * transaction to commit.
4146 * - Within sys_sync(), kupdate and such.
4147 * We wait on commit, if tol to.
4149 * - Within prune_icache() (PF_MEMALLOC == true)
4150 * Here we simply return. We can't afford to block kswapd on the
4153 * In all cases it is actually safe for us to return without doing anything,
4154 * because the inode has been copied into a raw inode buffer in
4155 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4158 * Note that we are absolutely dependent upon all inode dirtiers doing the
4159 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4160 * which we are interested.
4162 * It would be a bug for them to not do this. The code:
4164 * mark_inode_dirty(inode)
4166 * inode->i_size = expr;
4168 * is in error because a kswapd-driven write_inode() could occur while
4169 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4170 * will no longer be on the superblock's dirty inode list.
4172 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4176 if (current->flags & PF_MEMALLOC)
4179 if (EXT4_SB(inode->i_sb)->s_journal) {
4180 if (ext4_journal_current_handle()) {
4181 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4186 if (wbc->sync_mode != WB_SYNC_ALL)
4189 err = ext4_force_commit(inode->i_sb);
4191 struct ext4_iloc iloc;
4193 err = __ext4_get_inode_loc(inode, &iloc, 0);
4196 if (wbc->sync_mode == WB_SYNC_ALL)
4197 sync_dirty_buffer(iloc.bh);
4198 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4199 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4200 "IO error syncing inode");
4211 * Called from notify_change.
4213 * We want to trap VFS attempts to truncate the file as soon as
4214 * possible. In particular, we want to make sure that when the VFS
4215 * shrinks i_size, we put the inode on the orphan list and modify
4216 * i_disksize immediately, so that during the subsequent flushing of
4217 * dirty pages and freeing of disk blocks, we can guarantee that any
4218 * commit will leave the blocks being flushed in an unused state on
4219 * disk. (On recovery, the inode will get truncated and the blocks will
4220 * be freed, so we have a strong guarantee that no future commit will
4221 * leave these blocks visible to the user.)
4223 * Another thing we have to assure is that if we are in ordered mode
4224 * and inode is still attached to the committing transaction, we must
4225 * we start writeout of all the dirty pages which are being truncated.
4226 * This way we are sure that all the data written in the previous
4227 * transaction are already on disk (truncate waits for pages under
4230 * Called with inode->i_mutex down.
4232 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4234 struct inode *inode = dentry->d_inode;
4237 const unsigned int ia_valid = attr->ia_valid;
4239 error = inode_change_ok(inode, attr);
4243 if (is_quota_modification(inode, attr))
4244 dquot_initialize(inode);
4245 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4246 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4249 /* (user+group)*(old+new) structure, inode write (sb,
4250 * inode block, ? - but truncate inode update has it) */
4251 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4252 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4253 if (IS_ERR(handle)) {
4254 error = PTR_ERR(handle);
4257 error = dquot_transfer(inode, attr);
4259 ext4_journal_stop(handle);
4262 /* Update corresponding info in inode so that everything is in
4263 * one transaction */
4264 if (attr->ia_valid & ATTR_UID)
4265 inode->i_uid = attr->ia_uid;
4266 if (attr->ia_valid & ATTR_GID)
4267 inode->i_gid = attr->ia_gid;
4268 error = ext4_mark_inode_dirty(handle, inode);
4269 ext4_journal_stop(handle);
4272 if (attr->ia_valid & ATTR_SIZE) {
4274 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4275 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4277 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4282 if (S_ISREG(inode->i_mode) &&
4283 attr->ia_valid & ATTR_SIZE &&
4284 (attr->ia_size < inode->i_size)) {
4287 handle = ext4_journal_start(inode, 3);
4288 if (IS_ERR(handle)) {
4289 error = PTR_ERR(handle);
4292 if (ext4_handle_valid(handle)) {
4293 error = ext4_orphan_add(handle, inode);
4296 EXT4_I(inode)->i_disksize = attr->ia_size;
4297 rc = ext4_mark_inode_dirty(handle, inode);
4300 ext4_journal_stop(handle);
4302 if (ext4_should_order_data(inode)) {
4303 error = ext4_begin_ordered_truncate(inode,
4306 /* Do as much error cleanup as possible */
4307 handle = ext4_journal_start(inode, 3);
4308 if (IS_ERR(handle)) {
4309 ext4_orphan_del(NULL, inode);
4312 ext4_orphan_del(handle, inode);
4314 ext4_journal_stop(handle);
4320 if (attr->ia_valid & ATTR_SIZE) {
4321 if (attr->ia_size != i_size_read(inode)) {
4322 truncate_setsize(inode, attr->ia_size);
4323 /* Inode size will be reduced, wait for dio in flight.
4324 * Temporarily disable dioread_nolock to prevent
4327 ext4_inode_block_unlocked_dio(inode);
4328 inode_dio_wait(inode);
4329 ext4_inode_resume_unlocked_dio(inode);
4332 ext4_truncate(inode);
4336 setattr_copy(inode, attr);
4337 mark_inode_dirty(inode);
4341 * If the call to ext4_truncate failed to get a transaction handle at
4342 * all, we need to clean up the in-core orphan list manually.
4344 if (orphan && inode->i_nlink)
4345 ext4_orphan_del(NULL, inode);
4347 if (!rc && (ia_valid & ATTR_MODE))
4348 rc = ext4_acl_chmod(inode);
4351 ext4_std_error(inode->i_sb, error);
4357 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4360 struct inode *inode;
4361 unsigned long delalloc_blocks;
4363 inode = dentry->d_inode;
4364 generic_fillattr(inode, stat);
4367 * We can't update i_blocks if the block allocation is delayed
4368 * otherwise in the case of system crash before the real block
4369 * allocation is done, we will have i_blocks inconsistent with
4370 * on-disk file blocks.
4371 * We always keep i_blocks updated together with real
4372 * allocation. But to not confuse with user, stat
4373 * will return the blocks that include the delayed allocation
4374 * blocks for this file.
4376 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4377 EXT4_I(inode)->i_reserved_data_blocks);
4379 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4383 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4385 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4386 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4387 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4391 * Account for index blocks, block groups bitmaps and block group
4392 * descriptor blocks if modify datablocks and index blocks
4393 * worse case, the indexs blocks spread over different block groups
4395 * If datablocks are discontiguous, they are possible to spread over
4396 * different block groups too. If they are contiguous, with flexbg,
4397 * they could still across block group boundary.
4399 * Also account for superblock, inode, quota and xattr blocks
4401 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4403 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4409 * How many index blocks need to touch to modify nrblocks?
4410 * The "Chunk" flag indicating whether the nrblocks is
4411 * physically contiguous on disk
4413 * For Direct IO and fallocate, they calls get_block to allocate
4414 * one single extent at a time, so they could set the "Chunk" flag
4416 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4421 * Now let's see how many group bitmaps and group descriptors need
4431 if (groups > ngroups)
4433 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4434 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4436 /* bitmaps and block group descriptor blocks */
4437 ret += groups + gdpblocks;
4439 /* Blocks for super block, inode, quota and xattr blocks */
4440 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4446 * Calculate the total number of credits to reserve to fit
4447 * the modification of a single pages into a single transaction,
4448 * which may include multiple chunks of block allocations.
4450 * This could be called via ext4_write_begin()
4452 * We need to consider the worse case, when
4453 * one new block per extent.
4455 int ext4_writepage_trans_blocks(struct inode *inode)
4457 int bpp = ext4_journal_blocks_per_page(inode);
4460 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4462 /* Account for data blocks for journalled mode */
4463 if (ext4_should_journal_data(inode))
4469 * Calculate the journal credits for a chunk of data modification.
4471 * This is called from DIO, fallocate or whoever calling
4472 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4474 * journal buffers for data blocks are not included here, as DIO
4475 * and fallocate do no need to journal data buffers.
4477 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4479 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4483 * The caller must have previously called ext4_reserve_inode_write().
4484 * Give this, we know that the caller already has write access to iloc->bh.
4486 int ext4_mark_iloc_dirty(handle_t *handle,
4487 struct inode *inode, struct ext4_iloc *iloc)
4491 if (IS_I_VERSION(inode))
4492 inode_inc_iversion(inode);
4494 /* the do_update_inode consumes one bh->b_count */
4497 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4498 err = ext4_do_update_inode(handle, inode, iloc);
4504 * On success, We end up with an outstanding reference count against
4505 * iloc->bh. This _must_ be cleaned up later.
4509 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4510 struct ext4_iloc *iloc)
4514 err = ext4_get_inode_loc(inode, iloc);
4516 BUFFER_TRACE(iloc->bh, "get_write_access");
4517 err = ext4_journal_get_write_access(handle, iloc->bh);
4523 ext4_std_error(inode->i_sb, err);
4528 * Expand an inode by new_extra_isize bytes.
4529 * Returns 0 on success or negative error number on failure.
4531 static int ext4_expand_extra_isize(struct inode *inode,
4532 unsigned int new_extra_isize,
4533 struct ext4_iloc iloc,
4536 struct ext4_inode *raw_inode;
4537 struct ext4_xattr_ibody_header *header;
4539 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4542 raw_inode = ext4_raw_inode(&iloc);
4544 header = IHDR(inode, raw_inode);
4546 /* No extended attributes present */
4547 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4548 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4549 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4551 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4555 /* try to expand with EAs present */
4556 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4561 * What we do here is to mark the in-core inode as clean with respect to inode
4562 * dirtiness (it may still be data-dirty).
4563 * This means that the in-core inode may be reaped by prune_icache
4564 * without having to perform any I/O. This is a very good thing,
4565 * because *any* task may call prune_icache - even ones which
4566 * have a transaction open against a different journal.
4568 * Is this cheating? Not really. Sure, we haven't written the
4569 * inode out, but prune_icache isn't a user-visible syncing function.
4570 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4571 * we start and wait on commits.
4573 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4575 struct ext4_iloc iloc;
4576 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4577 static unsigned int mnt_count;
4581 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4582 err = ext4_reserve_inode_write(handle, inode, &iloc);
4583 if (ext4_handle_valid(handle) &&
4584 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4585 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4587 * We need extra buffer credits since we may write into EA block
4588 * with this same handle. If journal_extend fails, then it will
4589 * only result in a minor loss of functionality for that inode.
4590 * If this is felt to be critical, then e2fsck should be run to
4591 * force a large enough s_min_extra_isize.
4593 if ((jbd2_journal_extend(handle,
4594 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4595 ret = ext4_expand_extra_isize(inode,
4596 sbi->s_want_extra_isize,
4599 ext4_set_inode_state(inode,
4600 EXT4_STATE_NO_EXPAND);
4602 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4603 ext4_warning(inode->i_sb,
4604 "Unable to expand inode %lu. Delete"
4605 " some EAs or run e2fsck.",
4608 le16_to_cpu(sbi->s_es->s_mnt_count);
4614 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4619 * ext4_dirty_inode() is called from __mark_inode_dirty()
4621 * We're really interested in the case where a file is being extended.
4622 * i_size has been changed by generic_commit_write() and we thus need
4623 * to include the updated inode in the current transaction.
4625 * Also, dquot_alloc_block() will always dirty the inode when blocks
4626 * are allocated to the file.
4628 * If the inode is marked synchronous, we don't honour that here - doing
4629 * so would cause a commit on atime updates, which we don't bother doing.
4630 * We handle synchronous inodes at the highest possible level.
4632 void ext4_dirty_inode(struct inode *inode, int flags)
4636 handle = ext4_journal_start(inode, 2);
4640 ext4_mark_inode_dirty(handle, inode);
4642 ext4_journal_stop(handle);
4649 * Bind an inode's backing buffer_head into this transaction, to prevent
4650 * it from being flushed to disk early. Unlike
4651 * ext4_reserve_inode_write, this leaves behind no bh reference and
4652 * returns no iloc structure, so the caller needs to repeat the iloc
4653 * lookup to mark the inode dirty later.
4655 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4657 struct ext4_iloc iloc;
4661 err = ext4_get_inode_loc(inode, &iloc);
4663 BUFFER_TRACE(iloc.bh, "get_write_access");
4664 err = jbd2_journal_get_write_access(handle, iloc.bh);
4666 err = ext4_handle_dirty_metadata(handle,
4672 ext4_std_error(inode->i_sb, err);
4677 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4684 * We have to be very careful here: changing a data block's
4685 * journaling status dynamically is dangerous. If we write a
4686 * data block to the journal, change the status and then delete
4687 * that block, we risk forgetting to revoke the old log record
4688 * from the journal and so a subsequent replay can corrupt data.
4689 * So, first we make sure that the journal is empty and that
4690 * nobody is changing anything.
4693 journal = EXT4_JOURNAL(inode);
4696 if (is_journal_aborted(journal))
4698 /* We have to allocate physical blocks for delalloc blocks
4699 * before flushing journal. otherwise delalloc blocks can not
4700 * be allocated any more. even more truncate on delalloc blocks
4701 * could trigger BUG by flushing delalloc blocks in journal.
4702 * There is no delalloc block in non-journal data mode.
4704 if (val && test_opt(inode->i_sb, DELALLOC)) {
4705 err = ext4_alloc_da_blocks(inode);
4710 /* Wait for all existing dio workers */
4711 ext4_inode_block_unlocked_dio(inode);
4712 inode_dio_wait(inode);
4714 jbd2_journal_lock_updates(journal);
4717 * OK, there are no updates running now, and all cached data is
4718 * synced to disk. We are now in a completely consistent state
4719 * which doesn't have anything in the journal, and we know that
4720 * no filesystem updates are running, so it is safe to modify
4721 * the inode's in-core data-journaling state flag now.
4725 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4727 jbd2_journal_flush(journal);
4728 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4730 ext4_set_aops(inode);
4732 jbd2_journal_unlock_updates(journal);
4733 ext4_inode_resume_unlocked_dio(inode);
4735 /* Finally we can mark the inode as dirty. */
4737 handle = ext4_journal_start(inode, 1);
4739 return PTR_ERR(handle);
4741 err = ext4_mark_inode_dirty(handle, inode);
4742 ext4_handle_sync(handle);
4743 ext4_journal_stop(handle);
4744 ext4_std_error(inode->i_sb, err);
4749 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4751 return !buffer_mapped(bh);
4754 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4756 struct page *page = vmf->page;
4760 struct file *file = vma->vm_file;
4761 struct inode *inode = file->f_path.dentry->d_inode;
4762 struct address_space *mapping = inode->i_mapping;
4764 get_block_t *get_block;
4767 sb_start_pagefault(inode->i_sb);
4768 file_update_time(vma->vm_file);
4769 /* Delalloc case is easy... */
4770 if (test_opt(inode->i_sb, DELALLOC) &&
4771 !ext4_should_journal_data(inode) &&
4772 !ext4_nonda_switch(inode->i_sb)) {
4774 ret = __block_page_mkwrite(vma, vmf,
4775 ext4_da_get_block_prep);
4776 } while (ret == -ENOSPC &&
4777 ext4_should_retry_alloc(inode->i_sb, &retries));
4782 size = i_size_read(inode);
4783 /* Page got truncated from under us? */
4784 if (page->mapping != mapping || page_offset(page) > size) {
4786 ret = VM_FAULT_NOPAGE;
4790 if (page->index == size >> PAGE_CACHE_SHIFT)
4791 len = size & ~PAGE_CACHE_MASK;
4793 len = PAGE_CACHE_SIZE;
4795 * Return if we have all the buffers mapped. This avoids the need to do
4796 * journal_start/journal_stop which can block and take a long time
4798 if (page_has_buffers(page)) {
4799 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4800 ext4_bh_unmapped)) {
4801 /* Wait so that we don't change page under IO */
4802 wait_on_page_writeback(page);
4803 ret = VM_FAULT_LOCKED;
4808 /* OK, we need to fill the hole... */
4809 if (ext4_should_dioread_nolock(inode))
4810 get_block = ext4_get_block_write;
4812 get_block = ext4_get_block;
4814 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4815 if (IS_ERR(handle)) {
4816 ret = VM_FAULT_SIGBUS;
4819 ret = __block_page_mkwrite(vma, vmf, get_block);
4820 if (!ret && ext4_should_journal_data(inode)) {
4821 if (walk_page_buffers(handle, page_buffers(page), 0,
4822 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4824 ret = VM_FAULT_SIGBUS;
4825 ext4_journal_stop(handle);
4828 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4830 ext4_journal_stop(handle);
4831 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4834 ret = block_page_mkwrite_return(ret);
4836 sb_end_pagefault(inode->i_sb);