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>
40 #include <linux/aio.h>
42 #include "ext4_jbd2.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 struct ext4_inode_info *ei)
54 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
59 csum_lo = le16_to_cpu(raw->i_checksum_lo);
60 raw->i_checksum_lo = 0;
61 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
62 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
63 csum_hi = le16_to_cpu(raw->i_checksum_hi);
64 raw->i_checksum_hi = 0;
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
68 EXT4_INODE_SIZE(inode->i_sb));
70 raw->i_checksum_lo = cpu_to_le16(csum_lo);
71 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
72 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
73 raw->i_checksum_hi = cpu_to_le16(csum_hi);
78 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
79 struct ext4_inode_info *ei)
81 __u32 provided, calculated;
83 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
84 cpu_to_le32(EXT4_OS_LINUX) ||
85 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
86 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
89 provided = le16_to_cpu(raw->i_checksum_lo);
90 calculated = ext4_inode_csum(inode, raw, ei);
91 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
92 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
93 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
97 return provided == calculated;
100 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
101 struct ext4_inode_info *ei)
105 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
106 cpu_to_le32(EXT4_OS_LINUX) ||
107 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
108 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
111 csum = ext4_inode_csum(inode, raw, ei);
112 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
113 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
114 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
115 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
118 static inline int ext4_begin_ordered_truncate(struct inode *inode,
121 trace_ext4_begin_ordered_truncate(inode, new_size);
123 * If jinode is zero, then we never opened the file for
124 * writing, so there's no need to call
125 * jbd2_journal_begin_ordered_truncate() since there's no
126 * outstanding writes we need to flush.
128 if (!EXT4_I(inode)->jinode)
130 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
131 EXT4_I(inode)->jinode,
135 static void ext4_invalidatepage(struct page *page, unsigned long offset);
136 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
137 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
138 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
139 struct inode *inode, struct page *page, loff_t from,
140 loff_t length, int flags);
143 * Test whether an inode is a fast symlink.
145 static int ext4_inode_is_fast_symlink(struct inode *inode)
147 int ea_blocks = EXT4_I(inode)->i_file_acl ?
148 (inode->i_sb->s_blocksize >> 9) : 0;
150 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode) == NULL);
170 jbd_debug(2, "restarting handle %p\n", handle);
171 up_write(&EXT4_I(inode)->i_data_sem);
172 ret = ext4_journal_restart(handle, nblocks);
173 down_write(&EXT4_I(inode)->i_data_sem);
174 ext4_discard_preallocations(inode);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode *inode)
187 trace_ext4_evict_inode(inode);
189 if (inode->i_nlink) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode) &&
209 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
210 inode->i_ino != EXT4_JOURNAL_INO) {
211 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
212 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214 jbd2_complete_transaction(journal, commit_tid);
215 filemap_write_and_wait(&inode->i_data);
217 truncate_inode_pages(&inode->i_data, 0);
218 ext4_ioend_shutdown(inode);
222 if (!is_bad_inode(inode))
223 dquot_initialize(inode);
225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
227 truncate_inode_pages(&inode->i_data, 0);
228 ext4_ioend_shutdown(inode);
230 if (is_bad_inode(inode))
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode->i_sb);
238 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
239 ext4_blocks_for_truncate(inode)+3);
240 if (IS_ERR(handle)) {
241 ext4_std_error(inode->i_sb, PTR_ERR(handle));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL, inode);
248 sb_end_intwrite(inode->i_sb);
253 ext4_handle_sync(handle);
255 err = ext4_mark_inode_dirty(handle, inode);
257 ext4_warning(inode->i_sb,
258 "couldn't mark inode dirty (err %d)", err);
262 ext4_truncate(inode);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle, 3)) {
271 err = ext4_journal_extend(handle, 3);
273 err = ext4_journal_restart(handle, 3);
275 ext4_warning(inode->i_sb,
276 "couldn't extend journal (err %d)", err);
278 ext4_journal_stop(handle);
279 ext4_orphan_del(NULL, inode);
280 sb_end_intwrite(inode->i_sb);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle, inode);
294 EXT4_I(inode)->i_dtime = get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle, inode))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode);
307 ext4_free_inode(handle, inode);
308 ext4_journal_stop(handle);
309 sb_end_intwrite(inode->i_sb);
312 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
316 qsize_t *ext4_get_reserved_space(struct inode *inode)
318 return &EXT4_I(inode)->i_reserved_quota;
323 * Calculate the number of metadata blocks need to reserve
324 * to allocate a block located at @lblock
326 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
328 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
329 return ext4_ext_calc_metadata_amount(inode, lblock);
331 return ext4_ind_calc_metadata_amount(inode, lblock);
335 * Called with i_data_sem down, which is important since we can call
336 * ext4_discard_preallocations() from here.
338 void ext4_da_update_reserve_space(struct inode *inode,
339 int used, int quota_claim)
341 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
342 struct ext4_inode_info *ei = EXT4_I(inode);
344 spin_lock(&ei->i_block_reservation_lock);
345 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
346 if (unlikely(used > ei->i_reserved_data_blocks)) {
347 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
348 "with only %d reserved data blocks",
349 __func__, inode->i_ino, used,
350 ei->i_reserved_data_blocks);
352 used = ei->i_reserved_data_blocks;
355 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
356 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
357 "with only %d reserved metadata blocks "
358 "(releasing %d blocks with reserved %d data blocks)",
359 inode->i_ino, ei->i_allocated_meta_blocks,
360 ei->i_reserved_meta_blocks, used,
361 ei->i_reserved_data_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);
486 #ifdef ES_AGGRESSIVE_TEST
487 static void ext4_map_blocks_es_recheck(handle_t *handle,
489 struct ext4_map_blocks *es_map,
490 struct ext4_map_blocks *map,
497 * There is a race window that the result is not the same.
498 * e.g. xfstests #223 when dioread_nolock enables. The reason
499 * is that we lookup a block mapping in extent status tree with
500 * out taking i_data_sem. So at the time the unwritten extent
501 * could be converted.
503 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
504 down_read((&EXT4_I(inode)->i_data_sem));
505 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
506 retval = ext4_ext_map_blocks(handle, inode, map, flags &
507 EXT4_GET_BLOCKS_KEEP_SIZE);
509 retval = ext4_ind_map_blocks(handle, inode, map, flags &
510 EXT4_GET_BLOCKS_KEEP_SIZE);
512 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
513 up_read((&EXT4_I(inode)->i_data_sem));
515 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
516 * because it shouldn't be marked in es_map->m_flags.
518 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
521 * We don't check m_len because extent will be collpased in status
522 * tree. So the m_len might not equal.
524 if (es_map->m_lblk != map->m_lblk ||
525 es_map->m_flags != map->m_flags ||
526 es_map->m_pblk != map->m_pblk) {
527 printk("ES cache assertation failed for inode: %lu "
528 "es_cached ex [%d/%d/%llu/%x] != "
529 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
530 inode->i_ino, es_map->m_lblk, es_map->m_len,
531 es_map->m_pblk, es_map->m_flags, map->m_lblk,
532 map->m_len, map->m_pblk, map->m_flags,
536 #endif /* ES_AGGRESSIVE_TEST */
539 * The ext4_map_blocks() function tries to look up the requested blocks,
540 * and returns if the blocks are already mapped.
542 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
543 * and store the allocated blocks in the result buffer head and mark it
546 * If file type is extents based, it will call ext4_ext_map_blocks(),
547 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
550 * On success, it returns the number of blocks being mapped or allocate.
551 * if create==0 and the blocks are pre-allocated and uninitialized block,
552 * the result buffer head is unmapped. If the create ==1, it will make sure
553 * the buffer head is mapped.
555 * It returns 0 if plain look up failed (blocks have not been allocated), in
556 * that case, buffer head is unmapped
558 * It returns the error in case of allocation failure.
560 int ext4_map_blocks(handle_t *handle, struct inode *inode,
561 struct ext4_map_blocks *map, int flags)
563 struct extent_status es;
565 #ifdef ES_AGGRESSIVE_TEST
566 struct ext4_map_blocks orig_map;
568 memcpy(&orig_map, map, sizeof(*map));
572 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
573 "logical block %lu\n", inode->i_ino, flags, map->m_len,
574 (unsigned long) map->m_lblk);
576 /* Lookup extent status tree firstly */
577 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
578 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
579 map->m_pblk = ext4_es_pblock(&es) +
580 map->m_lblk - es.es_lblk;
581 map->m_flags |= ext4_es_is_written(&es) ?
582 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
583 retval = es.es_len - (map->m_lblk - es.es_lblk);
584 if (retval > map->m_len)
587 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
592 #ifdef ES_AGGRESSIVE_TEST
593 ext4_map_blocks_es_recheck(handle, inode, map,
600 * Try to see if we can get the block without requesting a new
603 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
604 down_read((&EXT4_I(inode)->i_data_sem));
605 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
606 retval = ext4_ext_map_blocks(handle, inode, map, flags &
607 EXT4_GET_BLOCKS_KEEP_SIZE);
609 retval = ext4_ind_map_blocks(handle, inode, map, flags &
610 EXT4_GET_BLOCKS_KEEP_SIZE);
614 unsigned long long status;
616 #ifdef ES_AGGRESSIVE_TEST
617 if (retval != map->m_len) {
618 printk("ES len assertation failed for inode: %lu "
619 "retval %d != map->m_len %d "
620 "in %s (lookup)\n", inode->i_ino, retval,
621 map->m_len, __func__);
625 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
626 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
627 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
628 ext4_find_delalloc_range(inode, map->m_lblk,
629 map->m_lblk + map->m_len - 1))
630 status |= EXTENT_STATUS_DELAYED;
631 ret = ext4_es_insert_extent(inode, map->m_lblk,
632 map->m_len, map->m_pblk, status);
636 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
637 up_read((&EXT4_I(inode)->i_data_sem));
640 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
641 int ret = check_block_validity(inode, map);
646 /* If it is only a block(s) look up */
647 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
651 * Returns if the blocks have already allocated
653 * Note that if blocks have been preallocated
654 * ext4_ext_get_block() returns the create = 0
655 * with buffer head unmapped.
657 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
661 * Here we clear m_flags because after allocating an new extent,
662 * it will be set again.
664 map->m_flags &= ~EXT4_MAP_FLAGS;
667 * New blocks allocate and/or writing to uninitialized extent
668 * will possibly result in updating i_data, so we take
669 * the write lock of i_data_sem, and call get_blocks()
670 * with create == 1 flag.
672 down_write((&EXT4_I(inode)->i_data_sem));
675 * if the caller is from delayed allocation writeout path
676 * we have already reserved fs blocks for allocation
677 * let the underlying get_block() function know to
678 * avoid double accounting
680 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
681 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
683 * We need to check for EXT4 here because migrate
684 * could have changed the inode type in between
686 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
687 retval = ext4_ext_map_blocks(handle, inode, map, flags);
689 retval = ext4_ind_map_blocks(handle, inode, map, flags);
691 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
693 * We allocated new blocks which will result in
694 * i_data's format changing. Force the migrate
695 * to fail by clearing migrate flags
697 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
701 * Update reserved blocks/metadata blocks after successful
702 * block allocation which had been deferred till now. We don't
703 * support fallocate for non extent files. So we can update
704 * reserve space here.
707 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
708 ext4_da_update_reserve_space(inode, retval, 1);
710 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
711 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
715 unsigned long long status;
717 #ifdef ES_AGGRESSIVE_TEST
718 if (retval != map->m_len) {
719 printk("ES len assertation failed for inode: %lu "
720 "retval %d != map->m_len %d "
721 "in %s (allocation)\n", inode->i_ino, retval,
722 map->m_len, __func__);
727 * If the extent has been zeroed out, we don't need to update
728 * extent status tree.
730 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
731 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
732 if (ext4_es_is_written(&es))
735 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
736 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
737 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
738 ext4_find_delalloc_range(inode, map->m_lblk,
739 map->m_lblk + map->m_len - 1))
740 status |= EXTENT_STATUS_DELAYED;
741 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
742 map->m_pblk, status);
748 up_write((&EXT4_I(inode)->i_data_sem));
749 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
750 int ret = check_block_validity(inode, map);
757 /* Maximum number of blocks we map for direct IO at once. */
758 #define DIO_MAX_BLOCKS 4096
760 static int _ext4_get_block(struct inode *inode, sector_t iblock,
761 struct buffer_head *bh, int flags)
763 handle_t *handle = ext4_journal_current_handle();
764 struct ext4_map_blocks map;
765 int ret = 0, started = 0;
768 if (ext4_has_inline_data(inode))
772 map.m_len = bh->b_size >> inode->i_blkbits;
774 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
775 /* Direct IO write... */
776 if (map.m_len > DIO_MAX_BLOCKS)
777 map.m_len = DIO_MAX_BLOCKS;
778 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
779 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
781 if (IS_ERR(handle)) {
782 ret = PTR_ERR(handle);
788 ret = ext4_map_blocks(handle, inode, &map, flags);
790 map_bh(bh, inode->i_sb, map.m_pblk);
791 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
792 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
796 ext4_journal_stop(handle);
800 int ext4_get_block(struct inode *inode, sector_t iblock,
801 struct buffer_head *bh, int create)
803 return _ext4_get_block(inode, iblock, bh,
804 create ? EXT4_GET_BLOCKS_CREATE : 0);
808 * `handle' can be NULL if create is zero
810 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
811 ext4_lblk_t block, int create, int *errp)
813 struct ext4_map_blocks map;
814 struct buffer_head *bh;
817 J_ASSERT(handle != NULL || create == 0);
821 err = ext4_map_blocks(handle, inode, &map,
822 create ? EXT4_GET_BLOCKS_CREATE : 0);
824 /* ensure we send some value back into *errp */
827 if (create && err == 0)
828 err = -ENOSPC; /* should never happen */
834 bh = sb_getblk(inode->i_sb, map.m_pblk);
839 if (map.m_flags & EXT4_MAP_NEW) {
840 J_ASSERT(create != 0);
841 J_ASSERT(handle != NULL);
844 * Now that we do not always journal data, we should
845 * keep in mind whether this should always journal the
846 * new buffer as metadata. For now, regular file
847 * writes use ext4_get_block instead, so it's not a
851 BUFFER_TRACE(bh, "call get_create_access");
852 fatal = ext4_journal_get_create_access(handle, bh);
853 if (!fatal && !buffer_uptodate(bh)) {
854 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
855 set_buffer_uptodate(bh);
858 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
859 err = ext4_handle_dirty_metadata(handle, inode, bh);
863 BUFFER_TRACE(bh, "not a new buffer");
873 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
874 ext4_lblk_t block, int create, int *err)
876 struct buffer_head *bh;
878 bh = ext4_getblk(handle, inode, block, create, err);
881 if (buffer_uptodate(bh))
883 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
885 if (buffer_uptodate(bh))
892 int ext4_walk_page_buffers(handle_t *handle,
893 struct buffer_head *head,
897 int (*fn)(handle_t *handle,
898 struct buffer_head *bh))
900 struct buffer_head *bh;
901 unsigned block_start, block_end;
902 unsigned blocksize = head->b_size;
904 struct buffer_head *next;
906 for (bh = head, block_start = 0;
907 ret == 0 && (bh != head || !block_start);
908 block_start = block_end, bh = next) {
909 next = bh->b_this_page;
910 block_end = block_start + blocksize;
911 if (block_end <= from || block_start >= to) {
912 if (partial && !buffer_uptodate(bh))
916 err = (*fn)(handle, bh);
924 * To preserve ordering, it is essential that the hole instantiation and
925 * the data write be encapsulated in a single transaction. We cannot
926 * close off a transaction and start a new one between the ext4_get_block()
927 * and the commit_write(). So doing the jbd2_journal_start at the start of
928 * prepare_write() is the right place.
930 * Also, this function can nest inside ext4_writepage(). In that case, we
931 * *know* that ext4_writepage() has generated enough buffer credits to do the
932 * whole page. So we won't block on the journal in that case, which is good,
933 * because the caller may be PF_MEMALLOC.
935 * By accident, ext4 can be reentered when a transaction is open via
936 * quota file writes. If we were to commit the transaction while thus
937 * reentered, there can be a deadlock - we would be holding a quota
938 * lock, and the commit would never complete if another thread had a
939 * transaction open and was blocking on the quota lock - a ranking
942 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
943 * will _not_ run commit under these circumstances because handle->h_ref
944 * is elevated. We'll still have enough credits for the tiny quotafile
947 int do_journal_get_write_access(handle_t *handle,
948 struct buffer_head *bh)
950 int dirty = buffer_dirty(bh);
953 if (!buffer_mapped(bh) || buffer_freed(bh))
956 * __block_write_begin() could have dirtied some buffers. Clean
957 * the dirty bit as jbd2_journal_get_write_access() could complain
958 * otherwise about fs integrity issues. Setting of the dirty bit
959 * by __block_write_begin() isn't a real problem here as we clear
960 * the bit before releasing a page lock and thus writeback cannot
961 * ever write the buffer.
964 clear_buffer_dirty(bh);
965 ret = ext4_journal_get_write_access(handle, bh);
967 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
971 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
972 struct buffer_head *bh_result, int create);
973 static int ext4_write_begin(struct file *file, struct address_space *mapping,
974 loff_t pos, unsigned len, unsigned flags,
975 struct page **pagep, void **fsdata)
977 struct inode *inode = mapping->host;
978 int ret, needed_blocks;
985 trace_ext4_write_begin(inode, pos, len, flags);
987 * Reserve one block more for addition to orphan list in case
988 * we allocate blocks but write fails for some reason
990 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
991 index = pos >> PAGE_CACHE_SHIFT;
992 from = pos & (PAGE_CACHE_SIZE - 1);
995 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
996 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1005 * grab_cache_page_write_begin() can take a long time if the
1006 * system is thrashing due to memory pressure, or if the page
1007 * is being written back. So grab it first before we start
1008 * the transaction handle. This also allows us to allocate
1009 * the page (if needed) without using GFP_NOFS.
1012 page = grab_cache_page_write_begin(mapping, index, flags);
1018 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1019 if (IS_ERR(handle)) {
1020 page_cache_release(page);
1021 return PTR_ERR(handle);
1025 if (page->mapping != mapping) {
1026 /* The page got truncated from under us */
1028 page_cache_release(page);
1029 ext4_journal_stop(handle);
1032 wait_on_page_writeback(page);
1034 if (ext4_should_dioread_nolock(inode))
1035 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1037 ret = __block_write_begin(page, pos, len, ext4_get_block);
1039 if (!ret && ext4_should_journal_data(inode)) {
1040 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1042 do_journal_get_write_access);
1048 * __block_write_begin may have instantiated a few blocks
1049 * outside i_size. Trim these off again. Don't need
1050 * i_size_read because we hold i_mutex.
1052 * Add inode to orphan list in case we crash before
1055 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1056 ext4_orphan_add(handle, inode);
1058 ext4_journal_stop(handle);
1059 if (pos + len > inode->i_size) {
1060 ext4_truncate_failed_write(inode);
1062 * If truncate failed early the inode might
1063 * still be on the orphan list; we need to
1064 * make sure the inode is removed from the
1065 * orphan list in that case.
1068 ext4_orphan_del(NULL, inode);
1071 if (ret == -ENOSPC &&
1072 ext4_should_retry_alloc(inode->i_sb, &retries))
1074 page_cache_release(page);
1081 /* For write_end() in data=journal mode */
1082 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1085 if (!buffer_mapped(bh) || buffer_freed(bh))
1087 set_buffer_uptodate(bh);
1088 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1089 clear_buffer_meta(bh);
1090 clear_buffer_prio(bh);
1095 * We need to pick up the new inode size which generic_commit_write gave us
1096 * `file' can be NULL - eg, when called from page_symlink().
1098 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1099 * buffers are managed internally.
1101 static int ext4_write_end(struct file *file,
1102 struct address_space *mapping,
1103 loff_t pos, unsigned len, unsigned copied,
1104 struct page *page, void *fsdata)
1106 handle_t *handle = ext4_journal_current_handle();
1107 struct inode *inode = mapping->host;
1109 int i_size_changed = 0;
1111 trace_ext4_write_end(inode, pos, len, copied);
1112 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1113 ret = ext4_jbd2_file_inode(handle, inode);
1116 page_cache_release(page);
1121 if (ext4_has_inline_data(inode)) {
1122 ret = ext4_write_inline_data_end(inode, pos, len,
1128 copied = block_write_end(file, mapping, pos,
1129 len, copied, page, fsdata);
1132 * No need to use i_size_read() here, the i_size
1133 * cannot change under us because we hole i_mutex.
1135 * But it's important to update i_size while still holding page lock:
1136 * page writeout could otherwise come in and zero beyond i_size.
1138 if (pos + copied > inode->i_size) {
1139 i_size_write(inode, pos + copied);
1143 if (pos + copied > EXT4_I(inode)->i_disksize) {
1144 /* We need to mark inode dirty even if
1145 * new_i_size is less that inode->i_size
1146 * but greater than i_disksize. (hint delalloc)
1148 ext4_update_i_disksize(inode, (pos + copied));
1152 page_cache_release(page);
1155 * Don't mark the inode dirty under page lock. First, it unnecessarily
1156 * makes the holding time of page lock longer. Second, it forces lock
1157 * ordering of page lock and transaction start for journaling
1161 ext4_mark_inode_dirty(handle, inode);
1165 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1166 /* if we have allocated more blocks and copied
1167 * less. We will have blocks allocated outside
1168 * inode->i_size. So truncate them
1170 ext4_orphan_add(handle, inode);
1172 ret2 = ext4_journal_stop(handle);
1176 if (pos + len > inode->i_size) {
1177 ext4_truncate_failed_write(inode);
1179 * If truncate failed early the inode might still be
1180 * on the orphan list; we need to make sure the inode
1181 * is removed from the orphan list in that case.
1184 ext4_orphan_del(NULL, inode);
1187 return ret ? ret : copied;
1190 static int ext4_journalled_write_end(struct file *file,
1191 struct address_space *mapping,
1192 loff_t pos, unsigned len, unsigned copied,
1193 struct page *page, void *fsdata)
1195 handle_t *handle = ext4_journal_current_handle();
1196 struct inode *inode = mapping->host;
1202 trace_ext4_journalled_write_end(inode, pos, len, copied);
1203 from = pos & (PAGE_CACHE_SIZE - 1);
1206 BUG_ON(!ext4_handle_valid(handle));
1208 if (ext4_has_inline_data(inode))
1209 copied = ext4_write_inline_data_end(inode, pos, len,
1213 if (!PageUptodate(page))
1215 page_zero_new_buffers(page, from+copied, to);
1218 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1219 to, &partial, write_end_fn);
1221 SetPageUptodate(page);
1223 new_i_size = pos + copied;
1224 if (new_i_size > inode->i_size)
1225 i_size_write(inode, pos+copied);
1226 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1227 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1228 if (new_i_size > EXT4_I(inode)->i_disksize) {
1229 ext4_update_i_disksize(inode, new_i_size);
1230 ret2 = ext4_mark_inode_dirty(handle, inode);
1236 page_cache_release(page);
1237 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1238 /* if we have allocated more blocks and copied
1239 * less. We will have blocks allocated outside
1240 * inode->i_size. So truncate them
1242 ext4_orphan_add(handle, inode);
1244 ret2 = ext4_journal_stop(handle);
1247 if (pos + len > inode->i_size) {
1248 ext4_truncate_failed_write(inode);
1250 * If truncate failed early the inode might still be
1251 * on the orphan list; we need to make sure the inode
1252 * is removed from the orphan list in that case.
1255 ext4_orphan_del(NULL, inode);
1258 return ret ? ret : copied;
1262 * Reserve a metadata for a single block located at lblock
1264 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1266 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1267 struct ext4_inode_info *ei = EXT4_I(inode);
1268 unsigned int md_needed;
1269 ext4_lblk_t save_last_lblock;
1273 * recalculate the amount of metadata blocks to reserve
1274 * in order to allocate nrblocks
1275 * worse case is one extent per block
1277 spin_lock(&ei->i_block_reservation_lock);
1279 * ext4_calc_metadata_amount() has side effects, which we have
1280 * to be prepared undo if we fail to claim space.
1282 save_len = ei->i_da_metadata_calc_len;
1283 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1284 md_needed = EXT4_NUM_B2C(sbi,
1285 ext4_calc_metadata_amount(inode, lblock));
1286 trace_ext4_da_reserve_space(inode, md_needed);
1289 * We do still charge estimated metadata to the sb though;
1290 * we cannot afford to run out of free blocks.
1292 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1293 ei->i_da_metadata_calc_len = save_len;
1294 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1295 spin_unlock(&ei->i_block_reservation_lock);
1298 ei->i_reserved_meta_blocks += md_needed;
1299 spin_unlock(&ei->i_block_reservation_lock);
1301 return 0; /* success */
1305 * Reserve a single cluster located at lblock
1307 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1309 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1310 struct ext4_inode_info *ei = EXT4_I(inode);
1311 unsigned int md_needed;
1313 ext4_lblk_t save_last_lblock;
1317 * We will charge metadata quota at writeout time; this saves
1318 * us from metadata over-estimation, though we may go over by
1319 * a small amount in the end. Here we just reserve for data.
1321 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1326 * recalculate the amount of metadata blocks to reserve
1327 * in order to allocate nrblocks
1328 * worse case is one extent per block
1330 spin_lock(&ei->i_block_reservation_lock);
1332 * ext4_calc_metadata_amount() has side effects, which we have
1333 * to be prepared undo if we fail to claim space.
1335 save_len = ei->i_da_metadata_calc_len;
1336 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1337 md_needed = EXT4_NUM_B2C(sbi,
1338 ext4_calc_metadata_amount(inode, lblock));
1339 trace_ext4_da_reserve_space(inode, md_needed);
1342 * We do still charge estimated metadata to the sb though;
1343 * we cannot afford to run out of free blocks.
1345 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1346 ei->i_da_metadata_calc_len = save_len;
1347 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1348 spin_unlock(&ei->i_block_reservation_lock);
1349 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1352 ei->i_reserved_data_blocks++;
1353 ei->i_reserved_meta_blocks += md_needed;
1354 spin_unlock(&ei->i_block_reservation_lock);
1356 return 0; /* success */
1359 static void ext4_da_release_space(struct inode *inode, int to_free)
1361 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1362 struct ext4_inode_info *ei = EXT4_I(inode);
1365 return; /* Nothing to release, exit */
1367 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1369 trace_ext4_da_release_space(inode, to_free);
1370 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1372 * if there aren't enough reserved blocks, then the
1373 * counter is messed up somewhere. Since this
1374 * function is called from invalidate page, it's
1375 * harmless to return without any action.
1377 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1378 "ino %lu, to_free %d with only %d reserved "
1379 "data blocks", inode->i_ino, to_free,
1380 ei->i_reserved_data_blocks);
1382 to_free = ei->i_reserved_data_blocks;
1384 ei->i_reserved_data_blocks -= to_free;
1386 if (ei->i_reserved_data_blocks == 0) {
1388 * We can release all of the reserved metadata blocks
1389 * only when we have written all of the delayed
1390 * allocation blocks.
1391 * Note that in case of bigalloc, i_reserved_meta_blocks,
1392 * i_reserved_data_blocks, etc. refer to number of clusters.
1394 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1395 ei->i_reserved_meta_blocks);
1396 ei->i_reserved_meta_blocks = 0;
1397 ei->i_da_metadata_calc_len = 0;
1400 /* update fs dirty data blocks counter */
1401 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1403 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1405 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1408 static void ext4_da_page_release_reservation(struct page *page,
1409 unsigned long offset)
1412 struct buffer_head *head, *bh;
1413 unsigned int curr_off = 0;
1414 struct inode *inode = page->mapping->host;
1415 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1419 head = page_buffers(page);
1422 unsigned int next_off = curr_off + bh->b_size;
1424 if ((offset <= curr_off) && (buffer_delay(bh))) {
1426 clear_buffer_delay(bh);
1428 curr_off = next_off;
1429 } while ((bh = bh->b_this_page) != head);
1432 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1433 ext4_es_remove_extent(inode, lblk, to_release);
1436 /* If we have released all the blocks belonging to a cluster, then we
1437 * need to release the reserved space for that cluster. */
1438 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1439 while (num_clusters > 0) {
1440 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1441 ((num_clusters - 1) << sbi->s_cluster_bits);
1442 if (sbi->s_cluster_ratio == 1 ||
1443 !ext4_find_delalloc_cluster(inode, lblk))
1444 ext4_da_release_space(inode, 1);
1451 * Delayed allocation stuff
1455 * mpage_da_submit_io - walks through extent of pages and try to write
1456 * them with writepage() call back
1458 * @mpd->inode: inode
1459 * @mpd->first_page: first page of the extent
1460 * @mpd->next_page: page after the last page of the extent
1462 * By the time mpage_da_submit_io() is called we expect all blocks
1463 * to be allocated. this may be wrong if allocation failed.
1465 * As pages are already locked by write_cache_pages(), we can't use it
1467 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1468 struct ext4_map_blocks *map)
1470 struct pagevec pvec;
1471 unsigned long index, end;
1472 int ret = 0, err, nr_pages, i;
1473 struct inode *inode = mpd->inode;
1474 struct address_space *mapping = inode->i_mapping;
1475 loff_t size = i_size_read(inode);
1476 unsigned int len, block_start;
1477 struct buffer_head *bh, *page_bufs = NULL;
1478 sector_t pblock = 0, cur_logical = 0;
1479 struct ext4_io_submit io_submit;
1481 BUG_ON(mpd->next_page <= mpd->first_page);
1482 memset(&io_submit, 0, sizeof(io_submit));
1484 * We need to start from the first_page to the next_page - 1
1485 * to make sure we also write the mapped dirty buffer_heads.
1486 * If we look at mpd->b_blocknr we would only be looking
1487 * at the currently mapped buffer_heads.
1489 index = mpd->first_page;
1490 end = mpd->next_page - 1;
1492 pagevec_init(&pvec, 0);
1493 while (index <= end) {
1494 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1497 for (i = 0; i < nr_pages; i++) {
1499 struct page *page = pvec.pages[i];
1501 index = page->index;
1505 if (index == size >> PAGE_CACHE_SHIFT)
1506 len = size & ~PAGE_CACHE_MASK;
1508 len = PAGE_CACHE_SIZE;
1510 cur_logical = index << (PAGE_CACHE_SHIFT -
1512 pblock = map->m_pblk + (cur_logical -
1517 BUG_ON(!PageLocked(page));
1518 BUG_ON(PageWriteback(page));
1520 bh = page_bufs = page_buffers(page);
1523 if (map && (cur_logical >= map->m_lblk) &&
1524 (cur_logical <= (map->m_lblk +
1525 (map->m_len - 1)))) {
1526 if (buffer_delay(bh)) {
1527 clear_buffer_delay(bh);
1528 bh->b_blocknr = pblock;
1530 if (buffer_unwritten(bh) ||
1532 BUG_ON(bh->b_blocknr != pblock);
1533 if (map->m_flags & EXT4_MAP_UNINIT)
1534 set_buffer_uninit(bh);
1535 clear_buffer_unwritten(bh);
1539 * skip page if block allocation undone and
1542 if (ext4_bh_delay_or_unwritten(NULL, bh))
1544 bh = bh->b_this_page;
1545 block_start += bh->b_size;
1548 } while (bh != page_bufs);
1555 clear_page_dirty_for_io(page);
1556 err = ext4_bio_write_page(&io_submit, page, len,
1559 mpd->pages_written++;
1561 * In error case, we have to continue because
1562 * remaining pages are still locked
1567 pagevec_release(&pvec);
1569 ext4_io_submit(&io_submit);
1573 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1577 struct pagevec pvec;
1578 struct inode *inode = mpd->inode;
1579 struct address_space *mapping = inode->i_mapping;
1580 ext4_lblk_t start, last;
1582 index = mpd->first_page;
1583 end = mpd->next_page - 1;
1585 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1586 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1587 ext4_es_remove_extent(inode, start, last - start + 1);
1589 pagevec_init(&pvec, 0);
1590 while (index <= end) {
1591 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1594 for (i = 0; i < nr_pages; i++) {
1595 struct page *page = pvec.pages[i];
1596 if (page->index > end)
1598 BUG_ON(!PageLocked(page));
1599 BUG_ON(PageWriteback(page));
1600 block_invalidatepage(page, 0);
1601 ClearPageUptodate(page);
1604 index = pvec.pages[nr_pages - 1]->index + 1;
1605 pagevec_release(&pvec);
1610 static void ext4_print_free_blocks(struct inode *inode)
1612 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1613 struct super_block *sb = inode->i_sb;
1614 struct ext4_inode_info *ei = EXT4_I(inode);
1616 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1617 EXT4_C2B(EXT4_SB(inode->i_sb),
1618 ext4_count_free_clusters(sb)));
1619 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1620 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1621 (long long) EXT4_C2B(EXT4_SB(sb),
1622 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1623 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1624 (long long) EXT4_C2B(EXT4_SB(sb),
1625 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1626 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1627 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1628 ei->i_reserved_data_blocks);
1629 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1630 ei->i_reserved_meta_blocks);
1631 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1632 ei->i_allocated_meta_blocks);
1637 * mpage_da_map_and_submit - go through given space, map them
1638 * if necessary, and then submit them for I/O
1640 * @mpd - bh describing space
1642 * The function skips space we know is already mapped to disk blocks.
1645 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1647 int err, blks, get_blocks_flags;
1648 struct ext4_map_blocks map, *mapp = NULL;
1649 sector_t next = mpd->b_blocknr;
1650 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1651 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1652 handle_t *handle = NULL;
1655 * If the blocks are mapped already, or we couldn't accumulate
1656 * any blocks, then proceed immediately to the submission stage.
1658 if ((mpd->b_size == 0) ||
1659 ((mpd->b_state & (1 << BH_Mapped)) &&
1660 !(mpd->b_state & (1 << BH_Delay)) &&
1661 !(mpd->b_state & (1 << BH_Unwritten))))
1664 handle = ext4_journal_current_handle();
1668 * Call ext4_map_blocks() to allocate any delayed allocation
1669 * blocks, or to convert an uninitialized extent to be
1670 * initialized (in the case where we have written into
1671 * one or more preallocated blocks).
1673 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1674 * indicate that we are on the delayed allocation path. This
1675 * affects functions in many different parts of the allocation
1676 * call path. This flag exists primarily because we don't
1677 * want to change *many* call functions, so ext4_map_blocks()
1678 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1679 * inode's allocation semaphore is taken.
1681 * If the blocks in questions were delalloc blocks, set
1682 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1683 * variables are updated after the blocks have been allocated.
1686 map.m_len = max_blocks;
1688 * We're in delalloc path and it is possible that we're going to
1689 * need more metadata blocks than previously reserved. However
1690 * we must not fail because we're in writeback and there is
1691 * nothing we can do about it so it might result in data loss.
1692 * So use reserved blocks to allocate metadata if possible.
1694 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1695 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1696 if (ext4_should_dioread_nolock(mpd->inode))
1697 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1698 if (mpd->b_state & (1 << BH_Delay))
1699 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1702 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1704 struct super_block *sb = mpd->inode->i_sb;
1708 * If get block returns EAGAIN or ENOSPC and there
1709 * appears to be free blocks we will just let
1710 * mpage_da_submit_io() unlock all of the pages.
1715 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1721 * get block failure will cause us to loop in
1722 * writepages, because a_ops->writepage won't be able
1723 * to make progress. The page will be redirtied by
1724 * writepage and writepages will again try to write
1727 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1728 ext4_msg(sb, KERN_CRIT,
1729 "delayed block allocation failed for inode %lu "
1730 "at logical offset %llu with max blocks %zd "
1731 "with error %d", mpd->inode->i_ino,
1732 (unsigned long long) next,
1733 mpd->b_size >> mpd->inode->i_blkbits, err);
1734 ext4_msg(sb, KERN_CRIT,
1735 "This should not happen!! Data will be lost");
1737 ext4_print_free_blocks(mpd->inode);
1739 /* invalidate all the pages */
1740 ext4_da_block_invalidatepages(mpd);
1742 /* Mark this page range as having been completed */
1749 if (map.m_flags & EXT4_MAP_NEW) {
1750 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1753 for (i = 0; i < map.m_len; i++)
1754 unmap_underlying_metadata(bdev, map.m_pblk + i);
1758 * Update on-disk size along with block allocation.
1760 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1761 if (disksize > i_size_read(mpd->inode))
1762 disksize = i_size_read(mpd->inode);
1763 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1764 ext4_update_i_disksize(mpd->inode, disksize);
1765 err = ext4_mark_inode_dirty(handle, mpd->inode);
1767 ext4_error(mpd->inode->i_sb,
1768 "Failed to mark inode %lu dirty",
1773 mpage_da_submit_io(mpd, mapp);
1777 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1778 (1 << BH_Delay) | (1 << BH_Unwritten))
1781 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1783 * @mpd->lbh - extent of blocks
1784 * @logical - logical number of the block in the file
1785 * @b_state - b_state of the buffer head added
1787 * the function is used to collect contig. blocks in same state
1789 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1790 unsigned long b_state)
1793 int blkbits = mpd->inode->i_blkbits;
1794 int nrblocks = mpd->b_size >> blkbits;
1797 * XXX Don't go larger than mballoc is willing to allocate
1798 * This is a stopgap solution. We eventually need to fold
1799 * mpage_da_submit_io() into this function and then call
1800 * ext4_map_blocks() multiple times in a loop
1802 if (nrblocks >= (8*1024*1024 >> blkbits))
1805 /* check if the reserved journal credits might overflow */
1806 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1807 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1809 * With non-extent format we are limited by the journal
1810 * credit available. Total credit needed to insert
1811 * nrblocks contiguous blocks is dependent on the
1812 * nrblocks. So limit nrblocks.
1818 * First block in the extent
1820 if (mpd->b_size == 0) {
1821 mpd->b_blocknr = logical;
1822 mpd->b_size = 1 << blkbits;
1823 mpd->b_state = b_state & BH_FLAGS;
1827 next = mpd->b_blocknr + nrblocks;
1829 * Can we merge the block to our big extent?
1831 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1832 mpd->b_size += 1 << blkbits;
1838 * We couldn't merge the block to our extent, so we
1839 * need to flush current extent and start new one
1841 mpage_da_map_and_submit(mpd);
1845 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1847 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1851 * This function is grabs code from the very beginning of
1852 * ext4_map_blocks, but assumes that the caller is from delayed write
1853 * time. This function looks up the requested blocks and sets the
1854 * buffer delay bit under the protection of i_data_sem.
1856 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1857 struct ext4_map_blocks *map,
1858 struct buffer_head *bh)
1860 struct extent_status es;
1862 sector_t invalid_block = ~((sector_t) 0xffff);
1863 #ifdef ES_AGGRESSIVE_TEST
1864 struct ext4_map_blocks orig_map;
1866 memcpy(&orig_map, map, sizeof(*map));
1869 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1873 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1874 "logical block %lu\n", inode->i_ino, map->m_len,
1875 (unsigned long) map->m_lblk);
1877 /* Lookup extent status tree firstly */
1878 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1880 if (ext4_es_is_hole(&es)) {
1882 down_read((&EXT4_I(inode)->i_data_sem));
1887 * Delayed extent could be allocated by fallocate.
1888 * So we need to check it.
1890 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1891 map_bh(bh, inode->i_sb, invalid_block);
1893 set_buffer_delay(bh);
1897 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1898 retval = es.es_len - (iblock - es.es_lblk);
1899 if (retval > map->m_len)
1900 retval = map->m_len;
1901 map->m_len = retval;
1902 if (ext4_es_is_written(&es))
1903 map->m_flags |= EXT4_MAP_MAPPED;
1904 else if (ext4_es_is_unwritten(&es))
1905 map->m_flags |= EXT4_MAP_UNWRITTEN;
1909 #ifdef ES_AGGRESSIVE_TEST
1910 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1916 * Try to see if we can get the block without requesting a new
1917 * file system block.
1919 down_read((&EXT4_I(inode)->i_data_sem));
1920 if (ext4_has_inline_data(inode)) {
1922 * We will soon create blocks for this page, and let
1923 * us pretend as if the blocks aren't allocated yet.
1924 * In case of clusters, we have to handle the work
1925 * of mapping from cluster so that the reserved space
1926 * is calculated properly.
1928 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1929 ext4_find_delalloc_cluster(inode, map->m_lblk))
1930 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1932 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1933 retval = ext4_ext_map_blocks(NULL, inode, map,
1934 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1936 retval = ext4_ind_map_blocks(NULL, inode, map,
1937 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1943 * XXX: __block_prepare_write() unmaps passed block,
1947 * If the block was allocated from previously allocated cluster,
1948 * then we don't need to reserve it again. However we still need
1949 * to reserve metadata for every block we're going to write.
1951 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1952 ret = ext4_da_reserve_space(inode, iblock);
1954 /* not enough space to reserve */
1959 ret = ext4_da_reserve_metadata(inode, iblock);
1961 /* not enough space to reserve */
1967 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1968 ~0, EXTENT_STATUS_DELAYED);
1974 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1975 * and it should not appear on the bh->b_state.
1977 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1979 map_bh(bh, inode->i_sb, invalid_block);
1981 set_buffer_delay(bh);
1982 } else if (retval > 0) {
1984 unsigned long long status;
1986 #ifdef ES_AGGRESSIVE_TEST
1987 if (retval != map->m_len) {
1988 printk("ES len assertation failed for inode: %lu "
1989 "retval %d != map->m_len %d "
1990 "in %s (lookup)\n", inode->i_ino, retval,
1991 map->m_len, __func__);
1995 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1996 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1997 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1998 map->m_pblk, status);
2004 up_read((&EXT4_I(inode)->i_data_sem));
2010 * This is a special get_blocks_t callback which is used by
2011 * ext4_da_write_begin(). It will either return mapped block or
2012 * reserve space for a single block.
2014 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2015 * We also have b_blocknr = -1 and b_bdev initialized properly
2017 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2018 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2019 * initialized properly.
2021 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2022 struct buffer_head *bh, int create)
2024 struct ext4_map_blocks map;
2027 BUG_ON(create == 0);
2028 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2030 map.m_lblk = iblock;
2034 * first, we need to know whether the block is allocated already
2035 * preallocated blocks are unmapped but should treated
2036 * the same as allocated blocks.
2038 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2042 map_bh(bh, inode->i_sb, map.m_pblk);
2043 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2045 if (buffer_unwritten(bh)) {
2046 /* A delayed write to unwritten bh should be marked
2047 * new and mapped. Mapped ensures that we don't do
2048 * get_block multiple times when we write to the same
2049 * offset and new ensures that we do proper zero out
2050 * for partial write.
2053 set_buffer_mapped(bh);
2058 static int bget_one(handle_t *handle, struct buffer_head *bh)
2064 static int bput_one(handle_t *handle, struct buffer_head *bh)
2070 static int __ext4_journalled_writepage(struct page *page,
2073 struct address_space *mapping = page->mapping;
2074 struct inode *inode = mapping->host;
2075 struct buffer_head *page_bufs = NULL;
2076 handle_t *handle = NULL;
2077 int ret = 0, err = 0;
2078 int inline_data = ext4_has_inline_data(inode);
2079 struct buffer_head *inode_bh = NULL;
2081 ClearPageChecked(page);
2084 BUG_ON(page->index != 0);
2085 BUG_ON(len > ext4_get_max_inline_size(inode));
2086 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2087 if (inode_bh == NULL)
2090 page_bufs = page_buffers(page);
2095 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2098 /* As soon as we unlock the page, it can go away, but we have
2099 * references to buffers so we are safe */
2102 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2103 ext4_writepage_trans_blocks(inode));
2104 if (IS_ERR(handle)) {
2105 ret = PTR_ERR(handle);
2109 BUG_ON(!ext4_handle_valid(handle));
2112 ret = ext4_journal_get_write_access(handle, inode_bh);
2114 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2117 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2118 do_journal_get_write_access);
2120 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2125 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2126 err = ext4_journal_stop(handle);
2130 if (!ext4_has_inline_data(inode))
2131 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2133 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2140 * Note that we don't need to start a transaction unless we're journaling data
2141 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2142 * need to file the inode to the transaction's list in ordered mode because if
2143 * we are writing back data added by write(), the inode is already there and if
2144 * we are writing back data modified via mmap(), no one guarantees in which
2145 * transaction the data will hit the disk. In case we are journaling data, we
2146 * cannot start transaction directly because transaction start ranks above page
2147 * lock so we have to do some magic.
2149 * This function can get called via...
2150 * - ext4_da_writepages after taking page lock (have journal handle)
2151 * - journal_submit_inode_data_buffers (no journal handle)
2152 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2153 * - grab_page_cache when doing write_begin (have journal handle)
2155 * We don't do any block allocation in this function. If we have page with
2156 * multiple blocks we need to write those buffer_heads that are mapped. This
2157 * is important for mmaped based write. So if we do with blocksize 1K
2158 * truncate(f, 1024);
2159 * a = mmap(f, 0, 4096);
2161 * truncate(f, 4096);
2162 * we have in the page first buffer_head mapped via page_mkwrite call back
2163 * but other buffer_heads would be unmapped but dirty (dirty done via the
2164 * do_wp_page). So writepage should write the first block. If we modify
2165 * the mmap area beyond 1024 we will again get a page_fault and the
2166 * page_mkwrite callback will do the block allocation and mark the
2167 * buffer_heads mapped.
2169 * We redirty the page if we have any buffer_heads that is either delay or
2170 * unwritten in the page.
2172 * We can get recursively called as show below.
2174 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2177 * But since we don't do any block allocation we should not deadlock.
2178 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2180 static int ext4_writepage(struct page *page,
2181 struct writeback_control *wbc)
2186 struct buffer_head *page_bufs = NULL;
2187 struct inode *inode = page->mapping->host;
2188 struct ext4_io_submit io_submit;
2190 trace_ext4_writepage(page);
2191 size = i_size_read(inode);
2192 if (page->index == size >> PAGE_CACHE_SHIFT)
2193 len = size & ~PAGE_CACHE_MASK;
2195 len = PAGE_CACHE_SIZE;
2197 page_bufs = page_buffers(page);
2199 * We cannot do block allocation or other extent handling in this
2200 * function. If there are buffers needing that, we have to redirty
2201 * the page. But we may reach here when we do a journal commit via
2202 * journal_submit_inode_data_buffers() and in that case we must write
2203 * allocated buffers to achieve data=ordered mode guarantees.
2205 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2206 ext4_bh_delay_or_unwritten)) {
2207 redirty_page_for_writepage(wbc, page);
2208 if (current->flags & PF_MEMALLOC) {
2210 * For memory cleaning there's no point in writing only
2211 * some buffers. So just bail out. Warn if we came here
2212 * from direct reclaim.
2214 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2221 if (PageChecked(page) && ext4_should_journal_data(inode))
2223 * It's mmapped pagecache. Add buffers and journal it. There
2224 * doesn't seem much point in redirtying the page here.
2226 return __ext4_journalled_writepage(page, len);
2228 memset(&io_submit, 0, sizeof(io_submit));
2229 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2230 ext4_io_submit(&io_submit);
2235 * This is called via ext4_da_writepages() to
2236 * calculate the total number of credits to reserve to fit
2237 * a single extent allocation into a single transaction,
2238 * ext4_da_writpeages() will loop calling this before
2239 * the block allocation.
2242 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2244 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2247 * With non-extent format the journal credit needed to
2248 * insert nrblocks contiguous block is dependent on
2249 * number of contiguous block. So we will limit
2250 * number of contiguous block to a sane value
2252 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2253 (max_blocks > EXT4_MAX_TRANS_DATA))
2254 max_blocks = EXT4_MAX_TRANS_DATA;
2256 return ext4_chunk_trans_blocks(inode, max_blocks);
2260 * write_cache_pages_da - walk the list of dirty pages of the given
2261 * address space and accumulate pages that need writing, and call
2262 * mpage_da_map_and_submit to map a single contiguous memory region
2263 * and then write them.
2265 static int write_cache_pages_da(handle_t *handle,
2266 struct address_space *mapping,
2267 struct writeback_control *wbc,
2268 struct mpage_da_data *mpd,
2269 pgoff_t *done_index)
2271 struct buffer_head *bh, *head;
2272 struct inode *inode = mapping->host;
2273 struct pagevec pvec;
2274 unsigned int nr_pages;
2277 long nr_to_write = wbc->nr_to_write;
2278 int i, tag, ret = 0;
2280 memset(mpd, 0, sizeof(struct mpage_da_data));
2283 pagevec_init(&pvec, 0);
2284 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2285 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2287 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2288 tag = PAGECACHE_TAG_TOWRITE;
2290 tag = PAGECACHE_TAG_DIRTY;
2292 *done_index = index;
2293 while (index <= end) {
2294 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2295 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2299 for (i = 0; i < nr_pages; i++) {
2300 struct page *page = pvec.pages[i];
2303 * At this point, the page may be truncated or
2304 * invalidated (changing page->mapping to NULL), or
2305 * even swizzled back from swapper_space to tmpfs file
2306 * mapping. However, page->index will not change
2307 * because we have a reference on the page.
2309 if (page->index > end)
2312 *done_index = page->index + 1;
2315 * If we can't merge this page, and we have
2316 * accumulated an contiguous region, write it
2318 if ((mpd->next_page != page->index) &&
2319 (mpd->next_page != mpd->first_page)) {
2320 mpage_da_map_and_submit(mpd);
2321 goto ret_extent_tail;
2327 * If the page is no longer dirty, or its
2328 * mapping no longer corresponds to inode we
2329 * are writing (which means it has been
2330 * truncated or invalidated), or the page is
2331 * already under writeback and we are not
2332 * doing a data integrity writeback, skip the page
2334 if (!PageDirty(page) ||
2335 (PageWriteback(page) &&
2336 (wbc->sync_mode == WB_SYNC_NONE)) ||
2337 unlikely(page->mapping != mapping)) {
2342 wait_on_page_writeback(page);
2343 BUG_ON(PageWriteback(page));
2346 * If we have inline data and arrive here, it means that
2347 * we will soon create the block for the 1st page, so
2348 * we'd better clear the inline data here.
2350 if (ext4_has_inline_data(inode)) {
2351 BUG_ON(ext4_test_inode_state(inode,
2352 EXT4_STATE_MAY_INLINE_DATA));
2353 ext4_destroy_inline_data(handle, inode);
2356 if (mpd->next_page != page->index)
2357 mpd->first_page = page->index;
2358 mpd->next_page = page->index + 1;
2359 logical = (sector_t) page->index <<
2360 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2362 /* Add all dirty buffers to mpd */
2363 head = page_buffers(page);
2366 BUG_ON(buffer_locked(bh));
2368 * We need to try to allocate unmapped blocks
2369 * in the same page. Otherwise we won't make
2370 * progress with the page in ext4_writepage
2372 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2373 mpage_add_bh_to_extent(mpd, logical,
2376 goto ret_extent_tail;
2377 } else if (buffer_dirty(bh) &&
2378 buffer_mapped(bh)) {
2380 * mapped dirty buffer. We need to
2381 * update the b_state because we look
2382 * at b_state in mpage_da_map_blocks.
2383 * We don't update b_size because if we
2384 * find an unmapped buffer_head later
2385 * we need to use the b_state flag of
2388 if (mpd->b_size == 0)
2390 bh->b_state & BH_FLAGS;
2393 } while ((bh = bh->b_this_page) != head);
2395 if (nr_to_write > 0) {
2397 if (nr_to_write == 0 &&
2398 wbc->sync_mode == WB_SYNC_NONE)
2400 * We stop writing back only if we are
2401 * not doing integrity sync. In case of
2402 * integrity sync we have to keep going
2403 * because someone may be concurrently
2404 * dirtying pages, and we might have
2405 * synced a lot of newly appeared dirty
2406 * pages, but have not synced all of the
2412 pagevec_release(&pvec);
2417 ret = MPAGE_DA_EXTENT_TAIL;
2419 pagevec_release(&pvec);
2425 static int ext4_da_writepages(struct address_space *mapping,
2426 struct writeback_control *wbc)
2429 int range_whole = 0;
2430 handle_t *handle = NULL;
2431 struct mpage_da_data mpd;
2432 struct inode *inode = mapping->host;
2433 int pages_written = 0;
2434 unsigned int max_pages;
2435 int range_cyclic, cycled = 1, io_done = 0;
2436 int needed_blocks, ret = 0;
2437 long desired_nr_to_write, nr_to_writebump = 0;
2438 loff_t range_start = wbc->range_start;
2439 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2440 pgoff_t done_index = 0;
2442 struct blk_plug plug;
2444 trace_ext4_da_writepages(inode, wbc);
2447 * No pages to write? This is mainly a kludge to avoid starting
2448 * a transaction for special inodes like journal inode on last iput()
2449 * because that could violate lock ordering on umount
2451 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2455 * If the filesystem has aborted, it is read-only, so return
2456 * right away instead of dumping stack traces later on that
2457 * will obscure the real source of the problem. We test
2458 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2459 * the latter could be true if the filesystem is mounted
2460 * read-only, and in that case, ext4_da_writepages should
2461 * *never* be called, so if that ever happens, we would want
2464 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2467 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2470 range_cyclic = wbc->range_cyclic;
2471 if (wbc->range_cyclic) {
2472 index = mapping->writeback_index;
2475 wbc->range_start = index << PAGE_CACHE_SHIFT;
2476 wbc->range_end = LLONG_MAX;
2477 wbc->range_cyclic = 0;
2480 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2481 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2485 * This works around two forms of stupidity. The first is in
2486 * the writeback code, which caps the maximum number of pages
2487 * written to be 1024 pages. This is wrong on multiple
2488 * levels; different architectues have a different page size,
2489 * which changes the maximum amount of data which gets
2490 * written. Secondly, 4 megabytes is way too small. XFS
2491 * forces this value to be 16 megabytes by multiplying
2492 * nr_to_write parameter by four, and then relies on its
2493 * allocator to allocate larger extents to make them
2494 * contiguous. Unfortunately this brings us to the second
2495 * stupidity, which is that ext4's mballoc code only allocates
2496 * at most 2048 blocks. So we force contiguous writes up to
2497 * the number of dirty blocks in the inode, or
2498 * sbi->max_writeback_mb_bump whichever is smaller.
2500 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2501 if (!range_cyclic && range_whole) {
2502 if (wbc->nr_to_write == LONG_MAX)
2503 desired_nr_to_write = wbc->nr_to_write;
2505 desired_nr_to_write = wbc->nr_to_write * 8;
2507 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2509 if (desired_nr_to_write > max_pages)
2510 desired_nr_to_write = max_pages;
2512 if (wbc->nr_to_write < desired_nr_to_write) {
2513 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2514 wbc->nr_to_write = desired_nr_to_write;
2518 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2519 tag_pages_for_writeback(mapping, index, end);
2521 blk_start_plug(&plug);
2522 while (!ret && wbc->nr_to_write > 0) {
2525 * we insert one extent at a time. So we need
2526 * credit needed for single extent allocation.
2527 * journalled mode is currently not supported
2530 BUG_ON(ext4_should_journal_data(inode));
2531 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2533 /* start a new transaction*/
2534 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2536 if (IS_ERR(handle)) {
2537 ret = PTR_ERR(handle);
2538 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2539 "%ld pages, ino %lu; err %d", __func__,
2540 wbc->nr_to_write, inode->i_ino, ret);
2541 blk_finish_plug(&plug);
2542 goto out_writepages;
2546 * Now call write_cache_pages_da() to find the next
2547 * contiguous region of logical blocks that need
2548 * blocks to be allocated by ext4 and submit them.
2550 ret = write_cache_pages_da(handle, mapping,
2551 wbc, &mpd, &done_index);
2553 * If we have a contiguous extent of pages and we
2554 * haven't done the I/O yet, map the blocks and submit
2557 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2558 mpage_da_map_and_submit(&mpd);
2559 ret = MPAGE_DA_EXTENT_TAIL;
2561 trace_ext4_da_write_pages(inode, &mpd);
2562 wbc->nr_to_write -= mpd.pages_written;
2564 ext4_journal_stop(handle);
2566 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2567 /* commit the transaction which would
2568 * free blocks released in the transaction
2571 jbd2_journal_force_commit_nested(sbi->s_journal);
2573 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2575 * Got one extent now try with rest of the pages.
2576 * If mpd.retval is set -EIO, journal is aborted.
2577 * So we don't need to write any more.
2579 pages_written += mpd.pages_written;
2582 } else if (wbc->nr_to_write)
2584 * There is no more writeout needed
2585 * or we requested for a noblocking writeout
2586 * and we found the device congested
2590 blk_finish_plug(&plug);
2591 if (!io_done && !cycled) {
2594 wbc->range_start = index << PAGE_CACHE_SHIFT;
2595 wbc->range_end = mapping->writeback_index - 1;
2600 wbc->range_cyclic = range_cyclic;
2601 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2603 * set the writeback_index so that range_cyclic
2604 * mode will write it back later
2606 mapping->writeback_index = done_index;
2609 wbc->nr_to_write -= nr_to_writebump;
2610 wbc->range_start = range_start;
2611 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2615 static int ext4_nonda_switch(struct super_block *sb)
2617 s64 free_clusters, dirty_clusters;
2618 struct ext4_sb_info *sbi = EXT4_SB(sb);
2621 * switch to non delalloc mode if we are running low
2622 * on free block. The free block accounting via percpu
2623 * counters can get slightly wrong with percpu_counter_batch getting
2624 * accumulated on each CPU without updating global counters
2625 * Delalloc need an accurate free block accounting. So switch
2626 * to non delalloc when we are near to error range.
2629 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2631 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2633 * Start pushing delalloc when 1/2 of free blocks are dirty.
2635 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2636 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2638 if (2 * free_clusters < 3 * dirty_clusters ||
2639 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2641 * free block count is less than 150% of dirty blocks
2642 * or free blocks is less than watermark
2649 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2650 loff_t pos, unsigned len, unsigned flags,
2651 struct page **pagep, void **fsdata)
2653 int ret, retries = 0;
2656 struct inode *inode = mapping->host;
2659 index = pos >> PAGE_CACHE_SHIFT;
2661 if (ext4_nonda_switch(inode->i_sb)) {
2662 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2663 return ext4_write_begin(file, mapping, pos,
2664 len, flags, pagep, fsdata);
2666 *fsdata = (void *)0;
2667 trace_ext4_da_write_begin(inode, pos, len, flags);
2669 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2670 ret = ext4_da_write_inline_data_begin(mapping, inode,
2680 * grab_cache_page_write_begin() can take a long time if the
2681 * system is thrashing due to memory pressure, or if the page
2682 * is being written back. So grab it first before we start
2683 * the transaction handle. This also allows us to allocate
2684 * the page (if needed) without using GFP_NOFS.
2687 page = grab_cache_page_write_begin(mapping, index, flags);
2693 * With delayed allocation, we don't log the i_disksize update
2694 * if there is delayed block allocation. But we still need
2695 * to journalling the i_disksize update if writes to the end
2696 * of file which has an already mapped buffer.
2699 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2700 if (IS_ERR(handle)) {
2701 page_cache_release(page);
2702 return PTR_ERR(handle);
2706 if (page->mapping != mapping) {
2707 /* The page got truncated from under us */
2709 page_cache_release(page);
2710 ext4_journal_stop(handle);
2713 /* In case writeback began while the page was unlocked */
2714 wait_on_page_writeback(page);
2716 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2719 ext4_journal_stop(handle);
2721 * block_write_begin may have instantiated a few blocks
2722 * outside i_size. Trim these off again. Don't need
2723 * i_size_read because we hold i_mutex.
2725 if (pos + len > inode->i_size)
2726 ext4_truncate_failed_write(inode);
2728 if (ret == -ENOSPC &&
2729 ext4_should_retry_alloc(inode->i_sb, &retries))
2732 page_cache_release(page);
2741 * Check if we should update i_disksize
2742 * when write to the end of file but not require block allocation
2744 static int ext4_da_should_update_i_disksize(struct page *page,
2745 unsigned long offset)
2747 struct buffer_head *bh;
2748 struct inode *inode = page->mapping->host;
2752 bh = page_buffers(page);
2753 idx = offset >> inode->i_blkbits;
2755 for (i = 0; i < idx; i++)
2756 bh = bh->b_this_page;
2758 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2763 static int ext4_da_write_end(struct file *file,
2764 struct address_space *mapping,
2765 loff_t pos, unsigned len, unsigned copied,
2766 struct page *page, void *fsdata)
2768 struct inode *inode = mapping->host;
2770 handle_t *handle = ext4_journal_current_handle();
2772 unsigned long start, end;
2773 int write_mode = (int)(unsigned long)fsdata;
2775 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2776 return ext4_write_end(file, mapping, pos,
2777 len, copied, page, fsdata);
2779 trace_ext4_da_write_end(inode, pos, len, copied);
2780 start = pos & (PAGE_CACHE_SIZE - 1);
2781 end = start + copied - 1;
2784 * generic_write_end() will run mark_inode_dirty() if i_size
2785 * changes. So let's piggyback the i_disksize mark_inode_dirty
2788 new_i_size = pos + copied;
2789 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2790 if (ext4_has_inline_data(inode) ||
2791 ext4_da_should_update_i_disksize(page, end)) {
2792 down_write(&EXT4_I(inode)->i_data_sem);
2793 if (new_i_size > EXT4_I(inode)->i_disksize)
2794 EXT4_I(inode)->i_disksize = new_i_size;
2795 up_write(&EXT4_I(inode)->i_data_sem);
2796 /* We need to mark inode dirty even if
2797 * new_i_size is less that inode->i_size
2798 * bu greater than i_disksize.(hint delalloc)
2800 ext4_mark_inode_dirty(handle, inode);
2804 if (write_mode != CONVERT_INLINE_DATA &&
2805 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2806 ext4_has_inline_data(inode))
2807 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2810 ret2 = generic_write_end(file, mapping, pos, len, copied,
2816 ret2 = ext4_journal_stop(handle);
2820 return ret ? ret : copied;
2823 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2826 * Drop reserved blocks
2828 BUG_ON(!PageLocked(page));
2829 if (!page_has_buffers(page))
2832 ext4_da_page_release_reservation(page, offset);
2835 ext4_invalidatepage(page, offset);
2841 * Force all delayed allocation blocks to be allocated for a given inode.
2843 int ext4_alloc_da_blocks(struct inode *inode)
2845 trace_ext4_alloc_da_blocks(inode);
2847 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2848 !EXT4_I(inode)->i_reserved_meta_blocks)
2852 * We do something simple for now. The filemap_flush() will
2853 * also start triggering a write of the data blocks, which is
2854 * not strictly speaking necessary (and for users of
2855 * laptop_mode, not even desirable). However, to do otherwise
2856 * would require replicating code paths in:
2858 * ext4_da_writepages() ->
2859 * write_cache_pages() ---> (via passed in callback function)
2860 * __mpage_da_writepage() -->
2861 * mpage_add_bh_to_extent()
2862 * mpage_da_map_blocks()
2864 * The problem is that write_cache_pages(), located in
2865 * mm/page-writeback.c, marks pages clean in preparation for
2866 * doing I/O, which is not desirable if we're not planning on
2869 * We could call write_cache_pages(), and then redirty all of
2870 * the pages by calling redirty_page_for_writepage() but that
2871 * would be ugly in the extreme. So instead we would need to
2872 * replicate parts of the code in the above functions,
2873 * simplifying them because we wouldn't actually intend to
2874 * write out the pages, but rather only collect contiguous
2875 * logical block extents, call the multi-block allocator, and
2876 * then update the buffer heads with the block allocations.
2878 * For now, though, we'll cheat by calling filemap_flush(),
2879 * which will map the blocks, and start the I/O, but not
2880 * actually wait for the I/O to complete.
2882 return filemap_flush(inode->i_mapping);
2886 * bmap() is special. It gets used by applications such as lilo and by
2887 * the swapper to find the on-disk block of a specific piece of data.
2889 * Naturally, this is dangerous if the block concerned is still in the
2890 * journal. If somebody makes a swapfile on an ext4 data-journaling
2891 * filesystem and enables swap, then they may get a nasty shock when the
2892 * data getting swapped to that swapfile suddenly gets overwritten by
2893 * the original zero's written out previously to the journal and
2894 * awaiting writeback in the kernel's buffer cache.
2896 * So, if we see any bmap calls here on a modified, data-journaled file,
2897 * take extra steps to flush any blocks which might be in the cache.
2899 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2901 struct inode *inode = mapping->host;
2906 * We can get here for an inline file via the FIBMAP ioctl
2908 if (ext4_has_inline_data(inode))
2911 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2912 test_opt(inode->i_sb, DELALLOC)) {
2914 * With delalloc we want to sync the file
2915 * so that we can make sure we allocate
2918 filemap_write_and_wait(mapping);
2921 if (EXT4_JOURNAL(inode) &&
2922 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2924 * This is a REALLY heavyweight approach, but the use of
2925 * bmap on dirty files is expected to be extremely rare:
2926 * only if we run lilo or swapon on a freshly made file
2927 * do we expect this to happen.
2929 * (bmap requires CAP_SYS_RAWIO so this does not
2930 * represent an unprivileged user DOS attack --- we'd be
2931 * in trouble if mortal users could trigger this path at
2934 * NB. EXT4_STATE_JDATA is not set on files other than
2935 * regular files. If somebody wants to bmap a directory
2936 * or symlink and gets confused because the buffer
2937 * hasn't yet been flushed to disk, they deserve
2938 * everything they get.
2941 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2942 journal = EXT4_JOURNAL(inode);
2943 jbd2_journal_lock_updates(journal);
2944 err = jbd2_journal_flush(journal);
2945 jbd2_journal_unlock_updates(journal);
2951 return generic_block_bmap(mapping, block, ext4_get_block);
2954 static int ext4_readpage(struct file *file, struct page *page)
2957 struct inode *inode = page->mapping->host;
2959 trace_ext4_readpage(page);
2961 if (ext4_has_inline_data(inode))
2962 ret = ext4_readpage_inline(inode, page);
2965 return mpage_readpage(page, ext4_get_block);
2971 ext4_readpages(struct file *file, struct address_space *mapping,
2972 struct list_head *pages, unsigned nr_pages)
2974 struct inode *inode = mapping->host;
2976 /* If the file has inline data, no need to do readpages. */
2977 if (ext4_has_inline_data(inode))
2980 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2983 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2985 trace_ext4_invalidatepage(page, offset);
2987 /* No journalling happens on data buffers when this function is used */
2988 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2990 block_invalidatepage(page, offset);
2993 static int __ext4_journalled_invalidatepage(struct page *page,
2994 unsigned long offset)
2996 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2998 trace_ext4_journalled_invalidatepage(page, offset);
3001 * If it's a full truncate we just forget about the pending dirtying
3004 ClearPageChecked(page);
3006 return jbd2_journal_invalidatepage(journal, page, offset);
3009 /* Wrapper for aops... */
3010 static void ext4_journalled_invalidatepage(struct page *page,
3011 unsigned long offset)
3013 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3016 static int ext4_releasepage(struct page *page, gfp_t wait)
3018 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3020 trace_ext4_releasepage(page);
3022 /* Page has dirty journalled data -> cannot release */
3023 if (PageChecked(page))
3026 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3028 return try_to_free_buffers(page);
3032 * ext4_get_block used when preparing for a DIO write or buffer write.
3033 * We allocate an uinitialized extent if blocks haven't been allocated.
3034 * The extent will be converted to initialized after the IO is complete.
3036 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3037 struct buffer_head *bh_result, int create)
3039 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3040 inode->i_ino, create);
3041 return _ext4_get_block(inode, iblock, bh_result,
3042 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3045 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3046 struct buffer_head *bh_result, int create)
3048 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3049 inode->i_ino, create);
3050 return _ext4_get_block(inode, iblock, bh_result,
3051 EXT4_GET_BLOCKS_NO_LOCK);
3054 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3055 ssize_t size, void *private, int ret,
3058 struct inode *inode = file_inode(iocb->ki_filp);
3059 ext4_io_end_t *io_end = iocb->private;
3061 /* if not async direct IO or dio with 0 bytes write, just return */
3062 if (!io_end || !size)
3065 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3066 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3067 iocb->private, io_end->inode->i_ino, iocb, offset,
3070 iocb->private = NULL;
3072 /* if not aio dio with unwritten extents, just free io and return */
3073 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3074 ext4_free_io_end(io_end);
3076 inode_dio_done(inode);
3078 aio_complete(iocb, ret, 0);
3082 io_end->offset = offset;
3083 io_end->size = size;
3085 io_end->iocb = iocb;
3086 io_end->result = ret;
3089 ext4_add_complete_io(io_end);
3093 * For ext4 extent files, ext4 will do direct-io write to holes,
3094 * preallocated extents, and those write extend the file, no need to
3095 * fall back to buffered IO.
3097 * For holes, we fallocate those blocks, mark them as uninitialized
3098 * If those blocks were preallocated, we mark sure they are split, but
3099 * still keep the range to write as uninitialized.
3101 * The unwritten extents will be converted to written when DIO is completed.
3102 * For async direct IO, since the IO may still pending when return, we
3103 * set up an end_io call back function, which will do the conversion
3104 * when async direct IO completed.
3106 * If the O_DIRECT write will extend the file then add this inode to the
3107 * orphan list. So recovery will truncate it back to the original size
3108 * if the machine crashes during the write.
3111 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3112 const struct iovec *iov, loff_t offset,
3113 unsigned long nr_segs)
3115 struct file *file = iocb->ki_filp;
3116 struct inode *inode = file->f_mapping->host;
3118 size_t count = iov_length(iov, nr_segs);
3120 get_block_t *get_block_func = NULL;
3122 loff_t final_size = offset + count;
3124 /* Use the old path for reads and writes beyond i_size. */
3125 if (rw != WRITE || final_size > inode->i_size)
3126 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3128 BUG_ON(iocb->private == NULL);
3130 /* If we do a overwrite dio, i_mutex locking can be released */
3131 overwrite = *((int *)iocb->private);
3134 atomic_inc(&inode->i_dio_count);
3135 down_read(&EXT4_I(inode)->i_data_sem);
3136 mutex_unlock(&inode->i_mutex);
3140 * We could direct write to holes and fallocate.
3142 * Allocated blocks to fill the hole are marked as
3143 * uninitialized to prevent parallel buffered read to expose
3144 * the stale data before DIO complete the data IO.
3146 * As to previously fallocated extents, ext4 get_block will
3147 * just simply mark the buffer mapped but still keep the
3148 * extents uninitialized.
3150 * For non AIO case, we will convert those unwritten extents
3151 * to written after return back from blockdev_direct_IO.
3153 * For async DIO, the conversion needs to be deferred when the
3154 * IO is completed. The ext4 end_io callback function will be
3155 * called to take care of the conversion work. Here for async
3156 * case, we allocate an io_end structure to hook to the iocb.
3158 iocb->private = NULL;
3159 ext4_inode_aio_set(inode, NULL);
3160 if (!is_sync_kiocb(iocb)) {
3161 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3166 io_end->flag |= EXT4_IO_END_DIRECT;
3167 iocb->private = io_end;
3169 * we save the io structure for current async direct
3170 * IO, so that later ext4_map_blocks() could flag the
3171 * io structure whether there is a unwritten extents
3172 * needs to be converted when IO is completed.
3174 ext4_inode_aio_set(inode, io_end);
3178 get_block_func = ext4_get_block_write_nolock;
3180 get_block_func = ext4_get_block_write;
3181 dio_flags = DIO_LOCKING;
3183 ret = __blockdev_direct_IO(rw, iocb, inode,
3184 inode->i_sb->s_bdev, iov,
3192 ext4_inode_aio_set(inode, NULL);
3194 * The io_end structure takes a reference to the inode, that
3195 * structure needs to be destroyed and the reference to the
3196 * inode need to be dropped, when IO is complete, even with 0
3197 * byte write, or failed.
3199 * In the successful AIO DIO case, the io_end structure will
3200 * be destroyed and the reference to the inode will be dropped
3201 * after the end_io call back function is called.
3203 * In the case there is 0 byte write, or error case, since VFS
3204 * direct IO won't invoke the end_io call back function, we
3205 * need to free the end_io structure here.
3207 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3208 ext4_free_io_end(iocb->private);
3209 iocb->private = NULL;
3210 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3211 EXT4_STATE_DIO_UNWRITTEN)) {
3214 * for non AIO case, since the IO is already
3215 * completed, we could do the conversion right here
3217 err = ext4_convert_unwritten_extents(inode,
3221 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3225 /* take i_mutex locking again if we do a ovewrite dio */
3227 inode_dio_done(inode);
3228 up_read(&EXT4_I(inode)->i_data_sem);
3229 mutex_lock(&inode->i_mutex);
3235 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3236 const struct iovec *iov, loff_t offset,
3237 unsigned long nr_segs)
3239 struct file *file = iocb->ki_filp;
3240 struct inode *inode = file->f_mapping->host;
3244 * If we are doing data journalling we don't support O_DIRECT
3246 if (ext4_should_journal_data(inode))
3249 /* Let buffer I/O handle the inline data case. */
3250 if (ext4_has_inline_data(inode))
3253 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3254 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3255 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3257 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3258 trace_ext4_direct_IO_exit(inode, offset,
3259 iov_length(iov, nr_segs), rw, ret);
3264 * Pages can be marked dirty completely asynchronously from ext4's journalling
3265 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3266 * much here because ->set_page_dirty is called under VFS locks. The page is
3267 * not necessarily locked.
3269 * We cannot just dirty the page and leave attached buffers clean, because the
3270 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3271 * or jbddirty because all the journalling code will explode.
3273 * So what we do is to mark the page "pending dirty" and next time writepage
3274 * is called, propagate that into the buffers appropriately.
3276 static int ext4_journalled_set_page_dirty(struct page *page)
3278 SetPageChecked(page);
3279 return __set_page_dirty_nobuffers(page);
3282 static const struct address_space_operations ext4_aops = {
3283 .readpage = ext4_readpage,
3284 .readpages = ext4_readpages,
3285 .writepage = ext4_writepage,
3286 .write_begin = ext4_write_begin,
3287 .write_end = ext4_write_end,
3289 .invalidatepage = ext4_invalidatepage,
3290 .releasepage = ext4_releasepage,
3291 .direct_IO = ext4_direct_IO,
3292 .migratepage = buffer_migrate_page,
3293 .is_partially_uptodate = block_is_partially_uptodate,
3294 .error_remove_page = generic_error_remove_page,
3297 static const struct address_space_operations ext4_journalled_aops = {
3298 .readpage = ext4_readpage,
3299 .readpages = ext4_readpages,
3300 .writepage = ext4_writepage,
3301 .write_begin = ext4_write_begin,
3302 .write_end = ext4_journalled_write_end,
3303 .set_page_dirty = ext4_journalled_set_page_dirty,
3305 .invalidatepage = ext4_journalled_invalidatepage,
3306 .releasepage = ext4_releasepage,
3307 .direct_IO = ext4_direct_IO,
3308 .is_partially_uptodate = block_is_partially_uptodate,
3309 .error_remove_page = generic_error_remove_page,
3312 static const struct address_space_operations ext4_da_aops = {
3313 .readpage = ext4_readpage,
3314 .readpages = ext4_readpages,
3315 .writepage = ext4_writepage,
3316 .writepages = ext4_da_writepages,
3317 .write_begin = ext4_da_write_begin,
3318 .write_end = ext4_da_write_end,
3320 .invalidatepage = ext4_da_invalidatepage,
3321 .releasepage = ext4_releasepage,
3322 .direct_IO = ext4_direct_IO,
3323 .migratepage = buffer_migrate_page,
3324 .is_partially_uptodate = block_is_partially_uptodate,
3325 .error_remove_page = generic_error_remove_page,
3328 void ext4_set_aops(struct inode *inode)
3330 switch (ext4_inode_journal_mode(inode)) {
3331 case EXT4_INODE_ORDERED_DATA_MODE:
3332 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3334 case EXT4_INODE_WRITEBACK_DATA_MODE:
3335 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3337 case EXT4_INODE_JOURNAL_DATA_MODE:
3338 inode->i_mapping->a_ops = &ext4_journalled_aops;
3343 if (test_opt(inode->i_sb, DELALLOC))
3344 inode->i_mapping->a_ops = &ext4_da_aops;
3346 inode->i_mapping->a_ops = &ext4_aops;
3351 * ext4_discard_partial_page_buffers()
3352 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3353 * This function finds and locks the page containing the offset
3354 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3355 * Calling functions that already have the page locked should call
3356 * ext4_discard_partial_page_buffers_no_lock directly.
3358 int ext4_discard_partial_page_buffers(handle_t *handle,
3359 struct address_space *mapping, loff_t from,
3360 loff_t length, int flags)
3362 struct inode *inode = mapping->host;
3366 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3367 mapping_gfp_mask(mapping) & ~__GFP_FS);
3371 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3372 from, length, flags);
3375 page_cache_release(page);
3380 * ext4_discard_partial_page_buffers_no_lock()
3381 * Zeros a page range of length 'length' starting from offset 'from'.
3382 * Buffer heads that correspond to the block aligned regions of the
3383 * zeroed range will be unmapped. Unblock aligned regions
3384 * will have the corresponding buffer head mapped if needed so that
3385 * that region of the page can be updated with the partial zero out.
3387 * This function assumes that the page has already been locked. The
3388 * The range to be discarded must be contained with in the given page.
3389 * If the specified range exceeds the end of the page it will be shortened
3390 * to the end of the page that corresponds to 'from'. This function is
3391 * appropriate for updating a page and it buffer heads to be unmapped and
3392 * zeroed for blocks that have been either released, or are going to be
3395 * handle: The journal handle
3396 * inode: The files inode
3397 * page: A locked page that contains the offset "from"
3398 * from: The starting byte offset (from the beginning of the file)
3399 * to begin discarding
3400 * len: The length of bytes to discard
3401 * flags: Optional flags that may be used:
3403 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3404 * Only zero the regions of the page whose buffer heads
3405 * have already been unmapped. This flag is appropriate
3406 * for updating the contents of a page whose blocks may
3407 * have already been released, and we only want to zero
3408 * out the regions that correspond to those released blocks.
3410 * Returns zero on success or negative on failure.
3412 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3413 struct inode *inode, struct page *page, loff_t from,
3414 loff_t length, int flags)
3416 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3417 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3418 unsigned int blocksize, max, pos;
3420 struct buffer_head *bh;
3423 blocksize = inode->i_sb->s_blocksize;
3424 max = PAGE_CACHE_SIZE - offset;
3426 if (index != page->index)
3430 * correct length if it does not fall between
3431 * 'from' and the end of the page
3433 if (length > max || length < 0)
3436 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3438 if (!page_has_buffers(page))
3439 create_empty_buffers(page, blocksize, 0);
3441 /* Find the buffer that contains "offset" */
3442 bh = page_buffers(page);
3444 while (offset >= pos) {
3445 bh = bh->b_this_page;
3451 while (pos < offset + length) {
3452 unsigned int end_of_block, range_to_discard;
3456 /* The length of space left to zero and unmap */
3457 range_to_discard = offset + length - pos;
3459 /* The length of space until the end of the block */
3460 end_of_block = blocksize - (pos & (blocksize-1));
3463 * Do not unmap or zero past end of block
3464 * for this buffer head
3466 if (range_to_discard > end_of_block)
3467 range_to_discard = end_of_block;
3471 * Skip this buffer head if we are only zeroing unampped
3472 * regions of the page
3474 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3478 /* If the range is block aligned, unmap */
3479 if (range_to_discard == blocksize) {
3480 clear_buffer_dirty(bh);
3482 clear_buffer_mapped(bh);
3483 clear_buffer_req(bh);
3484 clear_buffer_new(bh);
3485 clear_buffer_delay(bh);
3486 clear_buffer_unwritten(bh);
3487 clear_buffer_uptodate(bh);
3488 zero_user(page, pos, range_to_discard);
3489 BUFFER_TRACE(bh, "Buffer discarded");
3494 * If this block is not completely contained in the range
3495 * to be discarded, then it is not going to be released. Because
3496 * we need to keep this block, we need to make sure this part
3497 * of the page is uptodate before we modify it by writeing
3498 * partial zeros on it.
3500 if (!buffer_mapped(bh)) {
3502 * Buffer head must be mapped before we can read
3505 BUFFER_TRACE(bh, "unmapped");
3506 ext4_get_block(inode, iblock, bh, 0);
3507 /* unmapped? It's a hole - nothing to do */
3508 if (!buffer_mapped(bh)) {
3509 BUFFER_TRACE(bh, "still unmapped");
3514 /* Ok, it's mapped. Make sure it's up-to-date */
3515 if (PageUptodate(page))
3516 set_buffer_uptodate(bh);
3518 if (!buffer_uptodate(bh)) {
3520 ll_rw_block(READ, 1, &bh);
3522 /* Uhhuh. Read error. Complain and punt.*/
3523 if (!buffer_uptodate(bh))
3527 if (ext4_should_journal_data(inode)) {
3528 BUFFER_TRACE(bh, "get write access");
3529 err = ext4_journal_get_write_access(handle, bh);
3534 zero_user(page, pos, range_to_discard);
3537 if (ext4_should_journal_data(inode)) {
3538 err = ext4_handle_dirty_metadata(handle, inode, bh);
3540 mark_buffer_dirty(bh);
3542 BUFFER_TRACE(bh, "Partial buffer zeroed");
3544 bh = bh->b_this_page;
3546 pos += range_to_discard;
3552 int ext4_can_truncate(struct inode *inode)
3554 if (S_ISREG(inode->i_mode))
3556 if (S_ISDIR(inode->i_mode))
3558 if (S_ISLNK(inode->i_mode))
3559 return !ext4_inode_is_fast_symlink(inode);
3564 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3565 * associated with the given offset and length
3567 * @inode: File inode
3568 * @offset: The offset where the hole will begin
3569 * @len: The length of the hole
3571 * Returns: 0 on success or negative on failure
3574 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3576 struct inode *inode = file_inode(file);
3577 struct super_block *sb = inode->i_sb;
3578 ext4_lblk_t first_block, stop_block;
3579 struct address_space *mapping = inode->i_mapping;
3580 loff_t first_page, last_page, page_len;
3581 loff_t first_page_offset, last_page_offset;
3583 unsigned int credits;
3586 if (!S_ISREG(inode->i_mode))
3589 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3590 /* TODO: Add support for bigalloc file systems */
3594 trace_ext4_punch_hole(inode, offset, length);
3597 * Write out all dirty pages to avoid race conditions
3598 * Then release them.
3600 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3601 ret = filemap_write_and_wait_range(mapping, offset,
3602 offset + length - 1);
3607 mutex_lock(&inode->i_mutex);
3608 /* It's not possible punch hole on append only file */
3609 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3613 if (IS_SWAPFILE(inode)) {
3618 /* No need to punch hole beyond i_size */
3619 if (offset >= inode->i_size)
3623 * If the hole extends beyond i_size, set the hole
3624 * to end after the page that contains i_size
3626 if (offset + length > inode->i_size) {
3627 length = inode->i_size +
3628 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3632 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
3633 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
3635 first_page_offset = first_page << PAGE_CACHE_SHIFT;
3636 last_page_offset = last_page << PAGE_CACHE_SHIFT;
3638 /* Now release the pages */
3639 if (last_page_offset > first_page_offset) {
3640 truncate_pagecache_range(inode, first_page_offset,
3641 last_page_offset - 1);
3644 /* Wait all existing dio workers, newcomers will block on i_mutex */
3645 ext4_inode_block_unlocked_dio(inode);
3646 ret = ext4_flush_unwritten_io(inode);
3649 inode_dio_wait(inode);
3651 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3652 credits = ext4_writepage_trans_blocks(inode);
3654 credits = ext4_blocks_for_truncate(inode);
3655 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3656 if (IS_ERR(handle)) {
3657 ret = PTR_ERR(handle);
3658 ext4_std_error(sb, ret);
3663 * Now we need to zero out the non-page-aligned data in the
3664 * pages at the start and tail of the hole, and unmap the
3665 * buffer heads for the block aligned regions of the page that
3666 * were completely zeroed.
3668 if (first_page > last_page) {
3670 * If the file space being truncated is contained
3671 * within a page just zero out and unmap the middle of
3674 ret = ext4_discard_partial_page_buffers(handle,
3675 mapping, offset, length, 0);
3681 * zero out and unmap the partial page that contains
3682 * the start of the hole
3684 page_len = first_page_offset - offset;
3686 ret = ext4_discard_partial_page_buffers(handle, mapping,
3687 offset, page_len, 0);
3693 * zero out and unmap the partial page that contains
3694 * the end of the hole
3696 page_len = offset + length - last_page_offset;
3698 ret = ext4_discard_partial_page_buffers(handle, mapping,
3699 last_page_offset, page_len, 0);
3706 * If i_size is contained in the last page, we need to
3707 * unmap and zero the partial page after i_size
3709 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
3710 inode->i_size % PAGE_CACHE_SIZE != 0) {
3711 page_len = PAGE_CACHE_SIZE -
3712 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3715 ret = ext4_discard_partial_page_buffers(handle,
3716 mapping, inode->i_size, page_len, 0);
3723 first_block = (offset + sb->s_blocksize - 1) >>
3724 EXT4_BLOCK_SIZE_BITS(sb);
3725 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3727 /* If there are no blocks to remove, return now */
3728 if (first_block >= stop_block)
3731 down_write(&EXT4_I(inode)->i_data_sem);
3732 ext4_discard_preallocations(inode);
3734 ret = ext4_es_remove_extent(inode, first_block,
3735 stop_block - first_block);
3737 up_write(&EXT4_I(inode)->i_data_sem);
3741 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3742 ret = ext4_ext_remove_space(inode, first_block,
3745 ret = ext4_free_hole_blocks(handle, inode, first_block,
3748 ext4_discard_preallocations(inode);
3749 up_write(&EXT4_I(inode)->i_data_sem);
3751 ext4_handle_sync(handle);
3752 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3753 ext4_mark_inode_dirty(handle, inode);
3755 ext4_journal_stop(handle);
3757 ext4_inode_resume_unlocked_dio(inode);
3759 mutex_unlock(&inode->i_mutex);
3766 * We block out ext4_get_block() block instantiations across the entire
3767 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3768 * simultaneously on behalf of the same inode.
3770 * As we work through the truncate and commit bits of it to the journal there
3771 * is one core, guiding principle: the file's tree must always be consistent on
3772 * disk. We must be able to restart the truncate after a crash.
3774 * The file's tree may be transiently inconsistent in memory (although it
3775 * probably isn't), but whenever we close off and commit a journal transaction,
3776 * the contents of (the filesystem + the journal) must be consistent and
3777 * restartable. It's pretty simple, really: bottom up, right to left (although
3778 * left-to-right works OK too).
3780 * Note that at recovery time, journal replay occurs *before* the restart of
3781 * truncate against the orphan inode list.
3783 * The committed inode has the new, desired i_size (which is the same as
3784 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3785 * that this inode's truncate did not complete and it will again call
3786 * ext4_truncate() to have another go. So there will be instantiated blocks
3787 * to the right of the truncation point in a crashed ext4 filesystem. But
3788 * that's fine - as long as they are linked from the inode, the post-crash
3789 * ext4_truncate() run will find them and release them.
3791 void ext4_truncate(struct inode *inode)
3793 struct ext4_inode_info *ei = EXT4_I(inode);
3794 unsigned int credits;
3796 struct address_space *mapping = inode->i_mapping;
3800 * There is a possibility that we're either freeing the inode
3801 * or it completely new indode. In those cases we might not
3802 * have i_mutex locked because it's not necessary.
3804 if (!(inode->i_state & (I_NEW|I_FREEING)))
3805 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3806 trace_ext4_truncate_enter(inode);
3808 if (!ext4_can_truncate(inode))
3811 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3813 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3814 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3816 if (ext4_has_inline_data(inode)) {
3819 ext4_inline_data_truncate(inode, &has_inline);
3825 * finish any pending end_io work so we won't run the risk of
3826 * converting any truncated blocks to initialized later
3828 ext4_flush_unwritten_io(inode);
3830 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3831 credits = ext4_writepage_trans_blocks(inode);
3833 credits = ext4_blocks_for_truncate(inode);
3835 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3836 if (IS_ERR(handle)) {
3837 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3841 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3842 page_len = PAGE_CACHE_SIZE -
3843 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3845 if (ext4_discard_partial_page_buffers(handle,
3846 mapping, inode->i_size, page_len, 0))
3851 * We add the inode to the orphan list, so that if this
3852 * truncate spans multiple transactions, and we crash, we will
3853 * resume the truncate when the filesystem recovers. It also
3854 * marks the inode dirty, to catch the new size.
3856 * Implication: the file must always be in a sane, consistent
3857 * truncatable state while each transaction commits.
3859 if (ext4_orphan_add(handle, inode))
3862 down_write(&EXT4_I(inode)->i_data_sem);
3864 ext4_discard_preallocations(inode);
3866 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3867 ext4_ext_truncate(handle, inode);
3869 ext4_ind_truncate(handle, inode);
3871 up_write(&ei->i_data_sem);
3874 ext4_handle_sync(handle);
3878 * If this was a simple ftruncate() and the file will remain alive,
3879 * then we need to clear up the orphan record which we created above.
3880 * However, if this was a real unlink then we were called by
3881 * ext4_delete_inode(), and we allow that function to clean up the
3882 * orphan info for us.
3885 ext4_orphan_del(handle, inode);
3887 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3888 ext4_mark_inode_dirty(handle, inode);
3889 ext4_journal_stop(handle);
3891 trace_ext4_truncate_exit(inode);
3895 * ext4_get_inode_loc returns with an extra refcount against the inode's
3896 * underlying buffer_head on success. If 'in_mem' is true, we have all
3897 * data in memory that is needed to recreate the on-disk version of this
3900 static int __ext4_get_inode_loc(struct inode *inode,
3901 struct ext4_iloc *iloc, int in_mem)
3903 struct ext4_group_desc *gdp;
3904 struct buffer_head *bh;
3905 struct super_block *sb = inode->i_sb;
3907 int inodes_per_block, inode_offset;
3910 if (!ext4_valid_inum(sb, inode->i_ino))
3913 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3914 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3919 * Figure out the offset within the block group inode table
3921 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3922 inode_offset = ((inode->i_ino - 1) %
3923 EXT4_INODES_PER_GROUP(sb));
3924 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3925 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3927 bh = sb_getblk(sb, block);
3930 if (!buffer_uptodate(bh)) {
3934 * If the buffer has the write error flag, we have failed
3935 * to write out another inode in the same block. In this
3936 * case, we don't have to read the block because we may
3937 * read the old inode data successfully.
3939 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3940 set_buffer_uptodate(bh);
3942 if (buffer_uptodate(bh)) {
3943 /* someone brought it uptodate while we waited */
3949 * If we have all information of the inode in memory and this
3950 * is the only valid inode in the block, we need not read the
3954 struct buffer_head *bitmap_bh;
3957 start = inode_offset & ~(inodes_per_block - 1);
3959 /* Is the inode bitmap in cache? */
3960 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3961 if (unlikely(!bitmap_bh))
3965 * If the inode bitmap isn't in cache then the
3966 * optimisation may end up performing two reads instead
3967 * of one, so skip it.
3969 if (!buffer_uptodate(bitmap_bh)) {
3973 for (i = start; i < start + inodes_per_block; i++) {
3974 if (i == inode_offset)
3976 if (ext4_test_bit(i, bitmap_bh->b_data))
3980 if (i == start + inodes_per_block) {
3981 /* all other inodes are free, so skip I/O */
3982 memset(bh->b_data, 0, bh->b_size);
3983 set_buffer_uptodate(bh);
3991 * If we need to do any I/O, try to pre-readahead extra
3992 * blocks from the inode table.
3994 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3995 ext4_fsblk_t b, end, table;
3997 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
3999 table = ext4_inode_table(sb, gdp);
4000 /* s_inode_readahead_blks is always a power of 2 */
4001 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4005 num = EXT4_INODES_PER_GROUP(sb);
4006 if (ext4_has_group_desc_csum(sb))
4007 num -= ext4_itable_unused_count(sb, gdp);
4008 table += num / inodes_per_block;
4012 sb_breadahead(sb, b++);
4016 * There are other valid inodes in the buffer, this inode
4017 * has in-inode xattrs, or we don't have this inode in memory.
4018 * Read the block from disk.
4020 trace_ext4_load_inode(inode);
4022 bh->b_end_io = end_buffer_read_sync;
4023 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4025 if (!buffer_uptodate(bh)) {
4026 EXT4_ERROR_INODE_BLOCK(inode, block,
4027 "unable to read itable block");
4037 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4039 /* We have all inode data except xattrs in memory here. */
4040 return __ext4_get_inode_loc(inode, iloc,
4041 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4044 void ext4_set_inode_flags(struct inode *inode)
4046 unsigned int flags = EXT4_I(inode)->i_flags;
4048 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4049 if (flags & EXT4_SYNC_FL)
4050 inode->i_flags |= S_SYNC;
4051 if (flags & EXT4_APPEND_FL)
4052 inode->i_flags |= S_APPEND;
4053 if (flags & EXT4_IMMUTABLE_FL)
4054 inode->i_flags |= S_IMMUTABLE;
4055 if (flags & EXT4_NOATIME_FL)
4056 inode->i_flags |= S_NOATIME;
4057 if (flags & EXT4_DIRSYNC_FL)
4058 inode->i_flags |= S_DIRSYNC;
4061 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4062 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4064 unsigned int vfs_fl;
4065 unsigned long old_fl, new_fl;
4068 vfs_fl = ei->vfs_inode.i_flags;
4069 old_fl = ei->i_flags;
4070 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4071 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4073 if (vfs_fl & S_SYNC)
4074 new_fl |= EXT4_SYNC_FL;
4075 if (vfs_fl & S_APPEND)
4076 new_fl |= EXT4_APPEND_FL;
4077 if (vfs_fl & S_IMMUTABLE)
4078 new_fl |= EXT4_IMMUTABLE_FL;
4079 if (vfs_fl & S_NOATIME)
4080 new_fl |= EXT4_NOATIME_FL;
4081 if (vfs_fl & S_DIRSYNC)
4082 new_fl |= EXT4_DIRSYNC_FL;
4083 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4086 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4087 struct ext4_inode_info *ei)
4090 struct inode *inode = &(ei->vfs_inode);
4091 struct super_block *sb = inode->i_sb;
4093 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4094 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4095 /* we are using combined 48 bit field */
4096 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4097 le32_to_cpu(raw_inode->i_blocks_lo);
4098 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4099 /* i_blocks represent file system block size */
4100 return i_blocks << (inode->i_blkbits - 9);
4105 return le32_to_cpu(raw_inode->i_blocks_lo);
4109 static inline void ext4_iget_extra_inode(struct inode *inode,
4110 struct ext4_inode *raw_inode,
4111 struct ext4_inode_info *ei)
4113 __le32 *magic = (void *)raw_inode +
4114 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4115 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4116 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4117 ext4_find_inline_data_nolock(inode);
4119 EXT4_I(inode)->i_inline_off = 0;
4122 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4124 struct ext4_iloc iloc;
4125 struct ext4_inode *raw_inode;
4126 struct ext4_inode_info *ei;
4127 struct inode *inode;
4128 journal_t *journal = EXT4_SB(sb)->s_journal;
4134 inode = iget_locked(sb, ino);
4136 return ERR_PTR(-ENOMEM);
4137 if (!(inode->i_state & I_NEW))
4143 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4146 raw_inode = ext4_raw_inode(&iloc);
4148 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4149 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4150 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4151 EXT4_INODE_SIZE(inode->i_sb)) {
4152 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4153 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4154 EXT4_INODE_SIZE(inode->i_sb));
4159 ei->i_extra_isize = 0;
4161 /* Precompute checksum seed for inode metadata */
4162 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4163 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4164 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4166 __le32 inum = cpu_to_le32(inode->i_ino);
4167 __le32 gen = raw_inode->i_generation;
4168 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4170 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4174 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4175 EXT4_ERROR_INODE(inode, "checksum invalid");
4180 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4181 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4182 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4183 if (!(test_opt(inode->i_sb, NO_UID32))) {
4184 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4185 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4187 i_uid_write(inode, i_uid);
4188 i_gid_write(inode, i_gid);
4189 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4191 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4192 ei->i_inline_off = 0;
4193 ei->i_dir_start_lookup = 0;
4194 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4195 /* We now have enough fields to check if the inode was active or not.
4196 * This is needed because nfsd might try to access dead inodes
4197 * the test is that same one that e2fsck uses
4198 * NeilBrown 1999oct15
4200 if (inode->i_nlink == 0) {
4201 if ((inode->i_mode == 0 ||
4202 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4203 ino != EXT4_BOOT_LOADER_INO) {
4204 /* this inode is deleted */
4208 /* The only unlinked inodes we let through here have
4209 * valid i_mode and are being read by the orphan
4210 * recovery code: that's fine, we're about to complete
4211 * the process of deleting those.
4212 * OR it is the EXT4_BOOT_LOADER_INO which is
4213 * not initialized on a new filesystem. */
4215 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4216 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4217 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4218 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4220 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4221 inode->i_size = ext4_isize(raw_inode);
4222 ei->i_disksize = inode->i_size;
4224 ei->i_reserved_quota = 0;
4226 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4227 ei->i_block_group = iloc.block_group;
4228 ei->i_last_alloc_group = ~0;
4230 * NOTE! The in-memory inode i_data array is in little-endian order
4231 * even on big-endian machines: we do NOT byteswap the block numbers!
4233 for (block = 0; block < EXT4_N_BLOCKS; block++)
4234 ei->i_data[block] = raw_inode->i_block[block];
4235 INIT_LIST_HEAD(&ei->i_orphan);
4238 * Set transaction id's of transactions that have to be committed
4239 * to finish f[data]sync. We set them to currently running transaction
4240 * as we cannot be sure that the inode or some of its metadata isn't
4241 * part of the transaction - the inode could have been reclaimed and
4242 * now it is reread from disk.
4245 transaction_t *transaction;
4248 read_lock(&journal->j_state_lock);
4249 if (journal->j_running_transaction)
4250 transaction = journal->j_running_transaction;
4252 transaction = journal->j_committing_transaction;
4254 tid = transaction->t_tid;
4256 tid = journal->j_commit_sequence;
4257 read_unlock(&journal->j_state_lock);
4258 ei->i_sync_tid = tid;
4259 ei->i_datasync_tid = tid;
4262 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4263 if (ei->i_extra_isize == 0) {
4264 /* The extra space is currently unused. Use it. */
4265 ei->i_extra_isize = sizeof(struct ext4_inode) -
4266 EXT4_GOOD_OLD_INODE_SIZE;
4268 ext4_iget_extra_inode(inode, raw_inode, ei);
4272 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4273 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4274 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4275 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4277 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4278 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4279 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4281 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4285 if (ei->i_file_acl &&
4286 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4287 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4291 } else if (!ext4_has_inline_data(inode)) {
4292 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4293 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4294 (S_ISLNK(inode->i_mode) &&
4295 !ext4_inode_is_fast_symlink(inode))))
4296 /* Validate extent which is part of inode */
4297 ret = ext4_ext_check_inode(inode);
4298 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4299 (S_ISLNK(inode->i_mode) &&
4300 !ext4_inode_is_fast_symlink(inode))) {
4301 /* Validate block references which are part of inode */
4302 ret = ext4_ind_check_inode(inode);
4308 if (S_ISREG(inode->i_mode)) {
4309 inode->i_op = &ext4_file_inode_operations;
4310 inode->i_fop = &ext4_file_operations;
4311 ext4_set_aops(inode);
4312 } else if (S_ISDIR(inode->i_mode)) {
4313 inode->i_op = &ext4_dir_inode_operations;
4314 inode->i_fop = &ext4_dir_operations;
4315 } else if (S_ISLNK(inode->i_mode)) {
4316 if (ext4_inode_is_fast_symlink(inode)) {
4317 inode->i_op = &ext4_fast_symlink_inode_operations;
4318 nd_terminate_link(ei->i_data, inode->i_size,
4319 sizeof(ei->i_data) - 1);
4321 inode->i_op = &ext4_symlink_inode_operations;
4322 ext4_set_aops(inode);
4324 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4325 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4326 inode->i_op = &ext4_special_inode_operations;
4327 if (raw_inode->i_block[0])
4328 init_special_inode(inode, inode->i_mode,
4329 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4331 init_special_inode(inode, inode->i_mode,
4332 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4333 } else if (ino == EXT4_BOOT_LOADER_INO) {
4334 make_bad_inode(inode);
4337 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4341 ext4_set_inode_flags(inode);
4342 unlock_new_inode(inode);
4348 return ERR_PTR(ret);
4351 static int ext4_inode_blocks_set(handle_t *handle,
4352 struct ext4_inode *raw_inode,
4353 struct ext4_inode_info *ei)
4355 struct inode *inode = &(ei->vfs_inode);
4356 u64 i_blocks = inode->i_blocks;
4357 struct super_block *sb = inode->i_sb;
4359 if (i_blocks <= ~0U) {
4361 * i_blocks can be represented in a 32 bit variable
4362 * as multiple of 512 bytes
4364 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4365 raw_inode->i_blocks_high = 0;
4366 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4369 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4372 if (i_blocks <= 0xffffffffffffULL) {
4374 * i_blocks can be represented in a 48 bit variable
4375 * as multiple of 512 bytes
4377 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4378 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4379 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4381 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4382 /* i_block is stored in file system block size */
4383 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4384 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4385 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4391 * Post the struct inode info into an on-disk inode location in the
4392 * buffer-cache. This gobbles the caller's reference to the
4393 * buffer_head in the inode location struct.
4395 * The caller must have write access to iloc->bh.
4397 static int ext4_do_update_inode(handle_t *handle,
4398 struct inode *inode,
4399 struct ext4_iloc *iloc)
4401 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4402 struct ext4_inode_info *ei = EXT4_I(inode);
4403 struct buffer_head *bh = iloc->bh;
4404 int err = 0, rc, block;
4405 int need_datasync = 0;
4409 /* For fields not not tracking in the in-memory inode,
4410 * initialise them to zero for new inodes. */
4411 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4412 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4414 ext4_get_inode_flags(ei);
4415 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4416 i_uid = i_uid_read(inode);
4417 i_gid = i_gid_read(inode);
4418 if (!(test_opt(inode->i_sb, NO_UID32))) {
4419 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4420 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4422 * Fix up interoperability with old kernels. Otherwise, old inodes get
4423 * re-used with the upper 16 bits of the uid/gid intact
4426 raw_inode->i_uid_high =
4427 cpu_to_le16(high_16_bits(i_uid));
4428 raw_inode->i_gid_high =
4429 cpu_to_le16(high_16_bits(i_gid));
4431 raw_inode->i_uid_high = 0;
4432 raw_inode->i_gid_high = 0;
4435 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4436 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4437 raw_inode->i_uid_high = 0;
4438 raw_inode->i_gid_high = 0;
4440 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4442 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4443 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4444 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4445 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4447 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4449 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4450 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4451 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4452 cpu_to_le32(EXT4_OS_HURD))
4453 raw_inode->i_file_acl_high =
4454 cpu_to_le16(ei->i_file_acl >> 32);
4455 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4456 if (ei->i_disksize != ext4_isize(raw_inode)) {
4457 ext4_isize_set(raw_inode, ei->i_disksize);
4460 if (ei->i_disksize > 0x7fffffffULL) {
4461 struct super_block *sb = inode->i_sb;
4462 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4463 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4464 EXT4_SB(sb)->s_es->s_rev_level ==
4465 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4466 /* If this is the first large file
4467 * created, add a flag to the superblock.
4469 err = ext4_journal_get_write_access(handle,
4470 EXT4_SB(sb)->s_sbh);
4473 ext4_update_dynamic_rev(sb);
4474 EXT4_SET_RO_COMPAT_FEATURE(sb,
4475 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4476 ext4_handle_sync(handle);
4477 err = ext4_handle_dirty_super(handle, sb);
4480 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4481 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4482 if (old_valid_dev(inode->i_rdev)) {
4483 raw_inode->i_block[0] =
4484 cpu_to_le32(old_encode_dev(inode->i_rdev));
4485 raw_inode->i_block[1] = 0;
4487 raw_inode->i_block[0] = 0;
4488 raw_inode->i_block[1] =
4489 cpu_to_le32(new_encode_dev(inode->i_rdev));
4490 raw_inode->i_block[2] = 0;
4492 } else if (!ext4_has_inline_data(inode)) {
4493 for (block = 0; block < EXT4_N_BLOCKS; block++)
4494 raw_inode->i_block[block] = ei->i_data[block];
4497 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4498 if (ei->i_extra_isize) {
4499 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4500 raw_inode->i_version_hi =
4501 cpu_to_le32(inode->i_version >> 32);
4502 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4505 ext4_inode_csum_set(inode, raw_inode, ei);
4507 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4508 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4511 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4513 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4516 ext4_std_error(inode->i_sb, err);
4521 * ext4_write_inode()
4523 * We are called from a few places:
4525 * - Within generic_file_write() for O_SYNC files.
4526 * Here, there will be no transaction running. We wait for any running
4527 * transaction to commit.
4529 * - Within sys_sync(), kupdate and such.
4530 * We wait on commit, if tol to.
4532 * - Within prune_icache() (PF_MEMALLOC == true)
4533 * Here we simply return. We can't afford to block kswapd on the
4536 * In all cases it is actually safe for us to return without doing anything,
4537 * because the inode has been copied into a raw inode buffer in
4538 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4541 * Note that we are absolutely dependent upon all inode dirtiers doing the
4542 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4543 * which we are interested.
4545 * It would be a bug for them to not do this. The code:
4547 * mark_inode_dirty(inode)
4549 * inode->i_size = expr;
4551 * is in error because a kswapd-driven write_inode() could occur while
4552 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4553 * will no longer be on the superblock's dirty inode list.
4555 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4559 if (current->flags & PF_MEMALLOC)
4562 if (EXT4_SB(inode->i_sb)->s_journal) {
4563 if (ext4_journal_current_handle()) {
4564 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4569 if (wbc->sync_mode != WB_SYNC_ALL)
4572 err = ext4_force_commit(inode->i_sb);
4574 struct ext4_iloc iloc;
4576 err = __ext4_get_inode_loc(inode, &iloc, 0);
4579 if (wbc->sync_mode == WB_SYNC_ALL)
4580 sync_dirty_buffer(iloc.bh);
4581 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4582 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4583 "IO error syncing inode");
4592 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4593 * buffers that are attached to a page stradding i_size and are undergoing
4594 * commit. In that case we have to wait for commit to finish and try again.
4596 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4600 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4601 tid_t commit_tid = 0;
4604 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4606 * All buffers in the last page remain valid? Then there's nothing to
4607 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4610 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4613 page = find_lock_page(inode->i_mapping,
4614 inode->i_size >> PAGE_CACHE_SHIFT);
4617 ret = __ext4_journalled_invalidatepage(page, offset);
4619 page_cache_release(page);
4623 read_lock(&journal->j_state_lock);
4624 if (journal->j_committing_transaction)
4625 commit_tid = journal->j_committing_transaction->t_tid;
4626 read_unlock(&journal->j_state_lock);
4628 jbd2_log_wait_commit(journal, commit_tid);
4635 * Called from notify_change.
4637 * We want to trap VFS attempts to truncate the file as soon as
4638 * possible. In particular, we want to make sure that when the VFS
4639 * shrinks i_size, we put the inode on the orphan list and modify
4640 * i_disksize immediately, so that during the subsequent flushing of
4641 * dirty pages and freeing of disk blocks, we can guarantee that any
4642 * commit will leave the blocks being flushed in an unused state on
4643 * disk. (On recovery, the inode will get truncated and the blocks will
4644 * be freed, so we have a strong guarantee that no future commit will
4645 * leave these blocks visible to the user.)
4647 * Another thing we have to assure is that if we are in ordered mode
4648 * and inode is still attached to the committing transaction, we must
4649 * we start writeout of all the dirty pages which are being truncated.
4650 * This way we are sure that all the data written in the previous
4651 * transaction are already on disk (truncate waits for pages under
4654 * Called with inode->i_mutex down.
4656 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4658 struct inode *inode = dentry->d_inode;
4661 const unsigned int ia_valid = attr->ia_valid;
4663 error = inode_change_ok(inode, attr);
4667 if (is_quota_modification(inode, attr))
4668 dquot_initialize(inode);
4669 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4670 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4673 /* (user+group)*(old+new) structure, inode write (sb,
4674 * inode block, ? - but truncate inode update has it) */
4675 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4676 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4677 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4678 if (IS_ERR(handle)) {
4679 error = PTR_ERR(handle);
4682 error = dquot_transfer(inode, attr);
4684 ext4_journal_stop(handle);
4687 /* Update corresponding info in inode so that everything is in
4688 * one transaction */
4689 if (attr->ia_valid & ATTR_UID)
4690 inode->i_uid = attr->ia_uid;
4691 if (attr->ia_valid & ATTR_GID)
4692 inode->i_gid = attr->ia_gid;
4693 error = ext4_mark_inode_dirty(handle, inode);
4694 ext4_journal_stop(handle);
4697 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4699 loff_t oldsize = inode->i_size;
4701 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4702 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4704 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4707 if (S_ISREG(inode->i_mode) &&
4708 (attr->ia_size < inode->i_size)) {
4709 if (ext4_should_order_data(inode)) {
4710 error = ext4_begin_ordered_truncate(inode,
4715 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4716 if (IS_ERR(handle)) {
4717 error = PTR_ERR(handle);
4720 if (ext4_handle_valid(handle)) {
4721 error = ext4_orphan_add(handle, inode);
4724 EXT4_I(inode)->i_disksize = attr->ia_size;
4725 rc = ext4_mark_inode_dirty(handle, inode);
4728 ext4_journal_stop(handle);
4730 ext4_orphan_del(NULL, inode);
4735 i_size_write(inode, attr->ia_size);
4737 * Blocks are going to be removed from the inode. Wait
4738 * for dio in flight. Temporarily disable
4739 * dioread_nolock to prevent livelock.
4742 if (!ext4_should_journal_data(inode)) {
4743 ext4_inode_block_unlocked_dio(inode);
4744 inode_dio_wait(inode);
4745 ext4_inode_resume_unlocked_dio(inode);
4747 ext4_wait_for_tail_page_commit(inode);
4750 * Truncate pagecache after we've waited for commit
4751 * in data=journal mode to make pages freeable.
4753 truncate_pagecache(inode, oldsize, inode->i_size);
4756 * We want to call ext4_truncate() even if attr->ia_size ==
4757 * inode->i_size for cases like truncation of fallocated space
4759 if (attr->ia_valid & ATTR_SIZE)
4760 ext4_truncate(inode);
4763 setattr_copy(inode, attr);
4764 mark_inode_dirty(inode);
4768 * If the call to ext4_truncate failed to get a transaction handle at
4769 * all, we need to clean up the in-core orphan list manually.
4771 if (orphan && inode->i_nlink)
4772 ext4_orphan_del(NULL, inode);
4774 if (!rc && (ia_valid & ATTR_MODE))
4775 rc = ext4_acl_chmod(inode);
4778 ext4_std_error(inode->i_sb, error);
4784 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4787 struct inode *inode;
4788 unsigned long long delalloc_blocks;
4790 inode = dentry->d_inode;
4791 generic_fillattr(inode, stat);
4794 * We can't update i_blocks if the block allocation is delayed
4795 * otherwise in the case of system crash before the real block
4796 * allocation is done, we will have i_blocks inconsistent with
4797 * on-disk file blocks.
4798 * We always keep i_blocks updated together with real
4799 * allocation. But to not confuse with user, stat
4800 * will return the blocks that include the delayed allocation
4801 * blocks for this file.
4803 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4804 EXT4_I(inode)->i_reserved_data_blocks);
4806 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4810 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4812 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4813 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4814 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4818 * Account for index blocks, block groups bitmaps and block group
4819 * descriptor blocks if modify datablocks and index blocks
4820 * worse case, the indexs blocks spread over different block groups
4822 * If datablocks are discontiguous, they are possible to spread over
4823 * different block groups too. If they are contiguous, with flexbg,
4824 * they could still across block group boundary.
4826 * Also account for superblock, inode, quota and xattr blocks
4828 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4830 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4836 * How many index blocks need to touch to modify nrblocks?
4837 * The "Chunk" flag indicating whether the nrblocks is
4838 * physically contiguous on disk
4840 * For Direct IO and fallocate, they calls get_block to allocate
4841 * one single extent at a time, so they could set the "Chunk" flag
4843 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4848 * Now let's see how many group bitmaps and group descriptors need
4858 if (groups > ngroups)
4860 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4861 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4863 /* bitmaps and block group descriptor blocks */
4864 ret += groups + gdpblocks;
4866 /* Blocks for super block, inode, quota and xattr blocks */
4867 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4873 * Calculate the total number of credits to reserve to fit
4874 * the modification of a single pages into a single transaction,
4875 * which may include multiple chunks of block allocations.
4877 * This could be called via ext4_write_begin()
4879 * We need to consider the worse case, when
4880 * one new block per extent.
4882 int ext4_writepage_trans_blocks(struct inode *inode)
4884 int bpp = ext4_journal_blocks_per_page(inode);
4887 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4889 /* Account for data blocks for journalled mode */
4890 if (ext4_should_journal_data(inode))
4896 * Calculate the journal credits for a chunk of data modification.
4898 * This is called from DIO, fallocate or whoever calling
4899 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4901 * journal buffers for data blocks are not included here, as DIO
4902 * and fallocate do no need to journal data buffers.
4904 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4906 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4910 * The caller must have previously called ext4_reserve_inode_write().
4911 * Give this, we know that the caller already has write access to iloc->bh.
4913 int ext4_mark_iloc_dirty(handle_t *handle,
4914 struct inode *inode, struct ext4_iloc *iloc)
4918 if (IS_I_VERSION(inode))
4919 inode_inc_iversion(inode);
4921 /* the do_update_inode consumes one bh->b_count */
4924 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4925 err = ext4_do_update_inode(handle, inode, iloc);
4931 * On success, We end up with an outstanding reference count against
4932 * iloc->bh. This _must_ be cleaned up later.
4936 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4937 struct ext4_iloc *iloc)
4941 err = ext4_get_inode_loc(inode, iloc);
4943 BUFFER_TRACE(iloc->bh, "get_write_access");
4944 err = ext4_journal_get_write_access(handle, iloc->bh);
4950 ext4_std_error(inode->i_sb, err);
4955 * Expand an inode by new_extra_isize bytes.
4956 * Returns 0 on success or negative error number on failure.
4958 static int ext4_expand_extra_isize(struct inode *inode,
4959 unsigned int new_extra_isize,
4960 struct ext4_iloc iloc,
4963 struct ext4_inode *raw_inode;
4964 struct ext4_xattr_ibody_header *header;
4966 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4969 raw_inode = ext4_raw_inode(&iloc);
4971 header = IHDR(inode, raw_inode);
4973 /* No extended attributes present */
4974 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4975 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4976 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4978 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4982 /* try to expand with EAs present */
4983 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4988 * What we do here is to mark the in-core inode as clean with respect to inode
4989 * dirtiness (it may still be data-dirty).
4990 * This means that the in-core inode may be reaped by prune_icache
4991 * without having to perform any I/O. This is a very good thing,
4992 * because *any* task may call prune_icache - even ones which
4993 * have a transaction open against a different journal.
4995 * Is this cheating? Not really. Sure, we haven't written the
4996 * inode out, but prune_icache isn't a user-visible syncing function.
4997 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4998 * we start and wait on commits.
5000 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5002 struct ext4_iloc iloc;
5003 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5004 static unsigned int mnt_count;
5008 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5009 err = ext4_reserve_inode_write(handle, inode, &iloc);
5010 if (ext4_handle_valid(handle) &&
5011 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5012 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5014 * We need extra buffer credits since we may write into EA block
5015 * with this same handle. If journal_extend fails, then it will
5016 * only result in a minor loss of functionality for that inode.
5017 * If this is felt to be critical, then e2fsck should be run to
5018 * force a large enough s_min_extra_isize.
5020 if ((jbd2_journal_extend(handle,
5021 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5022 ret = ext4_expand_extra_isize(inode,
5023 sbi->s_want_extra_isize,
5026 ext4_set_inode_state(inode,
5027 EXT4_STATE_NO_EXPAND);
5029 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5030 ext4_warning(inode->i_sb,
5031 "Unable to expand inode %lu. Delete"
5032 " some EAs or run e2fsck.",
5035 le16_to_cpu(sbi->s_es->s_mnt_count);
5041 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5046 * ext4_dirty_inode() is called from __mark_inode_dirty()
5048 * We're really interested in the case where a file is being extended.
5049 * i_size has been changed by generic_commit_write() and we thus need
5050 * to include the updated inode in the current transaction.
5052 * Also, dquot_alloc_block() will always dirty the inode when blocks
5053 * are allocated to the file.
5055 * If the inode is marked synchronous, we don't honour that here - doing
5056 * so would cause a commit on atime updates, which we don't bother doing.
5057 * We handle synchronous inodes at the highest possible level.
5059 void ext4_dirty_inode(struct inode *inode, int flags)
5063 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5067 ext4_mark_inode_dirty(handle, inode);
5069 ext4_journal_stop(handle);
5076 * Bind an inode's backing buffer_head into this transaction, to prevent
5077 * it from being flushed to disk early. Unlike
5078 * ext4_reserve_inode_write, this leaves behind no bh reference and
5079 * returns no iloc structure, so the caller needs to repeat the iloc
5080 * lookup to mark the inode dirty later.
5082 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5084 struct ext4_iloc iloc;
5088 err = ext4_get_inode_loc(inode, &iloc);
5090 BUFFER_TRACE(iloc.bh, "get_write_access");
5091 err = jbd2_journal_get_write_access(handle, iloc.bh);
5093 err = ext4_handle_dirty_metadata(handle,
5099 ext4_std_error(inode->i_sb, err);
5104 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5111 * We have to be very careful here: changing a data block's
5112 * journaling status dynamically is dangerous. If we write a
5113 * data block to the journal, change the status and then delete
5114 * that block, we risk forgetting to revoke the old log record
5115 * from the journal and so a subsequent replay can corrupt data.
5116 * So, first we make sure that the journal is empty and that
5117 * nobody is changing anything.
5120 journal = EXT4_JOURNAL(inode);
5123 if (is_journal_aborted(journal))
5125 /* We have to allocate physical blocks for delalloc blocks
5126 * before flushing journal. otherwise delalloc blocks can not
5127 * be allocated any more. even more truncate on delalloc blocks
5128 * could trigger BUG by flushing delalloc blocks in journal.
5129 * There is no delalloc block in non-journal data mode.
5131 if (val && test_opt(inode->i_sb, DELALLOC)) {
5132 err = ext4_alloc_da_blocks(inode);
5137 /* Wait for all existing dio workers */
5138 ext4_inode_block_unlocked_dio(inode);
5139 inode_dio_wait(inode);
5141 jbd2_journal_lock_updates(journal);
5144 * OK, there are no updates running now, and all cached data is
5145 * synced to disk. We are now in a completely consistent state
5146 * which doesn't have anything in the journal, and we know that
5147 * no filesystem updates are running, so it is safe to modify
5148 * the inode's in-core data-journaling state flag now.
5152 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5154 jbd2_journal_flush(journal);
5155 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5157 ext4_set_aops(inode);
5159 jbd2_journal_unlock_updates(journal);
5160 ext4_inode_resume_unlocked_dio(inode);
5162 /* Finally we can mark the inode as dirty. */
5164 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5166 return PTR_ERR(handle);
5168 err = ext4_mark_inode_dirty(handle, inode);
5169 ext4_handle_sync(handle);
5170 ext4_journal_stop(handle);
5171 ext4_std_error(inode->i_sb, err);
5176 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5178 return !buffer_mapped(bh);
5181 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5183 struct page *page = vmf->page;
5187 struct file *file = vma->vm_file;
5188 struct inode *inode = file_inode(file);
5189 struct address_space *mapping = inode->i_mapping;
5191 get_block_t *get_block;
5194 sb_start_pagefault(inode->i_sb);
5195 file_update_time(vma->vm_file);
5196 /* Delalloc case is easy... */
5197 if (test_opt(inode->i_sb, DELALLOC) &&
5198 !ext4_should_journal_data(inode) &&
5199 !ext4_nonda_switch(inode->i_sb)) {
5201 ret = __block_page_mkwrite(vma, vmf,
5202 ext4_da_get_block_prep);
5203 } while (ret == -ENOSPC &&
5204 ext4_should_retry_alloc(inode->i_sb, &retries));
5209 size = i_size_read(inode);
5210 /* Page got truncated from under us? */
5211 if (page->mapping != mapping || page_offset(page) > size) {
5213 ret = VM_FAULT_NOPAGE;
5217 if (page->index == size >> PAGE_CACHE_SHIFT)
5218 len = size & ~PAGE_CACHE_MASK;
5220 len = PAGE_CACHE_SIZE;
5222 * Return if we have all the buffers mapped. This avoids the need to do
5223 * journal_start/journal_stop which can block and take a long time
5225 if (page_has_buffers(page)) {
5226 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5228 ext4_bh_unmapped)) {
5229 /* Wait so that we don't change page under IO */
5230 wait_for_stable_page(page);
5231 ret = VM_FAULT_LOCKED;
5236 /* OK, we need to fill the hole... */
5237 if (ext4_should_dioread_nolock(inode))
5238 get_block = ext4_get_block_write;
5240 get_block = ext4_get_block;
5242 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5243 ext4_writepage_trans_blocks(inode));
5244 if (IS_ERR(handle)) {
5245 ret = VM_FAULT_SIGBUS;
5248 ret = __block_page_mkwrite(vma, vmf, get_block);
5249 if (!ret && ext4_should_journal_data(inode)) {
5250 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5251 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5253 ret = VM_FAULT_SIGBUS;
5254 ext4_journal_stop(handle);
5257 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5259 ext4_journal_stop(handle);
5260 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5263 ret = block_page_mkwrite_return(ret);
5265 sb_end_pagefault(inode->i_sb);