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 copied = ext4_write_inline_data_end(inode, pos, len,
1125 copied = block_write_end(file, mapping, pos,
1126 len, copied, page, fsdata);
1129 * No need to use i_size_read() here, the i_size
1130 * cannot change under us because we hole i_mutex.
1132 * But it's important to update i_size while still holding page lock:
1133 * page writeout could otherwise come in and zero beyond i_size.
1135 if (pos + copied > inode->i_size) {
1136 i_size_write(inode, pos + copied);
1140 if (pos + copied > EXT4_I(inode)->i_disksize) {
1141 /* We need to mark inode dirty even if
1142 * new_i_size is less that inode->i_size
1143 * but greater than i_disksize. (hint delalloc)
1145 ext4_update_i_disksize(inode, (pos + copied));
1149 page_cache_release(page);
1152 * Don't mark the inode dirty under page lock. First, it unnecessarily
1153 * makes the holding time of page lock longer. Second, it forces lock
1154 * ordering of page lock and transaction start for journaling
1158 ext4_mark_inode_dirty(handle, inode);
1162 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1163 /* if we have allocated more blocks and copied
1164 * less. We will have blocks allocated outside
1165 * inode->i_size. So truncate them
1167 ext4_orphan_add(handle, inode);
1169 ret2 = ext4_journal_stop(handle);
1173 if (pos + len > inode->i_size) {
1174 ext4_truncate_failed_write(inode);
1176 * If truncate failed early the inode might still be
1177 * on the orphan list; we need to make sure the inode
1178 * is removed from the orphan list in that case.
1181 ext4_orphan_del(NULL, inode);
1184 return ret ? ret : copied;
1187 static int ext4_journalled_write_end(struct file *file,
1188 struct address_space *mapping,
1189 loff_t pos, unsigned len, unsigned copied,
1190 struct page *page, void *fsdata)
1192 handle_t *handle = ext4_journal_current_handle();
1193 struct inode *inode = mapping->host;
1199 trace_ext4_journalled_write_end(inode, pos, len, copied);
1200 from = pos & (PAGE_CACHE_SIZE - 1);
1203 BUG_ON(!ext4_handle_valid(handle));
1205 if (ext4_has_inline_data(inode))
1206 copied = ext4_write_inline_data_end(inode, pos, len,
1210 if (!PageUptodate(page))
1212 page_zero_new_buffers(page, from+copied, to);
1215 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1216 to, &partial, write_end_fn);
1218 SetPageUptodate(page);
1220 new_i_size = pos + copied;
1221 if (new_i_size > inode->i_size)
1222 i_size_write(inode, pos+copied);
1223 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1224 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1225 if (new_i_size > EXT4_I(inode)->i_disksize) {
1226 ext4_update_i_disksize(inode, new_i_size);
1227 ret2 = ext4_mark_inode_dirty(handle, inode);
1233 page_cache_release(page);
1234 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1235 /* if we have allocated more blocks and copied
1236 * less. We will have blocks allocated outside
1237 * inode->i_size. So truncate them
1239 ext4_orphan_add(handle, inode);
1241 ret2 = ext4_journal_stop(handle);
1244 if (pos + len > inode->i_size) {
1245 ext4_truncate_failed_write(inode);
1247 * If truncate failed early the inode might still be
1248 * on the orphan list; we need to make sure the inode
1249 * is removed from the orphan list in that case.
1252 ext4_orphan_del(NULL, inode);
1255 return ret ? ret : copied;
1259 * Reserve a metadata for a single block located at lblock
1261 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1264 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1265 struct ext4_inode_info *ei = EXT4_I(inode);
1266 unsigned int md_needed;
1267 ext4_lblk_t save_last_lblock;
1271 * recalculate the amount of metadata blocks to reserve
1272 * in order to allocate nrblocks
1273 * worse case is one extent per block
1276 spin_lock(&ei->i_block_reservation_lock);
1278 * ext4_calc_metadata_amount() has side effects, which we have
1279 * to be prepared undo if we fail to claim space.
1281 save_len = ei->i_da_metadata_calc_len;
1282 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1283 md_needed = EXT4_NUM_B2C(sbi,
1284 ext4_calc_metadata_amount(inode, lblock));
1285 trace_ext4_da_reserve_space(inode, md_needed);
1288 * We do still charge estimated metadata to the sb though;
1289 * we cannot afford to run out of free blocks.
1291 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1292 ei->i_da_metadata_calc_len = save_len;
1293 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1294 spin_unlock(&ei->i_block_reservation_lock);
1295 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1301 ei->i_reserved_meta_blocks += md_needed;
1302 spin_unlock(&ei->i_block_reservation_lock);
1304 return 0; /* success */
1308 * Reserve a single cluster located at lblock
1310 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1313 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1314 struct ext4_inode_info *ei = EXT4_I(inode);
1315 unsigned int md_needed;
1317 ext4_lblk_t save_last_lblock;
1321 * We will charge metadata quota at writeout time; this saves
1322 * us from metadata over-estimation, though we may go over by
1323 * a small amount in the end. Here we just reserve for data.
1325 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1330 * recalculate the amount of metadata blocks to reserve
1331 * in order to allocate nrblocks
1332 * worse case is one extent per block
1335 spin_lock(&ei->i_block_reservation_lock);
1337 * ext4_calc_metadata_amount() has side effects, which we have
1338 * to be prepared undo if we fail to claim space.
1340 save_len = ei->i_da_metadata_calc_len;
1341 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1342 md_needed = EXT4_NUM_B2C(sbi,
1343 ext4_calc_metadata_amount(inode, lblock));
1344 trace_ext4_da_reserve_space(inode, md_needed);
1347 * We do still charge estimated metadata to the sb though;
1348 * we cannot afford to run out of free blocks.
1350 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1351 ei->i_da_metadata_calc_len = save_len;
1352 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1353 spin_unlock(&ei->i_block_reservation_lock);
1354 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1358 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1361 ei->i_reserved_data_blocks++;
1362 ei->i_reserved_meta_blocks += md_needed;
1363 spin_unlock(&ei->i_block_reservation_lock);
1365 return 0; /* success */
1368 static void ext4_da_release_space(struct inode *inode, int to_free)
1370 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1371 struct ext4_inode_info *ei = EXT4_I(inode);
1374 return; /* Nothing to release, exit */
1376 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1378 trace_ext4_da_release_space(inode, to_free);
1379 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1381 * if there aren't enough reserved blocks, then the
1382 * counter is messed up somewhere. Since this
1383 * function is called from invalidate page, it's
1384 * harmless to return without any action.
1386 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1387 "ino %lu, to_free %d with only %d reserved "
1388 "data blocks", inode->i_ino, to_free,
1389 ei->i_reserved_data_blocks);
1391 to_free = ei->i_reserved_data_blocks;
1393 ei->i_reserved_data_blocks -= to_free;
1395 if (ei->i_reserved_data_blocks == 0) {
1397 * We can release all of the reserved metadata blocks
1398 * only when we have written all of the delayed
1399 * allocation blocks.
1400 * Note that in case of bigalloc, i_reserved_meta_blocks,
1401 * i_reserved_data_blocks, etc. refer to number of clusters.
1403 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1404 ei->i_reserved_meta_blocks);
1405 ei->i_reserved_meta_blocks = 0;
1406 ei->i_da_metadata_calc_len = 0;
1409 /* update fs dirty data blocks counter */
1410 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1412 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1414 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1417 static void ext4_da_page_release_reservation(struct page *page,
1418 unsigned long offset)
1421 struct buffer_head *head, *bh;
1422 unsigned int curr_off = 0;
1423 struct inode *inode = page->mapping->host;
1424 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1428 head = page_buffers(page);
1431 unsigned int next_off = curr_off + bh->b_size;
1433 if ((offset <= curr_off) && (buffer_delay(bh))) {
1435 clear_buffer_delay(bh);
1437 curr_off = next_off;
1438 } while ((bh = bh->b_this_page) != head);
1441 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1442 ext4_es_remove_extent(inode, lblk, to_release);
1445 /* If we have released all the blocks belonging to a cluster, then we
1446 * need to release the reserved space for that cluster. */
1447 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1448 while (num_clusters > 0) {
1449 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1450 ((num_clusters - 1) << sbi->s_cluster_bits);
1451 if (sbi->s_cluster_ratio == 1 ||
1452 !ext4_find_delalloc_cluster(inode, lblk))
1453 ext4_da_release_space(inode, 1);
1460 * Delayed allocation stuff
1464 * mpage_da_submit_io - walks through extent of pages and try to write
1465 * them with writepage() call back
1467 * @mpd->inode: inode
1468 * @mpd->first_page: first page of the extent
1469 * @mpd->next_page: page after the last page of the extent
1471 * By the time mpage_da_submit_io() is called we expect all blocks
1472 * to be allocated. this may be wrong if allocation failed.
1474 * As pages are already locked by write_cache_pages(), we can't use it
1476 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1477 struct ext4_map_blocks *map)
1479 struct pagevec pvec;
1480 unsigned long index, end;
1481 int ret = 0, err, nr_pages, i;
1482 struct inode *inode = mpd->inode;
1483 struct address_space *mapping = inode->i_mapping;
1484 loff_t size = i_size_read(inode);
1485 unsigned int len, block_start;
1486 struct buffer_head *bh, *page_bufs = NULL;
1487 sector_t pblock = 0, cur_logical = 0;
1488 struct ext4_io_submit io_submit;
1490 BUG_ON(mpd->next_page <= mpd->first_page);
1491 ext4_io_submit_init(&io_submit, mpd->wbc);
1492 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1493 if (!io_submit.io_end)
1496 * We need to start from the first_page to the next_page - 1
1497 * to make sure we also write the mapped dirty buffer_heads.
1498 * If we look at mpd->b_blocknr we would only be looking
1499 * at the currently mapped buffer_heads.
1501 index = mpd->first_page;
1502 end = mpd->next_page - 1;
1504 pagevec_init(&pvec, 0);
1505 while (index <= end) {
1506 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1509 for (i = 0; i < nr_pages; i++) {
1511 struct page *page = pvec.pages[i];
1513 index = page->index;
1517 if (index == size >> PAGE_CACHE_SHIFT)
1518 len = size & ~PAGE_CACHE_MASK;
1520 len = PAGE_CACHE_SIZE;
1522 cur_logical = index << (PAGE_CACHE_SHIFT -
1524 pblock = map->m_pblk + (cur_logical -
1529 BUG_ON(!PageLocked(page));
1530 BUG_ON(PageWriteback(page));
1532 bh = page_bufs = page_buffers(page);
1535 if (map && (cur_logical >= map->m_lblk) &&
1536 (cur_logical <= (map->m_lblk +
1537 (map->m_len - 1)))) {
1538 if (buffer_delay(bh)) {
1539 clear_buffer_delay(bh);
1540 bh->b_blocknr = pblock;
1542 if (buffer_unwritten(bh) ||
1544 BUG_ON(bh->b_blocknr != pblock);
1545 if (map->m_flags & EXT4_MAP_UNINIT)
1546 set_buffer_uninit(bh);
1547 clear_buffer_unwritten(bh);
1551 * skip page if block allocation undone and
1554 if (ext4_bh_delay_or_unwritten(NULL, bh))
1556 bh = bh->b_this_page;
1557 block_start += bh->b_size;
1560 } while (bh != page_bufs);
1567 clear_page_dirty_for_io(page);
1568 err = ext4_bio_write_page(&io_submit, page, len,
1571 mpd->pages_written++;
1573 * In error case, we have to continue because
1574 * remaining pages are still locked
1579 pagevec_release(&pvec);
1581 ext4_io_submit(&io_submit);
1582 /* Drop io_end reference we got from init */
1583 ext4_put_io_end_defer(io_submit.io_end);
1587 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1591 struct pagevec pvec;
1592 struct inode *inode = mpd->inode;
1593 struct address_space *mapping = inode->i_mapping;
1594 ext4_lblk_t start, last;
1596 index = mpd->first_page;
1597 end = mpd->next_page - 1;
1599 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1600 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1601 ext4_es_remove_extent(inode, start, last - start + 1);
1603 pagevec_init(&pvec, 0);
1604 while (index <= end) {
1605 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1608 for (i = 0; i < nr_pages; i++) {
1609 struct page *page = pvec.pages[i];
1610 if (page->index > end)
1612 BUG_ON(!PageLocked(page));
1613 BUG_ON(PageWriteback(page));
1614 block_invalidatepage(page, 0);
1615 ClearPageUptodate(page);
1618 index = pvec.pages[nr_pages - 1]->index + 1;
1619 pagevec_release(&pvec);
1624 static void ext4_print_free_blocks(struct inode *inode)
1626 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1627 struct super_block *sb = inode->i_sb;
1628 struct ext4_inode_info *ei = EXT4_I(inode);
1630 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1631 EXT4_C2B(EXT4_SB(inode->i_sb),
1632 ext4_count_free_clusters(sb)));
1633 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1634 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1635 (long long) EXT4_C2B(EXT4_SB(sb),
1636 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1637 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1638 (long long) EXT4_C2B(EXT4_SB(sb),
1639 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1640 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1641 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1642 ei->i_reserved_data_blocks);
1643 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1644 ei->i_reserved_meta_blocks);
1645 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1646 ei->i_allocated_meta_blocks);
1651 * mpage_da_map_and_submit - go through given space, map them
1652 * if necessary, and then submit them for I/O
1654 * @mpd - bh describing space
1656 * The function skips space we know is already mapped to disk blocks.
1659 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1661 int err, blks, get_blocks_flags;
1662 struct ext4_map_blocks map, *mapp = NULL;
1663 sector_t next = mpd->b_blocknr;
1664 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1665 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1666 handle_t *handle = NULL;
1669 * If the blocks are mapped already, or we couldn't accumulate
1670 * any blocks, then proceed immediately to the submission stage.
1672 if ((mpd->b_size == 0) ||
1673 ((mpd->b_state & (1 << BH_Mapped)) &&
1674 !(mpd->b_state & (1 << BH_Delay)) &&
1675 !(mpd->b_state & (1 << BH_Unwritten))))
1678 handle = ext4_journal_current_handle();
1682 * Call ext4_map_blocks() to allocate any delayed allocation
1683 * blocks, or to convert an uninitialized extent to be
1684 * initialized (in the case where we have written into
1685 * one or more preallocated blocks).
1687 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1688 * indicate that we are on the delayed allocation path. This
1689 * affects functions in many different parts of the allocation
1690 * call path. This flag exists primarily because we don't
1691 * want to change *many* call functions, so ext4_map_blocks()
1692 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1693 * inode's allocation semaphore is taken.
1695 * If the blocks in questions were delalloc blocks, set
1696 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1697 * variables are updated after the blocks have been allocated.
1700 map.m_len = max_blocks;
1702 * We're in delalloc path and it is possible that we're going to
1703 * need more metadata blocks than previously reserved. However
1704 * we must not fail because we're in writeback and there is
1705 * nothing we can do about it so it might result in data loss.
1706 * So use reserved blocks to allocate metadata if possible.
1708 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1709 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1710 if (ext4_should_dioread_nolock(mpd->inode))
1711 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1712 if (mpd->b_state & (1 << BH_Delay))
1713 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1716 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1718 struct super_block *sb = mpd->inode->i_sb;
1722 * If get block returns EAGAIN or ENOSPC and there
1723 * appears to be free blocks we will just let
1724 * mpage_da_submit_io() unlock all of the pages.
1729 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1735 * get block failure will cause us to loop in
1736 * writepages, because a_ops->writepage won't be able
1737 * to make progress. The page will be redirtied by
1738 * writepage and writepages will again try to write
1741 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1742 ext4_msg(sb, KERN_CRIT,
1743 "delayed block allocation failed for inode %lu "
1744 "at logical offset %llu with max blocks %zd "
1745 "with error %d", mpd->inode->i_ino,
1746 (unsigned long long) next,
1747 mpd->b_size >> mpd->inode->i_blkbits, err);
1748 ext4_msg(sb, KERN_CRIT,
1749 "This should not happen!! Data will be lost");
1751 ext4_print_free_blocks(mpd->inode);
1753 /* invalidate all the pages */
1754 ext4_da_block_invalidatepages(mpd);
1756 /* Mark this page range as having been completed */
1763 if (map.m_flags & EXT4_MAP_NEW) {
1764 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1767 for (i = 0; i < map.m_len; i++)
1768 unmap_underlying_metadata(bdev, map.m_pblk + i);
1772 * Update on-disk size along with block allocation.
1774 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1775 if (disksize > i_size_read(mpd->inode))
1776 disksize = i_size_read(mpd->inode);
1777 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1778 ext4_update_i_disksize(mpd->inode, disksize);
1779 err = ext4_mark_inode_dirty(handle, mpd->inode);
1781 ext4_error(mpd->inode->i_sb,
1782 "Failed to mark inode %lu dirty",
1787 mpage_da_submit_io(mpd, mapp);
1791 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1792 (1 << BH_Delay) | (1 << BH_Unwritten))
1795 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1797 * @mpd->lbh - extent of blocks
1798 * @logical - logical number of the block in the file
1799 * @b_state - b_state of the buffer head added
1801 * the function is used to collect contig. blocks in same state
1803 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1804 unsigned long b_state)
1807 int blkbits = mpd->inode->i_blkbits;
1808 int nrblocks = mpd->b_size >> blkbits;
1811 * XXX Don't go larger than mballoc is willing to allocate
1812 * This is a stopgap solution. We eventually need to fold
1813 * mpage_da_submit_io() into this function and then call
1814 * ext4_map_blocks() multiple times in a loop
1816 if (nrblocks >= (8*1024*1024 >> blkbits))
1819 /* check if the reserved journal credits might overflow */
1820 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1821 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1823 * With non-extent format we are limited by the journal
1824 * credit available. Total credit needed to insert
1825 * nrblocks contiguous blocks is dependent on the
1826 * nrblocks. So limit nrblocks.
1832 * First block in the extent
1834 if (mpd->b_size == 0) {
1835 mpd->b_blocknr = logical;
1836 mpd->b_size = 1 << blkbits;
1837 mpd->b_state = b_state & BH_FLAGS;
1841 next = mpd->b_blocknr + nrblocks;
1843 * Can we merge the block to our big extent?
1845 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1846 mpd->b_size += 1 << blkbits;
1852 * We couldn't merge the block to our extent, so we
1853 * need to flush current extent and start new one
1855 mpage_da_map_and_submit(mpd);
1859 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1861 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1865 * This function is grabs code from the very beginning of
1866 * ext4_map_blocks, but assumes that the caller is from delayed write
1867 * time. This function looks up the requested blocks and sets the
1868 * buffer delay bit under the protection of i_data_sem.
1870 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1871 struct ext4_map_blocks *map,
1872 struct buffer_head *bh)
1874 struct extent_status es;
1876 sector_t invalid_block = ~((sector_t) 0xffff);
1877 #ifdef ES_AGGRESSIVE_TEST
1878 struct ext4_map_blocks orig_map;
1880 memcpy(&orig_map, map, sizeof(*map));
1883 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1887 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1888 "logical block %lu\n", inode->i_ino, map->m_len,
1889 (unsigned long) map->m_lblk);
1891 /* Lookup extent status tree firstly */
1892 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1894 if (ext4_es_is_hole(&es)) {
1896 down_read((&EXT4_I(inode)->i_data_sem));
1901 * Delayed extent could be allocated by fallocate.
1902 * So we need to check it.
1904 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1905 map_bh(bh, inode->i_sb, invalid_block);
1907 set_buffer_delay(bh);
1911 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1912 retval = es.es_len - (iblock - es.es_lblk);
1913 if (retval > map->m_len)
1914 retval = map->m_len;
1915 map->m_len = retval;
1916 if (ext4_es_is_written(&es))
1917 map->m_flags |= EXT4_MAP_MAPPED;
1918 else if (ext4_es_is_unwritten(&es))
1919 map->m_flags |= EXT4_MAP_UNWRITTEN;
1923 #ifdef ES_AGGRESSIVE_TEST
1924 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1930 * Try to see if we can get the block without requesting a new
1931 * file system block.
1933 down_read((&EXT4_I(inode)->i_data_sem));
1934 if (ext4_has_inline_data(inode)) {
1936 * We will soon create blocks for this page, and let
1937 * us pretend as if the blocks aren't allocated yet.
1938 * In case of clusters, we have to handle the work
1939 * of mapping from cluster so that the reserved space
1940 * is calculated properly.
1942 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1943 ext4_find_delalloc_cluster(inode, map->m_lblk))
1944 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1946 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1947 retval = ext4_ext_map_blocks(NULL, inode, map,
1948 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1950 retval = ext4_ind_map_blocks(NULL, inode, map,
1951 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1957 * XXX: __block_prepare_write() unmaps passed block,
1961 * If the block was allocated from previously allocated cluster,
1962 * then we don't need to reserve it again. However we still need
1963 * to reserve metadata for every block we're going to write.
1965 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1966 ret = ext4_da_reserve_space(inode, iblock);
1968 /* not enough space to reserve */
1973 ret = ext4_da_reserve_metadata(inode, iblock);
1975 /* not enough space to reserve */
1981 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1982 ~0, EXTENT_STATUS_DELAYED);
1988 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1989 * and it should not appear on the bh->b_state.
1991 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1993 map_bh(bh, inode->i_sb, invalid_block);
1995 set_buffer_delay(bh);
1996 } else if (retval > 0) {
1998 unsigned long long status;
2000 #ifdef ES_AGGRESSIVE_TEST
2001 if (retval != map->m_len) {
2002 printk("ES len assertation failed for inode: %lu "
2003 "retval %d != map->m_len %d "
2004 "in %s (lookup)\n", inode->i_ino, retval,
2005 map->m_len, __func__);
2009 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
2010 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
2011 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2012 map->m_pblk, status);
2018 up_read((&EXT4_I(inode)->i_data_sem));
2024 * This is a special get_blocks_t callback which is used by
2025 * ext4_da_write_begin(). It will either return mapped block or
2026 * reserve space for a single block.
2028 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2029 * We also have b_blocknr = -1 and b_bdev initialized properly
2031 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2032 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2033 * initialized properly.
2035 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2036 struct buffer_head *bh, int create)
2038 struct ext4_map_blocks map;
2041 BUG_ON(create == 0);
2042 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2044 map.m_lblk = iblock;
2048 * first, we need to know whether the block is allocated already
2049 * preallocated blocks are unmapped but should treated
2050 * the same as allocated blocks.
2052 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2056 map_bh(bh, inode->i_sb, map.m_pblk);
2057 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2059 if (buffer_unwritten(bh)) {
2060 /* A delayed write to unwritten bh should be marked
2061 * new and mapped. Mapped ensures that we don't do
2062 * get_block multiple times when we write to the same
2063 * offset and new ensures that we do proper zero out
2064 * for partial write.
2067 set_buffer_mapped(bh);
2072 static int bget_one(handle_t *handle, struct buffer_head *bh)
2078 static int bput_one(handle_t *handle, struct buffer_head *bh)
2084 static int __ext4_journalled_writepage(struct page *page,
2087 struct address_space *mapping = page->mapping;
2088 struct inode *inode = mapping->host;
2089 struct buffer_head *page_bufs = NULL;
2090 handle_t *handle = NULL;
2091 int ret = 0, err = 0;
2092 int inline_data = ext4_has_inline_data(inode);
2093 struct buffer_head *inode_bh = NULL;
2095 ClearPageChecked(page);
2098 BUG_ON(page->index != 0);
2099 BUG_ON(len > ext4_get_max_inline_size(inode));
2100 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2101 if (inode_bh == NULL)
2104 page_bufs = page_buffers(page);
2109 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2112 /* As soon as we unlock the page, it can go away, but we have
2113 * references to buffers so we are safe */
2116 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2117 ext4_writepage_trans_blocks(inode));
2118 if (IS_ERR(handle)) {
2119 ret = PTR_ERR(handle);
2123 BUG_ON(!ext4_handle_valid(handle));
2126 ret = ext4_journal_get_write_access(handle, inode_bh);
2128 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2131 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2132 do_journal_get_write_access);
2134 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2139 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2140 err = ext4_journal_stop(handle);
2144 if (!ext4_has_inline_data(inode))
2145 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2147 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2154 * Note that we don't need to start a transaction unless we're journaling data
2155 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2156 * need to file the inode to the transaction's list in ordered mode because if
2157 * we are writing back data added by write(), the inode is already there and if
2158 * we are writing back data modified via mmap(), no one guarantees in which
2159 * transaction the data will hit the disk. In case we are journaling data, we
2160 * cannot start transaction directly because transaction start ranks above page
2161 * lock so we have to do some magic.
2163 * This function can get called via...
2164 * - ext4_da_writepages after taking page lock (have journal handle)
2165 * - journal_submit_inode_data_buffers (no journal handle)
2166 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2167 * - grab_page_cache when doing write_begin (have journal handle)
2169 * We don't do any block allocation in this function. If we have page with
2170 * multiple blocks we need to write those buffer_heads that are mapped. This
2171 * is important for mmaped based write. So if we do with blocksize 1K
2172 * truncate(f, 1024);
2173 * a = mmap(f, 0, 4096);
2175 * truncate(f, 4096);
2176 * we have in the page first buffer_head mapped via page_mkwrite call back
2177 * but other buffer_heads would be unmapped but dirty (dirty done via the
2178 * do_wp_page). So writepage should write the first block. If we modify
2179 * the mmap area beyond 1024 we will again get a page_fault and the
2180 * page_mkwrite callback will do the block allocation and mark the
2181 * buffer_heads mapped.
2183 * We redirty the page if we have any buffer_heads that is either delay or
2184 * unwritten in the page.
2186 * We can get recursively called as show below.
2188 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2191 * But since we don't do any block allocation we should not deadlock.
2192 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2194 static int ext4_writepage(struct page *page,
2195 struct writeback_control *wbc)
2200 struct buffer_head *page_bufs = NULL;
2201 struct inode *inode = page->mapping->host;
2202 struct ext4_io_submit io_submit;
2204 trace_ext4_writepage(page);
2205 size = i_size_read(inode);
2206 if (page->index == size >> PAGE_CACHE_SHIFT)
2207 len = size & ~PAGE_CACHE_MASK;
2209 len = PAGE_CACHE_SIZE;
2211 page_bufs = page_buffers(page);
2213 * We cannot do block allocation or other extent handling in this
2214 * function. If there are buffers needing that, we have to redirty
2215 * the page. But we may reach here when we do a journal commit via
2216 * journal_submit_inode_data_buffers() and in that case we must write
2217 * allocated buffers to achieve data=ordered mode guarantees.
2219 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2220 ext4_bh_delay_or_unwritten)) {
2221 redirty_page_for_writepage(wbc, page);
2222 if (current->flags & PF_MEMALLOC) {
2224 * For memory cleaning there's no point in writing only
2225 * some buffers. So just bail out. Warn if we came here
2226 * from direct reclaim.
2228 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2235 if (PageChecked(page) && ext4_should_journal_data(inode))
2237 * It's mmapped pagecache. Add buffers and journal it. There
2238 * doesn't seem much point in redirtying the page here.
2240 return __ext4_journalled_writepage(page, len);
2242 ext4_io_submit_init(&io_submit, wbc);
2243 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2244 if (!io_submit.io_end) {
2245 redirty_page_for_writepage(wbc, page);
2248 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2249 ext4_io_submit(&io_submit);
2250 /* Drop io_end reference we got from init */
2251 ext4_put_io_end_defer(io_submit.io_end);
2256 * This is called via ext4_da_writepages() to
2257 * calculate the total number of credits to reserve to fit
2258 * a single extent allocation into a single transaction,
2259 * ext4_da_writpeages() will loop calling this before
2260 * the block allocation.
2263 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2265 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2268 * With non-extent format the journal credit needed to
2269 * insert nrblocks contiguous block is dependent on
2270 * number of contiguous block. So we will limit
2271 * number of contiguous block to a sane value
2273 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2274 (max_blocks > EXT4_MAX_TRANS_DATA))
2275 max_blocks = EXT4_MAX_TRANS_DATA;
2277 return ext4_chunk_trans_blocks(inode, max_blocks);
2281 * write_cache_pages_da - walk the list of dirty pages of the given
2282 * address space and accumulate pages that need writing, and call
2283 * mpage_da_map_and_submit to map a single contiguous memory region
2284 * and then write them.
2286 static int write_cache_pages_da(handle_t *handle,
2287 struct address_space *mapping,
2288 struct writeback_control *wbc,
2289 struct mpage_da_data *mpd,
2290 pgoff_t *done_index)
2292 struct buffer_head *bh, *head;
2293 struct inode *inode = mapping->host;
2294 struct pagevec pvec;
2295 unsigned int nr_pages;
2298 long nr_to_write = wbc->nr_to_write;
2299 int i, tag, ret = 0;
2301 memset(mpd, 0, sizeof(struct mpage_da_data));
2304 pagevec_init(&pvec, 0);
2305 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2306 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2308 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2309 tag = PAGECACHE_TAG_TOWRITE;
2311 tag = PAGECACHE_TAG_DIRTY;
2313 *done_index = index;
2314 while (index <= end) {
2315 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2316 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2320 for (i = 0; i < nr_pages; i++) {
2321 struct page *page = pvec.pages[i];
2324 * At this point, the page may be truncated or
2325 * invalidated (changing page->mapping to NULL), or
2326 * even swizzled back from swapper_space to tmpfs file
2327 * mapping. However, page->index will not change
2328 * because we have a reference on the page.
2330 if (page->index > end)
2333 *done_index = page->index + 1;
2336 * If we can't merge this page, and we have
2337 * accumulated an contiguous region, write it
2339 if ((mpd->next_page != page->index) &&
2340 (mpd->next_page != mpd->first_page)) {
2341 mpage_da_map_and_submit(mpd);
2342 goto ret_extent_tail;
2348 * If the page is no longer dirty, or its
2349 * mapping no longer corresponds to inode we
2350 * are writing (which means it has been
2351 * truncated or invalidated), or the page is
2352 * already under writeback and we are not
2353 * doing a data integrity writeback, skip the page
2355 if (!PageDirty(page) ||
2356 (PageWriteback(page) &&
2357 (wbc->sync_mode == WB_SYNC_NONE)) ||
2358 unlikely(page->mapping != mapping)) {
2363 wait_on_page_writeback(page);
2364 BUG_ON(PageWriteback(page));
2367 * If we have inline data and arrive here, it means that
2368 * we will soon create the block for the 1st page, so
2369 * we'd better clear the inline data here.
2371 if (ext4_has_inline_data(inode)) {
2372 BUG_ON(ext4_test_inode_state(inode,
2373 EXT4_STATE_MAY_INLINE_DATA));
2374 ext4_destroy_inline_data(handle, inode);
2377 if (mpd->next_page != page->index)
2378 mpd->first_page = page->index;
2379 mpd->next_page = page->index + 1;
2380 logical = (sector_t) page->index <<
2381 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2383 /* Add all dirty buffers to mpd */
2384 head = page_buffers(page);
2387 BUG_ON(buffer_locked(bh));
2389 * We need to try to allocate unmapped blocks
2390 * in the same page. Otherwise we won't make
2391 * progress with the page in ext4_writepage
2393 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2394 mpage_add_bh_to_extent(mpd, logical,
2397 goto ret_extent_tail;
2398 } else if (buffer_dirty(bh) &&
2399 buffer_mapped(bh)) {
2401 * mapped dirty buffer. We need to
2402 * update the b_state because we look
2403 * at b_state in mpage_da_map_blocks.
2404 * We don't update b_size because if we
2405 * find an unmapped buffer_head later
2406 * we need to use the b_state flag of
2409 if (mpd->b_size == 0)
2411 bh->b_state & BH_FLAGS;
2414 } while ((bh = bh->b_this_page) != head);
2416 if (nr_to_write > 0) {
2418 if (nr_to_write == 0 &&
2419 wbc->sync_mode == WB_SYNC_NONE)
2421 * We stop writing back only if we are
2422 * not doing integrity sync. In case of
2423 * integrity sync we have to keep going
2424 * because someone may be concurrently
2425 * dirtying pages, and we might have
2426 * synced a lot of newly appeared dirty
2427 * pages, but have not synced all of the
2433 pagevec_release(&pvec);
2438 ret = MPAGE_DA_EXTENT_TAIL;
2440 pagevec_release(&pvec);
2446 static int ext4_da_writepages(struct address_space *mapping,
2447 struct writeback_control *wbc)
2450 int range_whole = 0;
2451 handle_t *handle = NULL;
2452 struct mpage_da_data mpd;
2453 struct inode *inode = mapping->host;
2454 int pages_written = 0;
2455 unsigned int max_pages;
2456 int range_cyclic, cycled = 1, io_done = 0;
2457 int needed_blocks, ret = 0;
2458 long desired_nr_to_write, nr_to_writebump = 0;
2459 loff_t range_start = wbc->range_start;
2460 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2461 pgoff_t done_index = 0;
2463 struct blk_plug plug;
2465 trace_ext4_da_writepages(inode, wbc);
2468 * No pages to write? This is mainly a kludge to avoid starting
2469 * a transaction for special inodes like journal inode on last iput()
2470 * because that could violate lock ordering on umount
2472 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2476 * If the filesystem has aborted, it is read-only, so return
2477 * right away instead of dumping stack traces later on that
2478 * will obscure the real source of the problem. We test
2479 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2480 * the latter could be true if the filesystem is mounted
2481 * read-only, and in that case, ext4_da_writepages should
2482 * *never* be called, so if that ever happens, we would want
2485 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2488 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2491 range_cyclic = wbc->range_cyclic;
2492 if (wbc->range_cyclic) {
2493 index = mapping->writeback_index;
2496 wbc->range_start = index << PAGE_CACHE_SHIFT;
2497 wbc->range_end = LLONG_MAX;
2498 wbc->range_cyclic = 0;
2501 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2502 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2506 * This works around two forms of stupidity. The first is in
2507 * the writeback code, which caps the maximum number of pages
2508 * written to be 1024 pages. This is wrong on multiple
2509 * levels; different architectues have a different page size,
2510 * which changes the maximum amount of data which gets
2511 * written. Secondly, 4 megabytes is way too small. XFS
2512 * forces this value to be 16 megabytes by multiplying
2513 * nr_to_write parameter by four, and then relies on its
2514 * allocator to allocate larger extents to make them
2515 * contiguous. Unfortunately this brings us to the second
2516 * stupidity, which is that ext4's mballoc code only allocates
2517 * at most 2048 blocks. So we force contiguous writes up to
2518 * the number of dirty blocks in the inode, or
2519 * sbi->max_writeback_mb_bump whichever is smaller.
2521 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2522 if (!range_cyclic && range_whole) {
2523 if (wbc->nr_to_write == LONG_MAX)
2524 desired_nr_to_write = wbc->nr_to_write;
2526 desired_nr_to_write = wbc->nr_to_write * 8;
2528 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2530 if (desired_nr_to_write > max_pages)
2531 desired_nr_to_write = max_pages;
2533 if (wbc->nr_to_write < desired_nr_to_write) {
2534 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2535 wbc->nr_to_write = desired_nr_to_write;
2539 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2540 tag_pages_for_writeback(mapping, index, end);
2542 blk_start_plug(&plug);
2543 while (!ret && wbc->nr_to_write > 0) {
2546 * we insert one extent at a time. So we need
2547 * credit needed for single extent allocation.
2548 * journalled mode is currently not supported
2551 BUG_ON(ext4_should_journal_data(inode));
2552 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2554 /* start a new transaction*/
2555 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2557 if (IS_ERR(handle)) {
2558 ret = PTR_ERR(handle);
2559 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2560 "%ld pages, ino %lu; err %d", __func__,
2561 wbc->nr_to_write, inode->i_ino, ret);
2562 blk_finish_plug(&plug);
2563 goto out_writepages;
2567 * Now call write_cache_pages_da() to find the next
2568 * contiguous region of logical blocks that need
2569 * blocks to be allocated by ext4 and submit them.
2571 ret = write_cache_pages_da(handle, mapping,
2572 wbc, &mpd, &done_index);
2574 * If we have a contiguous extent of pages and we
2575 * haven't done the I/O yet, map the blocks and submit
2578 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2579 mpage_da_map_and_submit(&mpd);
2580 ret = MPAGE_DA_EXTENT_TAIL;
2582 trace_ext4_da_write_pages(inode, &mpd);
2583 wbc->nr_to_write -= mpd.pages_written;
2585 ext4_journal_stop(handle);
2587 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2588 /* commit the transaction which would
2589 * free blocks released in the transaction
2592 jbd2_journal_force_commit_nested(sbi->s_journal);
2594 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2596 * Got one extent now try with rest of the pages.
2597 * If mpd.retval is set -EIO, journal is aborted.
2598 * So we don't need to write any more.
2600 pages_written += mpd.pages_written;
2603 } else if (wbc->nr_to_write)
2605 * There is no more writeout needed
2606 * or we requested for a noblocking writeout
2607 * and we found the device congested
2611 blk_finish_plug(&plug);
2612 if (!io_done && !cycled) {
2615 wbc->range_start = index << PAGE_CACHE_SHIFT;
2616 wbc->range_end = mapping->writeback_index - 1;
2621 wbc->range_cyclic = range_cyclic;
2622 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2624 * set the writeback_index so that range_cyclic
2625 * mode will write it back later
2627 mapping->writeback_index = done_index;
2630 wbc->nr_to_write -= nr_to_writebump;
2631 wbc->range_start = range_start;
2632 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2636 static int ext4_nonda_switch(struct super_block *sb)
2638 s64 free_clusters, dirty_clusters;
2639 struct ext4_sb_info *sbi = EXT4_SB(sb);
2642 * switch to non delalloc mode if we are running low
2643 * on free block. The free block accounting via percpu
2644 * counters can get slightly wrong with percpu_counter_batch getting
2645 * accumulated on each CPU without updating global counters
2646 * Delalloc need an accurate free block accounting. So switch
2647 * to non delalloc when we are near to error range.
2650 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2652 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2654 * Start pushing delalloc when 1/2 of free blocks are dirty.
2656 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2657 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2659 if (2 * free_clusters < 3 * dirty_clusters ||
2660 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2662 * free block count is less than 150% of dirty blocks
2663 * or free blocks is less than watermark
2670 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2671 loff_t pos, unsigned len, unsigned flags,
2672 struct page **pagep, void **fsdata)
2674 int ret, retries = 0;
2677 struct inode *inode = mapping->host;
2680 index = pos >> PAGE_CACHE_SHIFT;
2682 if (ext4_nonda_switch(inode->i_sb)) {
2683 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2684 return ext4_write_begin(file, mapping, pos,
2685 len, flags, pagep, fsdata);
2687 *fsdata = (void *)0;
2688 trace_ext4_da_write_begin(inode, pos, len, flags);
2690 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2691 ret = ext4_da_write_inline_data_begin(mapping, inode,
2701 * grab_cache_page_write_begin() can take a long time if the
2702 * system is thrashing due to memory pressure, or if the page
2703 * is being written back. So grab it first before we start
2704 * the transaction handle. This also allows us to allocate
2705 * the page (if needed) without using GFP_NOFS.
2708 page = grab_cache_page_write_begin(mapping, index, flags);
2714 * With delayed allocation, we don't log the i_disksize update
2715 * if there is delayed block allocation. But we still need
2716 * to journalling the i_disksize update if writes to the end
2717 * of file which has an already mapped buffer.
2720 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2721 if (IS_ERR(handle)) {
2722 page_cache_release(page);
2723 return PTR_ERR(handle);
2727 if (page->mapping != mapping) {
2728 /* The page got truncated from under us */
2730 page_cache_release(page);
2731 ext4_journal_stop(handle);
2734 /* In case writeback began while the page was unlocked */
2735 wait_on_page_writeback(page);
2737 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2740 ext4_journal_stop(handle);
2742 * block_write_begin may have instantiated a few blocks
2743 * outside i_size. Trim these off again. Don't need
2744 * i_size_read because we hold i_mutex.
2746 if (pos + len > inode->i_size)
2747 ext4_truncate_failed_write(inode);
2749 if (ret == -ENOSPC &&
2750 ext4_should_retry_alloc(inode->i_sb, &retries))
2753 page_cache_release(page);
2762 * Check if we should update i_disksize
2763 * when write to the end of file but not require block allocation
2765 static int ext4_da_should_update_i_disksize(struct page *page,
2766 unsigned long offset)
2768 struct buffer_head *bh;
2769 struct inode *inode = page->mapping->host;
2773 bh = page_buffers(page);
2774 idx = offset >> inode->i_blkbits;
2776 for (i = 0; i < idx; i++)
2777 bh = bh->b_this_page;
2779 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2784 static int ext4_da_write_end(struct file *file,
2785 struct address_space *mapping,
2786 loff_t pos, unsigned len, unsigned copied,
2787 struct page *page, void *fsdata)
2789 struct inode *inode = mapping->host;
2791 handle_t *handle = ext4_journal_current_handle();
2793 unsigned long start, end;
2794 int write_mode = (int)(unsigned long)fsdata;
2796 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2797 return ext4_write_end(file, mapping, pos,
2798 len, copied, page, fsdata);
2800 trace_ext4_da_write_end(inode, pos, len, copied);
2801 start = pos & (PAGE_CACHE_SIZE - 1);
2802 end = start + copied - 1;
2805 * generic_write_end() will run mark_inode_dirty() if i_size
2806 * changes. So let's piggyback the i_disksize mark_inode_dirty
2809 new_i_size = pos + copied;
2810 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2811 if (ext4_has_inline_data(inode) ||
2812 ext4_da_should_update_i_disksize(page, end)) {
2813 down_write(&EXT4_I(inode)->i_data_sem);
2814 if (new_i_size > EXT4_I(inode)->i_disksize)
2815 EXT4_I(inode)->i_disksize = new_i_size;
2816 up_write(&EXT4_I(inode)->i_data_sem);
2817 /* We need to mark inode dirty even if
2818 * new_i_size is less that inode->i_size
2819 * bu greater than i_disksize.(hint delalloc)
2821 ext4_mark_inode_dirty(handle, inode);
2825 if (write_mode != CONVERT_INLINE_DATA &&
2826 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2827 ext4_has_inline_data(inode))
2828 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2831 ret2 = generic_write_end(file, mapping, pos, len, copied,
2837 ret2 = ext4_journal_stop(handle);
2841 return ret ? ret : copied;
2844 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2847 * Drop reserved blocks
2849 BUG_ON(!PageLocked(page));
2850 if (!page_has_buffers(page))
2853 ext4_da_page_release_reservation(page, offset);
2856 ext4_invalidatepage(page, offset);
2862 * Force all delayed allocation blocks to be allocated for a given inode.
2864 int ext4_alloc_da_blocks(struct inode *inode)
2866 trace_ext4_alloc_da_blocks(inode);
2868 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2869 !EXT4_I(inode)->i_reserved_meta_blocks)
2873 * We do something simple for now. The filemap_flush() will
2874 * also start triggering a write of the data blocks, which is
2875 * not strictly speaking necessary (and for users of
2876 * laptop_mode, not even desirable). However, to do otherwise
2877 * would require replicating code paths in:
2879 * ext4_da_writepages() ->
2880 * write_cache_pages() ---> (via passed in callback function)
2881 * __mpage_da_writepage() -->
2882 * mpage_add_bh_to_extent()
2883 * mpage_da_map_blocks()
2885 * The problem is that write_cache_pages(), located in
2886 * mm/page-writeback.c, marks pages clean in preparation for
2887 * doing I/O, which is not desirable if we're not planning on
2890 * We could call write_cache_pages(), and then redirty all of
2891 * the pages by calling redirty_page_for_writepage() but that
2892 * would be ugly in the extreme. So instead we would need to
2893 * replicate parts of the code in the above functions,
2894 * simplifying them because we wouldn't actually intend to
2895 * write out the pages, but rather only collect contiguous
2896 * logical block extents, call the multi-block allocator, and
2897 * then update the buffer heads with the block allocations.
2899 * For now, though, we'll cheat by calling filemap_flush(),
2900 * which will map the blocks, and start the I/O, but not
2901 * actually wait for the I/O to complete.
2903 return filemap_flush(inode->i_mapping);
2907 * bmap() is special. It gets used by applications such as lilo and by
2908 * the swapper to find the on-disk block of a specific piece of data.
2910 * Naturally, this is dangerous if the block concerned is still in the
2911 * journal. If somebody makes a swapfile on an ext4 data-journaling
2912 * filesystem and enables swap, then they may get a nasty shock when the
2913 * data getting swapped to that swapfile suddenly gets overwritten by
2914 * the original zero's written out previously to the journal and
2915 * awaiting writeback in the kernel's buffer cache.
2917 * So, if we see any bmap calls here on a modified, data-journaled file,
2918 * take extra steps to flush any blocks which might be in the cache.
2920 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2922 struct inode *inode = mapping->host;
2927 * We can get here for an inline file via the FIBMAP ioctl
2929 if (ext4_has_inline_data(inode))
2932 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2933 test_opt(inode->i_sb, DELALLOC)) {
2935 * With delalloc we want to sync the file
2936 * so that we can make sure we allocate
2939 filemap_write_and_wait(mapping);
2942 if (EXT4_JOURNAL(inode) &&
2943 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2945 * This is a REALLY heavyweight approach, but the use of
2946 * bmap on dirty files is expected to be extremely rare:
2947 * only if we run lilo or swapon on a freshly made file
2948 * do we expect this to happen.
2950 * (bmap requires CAP_SYS_RAWIO so this does not
2951 * represent an unprivileged user DOS attack --- we'd be
2952 * in trouble if mortal users could trigger this path at
2955 * NB. EXT4_STATE_JDATA is not set on files other than
2956 * regular files. If somebody wants to bmap a directory
2957 * or symlink and gets confused because the buffer
2958 * hasn't yet been flushed to disk, they deserve
2959 * everything they get.
2962 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2963 journal = EXT4_JOURNAL(inode);
2964 jbd2_journal_lock_updates(journal);
2965 err = jbd2_journal_flush(journal);
2966 jbd2_journal_unlock_updates(journal);
2972 return generic_block_bmap(mapping, block, ext4_get_block);
2975 static int ext4_readpage(struct file *file, struct page *page)
2978 struct inode *inode = page->mapping->host;
2980 trace_ext4_readpage(page);
2982 if (ext4_has_inline_data(inode))
2983 ret = ext4_readpage_inline(inode, page);
2986 return mpage_readpage(page, ext4_get_block);
2992 ext4_readpages(struct file *file, struct address_space *mapping,
2993 struct list_head *pages, unsigned nr_pages)
2995 struct inode *inode = mapping->host;
2997 /* If the file has inline data, no need to do readpages. */
2998 if (ext4_has_inline_data(inode))
3001 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3004 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3006 trace_ext4_invalidatepage(page, offset);
3008 /* No journalling happens on data buffers when this function is used */
3009 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3011 block_invalidatepage(page, offset);
3014 static int __ext4_journalled_invalidatepage(struct page *page,
3015 unsigned long offset)
3017 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3019 trace_ext4_journalled_invalidatepage(page, offset);
3022 * If it's a full truncate we just forget about the pending dirtying
3025 ClearPageChecked(page);
3027 return jbd2_journal_invalidatepage(journal, page, offset);
3030 /* Wrapper for aops... */
3031 static void ext4_journalled_invalidatepage(struct page *page,
3032 unsigned long offset)
3034 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3037 static int ext4_releasepage(struct page *page, gfp_t wait)
3039 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3041 trace_ext4_releasepage(page);
3043 /* Page has dirty journalled data -> cannot release */
3044 if (PageChecked(page))
3047 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3049 return try_to_free_buffers(page);
3053 * ext4_get_block used when preparing for a DIO write or buffer write.
3054 * We allocate an uinitialized extent if blocks haven't been allocated.
3055 * The extent will be converted to initialized after the IO is complete.
3057 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3058 struct buffer_head *bh_result, int create)
3060 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3061 inode->i_ino, create);
3062 return _ext4_get_block(inode, iblock, bh_result,
3063 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3066 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3067 struct buffer_head *bh_result, int create)
3069 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3070 inode->i_ino, create);
3071 return _ext4_get_block(inode, iblock, bh_result,
3072 EXT4_GET_BLOCKS_NO_LOCK);
3075 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3076 ssize_t size, void *private, int ret,
3079 struct inode *inode = file_inode(iocb->ki_filp);
3080 ext4_io_end_t *io_end = iocb->private;
3082 /* if not async direct IO just return */
3084 inode_dio_done(inode);
3086 aio_complete(iocb, ret, 0);
3090 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3091 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3092 iocb->private, io_end->inode->i_ino, iocb, offset,
3095 iocb->private = NULL;
3096 io_end->offset = offset;
3097 io_end->size = size;
3099 io_end->iocb = iocb;
3100 io_end->result = ret;
3102 ext4_put_io_end_defer(io_end);
3106 * For ext4 extent files, ext4 will do direct-io write to holes,
3107 * preallocated extents, and those write extend the file, no need to
3108 * fall back to buffered IO.
3110 * For holes, we fallocate those blocks, mark them as uninitialized
3111 * If those blocks were preallocated, we mark sure they are split, but
3112 * still keep the range to write as uninitialized.
3114 * The unwritten extents will be converted to written when DIO is completed.
3115 * For async direct IO, since the IO may still pending when return, we
3116 * set up an end_io call back function, which will do the conversion
3117 * when async direct IO completed.
3119 * If the O_DIRECT write will extend the file then add this inode to the
3120 * orphan list. So recovery will truncate it back to the original size
3121 * if the machine crashes during the write.
3124 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3125 const struct iovec *iov, loff_t offset,
3126 unsigned long nr_segs)
3128 struct file *file = iocb->ki_filp;
3129 struct inode *inode = file->f_mapping->host;
3131 size_t count = iov_length(iov, nr_segs);
3133 get_block_t *get_block_func = NULL;
3135 loff_t final_size = offset + count;
3136 ext4_io_end_t *io_end = NULL;
3138 /* Use the old path for reads and writes beyond i_size. */
3139 if (rw != WRITE || final_size > inode->i_size)
3140 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3142 BUG_ON(iocb->private == NULL);
3144 /* If we do a overwrite dio, i_mutex locking can be released */
3145 overwrite = *((int *)iocb->private);
3148 atomic_inc(&inode->i_dio_count);
3149 down_read(&EXT4_I(inode)->i_data_sem);
3150 mutex_unlock(&inode->i_mutex);
3154 * We could direct write to holes and fallocate.
3156 * Allocated blocks to fill the hole are marked as
3157 * uninitialized to prevent parallel buffered read to expose
3158 * the stale data before DIO complete the data IO.
3160 * As to previously fallocated extents, ext4 get_block will
3161 * just simply mark the buffer mapped but still keep the
3162 * extents uninitialized.
3164 * For non AIO case, we will convert those unwritten extents
3165 * to written after return back from blockdev_direct_IO.
3167 * For async DIO, the conversion needs to be deferred when the
3168 * IO is completed. The ext4 end_io callback function will be
3169 * called to take care of the conversion work. Here for async
3170 * case, we allocate an io_end structure to hook to the iocb.
3172 iocb->private = NULL;
3173 ext4_inode_aio_set(inode, NULL);
3174 if (!is_sync_kiocb(iocb)) {
3175 io_end = ext4_init_io_end(inode, GFP_NOFS);
3180 io_end->flag |= EXT4_IO_END_DIRECT;
3182 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3184 iocb->private = ext4_get_io_end(io_end);
3186 * we save the io structure for current async direct
3187 * IO, so that later ext4_map_blocks() could flag the
3188 * io structure whether there is a unwritten extents
3189 * needs to be converted when IO is completed.
3191 ext4_inode_aio_set(inode, io_end);
3195 get_block_func = ext4_get_block_write_nolock;
3197 get_block_func = ext4_get_block_write;
3198 dio_flags = DIO_LOCKING;
3200 ret = __blockdev_direct_IO(rw, iocb, inode,
3201 inode->i_sb->s_bdev, iov,
3209 * Put our reference to io_end. This can free the io_end structure e.g.
3210 * in sync IO case or in case of error. It can even perform extent
3211 * conversion if all bios we submitted finished before we got here.
3212 * Note that in that case iocb->private can be already set to NULL
3216 ext4_inode_aio_set(inode, NULL);
3217 ext4_put_io_end(io_end);
3219 * In case of error or no write ext4_end_io_dio() was not
3220 * called so we have to put iocb's reference.
3222 if (ret <= 0 && ret != -EIOCBQUEUED) {
3223 WARN_ON(iocb->private != io_end);
3224 ext4_put_io_end(io_end);
3225 iocb->private = NULL;
3228 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3229 EXT4_STATE_DIO_UNWRITTEN)) {
3232 * for non AIO case, since the IO is already
3233 * completed, we could do the conversion right here
3235 err = ext4_convert_unwritten_extents(inode,
3239 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3243 /* take i_mutex locking again if we do a ovewrite dio */
3245 inode_dio_done(inode);
3246 up_read(&EXT4_I(inode)->i_data_sem);
3247 mutex_lock(&inode->i_mutex);
3253 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3254 const struct iovec *iov, loff_t offset,
3255 unsigned long nr_segs)
3257 struct file *file = iocb->ki_filp;
3258 struct inode *inode = file->f_mapping->host;
3262 * If we are doing data journalling we don't support O_DIRECT
3264 if (ext4_should_journal_data(inode))
3267 /* Let buffer I/O handle the inline data case. */
3268 if (ext4_has_inline_data(inode))
3271 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3272 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3273 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3275 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3276 trace_ext4_direct_IO_exit(inode, offset,
3277 iov_length(iov, nr_segs), rw, ret);
3282 * Pages can be marked dirty completely asynchronously from ext4's journalling
3283 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3284 * much here because ->set_page_dirty is called under VFS locks. The page is
3285 * not necessarily locked.
3287 * We cannot just dirty the page and leave attached buffers clean, because the
3288 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3289 * or jbddirty because all the journalling code will explode.
3291 * So what we do is to mark the page "pending dirty" and next time writepage
3292 * is called, propagate that into the buffers appropriately.
3294 static int ext4_journalled_set_page_dirty(struct page *page)
3296 SetPageChecked(page);
3297 return __set_page_dirty_nobuffers(page);
3300 static const struct address_space_operations ext4_aops = {
3301 .readpage = ext4_readpage,
3302 .readpages = ext4_readpages,
3303 .writepage = ext4_writepage,
3304 .write_begin = ext4_write_begin,
3305 .write_end = ext4_write_end,
3307 .invalidatepage = ext4_invalidatepage,
3308 .releasepage = ext4_releasepage,
3309 .direct_IO = ext4_direct_IO,
3310 .migratepage = buffer_migrate_page,
3311 .is_partially_uptodate = block_is_partially_uptodate,
3312 .error_remove_page = generic_error_remove_page,
3315 static const struct address_space_operations ext4_journalled_aops = {
3316 .readpage = ext4_readpage,
3317 .readpages = ext4_readpages,
3318 .writepage = ext4_writepage,
3319 .write_begin = ext4_write_begin,
3320 .write_end = ext4_journalled_write_end,
3321 .set_page_dirty = ext4_journalled_set_page_dirty,
3323 .invalidatepage = ext4_journalled_invalidatepage,
3324 .releasepage = ext4_releasepage,
3325 .direct_IO = ext4_direct_IO,
3326 .is_partially_uptodate = block_is_partially_uptodate,
3327 .error_remove_page = generic_error_remove_page,
3330 static const struct address_space_operations ext4_da_aops = {
3331 .readpage = ext4_readpage,
3332 .readpages = ext4_readpages,
3333 .writepage = ext4_writepage,
3334 .writepages = ext4_da_writepages,
3335 .write_begin = ext4_da_write_begin,
3336 .write_end = ext4_da_write_end,
3338 .invalidatepage = ext4_da_invalidatepage,
3339 .releasepage = ext4_releasepage,
3340 .direct_IO = ext4_direct_IO,
3341 .migratepage = buffer_migrate_page,
3342 .is_partially_uptodate = block_is_partially_uptodate,
3343 .error_remove_page = generic_error_remove_page,
3346 void ext4_set_aops(struct inode *inode)
3348 switch (ext4_inode_journal_mode(inode)) {
3349 case EXT4_INODE_ORDERED_DATA_MODE:
3350 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3352 case EXT4_INODE_WRITEBACK_DATA_MODE:
3353 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3355 case EXT4_INODE_JOURNAL_DATA_MODE:
3356 inode->i_mapping->a_ops = &ext4_journalled_aops;
3361 if (test_opt(inode->i_sb, DELALLOC))
3362 inode->i_mapping->a_ops = &ext4_da_aops;
3364 inode->i_mapping->a_ops = &ext4_aops;
3369 * ext4_discard_partial_page_buffers()
3370 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3371 * This function finds and locks the page containing the offset
3372 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3373 * Calling functions that already have the page locked should call
3374 * ext4_discard_partial_page_buffers_no_lock directly.
3376 int ext4_discard_partial_page_buffers(handle_t *handle,
3377 struct address_space *mapping, loff_t from,
3378 loff_t length, int flags)
3380 struct inode *inode = mapping->host;
3384 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3385 mapping_gfp_mask(mapping) & ~__GFP_FS);
3389 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3390 from, length, flags);
3393 page_cache_release(page);
3398 * ext4_discard_partial_page_buffers_no_lock()
3399 * Zeros a page range of length 'length' starting from offset 'from'.
3400 * Buffer heads that correspond to the block aligned regions of the
3401 * zeroed range will be unmapped. Unblock aligned regions
3402 * will have the corresponding buffer head mapped if needed so that
3403 * that region of the page can be updated with the partial zero out.
3405 * This function assumes that the page has already been locked. The
3406 * The range to be discarded must be contained with in the given page.
3407 * If the specified range exceeds the end of the page it will be shortened
3408 * to the end of the page that corresponds to 'from'. This function is
3409 * appropriate for updating a page and it buffer heads to be unmapped and
3410 * zeroed for blocks that have been either released, or are going to be
3413 * handle: The journal handle
3414 * inode: The files inode
3415 * page: A locked page that contains the offset "from"
3416 * from: The starting byte offset (from the beginning of the file)
3417 * to begin discarding
3418 * len: The length of bytes to discard
3419 * flags: Optional flags that may be used:
3421 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3422 * Only zero the regions of the page whose buffer heads
3423 * have already been unmapped. This flag is appropriate
3424 * for updating the contents of a page whose blocks may
3425 * have already been released, and we only want to zero
3426 * out the regions that correspond to those released blocks.
3428 * Returns zero on success or negative on failure.
3430 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3431 struct inode *inode, struct page *page, loff_t from,
3432 loff_t length, int flags)
3434 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3435 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3436 unsigned int blocksize, max, pos;
3438 struct buffer_head *bh;
3441 blocksize = inode->i_sb->s_blocksize;
3442 max = PAGE_CACHE_SIZE - offset;
3444 if (index != page->index)
3448 * correct length if it does not fall between
3449 * 'from' and the end of the page
3451 if (length > max || length < 0)
3454 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3456 if (!page_has_buffers(page))
3457 create_empty_buffers(page, blocksize, 0);
3459 /* Find the buffer that contains "offset" */
3460 bh = page_buffers(page);
3462 while (offset >= pos) {
3463 bh = bh->b_this_page;
3469 while (pos < offset + length) {
3470 unsigned int end_of_block, range_to_discard;
3474 /* The length of space left to zero and unmap */
3475 range_to_discard = offset + length - pos;
3477 /* The length of space until the end of the block */
3478 end_of_block = blocksize - (pos & (blocksize-1));
3481 * Do not unmap or zero past end of block
3482 * for this buffer head
3484 if (range_to_discard > end_of_block)
3485 range_to_discard = end_of_block;
3489 * Skip this buffer head if we are only zeroing unampped
3490 * regions of the page
3492 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3496 /* If the range is block aligned, unmap */
3497 if (range_to_discard == blocksize) {
3498 clear_buffer_dirty(bh);
3500 clear_buffer_mapped(bh);
3501 clear_buffer_req(bh);
3502 clear_buffer_new(bh);
3503 clear_buffer_delay(bh);
3504 clear_buffer_unwritten(bh);
3505 clear_buffer_uptodate(bh);
3506 zero_user(page, pos, range_to_discard);
3507 BUFFER_TRACE(bh, "Buffer discarded");
3512 * If this block is not completely contained in the range
3513 * to be discarded, then it is not going to be released. Because
3514 * we need to keep this block, we need to make sure this part
3515 * of the page is uptodate before we modify it by writeing
3516 * partial zeros on it.
3518 if (!buffer_mapped(bh)) {
3520 * Buffer head must be mapped before we can read
3523 BUFFER_TRACE(bh, "unmapped");
3524 ext4_get_block(inode, iblock, bh, 0);
3525 /* unmapped? It's a hole - nothing to do */
3526 if (!buffer_mapped(bh)) {
3527 BUFFER_TRACE(bh, "still unmapped");
3532 /* Ok, it's mapped. Make sure it's up-to-date */
3533 if (PageUptodate(page))
3534 set_buffer_uptodate(bh);
3536 if (!buffer_uptodate(bh)) {
3538 ll_rw_block(READ, 1, &bh);
3540 /* Uhhuh. Read error. Complain and punt.*/
3541 if (!buffer_uptodate(bh))
3545 if (ext4_should_journal_data(inode)) {
3546 BUFFER_TRACE(bh, "get write access");
3547 err = ext4_journal_get_write_access(handle, bh);
3552 zero_user(page, pos, range_to_discard);
3555 if (ext4_should_journal_data(inode)) {
3556 err = ext4_handle_dirty_metadata(handle, inode, bh);
3558 mark_buffer_dirty(bh);
3560 BUFFER_TRACE(bh, "Partial buffer zeroed");
3562 bh = bh->b_this_page;
3564 pos += range_to_discard;
3570 int ext4_can_truncate(struct inode *inode)
3572 if (S_ISREG(inode->i_mode))
3574 if (S_ISDIR(inode->i_mode))
3576 if (S_ISLNK(inode->i_mode))
3577 return !ext4_inode_is_fast_symlink(inode);
3582 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3583 * associated with the given offset and length
3585 * @inode: File inode
3586 * @offset: The offset where the hole will begin
3587 * @len: The length of the hole
3589 * Returns: 0 on success or negative on failure
3592 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3594 struct inode *inode = file_inode(file);
3595 struct super_block *sb = inode->i_sb;
3596 ext4_lblk_t first_block, stop_block;
3597 struct address_space *mapping = inode->i_mapping;
3598 loff_t first_page, last_page, page_len;
3599 loff_t first_page_offset, last_page_offset;
3601 unsigned int credits;
3604 if (!S_ISREG(inode->i_mode))
3607 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3608 /* TODO: Add support for bigalloc file systems */
3612 trace_ext4_punch_hole(inode, offset, length);
3615 * Write out all dirty pages to avoid race conditions
3616 * Then release them.
3618 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3619 ret = filemap_write_and_wait_range(mapping, offset,
3620 offset + length - 1);
3625 mutex_lock(&inode->i_mutex);
3626 /* It's not possible punch hole on append only file */
3627 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3631 if (IS_SWAPFILE(inode)) {
3636 /* No need to punch hole beyond i_size */
3637 if (offset >= inode->i_size)
3641 * If the hole extends beyond i_size, set the hole
3642 * to end after the page that contains i_size
3644 if (offset + length > inode->i_size) {
3645 length = inode->i_size +
3646 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3650 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
3651 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
3653 first_page_offset = first_page << PAGE_CACHE_SHIFT;
3654 last_page_offset = last_page << PAGE_CACHE_SHIFT;
3656 /* Now release the pages */
3657 if (last_page_offset > first_page_offset) {
3658 truncate_pagecache_range(inode, first_page_offset,
3659 last_page_offset - 1);
3662 /* Wait all existing dio workers, newcomers will block on i_mutex */
3663 ext4_inode_block_unlocked_dio(inode);
3664 ret = ext4_flush_unwritten_io(inode);
3667 inode_dio_wait(inode);
3669 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3670 credits = ext4_writepage_trans_blocks(inode);
3672 credits = ext4_blocks_for_truncate(inode);
3673 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3674 if (IS_ERR(handle)) {
3675 ret = PTR_ERR(handle);
3676 ext4_std_error(sb, ret);
3681 * Now we need to zero out the non-page-aligned data in the
3682 * pages at the start and tail of the hole, and unmap the
3683 * buffer heads for the block aligned regions of the page that
3684 * were completely zeroed.
3686 if (first_page > last_page) {
3688 * If the file space being truncated is contained
3689 * within a page just zero out and unmap the middle of
3692 ret = ext4_discard_partial_page_buffers(handle,
3693 mapping, offset, length, 0);
3699 * zero out and unmap the partial page that contains
3700 * the start of the hole
3702 page_len = first_page_offset - offset;
3704 ret = ext4_discard_partial_page_buffers(handle, mapping,
3705 offset, page_len, 0);
3711 * zero out and unmap the partial page that contains
3712 * the end of the hole
3714 page_len = offset + length - last_page_offset;
3716 ret = ext4_discard_partial_page_buffers(handle, mapping,
3717 last_page_offset, page_len, 0);
3724 * If i_size is contained in the last page, we need to
3725 * unmap and zero the partial page after i_size
3727 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
3728 inode->i_size % PAGE_CACHE_SIZE != 0) {
3729 page_len = PAGE_CACHE_SIZE -
3730 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3733 ret = ext4_discard_partial_page_buffers(handle,
3734 mapping, inode->i_size, page_len, 0);
3741 first_block = (offset + sb->s_blocksize - 1) >>
3742 EXT4_BLOCK_SIZE_BITS(sb);
3743 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3745 /* If there are no blocks to remove, return now */
3746 if (first_block >= stop_block)
3749 down_write(&EXT4_I(inode)->i_data_sem);
3750 ext4_discard_preallocations(inode);
3752 ret = ext4_es_remove_extent(inode, first_block,
3753 stop_block - first_block);
3755 up_write(&EXT4_I(inode)->i_data_sem);
3759 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3760 ret = ext4_ext_remove_space(inode, first_block,
3763 ret = ext4_free_hole_blocks(handle, inode, first_block,
3766 ext4_discard_preallocations(inode);
3767 up_write(&EXT4_I(inode)->i_data_sem);
3769 ext4_handle_sync(handle);
3770 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3771 ext4_mark_inode_dirty(handle, inode);
3773 ext4_journal_stop(handle);
3775 ext4_inode_resume_unlocked_dio(inode);
3777 mutex_unlock(&inode->i_mutex);
3784 * We block out ext4_get_block() block instantiations across the entire
3785 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3786 * simultaneously on behalf of the same inode.
3788 * As we work through the truncate and commit bits of it to the journal there
3789 * is one core, guiding principle: the file's tree must always be consistent on
3790 * disk. We must be able to restart the truncate after a crash.
3792 * The file's tree may be transiently inconsistent in memory (although it
3793 * probably isn't), but whenever we close off and commit a journal transaction,
3794 * the contents of (the filesystem + the journal) must be consistent and
3795 * restartable. It's pretty simple, really: bottom up, right to left (although
3796 * left-to-right works OK too).
3798 * Note that at recovery time, journal replay occurs *before* the restart of
3799 * truncate against the orphan inode list.
3801 * The committed inode has the new, desired i_size (which is the same as
3802 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3803 * that this inode's truncate did not complete and it will again call
3804 * ext4_truncate() to have another go. So there will be instantiated blocks
3805 * to the right of the truncation point in a crashed ext4 filesystem. But
3806 * that's fine - as long as they are linked from the inode, the post-crash
3807 * ext4_truncate() run will find them and release them.
3809 void ext4_truncate(struct inode *inode)
3811 struct ext4_inode_info *ei = EXT4_I(inode);
3812 unsigned int credits;
3814 struct address_space *mapping = inode->i_mapping;
3818 * There is a possibility that we're either freeing the inode
3819 * or it completely new indode. In those cases we might not
3820 * have i_mutex locked because it's not necessary.
3822 if (!(inode->i_state & (I_NEW|I_FREEING)))
3823 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3824 trace_ext4_truncate_enter(inode);
3826 if (!ext4_can_truncate(inode))
3829 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3831 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3832 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3834 if (ext4_has_inline_data(inode)) {
3837 ext4_inline_data_truncate(inode, &has_inline);
3843 * finish any pending end_io work so we won't run the risk of
3844 * converting any truncated blocks to initialized later
3846 ext4_flush_unwritten_io(inode);
3848 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3849 credits = ext4_writepage_trans_blocks(inode);
3851 credits = ext4_blocks_for_truncate(inode);
3853 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3854 if (IS_ERR(handle)) {
3855 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3859 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3860 page_len = PAGE_CACHE_SIZE -
3861 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3863 if (ext4_discard_partial_page_buffers(handle,
3864 mapping, inode->i_size, page_len, 0))
3869 * We add the inode to the orphan list, so that if this
3870 * truncate spans multiple transactions, and we crash, we will
3871 * resume the truncate when the filesystem recovers. It also
3872 * marks the inode dirty, to catch the new size.
3874 * Implication: the file must always be in a sane, consistent
3875 * truncatable state while each transaction commits.
3877 if (ext4_orphan_add(handle, inode))
3880 down_write(&EXT4_I(inode)->i_data_sem);
3882 ext4_discard_preallocations(inode);
3884 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3885 ext4_ext_truncate(handle, inode);
3887 ext4_ind_truncate(handle, inode);
3889 up_write(&ei->i_data_sem);
3892 ext4_handle_sync(handle);
3896 * If this was a simple ftruncate() and the file will remain alive,
3897 * then we need to clear up the orphan record which we created above.
3898 * However, if this was a real unlink then we were called by
3899 * ext4_delete_inode(), and we allow that function to clean up the
3900 * orphan info for us.
3903 ext4_orphan_del(handle, inode);
3905 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3906 ext4_mark_inode_dirty(handle, inode);
3907 ext4_journal_stop(handle);
3909 trace_ext4_truncate_exit(inode);
3913 * ext4_get_inode_loc returns with an extra refcount against the inode's
3914 * underlying buffer_head on success. If 'in_mem' is true, we have all
3915 * data in memory that is needed to recreate the on-disk version of this
3918 static int __ext4_get_inode_loc(struct inode *inode,
3919 struct ext4_iloc *iloc, int in_mem)
3921 struct ext4_group_desc *gdp;
3922 struct buffer_head *bh;
3923 struct super_block *sb = inode->i_sb;
3925 int inodes_per_block, inode_offset;
3928 if (!ext4_valid_inum(sb, inode->i_ino))
3931 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3932 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3937 * Figure out the offset within the block group inode table
3939 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3940 inode_offset = ((inode->i_ino - 1) %
3941 EXT4_INODES_PER_GROUP(sb));
3942 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3943 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3945 bh = sb_getblk(sb, block);
3948 if (!buffer_uptodate(bh)) {
3952 * If the buffer has the write error flag, we have failed
3953 * to write out another inode in the same block. In this
3954 * case, we don't have to read the block because we may
3955 * read the old inode data successfully.
3957 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3958 set_buffer_uptodate(bh);
3960 if (buffer_uptodate(bh)) {
3961 /* someone brought it uptodate while we waited */
3967 * If we have all information of the inode in memory and this
3968 * is the only valid inode in the block, we need not read the
3972 struct buffer_head *bitmap_bh;
3975 start = inode_offset & ~(inodes_per_block - 1);
3977 /* Is the inode bitmap in cache? */
3978 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3979 if (unlikely(!bitmap_bh))
3983 * If the inode bitmap isn't in cache then the
3984 * optimisation may end up performing two reads instead
3985 * of one, so skip it.
3987 if (!buffer_uptodate(bitmap_bh)) {
3991 for (i = start; i < start + inodes_per_block; i++) {
3992 if (i == inode_offset)
3994 if (ext4_test_bit(i, bitmap_bh->b_data))
3998 if (i == start + inodes_per_block) {
3999 /* all other inodes are free, so skip I/O */
4000 memset(bh->b_data, 0, bh->b_size);
4001 set_buffer_uptodate(bh);
4009 * If we need to do any I/O, try to pre-readahead extra
4010 * blocks from the inode table.
4012 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4013 ext4_fsblk_t b, end, table;
4015 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4017 table = ext4_inode_table(sb, gdp);
4018 /* s_inode_readahead_blks is always a power of 2 */
4019 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4023 num = EXT4_INODES_PER_GROUP(sb);
4024 if (ext4_has_group_desc_csum(sb))
4025 num -= ext4_itable_unused_count(sb, gdp);
4026 table += num / inodes_per_block;
4030 sb_breadahead(sb, b++);
4034 * There are other valid inodes in the buffer, this inode
4035 * has in-inode xattrs, or we don't have this inode in memory.
4036 * Read the block from disk.
4038 trace_ext4_load_inode(inode);
4040 bh->b_end_io = end_buffer_read_sync;
4041 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4043 if (!buffer_uptodate(bh)) {
4044 EXT4_ERROR_INODE_BLOCK(inode, block,
4045 "unable to read itable block");
4055 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4057 /* We have all inode data except xattrs in memory here. */
4058 return __ext4_get_inode_loc(inode, iloc,
4059 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4062 void ext4_set_inode_flags(struct inode *inode)
4064 unsigned int flags = EXT4_I(inode)->i_flags;
4066 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4067 if (flags & EXT4_SYNC_FL)
4068 inode->i_flags |= S_SYNC;
4069 if (flags & EXT4_APPEND_FL)
4070 inode->i_flags |= S_APPEND;
4071 if (flags & EXT4_IMMUTABLE_FL)
4072 inode->i_flags |= S_IMMUTABLE;
4073 if (flags & EXT4_NOATIME_FL)
4074 inode->i_flags |= S_NOATIME;
4075 if (flags & EXT4_DIRSYNC_FL)
4076 inode->i_flags |= S_DIRSYNC;
4079 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4080 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4082 unsigned int vfs_fl;
4083 unsigned long old_fl, new_fl;
4086 vfs_fl = ei->vfs_inode.i_flags;
4087 old_fl = ei->i_flags;
4088 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4089 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4091 if (vfs_fl & S_SYNC)
4092 new_fl |= EXT4_SYNC_FL;
4093 if (vfs_fl & S_APPEND)
4094 new_fl |= EXT4_APPEND_FL;
4095 if (vfs_fl & S_IMMUTABLE)
4096 new_fl |= EXT4_IMMUTABLE_FL;
4097 if (vfs_fl & S_NOATIME)
4098 new_fl |= EXT4_NOATIME_FL;
4099 if (vfs_fl & S_DIRSYNC)
4100 new_fl |= EXT4_DIRSYNC_FL;
4101 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4104 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4105 struct ext4_inode_info *ei)
4108 struct inode *inode = &(ei->vfs_inode);
4109 struct super_block *sb = inode->i_sb;
4111 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4112 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4113 /* we are using combined 48 bit field */
4114 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4115 le32_to_cpu(raw_inode->i_blocks_lo);
4116 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4117 /* i_blocks represent file system block size */
4118 return i_blocks << (inode->i_blkbits - 9);
4123 return le32_to_cpu(raw_inode->i_blocks_lo);
4127 static inline void ext4_iget_extra_inode(struct inode *inode,
4128 struct ext4_inode *raw_inode,
4129 struct ext4_inode_info *ei)
4131 __le32 *magic = (void *)raw_inode +
4132 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4133 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4134 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4135 ext4_find_inline_data_nolock(inode);
4137 EXT4_I(inode)->i_inline_off = 0;
4140 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4142 struct ext4_iloc iloc;
4143 struct ext4_inode *raw_inode;
4144 struct ext4_inode_info *ei;
4145 struct inode *inode;
4146 journal_t *journal = EXT4_SB(sb)->s_journal;
4152 inode = iget_locked(sb, ino);
4154 return ERR_PTR(-ENOMEM);
4155 if (!(inode->i_state & I_NEW))
4161 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4164 raw_inode = ext4_raw_inode(&iloc);
4166 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4167 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4168 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4169 EXT4_INODE_SIZE(inode->i_sb)) {
4170 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4171 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4172 EXT4_INODE_SIZE(inode->i_sb));
4177 ei->i_extra_isize = 0;
4179 /* Precompute checksum seed for inode metadata */
4180 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4181 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4182 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4184 __le32 inum = cpu_to_le32(inode->i_ino);
4185 __le32 gen = raw_inode->i_generation;
4186 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4188 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4192 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4193 EXT4_ERROR_INODE(inode, "checksum invalid");
4198 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4199 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4200 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4201 if (!(test_opt(inode->i_sb, NO_UID32))) {
4202 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4203 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4205 i_uid_write(inode, i_uid);
4206 i_gid_write(inode, i_gid);
4207 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4209 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4210 ei->i_inline_off = 0;
4211 ei->i_dir_start_lookup = 0;
4212 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4213 /* We now have enough fields to check if the inode was active or not.
4214 * This is needed because nfsd might try to access dead inodes
4215 * the test is that same one that e2fsck uses
4216 * NeilBrown 1999oct15
4218 if (inode->i_nlink == 0) {
4219 if ((inode->i_mode == 0 ||
4220 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4221 ino != EXT4_BOOT_LOADER_INO) {
4222 /* this inode is deleted */
4226 /* The only unlinked inodes we let through here have
4227 * valid i_mode and are being read by the orphan
4228 * recovery code: that's fine, we're about to complete
4229 * the process of deleting those.
4230 * OR it is the EXT4_BOOT_LOADER_INO which is
4231 * not initialized on a new filesystem. */
4233 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4234 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4235 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4236 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4238 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4239 inode->i_size = ext4_isize(raw_inode);
4240 ei->i_disksize = inode->i_size;
4242 ei->i_reserved_quota = 0;
4244 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4245 ei->i_block_group = iloc.block_group;
4246 ei->i_last_alloc_group = ~0;
4248 * NOTE! The in-memory inode i_data array is in little-endian order
4249 * even on big-endian machines: we do NOT byteswap the block numbers!
4251 for (block = 0; block < EXT4_N_BLOCKS; block++)
4252 ei->i_data[block] = raw_inode->i_block[block];
4253 INIT_LIST_HEAD(&ei->i_orphan);
4256 * Set transaction id's of transactions that have to be committed
4257 * to finish f[data]sync. We set them to currently running transaction
4258 * as we cannot be sure that the inode or some of its metadata isn't
4259 * part of the transaction - the inode could have been reclaimed and
4260 * now it is reread from disk.
4263 transaction_t *transaction;
4266 read_lock(&journal->j_state_lock);
4267 if (journal->j_running_transaction)
4268 transaction = journal->j_running_transaction;
4270 transaction = journal->j_committing_transaction;
4272 tid = transaction->t_tid;
4274 tid = journal->j_commit_sequence;
4275 read_unlock(&journal->j_state_lock);
4276 ei->i_sync_tid = tid;
4277 ei->i_datasync_tid = tid;
4280 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4281 if (ei->i_extra_isize == 0) {
4282 /* The extra space is currently unused. Use it. */
4283 ei->i_extra_isize = sizeof(struct ext4_inode) -
4284 EXT4_GOOD_OLD_INODE_SIZE;
4286 ext4_iget_extra_inode(inode, raw_inode, ei);
4290 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4291 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4292 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4293 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4295 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4296 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4297 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4299 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4303 if (ei->i_file_acl &&
4304 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4305 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4309 } else if (!ext4_has_inline_data(inode)) {
4310 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4311 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4312 (S_ISLNK(inode->i_mode) &&
4313 !ext4_inode_is_fast_symlink(inode))))
4314 /* Validate extent which is part of inode */
4315 ret = ext4_ext_check_inode(inode);
4316 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4317 (S_ISLNK(inode->i_mode) &&
4318 !ext4_inode_is_fast_symlink(inode))) {
4319 /* Validate block references which are part of inode */
4320 ret = ext4_ind_check_inode(inode);
4326 if (S_ISREG(inode->i_mode)) {
4327 inode->i_op = &ext4_file_inode_operations;
4328 inode->i_fop = &ext4_file_operations;
4329 ext4_set_aops(inode);
4330 } else if (S_ISDIR(inode->i_mode)) {
4331 inode->i_op = &ext4_dir_inode_operations;
4332 inode->i_fop = &ext4_dir_operations;
4333 } else if (S_ISLNK(inode->i_mode)) {
4334 if (ext4_inode_is_fast_symlink(inode)) {
4335 inode->i_op = &ext4_fast_symlink_inode_operations;
4336 nd_terminate_link(ei->i_data, inode->i_size,
4337 sizeof(ei->i_data) - 1);
4339 inode->i_op = &ext4_symlink_inode_operations;
4340 ext4_set_aops(inode);
4342 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4343 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4344 inode->i_op = &ext4_special_inode_operations;
4345 if (raw_inode->i_block[0])
4346 init_special_inode(inode, inode->i_mode,
4347 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4349 init_special_inode(inode, inode->i_mode,
4350 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4351 } else if (ino == EXT4_BOOT_LOADER_INO) {
4352 make_bad_inode(inode);
4355 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4359 ext4_set_inode_flags(inode);
4360 unlock_new_inode(inode);
4366 return ERR_PTR(ret);
4369 static int ext4_inode_blocks_set(handle_t *handle,
4370 struct ext4_inode *raw_inode,
4371 struct ext4_inode_info *ei)
4373 struct inode *inode = &(ei->vfs_inode);
4374 u64 i_blocks = inode->i_blocks;
4375 struct super_block *sb = inode->i_sb;
4377 if (i_blocks <= ~0U) {
4379 * i_blocks can be represented in a 32 bit variable
4380 * as multiple of 512 bytes
4382 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4383 raw_inode->i_blocks_high = 0;
4384 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4387 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4390 if (i_blocks <= 0xffffffffffffULL) {
4392 * i_blocks can be represented in a 48 bit variable
4393 * as multiple of 512 bytes
4395 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4396 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4397 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4399 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4400 /* i_block is stored in file system block size */
4401 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4402 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4403 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4409 * Post the struct inode info into an on-disk inode location in the
4410 * buffer-cache. This gobbles the caller's reference to the
4411 * buffer_head in the inode location struct.
4413 * The caller must have write access to iloc->bh.
4415 static int ext4_do_update_inode(handle_t *handle,
4416 struct inode *inode,
4417 struct ext4_iloc *iloc)
4419 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4420 struct ext4_inode_info *ei = EXT4_I(inode);
4421 struct buffer_head *bh = iloc->bh;
4422 int err = 0, rc, block;
4423 int need_datasync = 0;
4427 /* For fields not not tracking in the in-memory inode,
4428 * initialise them to zero for new inodes. */
4429 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4430 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4432 ext4_get_inode_flags(ei);
4433 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4434 i_uid = i_uid_read(inode);
4435 i_gid = i_gid_read(inode);
4436 if (!(test_opt(inode->i_sb, NO_UID32))) {
4437 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4438 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4440 * Fix up interoperability with old kernels. Otherwise, old inodes get
4441 * re-used with the upper 16 bits of the uid/gid intact
4444 raw_inode->i_uid_high =
4445 cpu_to_le16(high_16_bits(i_uid));
4446 raw_inode->i_gid_high =
4447 cpu_to_le16(high_16_bits(i_gid));
4449 raw_inode->i_uid_high = 0;
4450 raw_inode->i_gid_high = 0;
4453 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4454 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4455 raw_inode->i_uid_high = 0;
4456 raw_inode->i_gid_high = 0;
4458 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4460 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4461 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4462 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4463 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4465 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4467 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4468 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4469 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4470 cpu_to_le32(EXT4_OS_HURD))
4471 raw_inode->i_file_acl_high =
4472 cpu_to_le16(ei->i_file_acl >> 32);
4473 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4474 if (ei->i_disksize != ext4_isize(raw_inode)) {
4475 ext4_isize_set(raw_inode, ei->i_disksize);
4478 if (ei->i_disksize > 0x7fffffffULL) {
4479 struct super_block *sb = inode->i_sb;
4480 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4481 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4482 EXT4_SB(sb)->s_es->s_rev_level ==
4483 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4484 /* If this is the first large file
4485 * created, add a flag to the superblock.
4487 err = ext4_journal_get_write_access(handle,
4488 EXT4_SB(sb)->s_sbh);
4491 ext4_update_dynamic_rev(sb);
4492 EXT4_SET_RO_COMPAT_FEATURE(sb,
4493 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4494 ext4_handle_sync(handle);
4495 err = ext4_handle_dirty_super(handle, sb);
4498 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4499 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4500 if (old_valid_dev(inode->i_rdev)) {
4501 raw_inode->i_block[0] =
4502 cpu_to_le32(old_encode_dev(inode->i_rdev));
4503 raw_inode->i_block[1] = 0;
4505 raw_inode->i_block[0] = 0;
4506 raw_inode->i_block[1] =
4507 cpu_to_le32(new_encode_dev(inode->i_rdev));
4508 raw_inode->i_block[2] = 0;
4510 } else if (!ext4_has_inline_data(inode)) {
4511 for (block = 0; block < EXT4_N_BLOCKS; block++)
4512 raw_inode->i_block[block] = ei->i_data[block];
4515 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4516 if (ei->i_extra_isize) {
4517 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4518 raw_inode->i_version_hi =
4519 cpu_to_le32(inode->i_version >> 32);
4520 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4523 ext4_inode_csum_set(inode, raw_inode, ei);
4525 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4526 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4529 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4531 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4534 ext4_std_error(inode->i_sb, err);
4539 * ext4_write_inode()
4541 * We are called from a few places:
4543 * - Within generic_file_write() for O_SYNC files.
4544 * Here, there will be no transaction running. We wait for any running
4545 * transaction to commit.
4547 * - Within sys_sync(), kupdate and such.
4548 * We wait on commit, if tol to.
4550 * - Within prune_icache() (PF_MEMALLOC == true)
4551 * Here we simply return. We can't afford to block kswapd on the
4554 * In all cases it is actually safe for us to return without doing anything,
4555 * because the inode has been copied into a raw inode buffer in
4556 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4559 * Note that we are absolutely dependent upon all inode dirtiers doing the
4560 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4561 * which we are interested.
4563 * It would be a bug for them to not do this. The code:
4565 * mark_inode_dirty(inode)
4567 * inode->i_size = expr;
4569 * is in error because a kswapd-driven write_inode() could occur while
4570 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4571 * will no longer be on the superblock's dirty inode list.
4573 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4577 if (current->flags & PF_MEMALLOC)
4580 if (EXT4_SB(inode->i_sb)->s_journal) {
4581 if (ext4_journal_current_handle()) {
4582 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4587 if (wbc->sync_mode != WB_SYNC_ALL)
4590 err = ext4_force_commit(inode->i_sb);
4592 struct ext4_iloc iloc;
4594 err = __ext4_get_inode_loc(inode, &iloc, 0);
4597 if (wbc->sync_mode == WB_SYNC_ALL)
4598 sync_dirty_buffer(iloc.bh);
4599 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4600 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4601 "IO error syncing inode");
4610 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4611 * buffers that are attached to a page stradding i_size and are undergoing
4612 * commit. In that case we have to wait for commit to finish and try again.
4614 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4618 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4619 tid_t commit_tid = 0;
4622 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4624 * All buffers in the last page remain valid? Then there's nothing to
4625 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4628 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4631 page = find_lock_page(inode->i_mapping,
4632 inode->i_size >> PAGE_CACHE_SHIFT);
4635 ret = __ext4_journalled_invalidatepage(page, offset);
4637 page_cache_release(page);
4641 read_lock(&journal->j_state_lock);
4642 if (journal->j_committing_transaction)
4643 commit_tid = journal->j_committing_transaction->t_tid;
4644 read_unlock(&journal->j_state_lock);
4646 jbd2_log_wait_commit(journal, commit_tid);
4653 * Called from notify_change.
4655 * We want to trap VFS attempts to truncate the file as soon as
4656 * possible. In particular, we want to make sure that when the VFS
4657 * shrinks i_size, we put the inode on the orphan list and modify
4658 * i_disksize immediately, so that during the subsequent flushing of
4659 * dirty pages and freeing of disk blocks, we can guarantee that any
4660 * commit will leave the blocks being flushed in an unused state on
4661 * disk. (On recovery, the inode will get truncated and the blocks will
4662 * be freed, so we have a strong guarantee that no future commit will
4663 * leave these blocks visible to the user.)
4665 * Another thing we have to assure is that if we are in ordered mode
4666 * and inode is still attached to the committing transaction, we must
4667 * we start writeout of all the dirty pages which are being truncated.
4668 * This way we are sure that all the data written in the previous
4669 * transaction are already on disk (truncate waits for pages under
4672 * Called with inode->i_mutex down.
4674 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4676 struct inode *inode = dentry->d_inode;
4679 const unsigned int ia_valid = attr->ia_valid;
4681 error = inode_change_ok(inode, attr);
4685 if (is_quota_modification(inode, attr))
4686 dquot_initialize(inode);
4687 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4688 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4691 /* (user+group)*(old+new) structure, inode write (sb,
4692 * inode block, ? - but truncate inode update has it) */
4693 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4694 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4695 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4696 if (IS_ERR(handle)) {
4697 error = PTR_ERR(handle);
4700 error = dquot_transfer(inode, attr);
4702 ext4_journal_stop(handle);
4705 /* Update corresponding info in inode so that everything is in
4706 * one transaction */
4707 if (attr->ia_valid & ATTR_UID)
4708 inode->i_uid = attr->ia_uid;
4709 if (attr->ia_valid & ATTR_GID)
4710 inode->i_gid = attr->ia_gid;
4711 error = ext4_mark_inode_dirty(handle, inode);
4712 ext4_journal_stop(handle);
4715 if (attr->ia_valid & ATTR_SIZE) {
4717 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4718 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4720 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4725 if (S_ISREG(inode->i_mode) &&
4726 attr->ia_valid & ATTR_SIZE &&
4727 (attr->ia_size < inode->i_size)) {
4730 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4731 if (IS_ERR(handle)) {
4732 error = PTR_ERR(handle);
4735 if (ext4_handle_valid(handle)) {
4736 error = ext4_orphan_add(handle, inode);
4739 EXT4_I(inode)->i_disksize = attr->ia_size;
4740 rc = ext4_mark_inode_dirty(handle, inode);
4743 ext4_journal_stop(handle);
4745 if (ext4_should_order_data(inode)) {
4746 error = ext4_begin_ordered_truncate(inode,
4749 /* Do as much error cleanup as possible */
4750 handle = ext4_journal_start(inode,
4752 if (IS_ERR(handle)) {
4753 ext4_orphan_del(NULL, inode);
4756 ext4_orphan_del(handle, inode);
4758 ext4_journal_stop(handle);
4764 if (attr->ia_valid & ATTR_SIZE) {
4765 if (attr->ia_size != inode->i_size) {
4766 loff_t oldsize = inode->i_size;
4768 i_size_write(inode, attr->ia_size);
4770 * Blocks are going to be removed from the inode. Wait
4771 * for dio in flight. Temporarily disable
4772 * dioread_nolock to prevent livelock.
4775 if (!ext4_should_journal_data(inode)) {
4776 ext4_inode_block_unlocked_dio(inode);
4777 inode_dio_wait(inode);
4778 ext4_inode_resume_unlocked_dio(inode);
4780 ext4_wait_for_tail_page_commit(inode);
4783 * Truncate pagecache after we've waited for commit
4784 * in data=journal mode to make pages freeable.
4786 truncate_pagecache(inode, oldsize, inode->i_size);
4788 ext4_truncate(inode);
4792 setattr_copy(inode, attr);
4793 mark_inode_dirty(inode);
4797 * If the call to ext4_truncate failed to get a transaction handle at
4798 * all, we need to clean up the in-core orphan list manually.
4800 if (orphan && inode->i_nlink)
4801 ext4_orphan_del(NULL, inode);
4803 if (!rc && (ia_valid & ATTR_MODE))
4804 rc = ext4_acl_chmod(inode);
4807 ext4_std_error(inode->i_sb, error);
4813 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4816 struct inode *inode;
4817 unsigned long delalloc_blocks;
4819 inode = dentry->d_inode;
4820 generic_fillattr(inode, stat);
4823 * We can't update i_blocks if the block allocation is delayed
4824 * otherwise in the case of system crash before the real block
4825 * allocation is done, we will have i_blocks inconsistent with
4826 * on-disk file blocks.
4827 * We always keep i_blocks updated together with real
4828 * allocation. But to not confuse with user, stat
4829 * will return the blocks that include the delayed allocation
4830 * blocks for this file.
4832 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4833 EXT4_I(inode)->i_reserved_data_blocks);
4835 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4839 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4841 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4842 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4843 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4847 * Account for index blocks, block groups bitmaps and block group
4848 * descriptor blocks if modify datablocks and index blocks
4849 * worse case, the indexs blocks spread over different block groups
4851 * If datablocks are discontiguous, they are possible to spread over
4852 * different block groups too. If they are contiguous, with flexbg,
4853 * they could still across block group boundary.
4855 * Also account for superblock, inode, quota and xattr blocks
4857 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4859 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4865 * How many index blocks need to touch to modify nrblocks?
4866 * The "Chunk" flag indicating whether the nrblocks is
4867 * physically contiguous on disk
4869 * For Direct IO and fallocate, they calls get_block to allocate
4870 * one single extent at a time, so they could set the "Chunk" flag
4872 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4877 * Now let's see how many group bitmaps and group descriptors need
4887 if (groups > ngroups)
4889 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4890 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4892 /* bitmaps and block group descriptor blocks */
4893 ret += groups + gdpblocks;
4895 /* Blocks for super block, inode, quota and xattr blocks */
4896 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4902 * Calculate the total number of credits to reserve to fit
4903 * the modification of a single pages into a single transaction,
4904 * which may include multiple chunks of block allocations.
4906 * This could be called via ext4_write_begin()
4908 * We need to consider the worse case, when
4909 * one new block per extent.
4911 int ext4_writepage_trans_blocks(struct inode *inode)
4913 int bpp = ext4_journal_blocks_per_page(inode);
4916 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4918 /* Account for data blocks for journalled mode */
4919 if (ext4_should_journal_data(inode))
4925 * Calculate the journal credits for a chunk of data modification.
4927 * This is called from DIO, fallocate or whoever calling
4928 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4930 * journal buffers for data blocks are not included here, as DIO
4931 * and fallocate do no need to journal data buffers.
4933 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4935 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4939 * The caller must have previously called ext4_reserve_inode_write().
4940 * Give this, we know that the caller already has write access to iloc->bh.
4942 int ext4_mark_iloc_dirty(handle_t *handle,
4943 struct inode *inode, struct ext4_iloc *iloc)
4947 if (IS_I_VERSION(inode))
4948 inode_inc_iversion(inode);
4950 /* the do_update_inode consumes one bh->b_count */
4953 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4954 err = ext4_do_update_inode(handle, inode, iloc);
4960 * On success, We end up with an outstanding reference count against
4961 * iloc->bh. This _must_ be cleaned up later.
4965 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4966 struct ext4_iloc *iloc)
4970 err = ext4_get_inode_loc(inode, iloc);
4972 BUFFER_TRACE(iloc->bh, "get_write_access");
4973 err = ext4_journal_get_write_access(handle, iloc->bh);
4979 ext4_std_error(inode->i_sb, err);
4984 * Expand an inode by new_extra_isize bytes.
4985 * Returns 0 on success or negative error number on failure.
4987 static int ext4_expand_extra_isize(struct inode *inode,
4988 unsigned int new_extra_isize,
4989 struct ext4_iloc iloc,
4992 struct ext4_inode *raw_inode;
4993 struct ext4_xattr_ibody_header *header;
4995 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4998 raw_inode = ext4_raw_inode(&iloc);
5000 header = IHDR(inode, raw_inode);
5002 /* No extended attributes present */
5003 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5004 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5005 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5007 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5011 /* try to expand with EAs present */
5012 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5017 * What we do here is to mark the in-core inode as clean with respect to inode
5018 * dirtiness (it may still be data-dirty).
5019 * This means that the in-core inode may be reaped by prune_icache
5020 * without having to perform any I/O. This is a very good thing,
5021 * because *any* task may call prune_icache - even ones which
5022 * have a transaction open against a different journal.
5024 * Is this cheating? Not really. Sure, we haven't written the
5025 * inode out, but prune_icache isn't a user-visible syncing function.
5026 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5027 * we start and wait on commits.
5029 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5031 struct ext4_iloc iloc;
5032 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5033 static unsigned int mnt_count;
5037 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5038 err = ext4_reserve_inode_write(handle, inode, &iloc);
5039 if (ext4_handle_valid(handle) &&
5040 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5041 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5043 * We need extra buffer credits since we may write into EA block
5044 * with this same handle. If journal_extend fails, then it will
5045 * only result in a minor loss of functionality for that inode.
5046 * If this is felt to be critical, then e2fsck should be run to
5047 * force a large enough s_min_extra_isize.
5049 if ((jbd2_journal_extend(handle,
5050 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5051 ret = ext4_expand_extra_isize(inode,
5052 sbi->s_want_extra_isize,
5055 ext4_set_inode_state(inode,
5056 EXT4_STATE_NO_EXPAND);
5058 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5059 ext4_warning(inode->i_sb,
5060 "Unable to expand inode %lu. Delete"
5061 " some EAs or run e2fsck.",
5064 le16_to_cpu(sbi->s_es->s_mnt_count);
5070 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5075 * ext4_dirty_inode() is called from __mark_inode_dirty()
5077 * We're really interested in the case where a file is being extended.
5078 * i_size has been changed by generic_commit_write() and we thus need
5079 * to include the updated inode in the current transaction.
5081 * Also, dquot_alloc_block() will always dirty the inode when blocks
5082 * are allocated to the file.
5084 * If the inode is marked synchronous, we don't honour that here - doing
5085 * so would cause a commit on atime updates, which we don't bother doing.
5086 * We handle synchronous inodes at the highest possible level.
5088 void ext4_dirty_inode(struct inode *inode, int flags)
5092 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5096 ext4_mark_inode_dirty(handle, inode);
5098 ext4_journal_stop(handle);
5105 * Bind an inode's backing buffer_head into this transaction, to prevent
5106 * it from being flushed to disk early. Unlike
5107 * ext4_reserve_inode_write, this leaves behind no bh reference and
5108 * returns no iloc structure, so the caller needs to repeat the iloc
5109 * lookup to mark the inode dirty later.
5111 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5113 struct ext4_iloc iloc;
5117 err = ext4_get_inode_loc(inode, &iloc);
5119 BUFFER_TRACE(iloc.bh, "get_write_access");
5120 err = jbd2_journal_get_write_access(handle, iloc.bh);
5122 err = ext4_handle_dirty_metadata(handle,
5128 ext4_std_error(inode->i_sb, err);
5133 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5140 * We have to be very careful here: changing a data block's
5141 * journaling status dynamically is dangerous. If we write a
5142 * data block to the journal, change the status and then delete
5143 * that block, we risk forgetting to revoke the old log record
5144 * from the journal and so a subsequent replay can corrupt data.
5145 * So, first we make sure that the journal is empty and that
5146 * nobody is changing anything.
5149 journal = EXT4_JOURNAL(inode);
5152 if (is_journal_aborted(journal))
5154 /* We have to allocate physical blocks for delalloc blocks
5155 * before flushing journal. otherwise delalloc blocks can not
5156 * be allocated any more. even more truncate on delalloc blocks
5157 * could trigger BUG by flushing delalloc blocks in journal.
5158 * There is no delalloc block in non-journal data mode.
5160 if (val && test_opt(inode->i_sb, DELALLOC)) {
5161 err = ext4_alloc_da_blocks(inode);
5166 /* Wait for all existing dio workers */
5167 ext4_inode_block_unlocked_dio(inode);
5168 inode_dio_wait(inode);
5170 jbd2_journal_lock_updates(journal);
5173 * OK, there are no updates running now, and all cached data is
5174 * synced to disk. We are now in a completely consistent state
5175 * which doesn't have anything in the journal, and we know that
5176 * no filesystem updates are running, so it is safe to modify
5177 * the inode's in-core data-journaling state flag now.
5181 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5183 jbd2_journal_flush(journal);
5184 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5186 ext4_set_aops(inode);
5188 jbd2_journal_unlock_updates(journal);
5189 ext4_inode_resume_unlocked_dio(inode);
5191 /* Finally we can mark the inode as dirty. */
5193 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5195 return PTR_ERR(handle);
5197 err = ext4_mark_inode_dirty(handle, inode);
5198 ext4_handle_sync(handle);
5199 ext4_journal_stop(handle);
5200 ext4_std_error(inode->i_sb, err);
5205 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5207 return !buffer_mapped(bh);
5210 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5212 struct page *page = vmf->page;
5216 struct file *file = vma->vm_file;
5217 struct inode *inode = file_inode(file);
5218 struct address_space *mapping = inode->i_mapping;
5220 get_block_t *get_block;
5223 sb_start_pagefault(inode->i_sb);
5224 file_update_time(vma->vm_file);
5225 /* Delalloc case is easy... */
5226 if (test_opt(inode->i_sb, DELALLOC) &&
5227 !ext4_should_journal_data(inode) &&
5228 !ext4_nonda_switch(inode->i_sb)) {
5230 ret = __block_page_mkwrite(vma, vmf,
5231 ext4_da_get_block_prep);
5232 } while (ret == -ENOSPC &&
5233 ext4_should_retry_alloc(inode->i_sb, &retries));
5238 size = i_size_read(inode);
5239 /* Page got truncated from under us? */
5240 if (page->mapping != mapping || page_offset(page) > size) {
5242 ret = VM_FAULT_NOPAGE;
5246 if (page->index == size >> PAGE_CACHE_SHIFT)
5247 len = size & ~PAGE_CACHE_MASK;
5249 len = PAGE_CACHE_SIZE;
5251 * Return if we have all the buffers mapped. This avoids the need to do
5252 * journal_start/journal_stop which can block and take a long time
5254 if (page_has_buffers(page)) {
5255 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5257 ext4_bh_unmapped)) {
5258 /* Wait so that we don't change page under IO */
5259 wait_for_stable_page(page);
5260 ret = VM_FAULT_LOCKED;
5265 /* OK, we need to fill the hole... */
5266 if (ext4_should_dioread_nolock(inode))
5267 get_block = ext4_get_block_write;
5269 get_block = ext4_get_block;
5271 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5272 ext4_writepage_trans_blocks(inode));
5273 if (IS_ERR(handle)) {
5274 ret = VM_FAULT_SIGBUS;
5277 ret = __block_page_mkwrite(vma, vmf, get_block);
5278 if (!ret && ext4_should_journal_data(inode)) {
5279 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5280 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5282 ret = VM_FAULT_SIGBUS;
5283 ext4_journal_stop(handle);
5286 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5288 ext4_journal_stop(handle);
5289 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5292 ret = block_page_mkwrite_return(ret);
5294 sb_end_pagefault(inode->i_sb);