2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
96 return provided == calculated;
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 trace_ext4_begin_ordered_truncate(inode, new_size);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode)->jinode)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
134 static void ext4_invalidatepage(struct page *page, unsigned long offset);
135 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
136 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
137 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
138 struct inode *inode, struct page *page, loff_t from,
139 loff_t length, int flags);
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode *inode)
146 int ea_blocks = EXT4_I(inode)->i_file_acl ?
147 (inode->i_sb->s_blocksize >> 9) : 0;
149 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
157 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
168 BUG_ON(EXT4_JOURNAL(inode) == NULL);
169 jbd_debug(2, "restarting handle %p\n", handle);
170 up_write(&EXT4_I(inode)->i_data_sem);
171 ret = ext4_journal_restart(handle, nblocks);
172 down_write(&EXT4_I(inode)->i_data_sem);
173 ext4_discard_preallocations(inode);
179 * Called at the last iput() if i_nlink is zero.
181 void ext4_evict_inode(struct inode *inode)
186 trace_ext4_evict_inode(inode);
188 if (inode->i_nlink) {
190 * When journalling data dirty buffers are tracked only in the
191 * journal. So although mm thinks everything is clean and
192 * ready for reaping the inode might still have some pages to
193 * write in the running transaction or waiting to be
194 * checkpointed. Thus calling jbd2_journal_invalidatepage()
195 * (via truncate_inode_pages()) to discard these buffers can
196 * cause data loss. Also even if we did not discard these
197 * buffers, we would have no way to find them after the inode
198 * is reaped and thus user could see stale data if he tries to
199 * read them before the transaction is checkpointed. So be
200 * careful and force everything to disk here... We use
201 * ei->i_datasync_tid to store the newest transaction
202 * containing inode's data.
204 * Note that directories do not have this problem because they
205 * don't use page cache.
207 if (ext4_should_journal_data(inode) &&
208 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
209 inode->i_ino != EXT4_JOURNAL_INO) {
210 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
211 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
213 jbd2_complete_transaction(journal, commit_tid);
214 filemap_write_and_wait(&inode->i_data);
216 truncate_inode_pages(&inode->i_data, 0);
217 ext4_ioend_shutdown(inode);
221 if (!is_bad_inode(inode))
222 dquot_initialize(inode);
224 if (ext4_should_order_data(inode))
225 ext4_begin_ordered_truncate(inode, 0);
226 truncate_inode_pages(&inode->i_data, 0);
227 ext4_ioend_shutdown(inode);
229 if (is_bad_inode(inode))
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
236 sb_start_intwrite(inode->i_sb);
237 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
238 ext4_blocks_for_truncate(inode)+3);
239 if (IS_ERR(handle)) {
240 ext4_std_error(inode->i_sb, PTR_ERR(handle));
242 * If we're going to skip the normal cleanup, we still need to
243 * make sure that the in-core orphan linked list is properly
246 ext4_orphan_del(NULL, inode);
247 sb_end_intwrite(inode->i_sb);
252 ext4_handle_sync(handle);
254 err = ext4_mark_inode_dirty(handle, inode);
256 ext4_warning(inode->i_sb,
257 "couldn't mark inode dirty (err %d)", err);
261 ext4_truncate(inode);
264 * ext4_ext_truncate() doesn't reserve any slop when it
265 * restarts journal transactions; therefore there may not be
266 * enough credits left in the handle to remove the inode from
267 * the orphan list and set the dtime field.
269 if (!ext4_handle_has_enough_credits(handle, 3)) {
270 err = ext4_journal_extend(handle, 3);
272 err = ext4_journal_restart(handle, 3);
274 ext4_warning(inode->i_sb,
275 "couldn't extend journal (err %d)", err);
277 ext4_journal_stop(handle);
278 ext4_orphan_del(NULL, inode);
279 sb_end_intwrite(inode->i_sb);
285 * Kill off the orphan record which ext4_truncate created.
286 * AKPM: I think this can be inside the above `if'.
287 * Note that ext4_orphan_del() has to be able to cope with the
288 * deletion of a non-existent orphan - this is because we don't
289 * know if ext4_truncate() actually created an orphan record.
290 * (Well, we could do this if we need to, but heck - it works)
292 ext4_orphan_del(handle, inode);
293 EXT4_I(inode)->i_dtime = get_seconds();
296 * One subtle ordering requirement: if anything has gone wrong
297 * (transaction abort, IO errors, whatever), then we can still
298 * do these next steps (the fs will already have been marked as
299 * having errors), but we can't free the inode if the mark_dirty
302 if (ext4_mark_inode_dirty(handle, inode))
303 /* If that failed, just do the required in-core inode clear. */
304 ext4_clear_inode(inode);
306 ext4_free_inode(handle, inode);
307 ext4_journal_stop(handle);
308 sb_end_intwrite(inode->i_sb);
311 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
315 qsize_t *ext4_get_reserved_space(struct inode *inode)
317 return &EXT4_I(inode)->i_reserved_quota;
322 * Calculate the number of metadata blocks need to reserve
323 * to allocate a block located at @lblock
325 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
327 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
328 return ext4_ext_calc_metadata_amount(inode, lblock);
330 return ext4_ind_calc_metadata_amount(inode, lblock);
334 * Called with i_data_sem down, which is important since we can call
335 * ext4_discard_preallocations() from here.
337 void ext4_da_update_reserve_space(struct inode *inode,
338 int used, int quota_claim)
340 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
341 struct ext4_inode_info *ei = EXT4_I(inode);
343 spin_lock(&ei->i_block_reservation_lock);
344 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
345 if (unlikely(used > ei->i_reserved_data_blocks)) {
346 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
347 "with only %d reserved data blocks",
348 __func__, inode->i_ino, used,
349 ei->i_reserved_data_blocks);
351 used = ei->i_reserved_data_blocks;
354 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
355 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
356 "with only %d reserved metadata blocks "
357 "(releasing %d blocks with reserved %d data blocks)",
358 inode->i_ino, ei->i_allocated_meta_blocks,
359 ei->i_reserved_meta_blocks, used,
360 ei->i_reserved_data_blocks);
362 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
365 /* Update per-inode reservations */
366 ei->i_reserved_data_blocks -= used;
367 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
368 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
369 used + ei->i_allocated_meta_blocks);
370 ei->i_allocated_meta_blocks = 0;
372 if (ei->i_reserved_data_blocks == 0) {
374 * We can release all of the reserved metadata blocks
375 * only when we have written all of the delayed
378 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
379 ei->i_reserved_meta_blocks);
380 ei->i_reserved_meta_blocks = 0;
381 ei->i_da_metadata_calc_len = 0;
383 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
385 /* Update quota subsystem for data blocks */
387 dquot_claim_block(inode, EXT4_C2B(sbi, used));
390 * We did fallocate with an offset that is already delayed
391 * allocated. So on delayed allocated writeback we should
392 * not re-claim the quota for fallocated blocks.
394 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
398 * If we have done all the pending block allocations and if
399 * there aren't any writers on the inode, we can discard the
400 * inode's preallocations.
402 if ((ei->i_reserved_data_blocks == 0) &&
403 (atomic_read(&inode->i_writecount) == 0))
404 ext4_discard_preallocations(inode);
407 static int __check_block_validity(struct inode *inode, const char *func,
409 struct ext4_map_blocks *map)
411 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
413 ext4_error_inode(inode, func, line, map->m_pblk,
414 "lblock %lu mapped to illegal pblock "
415 "(length %d)", (unsigned long) map->m_lblk,
422 #define check_block_validity(inode, map) \
423 __check_block_validity((inode), __func__, __LINE__, (map))
426 * Return the number of contiguous dirty pages in a given inode
427 * starting at page frame idx.
429 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
430 unsigned int max_pages)
432 struct address_space *mapping = inode->i_mapping;
436 int i, nr_pages, done = 0;
440 pagevec_init(&pvec, 0);
443 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
445 (pgoff_t)PAGEVEC_SIZE);
448 for (i = 0; i < nr_pages; i++) {
449 struct page *page = pvec.pages[i];
450 struct buffer_head *bh, *head;
453 if (unlikely(page->mapping != mapping) ||
455 PageWriteback(page) ||
456 page->index != idx) {
461 if (page_has_buffers(page)) {
462 bh = head = page_buffers(page);
464 if (!buffer_delay(bh) &&
465 !buffer_unwritten(bh))
467 bh = bh->b_this_page;
468 } while (!done && (bh != head));
475 if (num >= max_pages) {
480 pagevec_release(&pvec);
485 #ifdef ES_AGGRESSIVE_TEST
486 static void ext4_map_blocks_es_recheck(handle_t *handle,
488 struct ext4_map_blocks *es_map,
489 struct ext4_map_blocks *map,
496 * There is a race window that the result is not the same.
497 * e.g. xfstests #223 when dioread_nolock enables. The reason
498 * is that we lookup a block mapping in extent status tree with
499 * out taking i_data_sem. So at the time the unwritten extent
500 * could be converted.
502 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
503 down_read((&EXT4_I(inode)->i_data_sem));
504 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
505 retval = ext4_ext_map_blocks(handle, inode, map, flags &
506 EXT4_GET_BLOCKS_KEEP_SIZE);
508 retval = ext4_ind_map_blocks(handle, inode, map, flags &
509 EXT4_GET_BLOCKS_KEEP_SIZE);
511 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
512 up_read((&EXT4_I(inode)->i_data_sem));
514 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
515 * because it shouldn't be marked in es_map->m_flags.
517 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
520 * We don't check m_len because extent will be collpased in status
521 * tree. So the m_len might not equal.
523 if (es_map->m_lblk != map->m_lblk ||
524 es_map->m_flags != map->m_flags ||
525 es_map->m_pblk != map->m_pblk) {
526 printk("ES cache assertation failed for inode: %lu "
527 "es_cached ex [%d/%d/%llu/%x] != "
528 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
529 inode->i_ino, es_map->m_lblk, es_map->m_len,
530 es_map->m_pblk, es_map->m_flags, map->m_lblk,
531 map->m_len, map->m_pblk, map->m_flags,
535 #endif /* ES_AGGRESSIVE_TEST */
538 * The ext4_map_blocks() function tries to look up the requested blocks,
539 * and returns if the blocks are already mapped.
541 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
542 * and store the allocated blocks in the result buffer head and mark it
545 * If file type is extents based, it will call ext4_ext_map_blocks(),
546 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
549 * On success, it returns the number of blocks being mapped or allocate.
550 * if create==0 and the blocks are pre-allocated and uninitialized block,
551 * the result buffer head is unmapped. If the create ==1, it will make sure
552 * the buffer head is mapped.
554 * It returns 0 if plain look up failed (blocks have not been allocated), in
555 * that case, buffer head is unmapped
557 * It returns the error in case of allocation failure.
559 int ext4_map_blocks(handle_t *handle, struct inode *inode,
560 struct ext4_map_blocks *map, int flags)
562 struct extent_status es;
564 #ifdef ES_AGGRESSIVE_TEST
565 struct ext4_map_blocks orig_map;
567 memcpy(&orig_map, map, sizeof(*map));
571 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
572 "logical block %lu\n", inode->i_ino, flags, map->m_len,
573 (unsigned long) map->m_lblk);
575 /* Lookup extent status tree firstly */
576 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
577 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
578 map->m_pblk = ext4_es_pblock(&es) +
579 map->m_lblk - es.es_lblk;
580 map->m_flags |= ext4_es_is_written(&es) ?
581 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
582 retval = es.es_len - (map->m_lblk - es.es_lblk);
583 if (retval > map->m_len)
586 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
591 #ifdef ES_AGGRESSIVE_TEST
592 ext4_map_blocks_es_recheck(handle, inode, map,
599 * Try to see if we can get the block without requesting a new
602 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
603 down_read((&EXT4_I(inode)->i_data_sem));
604 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
605 retval = ext4_ext_map_blocks(handle, inode, map, flags &
606 EXT4_GET_BLOCKS_KEEP_SIZE);
608 retval = ext4_ind_map_blocks(handle, inode, map, flags &
609 EXT4_GET_BLOCKS_KEEP_SIZE);
613 unsigned long long status;
615 #ifdef ES_AGGRESSIVE_TEST
616 if (retval != map->m_len) {
617 printk("ES len assertation failed for inode: %lu "
618 "retval %d != map->m_len %d "
619 "in %s (lookup)\n", inode->i_ino, retval,
620 map->m_len, __func__);
624 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
625 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
626 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
627 ext4_find_delalloc_range(inode, map->m_lblk,
628 map->m_lblk + map->m_len - 1))
629 status |= EXTENT_STATUS_DELAYED;
630 ret = ext4_es_insert_extent(inode, map->m_lblk,
631 map->m_len, map->m_pblk, status);
635 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
636 up_read((&EXT4_I(inode)->i_data_sem));
639 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
640 int ret = check_block_validity(inode, map);
645 /* If it is only a block(s) look up */
646 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
650 * Returns if the blocks have already allocated
652 * Note that if blocks have been preallocated
653 * ext4_ext_get_block() returns the create = 0
654 * with buffer head unmapped.
656 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
660 * Here we clear m_flags because after allocating an new extent,
661 * it will be set again.
663 map->m_flags &= ~EXT4_MAP_FLAGS;
666 * New blocks allocate and/or writing to uninitialized extent
667 * will possibly result in updating i_data, so we take
668 * the write lock of i_data_sem, and call get_blocks()
669 * with create == 1 flag.
671 down_write((&EXT4_I(inode)->i_data_sem));
674 * if the caller is from delayed allocation writeout path
675 * we have already reserved fs blocks for allocation
676 * let the underlying get_block() function know to
677 * avoid double accounting
679 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
680 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
682 * We need to check for EXT4 here because migrate
683 * could have changed the inode type in between
685 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
686 retval = ext4_ext_map_blocks(handle, inode, map, flags);
688 retval = ext4_ind_map_blocks(handle, inode, map, flags);
690 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
692 * We allocated new blocks which will result in
693 * i_data's format changing. Force the migrate
694 * to fail by clearing migrate flags
696 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
700 * Update reserved blocks/metadata blocks after successful
701 * block allocation which had been deferred till now. We don't
702 * support fallocate for non extent files. So we can update
703 * reserve space here.
706 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
707 ext4_da_update_reserve_space(inode, retval, 1);
709 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
710 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
714 unsigned long long status;
716 #ifdef ES_AGGRESSIVE_TEST
717 if (retval != map->m_len) {
718 printk("ES len assertation failed for inode: %lu "
719 "retval %d != map->m_len %d "
720 "in %s (allocation)\n", inode->i_ino, retval,
721 map->m_len, __func__);
726 * If the extent has been zeroed out, we don't need to update
727 * extent status tree.
729 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
730 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
731 if (ext4_es_is_written(&es))
734 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
735 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
736 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
737 ext4_find_delalloc_range(inode, map->m_lblk,
738 map->m_lblk + map->m_len - 1))
739 status |= EXTENT_STATUS_DELAYED;
740 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
741 map->m_pblk, status);
747 up_write((&EXT4_I(inode)->i_data_sem));
748 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
749 int ret = check_block_validity(inode, map);
756 /* Maximum number of blocks we map for direct IO at once. */
757 #define DIO_MAX_BLOCKS 4096
759 static int _ext4_get_block(struct inode *inode, sector_t iblock,
760 struct buffer_head *bh, int flags)
762 handle_t *handle = ext4_journal_current_handle();
763 struct ext4_map_blocks map;
764 int ret = 0, started = 0;
767 if (ext4_has_inline_data(inode))
771 map.m_len = bh->b_size >> inode->i_blkbits;
773 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
774 /* Direct IO write... */
775 if (map.m_len > DIO_MAX_BLOCKS)
776 map.m_len = DIO_MAX_BLOCKS;
777 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
778 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
780 if (IS_ERR(handle)) {
781 ret = PTR_ERR(handle);
787 ret = ext4_map_blocks(handle, inode, &map, flags);
789 map_bh(bh, inode->i_sb, map.m_pblk);
790 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
791 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
795 ext4_journal_stop(handle);
799 int ext4_get_block(struct inode *inode, sector_t iblock,
800 struct buffer_head *bh, int create)
802 return _ext4_get_block(inode, iblock, bh,
803 create ? EXT4_GET_BLOCKS_CREATE : 0);
807 * `handle' can be NULL if create is zero
809 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
810 ext4_lblk_t block, int create, int *errp)
812 struct ext4_map_blocks map;
813 struct buffer_head *bh;
816 J_ASSERT(handle != NULL || create == 0);
820 err = ext4_map_blocks(handle, inode, &map,
821 create ? EXT4_GET_BLOCKS_CREATE : 0);
823 /* ensure we send some value back into *errp */
826 if (create && err == 0)
827 err = -ENOSPC; /* should never happen */
833 bh = sb_getblk(inode->i_sb, map.m_pblk);
838 if (map.m_flags & EXT4_MAP_NEW) {
839 J_ASSERT(create != 0);
840 J_ASSERT(handle != NULL);
843 * Now that we do not always journal data, we should
844 * keep in mind whether this should always journal the
845 * new buffer as metadata. For now, regular file
846 * writes use ext4_get_block instead, so it's not a
850 BUFFER_TRACE(bh, "call get_create_access");
851 fatal = ext4_journal_get_create_access(handle, bh);
852 if (!fatal && !buffer_uptodate(bh)) {
853 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
854 set_buffer_uptodate(bh);
857 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
858 err = ext4_handle_dirty_metadata(handle, inode, bh);
862 BUFFER_TRACE(bh, "not a new buffer");
872 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
873 ext4_lblk_t block, int create, int *err)
875 struct buffer_head *bh;
877 bh = ext4_getblk(handle, inode, block, create, err);
880 if (buffer_uptodate(bh))
882 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
884 if (buffer_uptodate(bh))
891 int ext4_walk_page_buffers(handle_t *handle,
892 struct buffer_head *head,
896 int (*fn)(handle_t *handle,
897 struct buffer_head *bh))
899 struct buffer_head *bh;
900 unsigned block_start, block_end;
901 unsigned blocksize = head->b_size;
903 struct buffer_head *next;
905 for (bh = head, block_start = 0;
906 ret == 0 && (bh != head || !block_start);
907 block_start = block_end, bh = next) {
908 next = bh->b_this_page;
909 block_end = block_start + blocksize;
910 if (block_end <= from || block_start >= to) {
911 if (partial && !buffer_uptodate(bh))
915 err = (*fn)(handle, bh);
923 * To preserve ordering, it is essential that the hole instantiation and
924 * the data write be encapsulated in a single transaction. We cannot
925 * close off a transaction and start a new one between the ext4_get_block()
926 * and the commit_write(). So doing the jbd2_journal_start at the start of
927 * prepare_write() is the right place.
929 * Also, this function can nest inside ext4_writepage(). In that case, we
930 * *know* that ext4_writepage() has generated enough buffer credits to do the
931 * whole page. So we won't block on the journal in that case, which is good,
932 * because the caller may be PF_MEMALLOC.
934 * By accident, ext4 can be reentered when a transaction is open via
935 * quota file writes. If we were to commit the transaction while thus
936 * reentered, there can be a deadlock - we would be holding a quota
937 * lock, and the commit would never complete if another thread had a
938 * transaction open and was blocking on the quota lock - a ranking
941 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
942 * will _not_ run commit under these circumstances because handle->h_ref
943 * is elevated. We'll still have enough credits for the tiny quotafile
946 int do_journal_get_write_access(handle_t *handle,
947 struct buffer_head *bh)
949 int dirty = buffer_dirty(bh);
952 if (!buffer_mapped(bh) || buffer_freed(bh))
955 * __block_write_begin() could have dirtied some buffers. Clean
956 * the dirty bit as jbd2_journal_get_write_access() could complain
957 * otherwise about fs integrity issues. Setting of the dirty bit
958 * by __block_write_begin() isn't a real problem here as we clear
959 * the bit before releasing a page lock and thus writeback cannot
960 * ever write the buffer.
963 clear_buffer_dirty(bh);
964 ret = ext4_journal_get_write_access(handle, bh);
966 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
970 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
971 struct buffer_head *bh_result, int create);
972 static int ext4_write_begin(struct file *file, struct address_space *mapping,
973 loff_t pos, unsigned len, unsigned flags,
974 struct page **pagep, void **fsdata)
976 struct inode *inode = mapping->host;
977 int ret, needed_blocks;
984 trace_ext4_write_begin(inode, pos, len, flags);
986 * Reserve one block more for addition to orphan list in case
987 * we allocate blocks but write fails for some reason
989 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
990 index = pos >> PAGE_CACHE_SHIFT;
991 from = pos & (PAGE_CACHE_SIZE - 1);
994 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
995 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1004 * grab_cache_page_write_begin() can take a long time if the
1005 * system is thrashing due to memory pressure, or if the page
1006 * is being written back. So grab it first before we start
1007 * the transaction handle. This also allows us to allocate
1008 * the page (if needed) without using GFP_NOFS.
1011 page = grab_cache_page_write_begin(mapping, index, flags);
1017 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1018 if (IS_ERR(handle)) {
1019 page_cache_release(page);
1020 return PTR_ERR(handle);
1024 if (page->mapping != mapping) {
1025 /* The page got truncated from under us */
1027 page_cache_release(page);
1028 ext4_journal_stop(handle);
1031 wait_on_page_writeback(page);
1033 if (ext4_should_dioread_nolock(inode))
1034 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1036 ret = __block_write_begin(page, pos, len, ext4_get_block);
1038 if (!ret && ext4_should_journal_data(inode)) {
1039 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1041 do_journal_get_write_access);
1047 * __block_write_begin may have instantiated a few blocks
1048 * outside i_size. Trim these off again. Don't need
1049 * i_size_read because we hold i_mutex.
1051 * Add inode to orphan list in case we crash before
1054 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1055 ext4_orphan_add(handle, inode);
1057 ext4_journal_stop(handle);
1058 if (pos + len > inode->i_size) {
1059 ext4_truncate_failed_write(inode);
1061 * If truncate failed early the inode might
1062 * still be on the orphan list; we need to
1063 * make sure the inode is removed from the
1064 * orphan list in that case.
1067 ext4_orphan_del(NULL, inode);
1070 if (ret == -ENOSPC &&
1071 ext4_should_retry_alloc(inode->i_sb, &retries))
1073 page_cache_release(page);
1080 /* For write_end() in data=journal mode */
1081 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1083 if (!buffer_mapped(bh) || buffer_freed(bh))
1085 set_buffer_uptodate(bh);
1086 return ext4_handle_dirty_metadata(handle, NULL, bh);
1090 * We need to pick up the new inode size which generic_commit_write gave us
1091 * `file' can be NULL - eg, when called from page_symlink().
1093 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1094 * buffers are managed internally.
1096 static int ext4_write_end(struct file *file,
1097 struct address_space *mapping,
1098 loff_t pos, unsigned len, unsigned copied,
1099 struct page *page, void *fsdata)
1101 handle_t *handle = ext4_journal_current_handle();
1102 struct inode *inode = mapping->host;
1104 int i_size_changed = 0;
1106 trace_ext4_write_end(inode, pos, len, copied);
1107 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1108 ret = ext4_jbd2_file_inode(handle, inode);
1111 page_cache_release(page);
1116 if (ext4_has_inline_data(inode))
1117 copied = ext4_write_inline_data_end(inode, pos, len,
1120 copied = block_write_end(file, mapping, pos,
1121 len, copied, page, fsdata);
1124 * No need to use i_size_read() here, the i_size
1125 * cannot change under us because we hole i_mutex.
1127 * But it's important to update i_size while still holding page lock:
1128 * page writeout could otherwise come in and zero beyond i_size.
1130 if (pos + copied > inode->i_size) {
1131 i_size_write(inode, pos + copied);
1135 if (pos + copied > EXT4_I(inode)->i_disksize) {
1136 /* We need to mark inode dirty even if
1137 * new_i_size is less that inode->i_size
1138 * but greater than i_disksize. (hint delalloc)
1140 ext4_update_i_disksize(inode, (pos + copied));
1144 page_cache_release(page);
1147 * Don't mark the inode dirty under page lock. First, it unnecessarily
1148 * makes the holding time of page lock longer. Second, it forces lock
1149 * ordering of page lock and transaction start for journaling
1153 ext4_mark_inode_dirty(handle, inode);
1157 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1158 /* if we have allocated more blocks and copied
1159 * less. We will have blocks allocated outside
1160 * inode->i_size. So truncate them
1162 ext4_orphan_add(handle, inode);
1164 ret2 = ext4_journal_stop(handle);
1168 if (pos + len > inode->i_size) {
1169 ext4_truncate_failed_write(inode);
1171 * If truncate failed early the inode might still be
1172 * on the orphan list; we need to make sure the inode
1173 * is removed from the orphan list in that case.
1176 ext4_orphan_del(NULL, inode);
1179 return ret ? ret : copied;
1182 static int ext4_journalled_write_end(struct file *file,
1183 struct address_space *mapping,
1184 loff_t pos, unsigned len, unsigned copied,
1185 struct page *page, void *fsdata)
1187 handle_t *handle = ext4_journal_current_handle();
1188 struct inode *inode = mapping->host;
1194 trace_ext4_journalled_write_end(inode, pos, len, copied);
1195 from = pos & (PAGE_CACHE_SIZE - 1);
1198 BUG_ON(!ext4_handle_valid(handle));
1200 if (ext4_has_inline_data(inode))
1201 copied = ext4_write_inline_data_end(inode, pos, len,
1205 if (!PageUptodate(page))
1207 page_zero_new_buffers(page, from+copied, to);
1210 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1211 to, &partial, write_end_fn);
1213 SetPageUptodate(page);
1215 new_i_size = pos + copied;
1216 if (new_i_size > inode->i_size)
1217 i_size_write(inode, pos+copied);
1218 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1219 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1220 if (new_i_size > EXT4_I(inode)->i_disksize) {
1221 ext4_update_i_disksize(inode, new_i_size);
1222 ret2 = ext4_mark_inode_dirty(handle, inode);
1228 page_cache_release(page);
1229 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1230 /* if we have allocated more blocks and copied
1231 * less. We will have blocks allocated outside
1232 * inode->i_size. So truncate them
1234 ext4_orphan_add(handle, inode);
1236 ret2 = ext4_journal_stop(handle);
1239 if (pos + len > inode->i_size) {
1240 ext4_truncate_failed_write(inode);
1242 * If truncate failed early the inode might still be
1243 * on the orphan list; we need to make sure the inode
1244 * is removed from the orphan list in that case.
1247 ext4_orphan_del(NULL, inode);
1250 return ret ? ret : copied;
1254 * Reserve a metadata for a single block located at lblock
1256 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1259 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1260 struct ext4_inode_info *ei = EXT4_I(inode);
1261 unsigned int md_needed;
1262 ext4_lblk_t save_last_lblock;
1266 * recalculate the amount of metadata blocks to reserve
1267 * in order to allocate nrblocks
1268 * worse case is one extent per block
1271 spin_lock(&ei->i_block_reservation_lock);
1273 * ext4_calc_metadata_amount() has side effects, which we have
1274 * to be prepared undo if we fail to claim space.
1276 save_len = ei->i_da_metadata_calc_len;
1277 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1278 md_needed = EXT4_NUM_B2C(sbi,
1279 ext4_calc_metadata_amount(inode, lblock));
1280 trace_ext4_da_reserve_space(inode, md_needed);
1283 * We do still charge estimated metadata to the sb though;
1284 * we cannot afford to run out of free blocks.
1286 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1287 ei->i_da_metadata_calc_len = save_len;
1288 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1289 spin_unlock(&ei->i_block_reservation_lock);
1290 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1296 ei->i_reserved_meta_blocks += md_needed;
1297 spin_unlock(&ei->i_block_reservation_lock);
1299 return 0; /* success */
1303 * Reserve a single cluster located at lblock
1305 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1308 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1309 struct ext4_inode_info *ei = EXT4_I(inode);
1310 unsigned int md_needed;
1312 ext4_lblk_t save_last_lblock;
1316 * We will charge metadata quota at writeout time; this saves
1317 * us from metadata over-estimation, though we may go over by
1318 * a small amount in the end. Here we just reserve for data.
1320 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1325 * recalculate the amount of metadata blocks to reserve
1326 * in order to allocate nrblocks
1327 * worse case is one extent per block
1330 spin_lock(&ei->i_block_reservation_lock);
1332 * ext4_calc_metadata_amount() has side effects, which we have
1333 * to be prepared undo if we fail to claim space.
1335 save_len = ei->i_da_metadata_calc_len;
1336 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1337 md_needed = EXT4_NUM_B2C(sbi,
1338 ext4_calc_metadata_amount(inode, lblock));
1339 trace_ext4_da_reserve_space(inode, md_needed);
1342 * We do still charge estimated metadata to the sb though;
1343 * we cannot afford to run out of free blocks.
1345 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1346 ei->i_da_metadata_calc_len = save_len;
1347 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1348 spin_unlock(&ei->i_block_reservation_lock);
1349 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1353 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1356 ei->i_reserved_data_blocks++;
1357 ei->i_reserved_meta_blocks += md_needed;
1358 spin_unlock(&ei->i_block_reservation_lock);
1360 return 0; /* success */
1363 static void ext4_da_release_space(struct inode *inode, int to_free)
1365 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1366 struct ext4_inode_info *ei = EXT4_I(inode);
1369 return; /* Nothing to release, exit */
1371 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1373 trace_ext4_da_release_space(inode, to_free);
1374 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1376 * if there aren't enough reserved blocks, then the
1377 * counter is messed up somewhere. Since this
1378 * function is called from invalidate page, it's
1379 * harmless to return without any action.
1381 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1382 "ino %lu, to_free %d with only %d reserved "
1383 "data blocks", inode->i_ino, to_free,
1384 ei->i_reserved_data_blocks);
1386 to_free = ei->i_reserved_data_blocks;
1388 ei->i_reserved_data_blocks -= to_free;
1390 if (ei->i_reserved_data_blocks == 0) {
1392 * We can release all of the reserved metadata blocks
1393 * only when we have written all of the delayed
1394 * allocation blocks.
1395 * Note that in case of bigalloc, i_reserved_meta_blocks,
1396 * i_reserved_data_blocks, etc. refer to number of clusters.
1398 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1399 ei->i_reserved_meta_blocks);
1400 ei->i_reserved_meta_blocks = 0;
1401 ei->i_da_metadata_calc_len = 0;
1404 /* update fs dirty data blocks counter */
1405 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1407 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1409 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1412 static void ext4_da_page_release_reservation(struct page *page,
1413 unsigned long offset)
1416 struct buffer_head *head, *bh;
1417 unsigned int curr_off = 0;
1418 struct inode *inode = page->mapping->host;
1419 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1423 head = page_buffers(page);
1426 unsigned int next_off = curr_off + bh->b_size;
1428 if ((offset <= curr_off) && (buffer_delay(bh))) {
1430 clear_buffer_delay(bh);
1432 curr_off = next_off;
1433 } while ((bh = bh->b_this_page) != head);
1436 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1437 ext4_es_remove_extent(inode, lblk, to_release);
1440 /* If we have released all the blocks belonging to a cluster, then we
1441 * need to release the reserved space for that cluster. */
1442 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1443 while (num_clusters > 0) {
1444 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1445 ((num_clusters - 1) << sbi->s_cluster_bits);
1446 if (sbi->s_cluster_ratio == 1 ||
1447 !ext4_find_delalloc_cluster(inode, lblk))
1448 ext4_da_release_space(inode, 1);
1455 * Delayed allocation stuff
1459 * mpage_da_submit_io - walks through extent of pages and try to write
1460 * them with writepage() call back
1462 * @mpd->inode: inode
1463 * @mpd->first_page: first page of the extent
1464 * @mpd->next_page: page after the last page of the extent
1466 * By the time mpage_da_submit_io() is called we expect all blocks
1467 * to be allocated. this may be wrong if allocation failed.
1469 * As pages are already locked by write_cache_pages(), we can't use it
1471 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1472 struct ext4_map_blocks *map)
1474 struct pagevec pvec;
1475 unsigned long index, end;
1476 int ret = 0, err, nr_pages, i;
1477 struct inode *inode = mpd->inode;
1478 struct address_space *mapping = inode->i_mapping;
1479 loff_t size = i_size_read(inode);
1480 unsigned int len, block_start;
1481 struct buffer_head *bh, *page_bufs = NULL;
1482 sector_t pblock = 0, cur_logical = 0;
1483 struct ext4_io_submit io_submit;
1485 BUG_ON(mpd->next_page <= mpd->first_page);
1486 memset(&io_submit, 0, sizeof(io_submit));
1488 * We need to start from the first_page to the next_page - 1
1489 * to make sure we also write the mapped dirty buffer_heads.
1490 * If we look at mpd->b_blocknr we would only be looking
1491 * at the currently mapped buffer_heads.
1493 index = mpd->first_page;
1494 end = mpd->next_page - 1;
1496 pagevec_init(&pvec, 0);
1497 while (index <= end) {
1498 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1501 for (i = 0; i < nr_pages; i++) {
1503 struct page *page = pvec.pages[i];
1505 index = page->index;
1509 if (index == size >> PAGE_CACHE_SHIFT)
1510 len = size & ~PAGE_CACHE_MASK;
1512 len = PAGE_CACHE_SIZE;
1514 cur_logical = index << (PAGE_CACHE_SHIFT -
1516 pblock = map->m_pblk + (cur_logical -
1521 BUG_ON(!PageLocked(page));
1522 BUG_ON(PageWriteback(page));
1524 bh = page_bufs = page_buffers(page);
1527 if (map && (cur_logical >= map->m_lblk) &&
1528 (cur_logical <= (map->m_lblk +
1529 (map->m_len - 1)))) {
1530 if (buffer_delay(bh)) {
1531 clear_buffer_delay(bh);
1532 bh->b_blocknr = pblock;
1534 if (buffer_unwritten(bh) ||
1536 BUG_ON(bh->b_blocknr != pblock);
1537 if (map->m_flags & EXT4_MAP_UNINIT)
1538 set_buffer_uninit(bh);
1539 clear_buffer_unwritten(bh);
1543 * skip page if block allocation undone and
1546 if (ext4_bh_delay_or_unwritten(NULL, bh))
1548 bh = bh->b_this_page;
1549 block_start += bh->b_size;
1552 } while (bh != page_bufs);
1559 clear_page_dirty_for_io(page);
1560 err = ext4_bio_write_page(&io_submit, page, len,
1563 mpd->pages_written++;
1565 * In error case, we have to continue because
1566 * remaining pages are still locked
1571 pagevec_release(&pvec);
1573 ext4_io_submit(&io_submit);
1577 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1581 struct pagevec pvec;
1582 struct inode *inode = mpd->inode;
1583 struct address_space *mapping = inode->i_mapping;
1584 ext4_lblk_t start, last;
1586 index = mpd->first_page;
1587 end = mpd->next_page - 1;
1589 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1590 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1591 ext4_es_remove_extent(inode, start, last - start + 1);
1593 pagevec_init(&pvec, 0);
1594 while (index <= end) {
1595 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1598 for (i = 0; i < nr_pages; i++) {
1599 struct page *page = pvec.pages[i];
1600 if (page->index > end)
1602 BUG_ON(!PageLocked(page));
1603 BUG_ON(PageWriteback(page));
1604 block_invalidatepage(page, 0);
1605 ClearPageUptodate(page);
1608 index = pvec.pages[nr_pages - 1]->index + 1;
1609 pagevec_release(&pvec);
1614 static void ext4_print_free_blocks(struct inode *inode)
1616 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1617 struct super_block *sb = inode->i_sb;
1618 struct ext4_inode_info *ei = EXT4_I(inode);
1620 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1621 EXT4_C2B(EXT4_SB(inode->i_sb),
1622 ext4_count_free_clusters(sb)));
1623 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1624 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1625 (long long) EXT4_C2B(EXT4_SB(sb),
1626 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1627 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1628 (long long) EXT4_C2B(EXT4_SB(sb),
1629 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1630 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1631 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1632 ei->i_reserved_data_blocks);
1633 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1634 ei->i_reserved_meta_blocks);
1635 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1636 ei->i_allocated_meta_blocks);
1641 * mpage_da_map_and_submit - go through given space, map them
1642 * if necessary, and then submit them for I/O
1644 * @mpd - bh describing space
1646 * The function skips space we know is already mapped to disk blocks.
1649 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1651 int err, blks, get_blocks_flags;
1652 struct ext4_map_blocks map, *mapp = NULL;
1653 sector_t next = mpd->b_blocknr;
1654 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1655 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1656 handle_t *handle = NULL;
1659 * If the blocks are mapped already, or we couldn't accumulate
1660 * any blocks, then proceed immediately to the submission stage.
1662 if ((mpd->b_size == 0) ||
1663 ((mpd->b_state & (1 << BH_Mapped)) &&
1664 !(mpd->b_state & (1 << BH_Delay)) &&
1665 !(mpd->b_state & (1 << BH_Unwritten))))
1668 handle = ext4_journal_current_handle();
1672 * Call ext4_map_blocks() to allocate any delayed allocation
1673 * blocks, or to convert an uninitialized extent to be
1674 * initialized (in the case where we have written into
1675 * one or more preallocated blocks).
1677 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1678 * indicate that we are on the delayed allocation path. This
1679 * affects functions in many different parts of the allocation
1680 * call path. This flag exists primarily because we don't
1681 * want to change *many* call functions, so ext4_map_blocks()
1682 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1683 * inode's allocation semaphore is taken.
1685 * If the blocks in questions were delalloc blocks, set
1686 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1687 * variables are updated after the blocks have been allocated.
1690 map.m_len = max_blocks;
1691 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1692 if (ext4_should_dioread_nolock(mpd->inode))
1693 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1694 if (mpd->b_state & (1 << BH_Delay))
1695 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1697 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1699 struct super_block *sb = mpd->inode->i_sb;
1703 * If get block returns EAGAIN or ENOSPC and there
1704 * appears to be free blocks we will just let
1705 * mpage_da_submit_io() unlock all of the pages.
1710 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1716 * get block failure will cause us to loop in
1717 * writepages, because a_ops->writepage won't be able
1718 * to make progress. The page will be redirtied by
1719 * writepage and writepages will again try to write
1722 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1723 ext4_msg(sb, KERN_CRIT,
1724 "delayed block allocation failed for inode %lu "
1725 "at logical offset %llu with max blocks %zd "
1726 "with error %d", mpd->inode->i_ino,
1727 (unsigned long long) next,
1728 mpd->b_size >> mpd->inode->i_blkbits, err);
1729 ext4_msg(sb, KERN_CRIT,
1730 "This should not happen!! Data will be lost");
1732 ext4_print_free_blocks(mpd->inode);
1734 /* invalidate all the pages */
1735 ext4_da_block_invalidatepages(mpd);
1737 /* Mark this page range as having been completed */
1744 if (map.m_flags & EXT4_MAP_NEW) {
1745 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1748 for (i = 0; i < map.m_len; i++)
1749 unmap_underlying_metadata(bdev, map.m_pblk + i);
1753 * Update on-disk size along with block allocation.
1755 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1756 if (disksize > i_size_read(mpd->inode))
1757 disksize = i_size_read(mpd->inode);
1758 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1759 ext4_update_i_disksize(mpd->inode, disksize);
1760 err = ext4_mark_inode_dirty(handle, mpd->inode);
1762 ext4_error(mpd->inode->i_sb,
1763 "Failed to mark inode %lu dirty",
1768 mpage_da_submit_io(mpd, mapp);
1772 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1773 (1 << BH_Delay) | (1 << BH_Unwritten))
1776 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1778 * @mpd->lbh - extent of blocks
1779 * @logical - logical number of the block in the file
1780 * @b_state - b_state of the buffer head added
1782 * the function is used to collect contig. blocks in same state
1784 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1785 unsigned long b_state)
1788 int blkbits = mpd->inode->i_blkbits;
1789 int nrblocks = mpd->b_size >> blkbits;
1792 * XXX Don't go larger than mballoc is willing to allocate
1793 * This is a stopgap solution. We eventually need to fold
1794 * mpage_da_submit_io() into this function and then call
1795 * ext4_map_blocks() multiple times in a loop
1797 if (nrblocks >= (8*1024*1024 >> blkbits))
1800 /* check if the reserved journal credits might overflow */
1801 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1802 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1804 * With non-extent format we are limited by the journal
1805 * credit available. Total credit needed to insert
1806 * nrblocks contiguous blocks is dependent on the
1807 * nrblocks. So limit nrblocks.
1813 * First block in the extent
1815 if (mpd->b_size == 0) {
1816 mpd->b_blocknr = logical;
1817 mpd->b_size = 1 << blkbits;
1818 mpd->b_state = b_state & BH_FLAGS;
1822 next = mpd->b_blocknr + nrblocks;
1824 * Can we merge the block to our big extent?
1826 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1827 mpd->b_size += 1 << blkbits;
1833 * We couldn't merge the block to our extent, so we
1834 * need to flush current extent and start new one
1836 mpage_da_map_and_submit(mpd);
1840 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1842 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1846 * This function is grabs code from the very beginning of
1847 * ext4_map_blocks, but assumes that the caller is from delayed write
1848 * time. This function looks up the requested blocks and sets the
1849 * buffer delay bit under the protection of i_data_sem.
1851 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1852 struct ext4_map_blocks *map,
1853 struct buffer_head *bh)
1855 struct extent_status es;
1857 sector_t invalid_block = ~((sector_t) 0xffff);
1858 #ifdef ES_AGGRESSIVE_TEST
1859 struct ext4_map_blocks orig_map;
1861 memcpy(&orig_map, map, sizeof(*map));
1864 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1868 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1869 "logical block %lu\n", inode->i_ino, map->m_len,
1870 (unsigned long) map->m_lblk);
1872 /* Lookup extent status tree firstly */
1873 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1875 if (ext4_es_is_hole(&es)) {
1877 down_read((&EXT4_I(inode)->i_data_sem));
1882 * Delayed extent could be allocated by fallocate.
1883 * So we need to check it.
1885 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1886 map_bh(bh, inode->i_sb, invalid_block);
1888 set_buffer_delay(bh);
1892 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1893 retval = es.es_len - (iblock - es.es_lblk);
1894 if (retval > map->m_len)
1895 retval = map->m_len;
1896 map->m_len = retval;
1897 if (ext4_es_is_written(&es))
1898 map->m_flags |= EXT4_MAP_MAPPED;
1899 else if (ext4_es_is_unwritten(&es))
1900 map->m_flags |= EXT4_MAP_UNWRITTEN;
1904 #ifdef ES_AGGRESSIVE_TEST
1905 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1911 * Try to see if we can get the block without requesting a new
1912 * file system block.
1914 down_read((&EXT4_I(inode)->i_data_sem));
1915 if (ext4_has_inline_data(inode)) {
1917 * We will soon create blocks for this page, and let
1918 * us pretend as if the blocks aren't allocated yet.
1919 * In case of clusters, we have to handle the work
1920 * of mapping from cluster so that the reserved space
1921 * is calculated properly.
1923 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1924 ext4_find_delalloc_cluster(inode, map->m_lblk))
1925 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1927 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1928 retval = ext4_ext_map_blocks(NULL, inode, map,
1929 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1931 retval = ext4_ind_map_blocks(NULL, inode, map,
1932 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1938 * XXX: __block_prepare_write() unmaps passed block,
1942 * If the block was allocated from previously allocated cluster,
1943 * then we don't need to reserve it again. However we still need
1944 * to reserve metadata for every block we're going to write.
1946 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1947 ret = ext4_da_reserve_space(inode, iblock);
1949 /* not enough space to reserve */
1954 ret = ext4_da_reserve_metadata(inode, iblock);
1956 /* not enough space to reserve */
1962 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1963 ~0, EXTENT_STATUS_DELAYED);
1969 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1970 * and it should not appear on the bh->b_state.
1972 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1974 map_bh(bh, inode->i_sb, invalid_block);
1976 set_buffer_delay(bh);
1977 } else if (retval > 0) {
1979 unsigned long long status;
1981 #ifdef ES_AGGRESSIVE_TEST
1982 if (retval != map->m_len) {
1983 printk("ES len assertation failed for inode: %lu "
1984 "retval %d != map->m_len %d "
1985 "in %s (lookup)\n", inode->i_ino, retval,
1986 map->m_len, __func__);
1990 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1991 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1992 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1993 map->m_pblk, status);
1999 up_read((&EXT4_I(inode)->i_data_sem));
2005 * This is a special get_blocks_t callback which is used by
2006 * ext4_da_write_begin(). It will either return mapped block or
2007 * reserve space for a single block.
2009 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2010 * We also have b_blocknr = -1 and b_bdev initialized properly
2012 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2013 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2014 * initialized properly.
2016 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2017 struct buffer_head *bh, int create)
2019 struct ext4_map_blocks map;
2022 BUG_ON(create == 0);
2023 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2025 map.m_lblk = iblock;
2029 * first, we need to know whether the block is allocated already
2030 * preallocated blocks are unmapped but should treated
2031 * the same as allocated blocks.
2033 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2037 map_bh(bh, inode->i_sb, map.m_pblk);
2038 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2040 if (buffer_unwritten(bh)) {
2041 /* A delayed write to unwritten bh should be marked
2042 * new and mapped. Mapped ensures that we don't do
2043 * get_block multiple times when we write to the same
2044 * offset and new ensures that we do proper zero out
2045 * for partial write.
2048 set_buffer_mapped(bh);
2053 static int bget_one(handle_t *handle, struct buffer_head *bh)
2059 static int bput_one(handle_t *handle, struct buffer_head *bh)
2065 static int __ext4_journalled_writepage(struct page *page,
2068 struct address_space *mapping = page->mapping;
2069 struct inode *inode = mapping->host;
2070 struct buffer_head *page_bufs = NULL;
2071 handle_t *handle = NULL;
2072 int ret = 0, err = 0;
2073 int inline_data = ext4_has_inline_data(inode);
2074 struct buffer_head *inode_bh = NULL;
2076 ClearPageChecked(page);
2079 BUG_ON(page->index != 0);
2080 BUG_ON(len > ext4_get_max_inline_size(inode));
2081 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2082 if (inode_bh == NULL)
2085 page_bufs = page_buffers(page);
2090 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2093 /* As soon as we unlock the page, it can go away, but we have
2094 * references to buffers so we are safe */
2097 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2098 ext4_writepage_trans_blocks(inode));
2099 if (IS_ERR(handle)) {
2100 ret = PTR_ERR(handle);
2104 BUG_ON(!ext4_handle_valid(handle));
2107 ret = ext4_journal_get_write_access(handle, inode_bh);
2109 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2112 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2113 do_journal_get_write_access);
2115 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2120 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2121 err = ext4_journal_stop(handle);
2125 if (!ext4_has_inline_data(inode))
2126 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2128 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2135 * Note that we don't need to start a transaction unless we're journaling data
2136 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2137 * need to file the inode to the transaction's list in ordered mode because if
2138 * we are writing back data added by write(), the inode is already there and if
2139 * we are writing back data modified via mmap(), no one guarantees in which
2140 * transaction the data will hit the disk. In case we are journaling data, we
2141 * cannot start transaction directly because transaction start ranks above page
2142 * lock so we have to do some magic.
2144 * This function can get called via...
2145 * - ext4_da_writepages after taking page lock (have journal handle)
2146 * - journal_submit_inode_data_buffers (no journal handle)
2147 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2148 * - grab_page_cache when doing write_begin (have journal handle)
2150 * We don't do any block allocation in this function. If we have page with
2151 * multiple blocks we need to write those buffer_heads that are mapped. This
2152 * is important for mmaped based write. So if we do with blocksize 1K
2153 * truncate(f, 1024);
2154 * a = mmap(f, 0, 4096);
2156 * truncate(f, 4096);
2157 * we have in the page first buffer_head mapped via page_mkwrite call back
2158 * but other buffer_heads would be unmapped but dirty (dirty done via the
2159 * do_wp_page). So writepage should write the first block. If we modify
2160 * the mmap area beyond 1024 we will again get a page_fault and the
2161 * page_mkwrite callback will do the block allocation and mark the
2162 * buffer_heads mapped.
2164 * We redirty the page if we have any buffer_heads that is either delay or
2165 * unwritten in the page.
2167 * We can get recursively called as show below.
2169 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2172 * But since we don't do any block allocation we should not deadlock.
2173 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2175 static int ext4_writepage(struct page *page,
2176 struct writeback_control *wbc)
2181 struct buffer_head *page_bufs = NULL;
2182 struct inode *inode = page->mapping->host;
2183 struct ext4_io_submit io_submit;
2185 trace_ext4_writepage(page);
2186 size = i_size_read(inode);
2187 if (page->index == size >> PAGE_CACHE_SHIFT)
2188 len = size & ~PAGE_CACHE_MASK;
2190 len = PAGE_CACHE_SIZE;
2192 page_bufs = page_buffers(page);
2194 * We cannot do block allocation or other extent handling in this
2195 * function. If there are buffers needing that, we have to redirty
2196 * the page. But we may reach here when we do a journal commit via
2197 * journal_submit_inode_data_buffers() and in that case we must write
2198 * allocated buffers to achieve data=ordered mode guarantees.
2200 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2201 ext4_bh_delay_or_unwritten)) {
2202 redirty_page_for_writepage(wbc, page);
2203 if (current->flags & PF_MEMALLOC) {
2205 * For memory cleaning there's no point in writing only
2206 * some buffers. So just bail out. Warn if we came here
2207 * from direct reclaim.
2209 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2216 if (PageChecked(page) && ext4_should_journal_data(inode))
2218 * It's mmapped pagecache. Add buffers and journal it. There
2219 * doesn't seem much point in redirtying the page here.
2221 return __ext4_journalled_writepage(page, len);
2223 memset(&io_submit, 0, sizeof(io_submit));
2224 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2225 ext4_io_submit(&io_submit);
2230 * This is called via ext4_da_writepages() to
2231 * calculate the total number of credits to reserve to fit
2232 * a single extent allocation into a single transaction,
2233 * ext4_da_writpeages() will loop calling this before
2234 * the block allocation.
2237 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2239 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2242 * With non-extent format the journal credit needed to
2243 * insert nrblocks contiguous block is dependent on
2244 * number of contiguous block. So we will limit
2245 * number of contiguous block to a sane value
2247 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2248 (max_blocks > EXT4_MAX_TRANS_DATA))
2249 max_blocks = EXT4_MAX_TRANS_DATA;
2251 return ext4_chunk_trans_blocks(inode, max_blocks);
2255 * write_cache_pages_da - walk the list of dirty pages of the given
2256 * address space and accumulate pages that need writing, and call
2257 * mpage_da_map_and_submit to map a single contiguous memory region
2258 * and then write them.
2260 static int write_cache_pages_da(handle_t *handle,
2261 struct address_space *mapping,
2262 struct writeback_control *wbc,
2263 struct mpage_da_data *mpd,
2264 pgoff_t *done_index)
2266 struct buffer_head *bh, *head;
2267 struct inode *inode = mapping->host;
2268 struct pagevec pvec;
2269 unsigned int nr_pages;
2272 long nr_to_write = wbc->nr_to_write;
2273 int i, tag, ret = 0;
2275 memset(mpd, 0, sizeof(struct mpage_da_data));
2278 pagevec_init(&pvec, 0);
2279 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2280 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2282 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2283 tag = PAGECACHE_TAG_TOWRITE;
2285 tag = PAGECACHE_TAG_DIRTY;
2287 *done_index = index;
2288 while (index <= end) {
2289 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2290 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2294 for (i = 0; i < nr_pages; i++) {
2295 struct page *page = pvec.pages[i];
2298 * At this point, the page may be truncated or
2299 * invalidated (changing page->mapping to NULL), or
2300 * even swizzled back from swapper_space to tmpfs file
2301 * mapping. However, page->index will not change
2302 * because we have a reference on the page.
2304 if (page->index > end)
2307 *done_index = page->index + 1;
2310 * If we can't merge this page, and we have
2311 * accumulated an contiguous region, write it
2313 if ((mpd->next_page != page->index) &&
2314 (mpd->next_page != mpd->first_page)) {
2315 mpage_da_map_and_submit(mpd);
2316 goto ret_extent_tail;
2322 * If the page is no longer dirty, or its
2323 * mapping no longer corresponds to inode we
2324 * are writing (which means it has been
2325 * truncated or invalidated), or the page is
2326 * already under writeback and we are not
2327 * doing a data integrity writeback, skip the page
2329 if (!PageDirty(page) ||
2330 (PageWriteback(page) &&
2331 (wbc->sync_mode == WB_SYNC_NONE)) ||
2332 unlikely(page->mapping != mapping)) {
2337 wait_on_page_writeback(page);
2338 BUG_ON(PageWriteback(page));
2341 * If we have inline data and arrive here, it means that
2342 * we will soon create the block for the 1st page, so
2343 * we'd better clear the inline data here.
2345 if (ext4_has_inline_data(inode)) {
2346 BUG_ON(ext4_test_inode_state(inode,
2347 EXT4_STATE_MAY_INLINE_DATA));
2348 ext4_destroy_inline_data(handle, inode);
2351 if (mpd->next_page != page->index)
2352 mpd->first_page = page->index;
2353 mpd->next_page = page->index + 1;
2354 logical = (sector_t) page->index <<
2355 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2357 /* Add all dirty buffers to mpd */
2358 head = page_buffers(page);
2361 BUG_ON(buffer_locked(bh));
2363 * We need to try to allocate unmapped blocks
2364 * in the same page. Otherwise we won't make
2365 * progress with the page in ext4_writepage
2367 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2368 mpage_add_bh_to_extent(mpd, logical,
2371 goto ret_extent_tail;
2372 } else if (buffer_dirty(bh) &&
2373 buffer_mapped(bh)) {
2375 * mapped dirty buffer. We need to
2376 * update the b_state because we look
2377 * at b_state in mpage_da_map_blocks.
2378 * We don't update b_size because if we
2379 * find an unmapped buffer_head later
2380 * we need to use the b_state flag of
2383 if (mpd->b_size == 0)
2385 bh->b_state & BH_FLAGS;
2388 } while ((bh = bh->b_this_page) != head);
2390 if (nr_to_write > 0) {
2392 if (nr_to_write == 0 &&
2393 wbc->sync_mode == WB_SYNC_NONE)
2395 * We stop writing back only if we are
2396 * not doing integrity sync. In case of
2397 * integrity sync we have to keep going
2398 * because someone may be concurrently
2399 * dirtying pages, and we might have
2400 * synced a lot of newly appeared dirty
2401 * pages, but have not synced all of the
2407 pagevec_release(&pvec);
2412 ret = MPAGE_DA_EXTENT_TAIL;
2414 pagevec_release(&pvec);
2420 static int ext4_da_writepages(struct address_space *mapping,
2421 struct writeback_control *wbc)
2424 int range_whole = 0;
2425 handle_t *handle = NULL;
2426 struct mpage_da_data mpd;
2427 struct inode *inode = mapping->host;
2428 int pages_written = 0;
2429 unsigned int max_pages;
2430 int range_cyclic, cycled = 1, io_done = 0;
2431 int needed_blocks, ret = 0;
2432 long desired_nr_to_write, nr_to_writebump = 0;
2433 loff_t range_start = wbc->range_start;
2434 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2435 pgoff_t done_index = 0;
2437 struct blk_plug plug;
2439 trace_ext4_da_writepages(inode, wbc);
2442 * No pages to write? This is mainly a kludge to avoid starting
2443 * a transaction for special inodes like journal inode on last iput()
2444 * because that could violate lock ordering on umount
2446 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2450 * If the filesystem has aborted, it is read-only, so return
2451 * right away instead of dumping stack traces later on that
2452 * will obscure the real source of the problem. We test
2453 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2454 * the latter could be true if the filesystem is mounted
2455 * read-only, and in that case, ext4_da_writepages should
2456 * *never* be called, so if that ever happens, we would want
2459 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2462 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2465 range_cyclic = wbc->range_cyclic;
2466 if (wbc->range_cyclic) {
2467 index = mapping->writeback_index;
2470 wbc->range_start = index << PAGE_CACHE_SHIFT;
2471 wbc->range_end = LLONG_MAX;
2472 wbc->range_cyclic = 0;
2475 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2476 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2480 * This works around two forms of stupidity. The first is in
2481 * the writeback code, which caps the maximum number of pages
2482 * written to be 1024 pages. This is wrong on multiple
2483 * levels; different architectues have a different page size,
2484 * which changes the maximum amount of data which gets
2485 * written. Secondly, 4 megabytes is way too small. XFS
2486 * forces this value to be 16 megabytes by multiplying
2487 * nr_to_write parameter by four, and then relies on its
2488 * allocator to allocate larger extents to make them
2489 * contiguous. Unfortunately this brings us to the second
2490 * stupidity, which is that ext4's mballoc code only allocates
2491 * at most 2048 blocks. So we force contiguous writes up to
2492 * the number of dirty blocks in the inode, or
2493 * sbi->max_writeback_mb_bump whichever is smaller.
2495 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2496 if (!range_cyclic && range_whole) {
2497 if (wbc->nr_to_write == LONG_MAX)
2498 desired_nr_to_write = wbc->nr_to_write;
2500 desired_nr_to_write = wbc->nr_to_write * 8;
2502 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2504 if (desired_nr_to_write > max_pages)
2505 desired_nr_to_write = max_pages;
2507 if (wbc->nr_to_write < desired_nr_to_write) {
2508 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2509 wbc->nr_to_write = desired_nr_to_write;
2513 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2514 tag_pages_for_writeback(mapping, index, end);
2516 blk_start_plug(&plug);
2517 while (!ret && wbc->nr_to_write > 0) {
2520 * we insert one extent at a time. So we need
2521 * credit needed for single extent allocation.
2522 * journalled mode is currently not supported
2525 BUG_ON(ext4_should_journal_data(inode));
2526 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2528 /* start a new transaction*/
2529 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2531 if (IS_ERR(handle)) {
2532 ret = PTR_ERR(handle);
2533 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2534 "%ld pages, ino %lu; err %d", __func__,
2535 wbc->nr_to_write, inode->i_ino, ret);
2536 blk_finish_plug(&plug);
2537 goto out_writepages;
2541 * Now call write_cache_pages_da() to find the next
2542 * contiguous region of logical blocks that need
2543 * blocks to be allocated by ext4 and submit them.
2545 ret = write_cache_pages_da(handle, mapping,
2546 wbc, &mpd, &done_index);
2548 * If we have a contiguous extent of pages and we
2549 * haven't done the I/O yet, map the blocks and submit
2552 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2553 mpage_da_map_and_submit(&mpd);
2554 ret = MPAGE_DA_EXTENT_TAIL;
2556 trace_ext4_da_write_pages(inode, &mpd);
2557 wbc->nr_to_write -= mpd.pages_written;
2559 ext4_journal_stop(handle);
2561 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2562 /* commit the transaction which would
2563 * free blocks released in the transaction
2566 jbd2_journal_force_commit_nested(sbi->s_journal);
2568 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2570 * Got one extent now try with rest of the pages.
2571 * If mpd.retval is set -EIO, journal is aborted.
2572 * So we don't need to write any more.
2574 pages_written += mpd.pages_written;
2577 } else if (wbc->nr_to_write)
2579 * There is no more writeout needed
2580 * or we requested for a noblocking writeout
2581 * and we found the device congested
2585 blk_finish_plug(&plug);
2586 if (!io_done && !cycled) {
2589 wbc->range_start = index << PAGE_CACHE_SHIFT;
2590 wbc->range_end = mapping->writeback_index - 1;
2595 wbc->range_cyclic = range_cyclic;
2596 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2598 * set the writeback_index so that range_cyclic
2599 * mode will write it back later
2601 mapping->writeback_index = done_index;
2604 wbc->nr_to_write -= nr_to_writebump;
2605 wbc->range_start = range_start;
2606 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2610 static int ext4_nonda_switch(struct super_block *sb)
2612 s64 free_clusters, dirty_clusters;
2613 struct ext4_sb_info *sbi = EXT4_SB(sb);
2616 * switch to non delalloc mode if we are running low
2617 * on free block. The free block accounting via percpu
2618 * counters can get slightly wrong with percpu_counter_batch getting
2619 * accumulated on each CPU without updating global counters
2620 * Delalloc need an accurate free block accounting. So switch
2621 * to non delalloc when we are near to error range.
2624 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2626 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2628 * Start pushing delalloc when 1/2 of free blocks are dirty.
2630 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2631 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2633 if (2 * free_clusters < 3 * dirty_clusters ||
2634 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2636 * free block count is less than 150% of dirty blocks
2637 * or free blocks is less than watermark
2644 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2645 loff_t pos, unsigned len, unsigned flags,
2646 struct page **pagep, void **fsdata)
2648 int ret, retries = 0;
2651 struct inode *inode = mapping->host;
2654 index = pos >> PAGE_CACHE_SHIFT;
2656 if (ext4_nonda_switch(inode->i_sb)) {
2657 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2658 return ext4_write_begin(file, mapping, pos,
2659 len, flags, pagep, fsdata);
2661 *fsdata = (void *)0;
2662 trace_ext4_da_write_begin(inode, pos, len, flags);
2664 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2665 ret = ext4_da_write_inline_data_begin(mapping, inode,
2675 * grab_cache_page_write_begin() can take a long time if the
2676 * system is thrashing due to memory pressure, or if the page
2677 * is being written back. So grab it first before we start
2678 * the transaction handle. This also allows us to allocate
2679 * the page (if needed) without using GFP_NOFS.
2682 page = grab_cache_page_write_begin(mapping, index, flags);
2688 * With delayed allocation, we don't log the i_disksize update
2689 * if there is delayed block allocation. But we still need
2690 * to journalling the i_disksize update if writes to the end
2691 * of file which has an already mapped buffer.
2694 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2695 if (IS_ERR(handle)) {
2696 page_cache_release(page);
2697 return PTR_ERR(handle);
2701 if (page->mapping != mapping) {
2702 /* The page got truncated from under us */
2704 page_cache_release(page);
2705 ext4_journal_stop(handle);
2708 /* In case writeback began while the page was unlocked */
2709 wait_on_page_writeback(page);
2711 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2714 ext4_journal_stop(handle);
2716 * block_write_begin may have instantiated a few blocks
2717 * outside i_size. Trim these off again. Don't need
2718 * i_size_read because we hold i_mutex.
2720 if (pos + len > inode->i_size)
2721 ext4_truncate_failed_write(inode);
2723 if (ret == -ENOSPC &&
2724 ext4_should_retry_alloc(inode->i_sb, &retries))
2727 page_cache_release(page);
2736 * Check if we should update i_disksize
2737 * when write to the end of file but not require block allocation
2739 static int ext4_da_should_update_i_disksize(struct page *page,
2740 unsigned long offset)
2742 struct buffer_head *bh;
2743 struct inode *inode = page->mapping->host;
2747 bh = page_buffers(page);
2748 idx = offset >> inode->i_blkbits;
2750 for (i = 0; i < idx; i++)
2751 bh = bh->b_this_page;
2753 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2758 static int ext4_da_write_end(struct file *file,
2759 struct address_space *mapping,
2760 loff_t pos, unsigned len, unsigned copied,
2761 struct page *page, void *fsdata)
2763 struct inode *inode = mapping->host;
2765 handle_t *handle = ext4_journal_current_handle();
2767 unsigned long start, end;
2768 int write_mode = (int)(unsigned long)fsdata;
2770 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2771 return ext4_write_end(file, mapping, pos,
2772 len, copied, page, fsdata);
2774 trace_ext4_da_write_end(inode, pos, len, copied);
2775 start = pos & (PAGE_CACHE_SIZE - 1);
2776 end = start + copied - 1;
2779 * generic_write_end() will run mark_inode_dirty() if i_size
2780 * changes. So let's piggyback the i_disksize mark_inode_dirty
2783 new_i_size = pos + copied;
2784 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2785 if (ext4_has_inline_data(inode) ||
2786 ext4_da_should_update_i_disksize(page, end)) {
2787 down_write(&EXT4_I(inode)->i_data_sem);
2788 if (new_i_size > EXT4_I(inode)->i_disksize)
2789 EXT4_I(inode)->i_disksize = new_i_size;
2790 up_write(&EXT4_I(inode)->i_data_sem);
2791 /* We need to mark inode dirty even if
2792 * new_i_size is less that inode->i_size
2793 * bu greater than i_disksize.(hint delalloc)
2795 ext4_mark_inode_dirty(handle, inode);
2799 if (write_mode != CONVERT_INLINE_DATA &&
2800 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2801 ext4_has_inline_data(inode))
2802 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2805 ret2 = generic_write_end(file, mapping, pos, len, copied,
2811 ret2 = ext4_journal_stop(handle);
2815 return ret ? ret : copied;
2818 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2821 * Drop reserved blocks
2823 BUG_ON(!PageLocked(page));
2824 if (!page_has_buffers(page))
2827 ext4_da_page_release_reservation(page, offset);
2830 ext4_invalidatepage(page, offset);
2836 * Force all delayed allocation blocks to be allocated for a given inode.
2838 int ext4_alloc_da_blocks(struct inode *inode)
2840 trace_ext4_alloc_da_blocks(inode);
2842 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2843 !EXT4_I(inode)->i_reserved_meta_blocks)
2847 * We do something simple for now. The filemap_flush() will
2848 * also start triggering a write of the data blocks, which is
2849 * not strictly speaking necessary (and for users of
2850 * laptop_mode, not even desirable). However, to do otherwise
2851 * would require replicating code paths in:
2853 * ext4_da_writepages() ->
2854 * write_cache_pages() ---> (via passed in callback function)
2855 * __mpage_da_writepage() -->
2856 * mpage_add_bh_to_extent()
2857 * mpage_da_map_blocks()
2859 * The problem is that write_cache_pages(), located in
2860 * mm/page-writeback.c, marks pages clean in preparation for
2861 * doing I/O, which is not desirable if we're not planning on
2864 * We could call write_cache_pages(), and then redirty all of
2865 * the pages by calling redirty_page_for_writepage() but that
2866 * would be ugly in the extreme. So instead we would need to
2867 * replicate parts of the code in the above functions,
2868 * simplifying them because we wouldn't actually intend to
2869 * write out the pages, but rather only collect contiguous
2870 * logical block extents, call the multi-block allocator, and
2871 * then update the buffer heads with the block allocations.
2873 * For now, though, we'll cheat by calling filemap_flush(),
2874 * which will map the blocks, and start the I/O, but not
2875 * actually wait for the I/O to complete.
2877 return filemap_flush(inode->i_mapping);
2881 * bmap() is special. It gets used by applications such as lilo and by
2882 * the swapper to find the on-disk block of a specific piece of data.
2884 * Naturally, this is dangerous if the block concerned is still in the
2885 * journal. If somebody makes a swapfile on an ext4 data-journaling
2886 * filesystem and enables swap, then they may get a nasty shock when the
2887 * data getting swapped to that swapfile suddenly gets overwritten by
2888 * the original zero's written out previously to the journal and
2889 * awaiting writeback in the kernel's buffer cache.
2891 * So, if we see any bmap calls here on a modified, data-journaled file,
2892 * take extra steps to flush any blocks which might be in the cache.
2894 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2896 struct inode *inode = mapping->host;
2901 * We can get here for an inline file via the FIBMAP ioctl
2903 if (ext4_has_inline_data(inode))
2906 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2907 test_opt(inode->i_sb, DELALLOC)) {
2909 * With delalloc we want to sync the file
2910 * so that we can make sure we allocate
2913 filemap_write_and_wait(mapping);
2916 if (EXT4_JOURNAL(inode) &&
2917 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2919 * This is a REALLY heavyweight approach, but the use of
2920 * bmap on dirty files is expected to be extremely rare:
2921 * only if we run lilo or swapon on a freshly made file
2922 * do we expect this to happen.
2924 * (bmap requires CAP_SYS_RAWIO so this does not
2925 * represent an unprivileged user DOS attack --- we'd be
2926 * in trouble if mortal users could trigger this path at
2929 * NB. EXT4_STATE_JDATA is not set on files other than
2930 * regular files. If somebody wants to bmap a directory
2931 * or symlink and gets confused because the buffer
2932 * hasn't yet been flushed to disk, they deserve
2933 * everything they get.
2936 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2937 journal = EXT4_JOURNAL(inode);
2938 jbd2_journal_lock_updates(journal);
2939 err = jbd2_journal_flush(journal);
2940 jbd2_journal_unlock_updates(journal);
2946 return generic_block_bmap(mapping, block, ext4_get_block);
2949 static int ext4_readpage(struct file *file, struct page *page)
2952 struct inode *inode = page->mapping->host;
2954 trace_ext4_readpage(page);
2956 if (ext4_has_inline_data(inode))
2957 ret = ext4_readpage_inline(inode, page);
2960 return mpage_readpage(page, ext4_get_block);
2966 ext4_readpages(struct file *file, struct address_space *mapping,
2967 struct list_head *pages, unsigned nr_pages)
2969 struct inode *inode = mapping->host;
2971 /* If the file has inline data, no need to do readpages. */
2972 if (ext4_has_inline_data(inode))
2975 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2978 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2980 trace_ext4_invalidatepage(page, offset);
2982 /* No journalling happens on data buffers when this function is used */
2983 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2985 block_invalidatepage(page, offset);
2988 static int __ext4_journalled_invalidatepage(struct page *page,
2989 unsigned long offset)
2991 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2993 trace_ext4_journalled_invalidatepage(page, offset);
2996 * If it's a full truncate we just forget about the pending dirtying
2999 ClearPageChecked(page);
3001 return jbd2_journal_invalidatepage(journal, page, offset);
3004 /* Wrapper for aops... */
3005 static void ext4_journalled_invalidatepage(struct page *page,
3006 unsigned long offset)
3008 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3011 static int ext4_releasepage(struct page *page, gfp_t wait)
3013 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3015 trace_ext4_releasepage(page);
3017 /* Page has dirty journalled data -> cannot release */
3018 if (PageChecked(page))
3021 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3023 return try_to_free_buffers(page);
3027 * ext4_get_block used when preparing for a DIO write or buffer write.
3028 * We allocate an uinitialized extent if blocks haven't been allocated.
3029 * The extent will be converted to initialized after the IO is complete.
3031 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3032 struct buffer_head *bh_result, int create)
3034 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3035 inode->i_ino, create);
3036 return _ext4_get_block(inode, iblock, bh_result,
3037 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3040 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3041 struct buffer_head *bh_result, int create)
3043 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3044 inode->i_ino, create);
3045 return _ext4_get_block(inode, iblock, bh_result,
3046 EXT4_GET_BLOCKS_NO_LOCK);
3049 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3050 ssize_t size, void *private, int ret,
3053 struct inode *inode = file_inode(iocb->ki_filp);
3054 ext4_io_end_t *io_end = iocb->private;
3056 /* if not async direct IO or dio with 0 bytes write, just return */
3057 if (!io_end || !size)
3060 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3061 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3062 iocb->private, io_end->inode->i_ino, iocb, offset,
3065 iocb->private = NULL;
3067 /* if not aio dio with unwritten extents, just free io and return */
3068 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3069 ext4_free_io_end(io_end);
3071 inode_dio_done(inode);
3073 aio_complete(iocb, ret, 0);
3077 io_end->offset = offset;
3078 io_end->size = size;
3080 io_end->iocb = iocb;
3081 io_end->result = ret;
3084 ext4_add_complete_io(io_end);
3088 * For ext4 extent files, ext4 will do direct-io write to holes,
3089 * preallocated extents, and those write extend the file, no need to
3090 * fall back to buffered IO.
3092 * For holes, we fallocate those blocks, mark them as uninitialized
3093 * If those blocks were preallocated, we mark sure they are split, but
3094 * still keep the range to write as uninitialized.
3096 * The unwritten extents will be converted to written when DIO is completed.
3097 * For async direct IO, since the IO may still pending when return, we
3098 * set up an end_io call back function, which will do the conversion
3099 * when async direct IO completed.
3101 * If the O_DIRECT write will extend the file then add this inode to the
3102 * orphan list. So recovery will truncate it back to the original size
3103 * if the machine crashes during the write.
3106 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3107 const struct iovec *iov, loff_t offset,
3108 unsigned long nr_segs)
3110 struct file *file = iocb->ki_filp;
3111 struct inode *inode = file->f_mapping->host;
3113 size_t count = iov_length(iov, nr_segs);
3115 get_block_t *get_block_func = NULL;
3117 loff_t final_size = offset + count;
3119 /* Use the old path for reads and writes beyond i_size. */
3120 if (rw != WRITE || final_size > inode->i_size)
3121 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3123 BUG_ON(iocb->private == NULL);
3125 /* If we do a overwrite dio, i_mutex locking can be released */
3126 overwrite = *((int *)iocb->private);
3129 atomic_inc(&inode->i_dio_count);
3130 down_read(&EXT4_I(inode)->i_data_sem);
3131 mutex_unlock(&inode->i_mutex);
3135 * We could direct write to holes and fallocate.
3137 * Allocated blocks to fill the hole are marked as
3138 * uninitialized to prevent parallel buffered read to expose
3139 * the stale data before DIO complete the data IO.
3141 * As to previously fallocated extents, ext4 get_block will
3142 * just simply mark the buffer mapped but still keep the
3143 * extents uninitialized.
3145 * For non AIO case, we will convert those unwritten extents
3146 * to written after return back from blockdev_direct_IO.
3148 * For async DIO, the conversion needs to be deferred when the
3149 * IO is completed. The ext4 end_io callback function will be
3150 * called to take care of the conversion work. Here for async
3151 * case, we allocate an io_end structure to hook to the iocb.
3153 iocb->private = NULL;
3154 ext4_inode_aio_set(inode, NULL);
3155 if (!is_sync_kiocb(iocb)) {
3156 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3161 io_end->flag |= EXT4_IO_END_DIRECT;
3162 iocb->private = io_end;
3164 * we save the io structure for current async direct
3165 * IO, so that later ext4_map_blocks() could flag the
3166 * io structure whether there is a unwritten extents
3167 * needs to be converted when IO is completed.
3169 ext4_inode_aio_set(inode, io_end);
3173 get_block_func = ext4_get_block_write_nolock;
3175 get_block_func = ext4_get_block_write;
3176 dio_flags = DIO_LOCKING;
3178 ret = __blockdev_direct_IO(rw, iocb, inode,
3179 inode->i_sb->s_bdev, iov,
3187 ext4_inode_aio_set(inode, NULL);
3189 * The io_end structure takes a reference to the inode, that
3190 * structure needs to be destroyed and the reference to the
3191 * inode need to be dropped, when IO is complete, even with 0
3192 * byte write, or failed.
3194 * In the successful AIO DIO case, the io_end structure will
3195 * be destroyed and the reference to the inode will be dropped
3196 * after the end_io call back function is called.
3198 * In the case there is 0 byte write, or error case, since VFS
3199 * direct IO won't invoke the end_io call back function, we
3200 * need to free the end_io structure here.
3202 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3203 ext4_free_io_end(iocb->private);
3204 iocb->private = NULL;
3205 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3206 EXT4_STATE_DIO_UNWRITTEN)) {
3209 * for non AIO case, since the IO is already
3210 * completed, we could do the conversion right here
3212 err = ext4_convert_unwritten_extents(inode,
3216 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3220 /* take i_mutex locking again if we do a ovewrite dio */
3222 inode_dio_done(inode);
3223 up_read(&EXT4_I(inode)->i_data_sem);
3224 mutex_lock(&inode->i_mutex);
3230 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3231 const struct iovec *iov, loff_t offset,
3232 unsigned long nr_segs)
3234 struct file *file = iocb->ki_filp;
3235 struct inode *inode = file->f_mapping->host;
3239 * If we are doing data journalling we don't support O_DIRECT
3241 if (ext4_should_journal_data(inode))
3244 /* Let buffer I/O handle the inline data case. */
3245 if (ext4_has_inline_data(inode))
3248 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3249 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3250 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3252 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3253 trace_ext4_direct_IO_exit(inode, offset,
3254 iov_length(iov, nr_segs), rw, ret);
3259 * Pages can be marked dirty completely asynchronously from ext4's journalling
3260 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3261 * much here because ->set_page_dirty is called under VFS locks. The page is
3262 * not necessarily locked.
3264 * We cannot just dirty the page and leave attached buffers clean, because the
3265 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3266 * or jbddirty because all the journalling code will explode.
3268 * So what we do is to mark the page "pending dirty" and next time writepage
3269 * is called, propagate that into the buffers appropriately.
3271 static int ext4_journalled_set_page_dirty(struct page *page)
3273 SetPageChecked(page);
3274 return __set_page_dirty_nobuffers(page);
3277 static const struct address_space_operations ext4_aops = {
3278 .readpage = ext4_readpage,
3279 .readpages = ext4_readpages,
3280 .writepage = ext4_writepage,
3281 .write_begin = ext4_write_begin,
3282 .write_end = ext4_write_end,
3284 .invalidatepage = ext4_invalidatepage,
3285 .releasepage = ext4_releasepage,
3286 .direct_IO = ext4_direct_IO,
3287 .migratepage = buffer_migrate_page,
3288 .is_partially_uptodate = block_is_partially_uptodate,
3289 .error_remove_page = generic_error_remove_page,
3292 static const struct address_space_operations ext4_journalled_aops = {
3293 .readpage = ext4_readpage,
3294 .readpages = ext4_readpages,
3295 .writepage = ext4_writepage,
3296 .write_begin = ext4_write_begin,
3297 .write_end = ext4_journalled_write_end,
3298 .set_page_dirty = ext4_journalled_set_page_dirty,
3300 .invalidatepage = ext4_journalled_invalidatepage,
3301 .releasepage = ext4_releasepage,
3302 .direct_IO = ext4_direct_IO,
3303 .is_partially_uptodate = block_is_partially_uptodate,
3304 .error_remove_page = generic_error_remove_page,
3307 static const struct address_space_operations ext4_da_aops = {
3308 .readpage = ext4_readpage,
3309 .readpages = ext4_readpages,
3310 .writepage = ext4_writepage,
3311 .writepages = ext4_da_writepages,
3312 .write_begin = ext4_da_write_begin,
3313 .write_end = ext4_da_write_end,
3315 .invalidatepage = ext4_da_invalidatepage,
3316 .releasepage = ext4_releasepage,
3317 .direct_IO = ext4_direct_IO,
3318 .migratepage = buffer_migrate_page,
3319 .is_partially_uptodate = block_is_partially_uptodate,
3320 .error_remove_page = generic_error_remove_page,
3323 void ext4_set_aops(struct inode *inode)
3325 switch (ext4_inode_journal_mode(inode)) {
3326 case EXT4_INODE_ORDERED_DATA_MODE:
3327 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3329 case EXT4_INODE_WRITEBACK_DATA_MODE:
3330 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3332 case EXT4_INODE_JOURNAL_DATA_MODE:
3333 inode->i_mapping->a_ops = &ext4_journalled_aops;
3338 if (test_opt(inode->i_sb, DELALLOC))
3339 inode->i_mapping->a_ops = &ext4_da_aops;
3341 inode->i_mapping->a_ops = &ext4_aops;
3346 * ext4_discard_partial_page_buffers()
3347 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3348 * This function finds and locks the page containing the offset
3349 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3350 * Calling functions that already have the page locked should call
3351 * ext4_discard_partial_page_buffers_no_lock directly.
3353 int ext4_discard_partial_page_buffers(handle_t *handle,
3354 struct address_space *mapping, loff_t from,
3355 loff_t length, int flags)
3357 struct inode *inode = mapping->host;
3361 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3362 mapping_gfp_mask(mapping) & ~__GFP_FS);
3366 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3367 from, length, flags);
3370 page_cache_release(page);
3375 * ext4_discard_partial_page_buffers_no_lock()
3376 * Zeros a page range of length 'length' starting from offset 'from'.
3377 * Buffer heads that correspond to the block aligned regions of the
3378 * zeroed range will be unmapped. Unblock aligned regions
3379 * will have the corresponding buffer head mapped if needed so that
3380 * that region of the page can be updated with the partial zero out.
3382 * This function assumes that the page has already been locked. The
3383 * The range to be discarded must be contained with in the given page.
3384 * If the specified range exceeds the end of the page it will be shortened
3385 * to the end of the page that corresponds to 'from'. This function is
3386 * appropriate for updating a page and it buffer heads to be unmapped and
3387 * zeroed for blocks that have been either released, or are going to be
3390 * handle: The journal handle
3391 * inode: The files inode
3392 * page: A locked page that contains the offset "from"
3393 * from: The starting byte offset (from the beginning of the file)
3394 * to begin discarding
3395 * len: The length of bytes to discard
3396 * flags: Optional flags that may be used:
3398 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3399 * Only zero the regions of the page whose buffer heads
3400 * have already been unmapped. This flag is appropriate
3401 * for updating the contents of a page whose blocks may
3402 * have already been released, and we only want to zero
3403 * out the regions that correspond to those released blocks.
3405 * Returns zero on success or negative on failure.
3407 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3408 struct inode *inode, struct page *page, loff_t from,
3409 loff_t length, int flags)
3411 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3412 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3413 unsigned int blocksize, max, pos;
3415 struct buffer_head *bh;
3418 blocksize = inode->i_sb->s_blocksize;
3419 max = PAGE_CACHE_SIZE - offset;
3421 if (index != page->index)
3425 * correct length if it does not fall between
3426 * 'from' and the end of the page
3428 if (length > max || length < 0)
3431 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3433 if (!page_has_buffers(page))
3434 create_empty_buffers(page, blocksize, 0);
3436 /* Find the buffer that contains "offset" */
3437 bh = page_buffers(page);
3439 while (offset >= pos) {
3440 bh = bh->b_this_page;
3446 while (pos < offset + length) {
3447 unsigned int end_of_block, range_to_discard;
3451 /* The length of space left to zero and unmap */
3452 range_to_discard = offset + length - pos;
3454 /* The length of space until the end of the block */
3455 end_of_block = blocksize - (pos & (blocksize-1));
3458 * Do not unmap or zero past end of block
3459 * for this buffer head
3461 if (range_to_discard > end_of_block)
3462 range_to_discard = end_of_block;
3466 * Skip this buffer head if we are only zeroing unampped
3467 * regions of the page
3469 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3473 /* If the range is block aligned, unmap */
3474 if (range_to_discard == blocksize) {
3475 clear_buffer_dirty(bh);
3477 clear_buffer_mapped(bh);
3478 clear_buffer_req(bh);
3479 clear_buffer_new(bh);
3480 clear_buffer_delay(bh);
3481 clear_buffer_unwritten(bh);
3482 clear_buffer_uptodate(bh);
3483 zero_user(page, pos, range_to_discard);
3484 BUFFER_TRACE(bh, "Buffer discarded");
3489 * If this block is not completely contained in the range
3490 * to be discarded, then it is not going to be released. Because
3491 * we need to keep this block, we need to make sure this part
3492 * of the page is uptodate before we modify it by writeing
3493 * partial zeros on it.
3495 if (!buffer_mapped(bh)) {
3497 * Buffer head must be mapped before we can read
3500 BUFFER_TRACE(bh, "unmapped");
3501 ext4_get_block(inode, iblock, bh, 0);
3502 /* unmapped? It's a hole - nothing to do */
3503 if (!buffer_mapped(bh)) {
3504 BUFFER_TRACE(bh, "still unmapped");
3509 /* Ok, it's mapped. Make sure it's up-to-date */
3510 if (PageUptodate(page))
3511 set_buffer_uptodate(bh);
3513 if (!buffer_uptodate(bh)) {
3515 ll_rw_block(READ, 1, &bh);
3517 /* Uhhuh. Read error. Complain and punt.*/
3518 if (!buffer_uptodate(bh))
3522 if (ext4_should_journal_data(inode)) {
3523 BUFFER_TRACE(bh, "get write access");
3524 err = ext4_journal_get_write_access(handle, bh);
3529 zero_user(page, pos, range_to_discard);
3532 if (ext4_should_journal_data(inode)) {
3533 err = ext4_handle_dirty_metadata(handle, inode, bh);
3535 mark_buffer_dirty(bh);
3537 BUFFER_TRACE(bh, "Partial buffer zeroed");
3539 bh = bh->b_this_page;
3541 pos += range_to_discard;
3547 int ext4_can_truncate(struct inode *inode)
3549 if (S_ISREG(inode->i_mode))
3551 if (S_ISDIR(inode->i_mode))
3553 if (S_ISLNK(inode->i_mode))
3554 return !ext4_inode_is_fast_symlink(inode);
3559 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3560 * associated with the given offset and length
3562 * @inode: File inode
3563 * @offset: The offset where the hole will begin
3564 * @len: The length of the hole
3566 * Returns: 0 on success or negative on failure
3569 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3571 struct inode *inode = file_inode(file);
3572 struct super_block *sb = inode->i_sb;
3573 ext4_lblk_t first_block, stop_block;
3574 struct address_space *mapping = inode->i_mapping;
3575 loff_t first_page, last_page, page_len;
3576 loff_t first_page_offset, last_page_offset;
3578 unsigned int credits;
3581 if (!S_ISREG(inode->i_mode))
3584 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3585 /* TODO: Add support for bigalloc file systems */
3589 trace_ext4_punch_hole(inode, offset, length);
3592 * Write out all dirty pages to avoid race conditions
3593 * Then release them.
3595 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3596 ret = filemap_write_and_wait_range(mapping, offset,
3597 offset + length - 1);
3602 mutex_lock(&inode->i_mutex);
3603 /* It's not possible punch hole on append only file */
3604 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3608 if (IS_SWAPFILE(inode)) {
3613 /* No need to punch hole beyond i_size */
3614 if (offset >= inode->i_size)
3618 * If the hole extends beyond i_size, set the hole
3619 * to end after the page that contains i_size
3621 if (offset + length > inode->i_size) {
3622 length = inode->i_size +
3623 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3627 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
3628 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
3630 first_page_offset = first_page << PAGE_CACHE_SHIFT;
3631 last_page_offset = last_page << PAGE_CACHE_SHIFT;
3633 /* Now release the pages */
3634 if (last_page_offset > first_page_offset) {
3635 truncate_pagecache_range(inode, first_page_offset,
3636 last_page_offset - 1);
3639 /* Wait all existing dio workers, newcomers will block on i_mutex */
3640 ext4_inode_block_unlocked_dio(inode);
3641 ret = ext4_flush_unwritten_io(inode);
3644 inode_dio_wait(inode);
3646 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3647 credits = ext4_writepage_trans_blocks(inode);
3649 credits = ext4_blocks_for_truncate(inode);
3650 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3651 if (IS_ERR(handle)) {
3652 ret = PTR_ERR(handle);
3653 ext4_std_error(sb, ret);
3658 * Now we need to zero out the non-page-aligned data in the
3659 * pages at the start and tail of the hole, and unmap the
3660 * buffer heads for the block aligned regions of the page that
3661 * were completely zeroed.
3663 if (first_page > last_page) {
3665 * If the file space being truncated is contained
3666 * within a page just zero out and unmap the middle of
3669 ret = ext4_discard_partial_page_buffers(handle,
3670 mapping, offset, length, 0);
3676 * zero out and unmap the partial page that contains
3677 * the start of the hole
3679 page_len = first_page_offset - offset;
3681 ret = ext4_discard_partial_page_buffers(handle, mapping,
3682 offset, page_len, 0);
3688 * zero out and unmap the partial page that contains
3689 * the end of the hole
3691 page_len = offset + length - last_page_offset;
3693 ret = ext4_discard_partial_page_buffers(handle, mapping,
3694 last_page_offset, page_len, 0);
3701 * If i_size is contained in the last page, we need to
3702 * unmap and zero the partial page after i_size
3704 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
3705 inode->i_size % PAGE_CACHE_SIZE != 0) {
3706 page_len = PAGE_CACHE_SIZE -
3707 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3710 ret = ext4_discard_partial_page_buffers(handle,
3711 mapping, inode->i_size, page_len, 0);
3718 first_block = (offset + sb->s_blocksize - 1) >>
3719 EXT4_BLOCK_SIZE_BITS(sb);
3720 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3722 /* If there are no blocks to remove, return now */
3723 if (first_block >= stop_block)
3726 down_write(&EXT4_I(inode)->i_data_sem);
3727 ext4_discard_preallocations(inode);
3729 ret = ext4_es_remove_extent(inode, first_block,
3730 stop_block - first_block);
3732 up_write(&EXT4_I(inode)->i_data_sem);
3736 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3737 ret = ext4_ext_remove_space(inode, first_block,
3740 ret = ext4_free_hole_blocks(handle, inode, first_block,
3743 ext4_discard_preallocations(inode);
3744 up_write(&EXT4_I(inode)->i_data_sem);
3746 ext4_handle_sync(handle);
3747 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3748 ext4_mark_inode_dirty(handle, inode);
3750 ext4_journal_stop(handle);
3752 ext4_inode_resume_unlocked_dio(inode);
3754 mutex_unlock(&inode->i_mutex);
3761 * We block out ext4_get_block() block instantiations across the entire
3762 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3763 * simultaneously on behalf of the same inode.
3765 * As we work through the truncate and commit bits of it to the journal there
3766 * is one core, guiding principle: the file's tree must always be consistent on
3767 * disk. We must be able to restart the truncate after a crash.
3769 * The file's tree may be transiently inconsistent in memory (although it
3770 * probably isn't), but whenever we close off and commit a journal transaction,
3771 * the contents of (the filesystem + the journal) must be consistent and
3772 * restartable. It's pretty simple, really: bottom up, right to left (although
3773 * left-to-right works OK too).
3775 * Note that at recovery time, journal replay occurs *before* the restart of
3776 * truncate against the orphan inode list.
3778 * The committed inode has the new, desired i_size (which is the same as
3779 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3780 * that this inode's truncate did not complete and it will again call
3781 * ext4_truncate() to have another go. So there will be instantiated blocks
3782 * to the right of the truncation point in a crashed ext4 filesystem. But
3783 * that's fine - as long as they are linked from the inode, the post-crash
3784 * ext4_truncate() run will find them and release them.
3786 void ext4_truncate(struct inode *inode)
3788 struct ext4_inode_info *ei = EXT4_I(inode);
3789 unsigned int credits;
3791 struct address_space *mapping = inode->i_mapping;
3795 * There is a possibility that we're either freeing the inode
3796 * or it completely new indode. In those cases we might not
3797 * have i_mutex locked because it's not necessary.
3799 if (!(inode->i_state & (I_NEW|I_FREEING)))
3800 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3801 trace_ext4_truncate_enter(inode);
3803 if (!ext4_can_truncate(inode))
3806 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3808 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3809 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3811 if (ext4_has_inline_data(inode)) {
3814 ext4_inline_data_truncate(inode, &has_inline);
3820 * finish any pending end_io work so we won't run the risk of
3821 * converting any truncated blocks to initialized later
3823 ext4_flush_unwritten_io(inode);
3825 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3826 credits = ext4_writepage_trans_blocks(inode);
3828 credits = ext4_blocks_for_truncate(inode);
3830 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3831 if (IS_ERR(handle)) {
3832 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3836 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3837 page_len = PAGE_CACHE_SIZE -
3838 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3840 if (ext4_discard_partial_page_buffers(handle,
3841 mapping, inode->i_size, page_len, 0))
3846 * We add the inode to the orphan list, so that if this
3847 * truncate spans multiple transactions, and we crash, we will
3848 * resume the truncate when the filesystem recovers. It also
3849 * marks the inode dirty, to catch the new size.
3851 * Implication: the file must always be in a sane, consistent
3852 * truncatable state while each transaction commits.
3854 if (ext4_orphan_add(handle, inode))
3857 down_write(&EXT4_I(inode)->i_data_sem);
3859 ext4_discard_preallocations(inode);
3861 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3862 ext4_ext_truncate(handle, inode);
3864 ext4_ind_truncate(handle, inode);
3866 up_write(&ei->i_data_sem);
3869 ext4_handle_sync(handle);
3873 * If this was a simple ftruncate() and the file will remain alive,
3874 * then we need to clear up the orphan record which we created above.
3875 * However, if this was a real unlink then we were called by
3876 * ext4_delete_inode(), and we allow that function to clean up the
3877 * orphan info for us.
3880 ext4_orphan_del(handle, inode);
3882 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3883 ext4_mark_inode_dirty(handle, inode);
3884 ext4_journal_stop(handle);
3886 trace_ext4_truncate_exit(inode);
3890 * ext4_get_inode_loc returns with an extra refcount against the inode's
3891 * underlying buffer_head on success. If 'in_mem' is true, we have all
3892 * data in memory that is needed to recreate the on-disk version of this
3895 static int __ext4_get_inode_loc(struct inode *inode,
3896 struct ext4_iloc *iloc, int in_mem)
3898 struct ext4_group_desc *gdp;
3899 struct buffer_head *bh;
3900 struct super_block *sb = inode->i_sb;
3902 int inodes_per_block, inode_offset;
3905 if (!ext4_valid_inum(sb, inode->i_ino))
3908 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3909 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3914 * Figure out the offset within the block group inode table
3916 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3917 inode_offset = ((inode->i_ino - 1) %
3918 EXT4_INODES_PER_GROUP(sb));
3919 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3920 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3922 bh = sb_getblk(sb, block);
3925 if (!buffer_uptodate(bh)) {
3929 * If the buffer has the write error flag, we have failed
3930 * to write out another inode in the same block. In this
3931 * case, we don't have to read the block because we may
3932 * read the old inode data successfully.
3934 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3935 set_buffer_uptodate(bh);
3937 if (buffer_uptodate(bh)) {
3938 /* someone brought it uptodate while we waited */
3944 * If we have all information of the inode in memory and this
3945 * is the only valid inode in the block, we need not read the
3949 struct buffer_head *bitmap_bh;
3952 start = inode_offset & ~(inodes_per_block - 1);
3954 /* Is the inode bitmap in cache? */
3955 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3956 if (unlikely(!bitmap_bh))
3960 * If the inode bitmap isn't in cache then the
3961 * optimisation may end up performing two reads instead
3962 * of one, so skip it.
3964 if (!buffer_uptodate(bitmap_bh)) {
3968 for (i = start; i < start + inodes_per_block; i++) {
3969 if (i == inode_offset)
3971 if (ext4_test_bit(i, bitmap_bh->b_data))
3975 if (i == start + inodes_per_block) {
3976 /* all other inodes are free, so skip I/O */
3977 memset(bh->b_data, 0, bh->b_size);
3978 set_buffer_uptodate(bh);
3986 * If we need to do any I/O, try to pre-readahead extra
3987 * blocks from the inode table.
3989 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3990 ext4_fsblk_t b, end, table;
3993 table = ext4_inode_table(sb, gdp);
3994 /* s_inode_readahead_blks is always a power of 2 */
3995 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3998 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3999 num = EXT4_INODES_PER_GROUP(sb);
4000 if (ext4_has_group_desc_csum(sb))
4001 num -= ext4_itable_unused_count(sb, gdp);
4002 table += num / inodes_per_block;
4006 sb_breadahead(sb, b++);
4010 * There are other valid inodes in the buffer, this inode
4011 * has in-inode xattrs, or we don't have this inode in memory.
4012 * Read the block from disk.
4014 trace_ext4_load_inode(inode);
4016 bh->b_end_io = end_buffer_read_sync;
4017 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4019 if (!buffer_uptodate(bh)) {
4020 EXT4_ERROR_INODE_BLOCK(inode, block,
4021 "unable to read itable block");
4031 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4033 /* We have all inode data except xattrs in memory here. */
4034 return __ext4_get_inode_loc(inode, iloc,
4035 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4038 void ext4_set_inode_flags(struct inode *inode)
4040 unsigned int flags = EXT4_I(inode)->i_flags;
4042 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4043 if (flags & EXT4_SYNC_FL)
4044 inode->i_flags |= S_SYNC;
4045 if (flags & EXT4_APPEND_FL)
4046 inode->i_flags |= S_APPEND;
4047 if (flags & EXT4_IMMUTABLE_FL)
4048 inode->i_flags |= S_IMMUTABLE;
4049 if (flags & EXT4_NOATIME_FL)
4050 inode->i_flags |= S_NOATIME;
4051 if (flags & EXT4_DIRSYNC_FL)
4052 inode->i_flags |= S_DIRSYNC;
4055 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4056 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4058 unsigned int vfs_fl;
4059 unsigned long old_fl, new_fl;
4062 vfs_fl = ei->vfs_inode.i_flags;
4063 old_fl = ei->i_flags;
4064 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4065 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4067 if (vfs_fl & S_SYNC)
4068 new_fl |= EXT4_SYNC_FL;
4069 if (vfs_fl & S_APPEND)
4070 new_fl |= EXT4_APPEND_FL;
4071 if (vfs_fl & S_IMMUTABLE)
4072 new_fl |= EXT4_IMMUTABLE_FL;
4073 if (vfs_fl & S_NOATIME)
4074 new_fl |= EXT4_NOATIME_FL;
4075 if (vfs_fl & S_DIRSYNC)
4076 new_fl |= EXT4_DIRSYNC_FL;
4077 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4080 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4081 struct ext4_inode_info *ei)
4084 struct inode *inode = &(ei->vfs_inode);
4085 struct super_block *sb = inode->i_sb;
4087 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4088 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4089 /* we are using combined 48 bit field */
4090 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4091 le32_to_cpu(raw_inode->i_blocks_lo);
4092 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4093 /* i_blocks represent file system block size */
4094 return i_blocks << (inode->i_blkbits - 9);
4099 return le32_to_cpu(raw_inode->i_blocks_lo);
4103 static inline void ext4_iget_extra_inode(struct inode *inode,
4104 struct ext4_inode *raw_inode,
4105 struct ext4_inode_info *ei)
4107 __le32 *magic = (void *)raw_inode +
4108 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4109 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4110 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4111 ext4_find_inline_data_nolock(inode);
4113 EXT4_I(inode)->i_inline_off = 0;
4116 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4118 struct ext4_iloc iloc;
4119 struct ext4_inode *raw_inode;
4120 struct ext4_inode_info *ei;
4121 struct inode *inode;
4122 journal_t *journal = EXT4_SB(sb)->s_journal;
4128 inode = iget_locked(sb, ino);
4130 return ERR_PTR(-ENOMEM);
4131 if (!(inode->i_state & I_NEW))
4137 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4140 raw_inode = ext4_raw_inode(&iloc);
4142 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4143 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4144 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4145 EXT4_INODE_SIZE(inode->i_sb)) {
4146 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4147 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4148 EXT4_INODE_SIZE(inode->i_sb));
4153 ei->i_extra_isize = 0;
4155 /* Precompute checksum seed for inode metadata */
4156 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4157 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4158 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4160 __le32 inum = cpu_to_le32(inode->i_ino);
4161 __le32 gen = raw_inode->i_generation;
4162 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4164 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4168 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4169 EXT4_ERROR_INODE(inode, "checksum invalid");
4174 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4175 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4176 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4177 if (!(test_opt(inode->i_sb, NO_UID32))) {
4178 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4179 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4181 i_uid_write(inode, i_uid);
4182 i_gid_write(inode, i_gid);
4183 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4185 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4186 ei->i_inline_off = 0;
4187 ei->i_dir_start_lookup = 0;
4188 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4189 /* We now have enough fields to check if the inode was active or not.
4190 * This is needed because nfsd might try to access dead inodes
4191 * the test is that same one that e2fsck uses
4192 * NeilBrown 1999oct15
4194 if (inode->i_nlink == 0) {
4195 if ((inode->i_mode == 0 ||
4196 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4197 ino != EXT4_BOOT_LOADER_INO) {
4198 /* this inode is deleted */
4202 /* The only unlinked inodes we let through here have
4203 * valid i_mode and are being read by the orphan
4204 * recovery code: that's fine, we're about to complete
4205 * the process of deleting those.
4206 * OR it is the EXT4_BOOT_LOADER_INO which is
4207 * not initialized on a new filesystem. */
4209 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4210 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4211 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4212 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4214 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4215 inode->i_size = ext4_isize(raw_inode);
4216 ei->i_disksize = inode->i_size;
4218 ei->i_reserved_quota = 0;
4220 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4221 ei->i_block_group = iloc.block_group;
4222 ei->i_last_alloc_group = ~0;
4224 * NOTE! The in-memory inode i_data array is in little-endian order
4225 * even on big-endian machines: we do NOT byteswap the block numbers!
4227 for (block = 0; block < EXT4_N_BLOCKS; block++)
4228 ei->i_data[block] = raw_inode->i_block[block];
4229 INIT_LIST_HEAD(&ei->i_orphan);
4232 * Set transaction id's of transactions that have to be committed
4233 * to finish f[data]sync. We set them to currently running transaction
4234 * as we cannot be sure that the inode or some of its metadata isn't
4235 * part of the transaction - the inode could have been reclaimed and
4236 * now it is reread from disk.
4239 transaction_t *transaction;
4242 read_lock(&journal->j_state_lock);
4243 if (journal->j_running_transaction)
4244 transaction = journal->j_running_transaction;
4246 transaction = journal->j_committing_transaction;
4248 tid = transaction->t_tid;
4250 tid = journal->j_commit_sequence;
4251 read_unlock(&journal->j_state_lock);
4252 ei->i_sync_tid = tid;
4253 ei->i_datasync_tid = tid;
4256 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4257 if (ei->i_extra_isize == 0) {
4258 /* The extra space is currently unused. Use it. */
4259 ei->i_extra_isize = sizeof(struct ext4_inode) -
4260 EXT4_GOOD_OLD_INODE_SIZE;
4262 ext4_iget_extra_inode(inode, raw_inode, ei);
4266 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4267 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4268 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4269 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4271 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4272 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4273 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4275 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4279 if (ei->i_file_acl &&
4280 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4281 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4285 } else if (!ext4_has_inline_data(inode)) {
4286 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4287 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4288 (S_ISLNK(inode->i_mode) &&
4289 !ext4_inode_is_fast_symlink(inode))))
4290 /* Validate extent which is part of inode */
4291 ret = ext4_ext_check_inode(inode);
4292 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4293 (S_ISLNK(inode->i_mode) &&
4294 !ext4_inode_is_fast_symlink(inode))) {
4295 /* Validate block references which are part of inode */
4296 ret = ext4_ind_check_inode(inode);
4302 if (S_ISREG(inode->i_mode)) {
4303 inode->i_op = &ext4_file_inode_operations;
4304 inode->i_fop = &ext4_file_operations;
4305 ext4_set_aops(inode);
4306 } else if (S_ISDIR(inode->i_mode)) {
4307 inode->i_op = &ext4_dir_inode_operations;
4308 inode->i_fop = &ext4_dir_operations;
4309 } else if (S_ISLNK(inode->i_mode)) {
4310 if (ext4_inode_is_fast_symlink(inode)) {
4311 inode->i_op = &ext4_fast_symlink_inode_operations;
4312 nd_terminate_link(ei->i_data, inode->i_size,
4313 sizeof(ei->i_data) - 1);
4315 inode->i_op = &ext4_symlink_inode_operations;
4316 ext4_set_aops(inode);
4318 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4319 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4320 inode->i_op = &ext4_special_inode_operations;
4321 if (raw_inode->i_block[0])
4322 init_special_inode(inode, inode->i_mode,
4323 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4325 init_special_inode(inode, inode->i_mode,
4326 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4327 } else if (ino == EXT4_BOOT_LOADER_INO) {
4328 make_bad_inode(inode);
4331 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4335 ext4_set_inode_flags(inode);
4336 unlock_new_inode(inode);
4342 return ERR_PTR(ret);
4345 static int ext4_inode_blocks_set(handle_t *handle,
4346 struct ext4_inode *raw_inode,
4347 struct ext4_inode_info *ei)
4349 struct inode *inode = &(ei->vfs_inode);
4350 u64 i_blocks = inode->i_blocks;
4351 struct super_block *sb = inode->i_sb;
4353 if (i_blocks <= ~0U) {
4355 * i_blocks can be represented in a 32 bit variable
4356 * as multiple of 512 bytes
4358 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4359 raw_inode->i_blocks_high = 0;
4360 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4363 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4366 if (i_blocks <= 0xffffffffffffULL) {
4368 * i_blocks can be represented in a 48 bit variable
4369 * as multiple of 512 bytes
4371 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4372 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4373 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4375 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4376 /* i_block is stored in file system block size */
4377 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4378 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4379 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4385 * Post the struct inode info into an on-disk inode location in the
4386 * buffer-cache. This gobbles the caller's reference to the
4387 * buffer_head in the inode location struct.
4389 * The caller must have write access to iloc->bh.
4391 static int ext4_do_update_inode(handle_t *handle,
4392 struct inode *inode,
4393 struct ext4_iloc *iloc)
4395 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4396 struct ext4_inode_info *ei = EXT4_I(inode);
4397 struct buffer_head *bh = iloc->bh;
4398 int err = 0, rc, block;
4399 int need_datasync = 0;
4403 /* For fields not not tracking in the in-memory inode,
4404 * initialise them to zero for new inodes. */
4405 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4406 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4408 ext4_get_inode_flags(ei);
4409 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4410 i_uid = i_uid_read(inode);
4411 i_gid = i_gid_read(inode);
4412 if (!(test_opt(inode->i_sb, NO_UID32))) {
4413 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4414 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4416 * Fix up interoperability with old kernels. Otherwise, old inodes get
4417 * re-used with the upper 16 bits of the uid/gid intact
4420 raw_inode->i_uid_high =
4421 cpu_to_le16(high_16_bits(i_uid));
4422 raw_inode->i_gid_high =
4423 cpu_to_le16(high_16_bits(i_gid));
4425 raw_inode->i_uid_high = 0;
4426 raw_inode->i_gid_high = 0;
4429 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4430 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4431 raw_inode->i_uid_high = 0;
4432 raw_inode->i_gid_high = 0;
4434 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4436 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4437 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4438 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4439 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4441 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4443 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4444 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4445 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4446 cpu_to_le32(EXT4_OS_HURD))
4447 raw_inode->i_file_acl_high =
4448 cpu_to_le16(ei->i_file_acl >> 32);
4449 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4450 if (ei->i_disksize != ext4_isize(raw_inode)) {
4451 ext4_isize_set(raw_inode, ei->i_disksize);
4454 if (ei->i_disksize > 0x7fffffffULL) {
4455 struct super_block *sb = inode->i_sb;
4456 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4457 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4458 EXT4_SB(sb)->s_es->s_rev_level ==
4459 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4460 /* If this is the first large file
4461 * created, add a flag to the superblock.
4463 err = ext4_journal_get_write_access(handle,
4464 EXT4_SB(sb)->s_sbh);
4467 ext4_update_dynamic_rev(sb);
4468 EXT4_SET_RO_COMPAT_FEATURE(sb,
4469 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4470 ext4_handle_sync(handle);
4471 err = ext4_handle_dirty_super(handle, sb);
4474 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4475 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4476 if (old_valid_dev(inode->i_rdev)) {
4477 raw_inode->i_block[0] =
4478 cpu_to_le32(old_encode_dev(inode->i_rdev));
4479 raw_inode->i_block[1] = 0;
4481 raw_inode->i_block[0] = 0;
4482 raw_inode->i_block[1] =
4483 cpu_to_le32(new_encode_dev(inode->i_rdev));
4484 raw_inode->i_block[2] = 0;
4486 } else if (!ext4_has_inline_data(inode)) {
4487 for (block = 0; block < EXT4_N_BLOCKS; block++)
4488 raw_inode->i_block[block] = ei->i_data[block];
4491 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4492 if (ei->i_extra_isize) {
4493 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4494 raw_inode->i_version_hi =
4495 cpu_to_le32(inode->i_version >> 32);
4496 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4499 ext4_inode_csum_set(inode, raw_inode, ei);
4501 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4502 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4505 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4507 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4510 ext4_std_error(inode->i_sb, err);
4515 * ext4_write_inode()
4517 * We are called from a few places:
4519 * - Within generic_file_write() for O_SYNC files.
4520 * Here, there will be no transaction running. We wait for any running
4521 * transaction to commit.
4523 * - Within sys_sync(), kupdate and such.
4524 * We wait on commit, if tol to.
4526 * - Within prune_icache() (PF_MEMALLOC == true)
4527 * Here we simply return. We can't afford to block kswapd on the
4530 * In all cases it is actually safe for us to return without doing anything,
4531 * because the inode has been copied into a raw inode buffer in
4532 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4535 * Note that we are absolutely dependent upon all inode dirtiers doing the
4536 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4537 * which we are interested.
4539 * It would be a bug for them to not do this. The code:
4541 * mark_inode_dirty(inode)
4543 * inode->i_size = expr;
4545 * is in error because a kswapd-driven write_inode() could occur while
4546 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4547 * will no longer be on the superblock's dirty inode list.
4549 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4553 if (current->flags & PF_MEMALLOC)
4556 if (EXT4_SB(inode->i_sb)->s_journal) {
4557 if (ext4_journal_current_handle()) {
4558 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4563 if (wbc->sync_mode != WB_SYNC_ALL)
4566 err = ext4_force_commit(inode->i_sb);
4568 struct ext4_iloc iloc;
4570 err = __ext4_get_inode_loc(inode, &iloc, 0);
4573 if (wbc->sync_mode == WB_SYNC_ALL)
4574 sync_dirty_buffer(iloc.bh);
4575 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4576 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4577 "IO error syncing inode");
4586 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4587 * buffers that are attached to a page stradding i_size and are undergoing
4588 * commit. In that case we have to wait for commit to finish and try again.
4590 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4594 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4595 tid_t commit_tid = 0;
4598 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4600 * All buffers in the last page remain valid? Then there's nothing to
4601 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4604 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4607 page = find_lock_page(inode->i_mapping,
4608 inode->i_size >> PAGE_CACHE_SHIFT);
4611 ret = __ext4_journalled_invalidatepage(page, offset);
4613 page_cache_release(page);
4617 read_lock(&journal->j_state_lock);
4618 if (journal->j_committing_transaction)
4619 commit_tid = journal->j_committing_transaction->t_tid;
4620 read_unlock(&journal->j_state_lock);
4622 jbd2_log_wait_commit(journal, commit_tid);
4629 * Called from notify_change.
4631 * We want to trap VFS attempts to truncate the file as soon as
4632 * possible. In particular, we want to make sure that when the VFS
4633 * shrinks i_size, we put the inode on the orphan list and modify
4634 * i_disksize immediately, so that during the subsequent flushing of
4635 * dirty pages and freeing of disk blocks, we can guarantee that any
4636 * commit will leave the blocks being flushed in an unused state on
4637 * disk. (On recovery, the inode will get truncated and the blocks will
4638 * be freed, so we have a strong guarantee that no future commit will
4639 * leave these blocks visible to the user.)
4641 * Another thing we have to assure is that if we are in ordered mode
4642 * and inode is still attached to the committing transaction, we must
4643 * we start writeout of all the dirty pages which are being truncated.
4644 * This way we are sure that all the data written in the previous
4645 * transaction are already on disk (truncate waits for pages under
4648 * Called with inode->i_mutex down.
4650 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4652 struct inode *inode = dentry->d_inode;
4655 const unsigned int ia_valid = attr->ia_valid;
4657 error = inode_change_ok(inode, attr);
4661 if (is_quota_modification(inode, attr))
4662 dquot_initialize(inode);
4663 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4664 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4667 /* (user+group)*(old+new) structure, inode write (sb,
4668 * inode block, ? - but truncate inode update has it) */
4669 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4670 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4671 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4672 if (IS_ERR(handle)) {
4673 error = PTR_ERR(handle);
4676 error = dquot_transfer(inode, attr);
4678 ext4_journal_stop(handle);
4681 /* Update corresponding info in inode so that everything is in
4682 * one transaction */
4683 if (attr->ia_valid & ATTR_UID)
4684 inode->i_uid = attr->ia_uid;
4685 if (attr->ia_valid & ATTR_GID)
4686 inode->i_gid = attr->ia_gid;
4687 error = ext4_mark_inode_dirty(handle, inode);
4688 ext4_journal_stop(handle);
4691 if (attr->ia_valid & ATTR_SIZE) {
4693 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4694 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4696 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4701 if (S_ISREG(inode->i_mode) &&
4702 attr->ia_valid & ATTR_SIZE &&
4703 (attr->ia_size < inode->i_size)) {
4706 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4707 if (IS_ERR(handle)) {
4708 error = PTR_ERR(handle);
4711 if (ext4_handle_valid(handle)) {
4712 error = ext4_orphan_add(handle, inode);
4715 EXT4_I(inode)->i_disksize = attr->ia_size;
4716 rc = ext4_mark_inode_dirty(handle, inode);
4719 ext4_journal_stop(handle);
4721 if (ext4_should_order_data(inode)) {
4722 error = ext4_begin_ordered_truncate(inode,
4725 /* Do as much error cleanup as possible */
4726 handle = ext4_journal_start(inode,
4728 if (IS_ERR(handle)) {
4729 ext4_orphan_del(NULL, inode);
4732 ext4_orphan_del(handle, inode);
4734 ext4_journal_stop(handle);
4740 if (attr->ia_valid & ATTR_SIZE) {
4741 if (attr->ia_size != inode->i_size) {
4742 loff_t oldsize = inode->i_size;
4744 i_size_write(inode, attr->ia_size);
4746 * Blocks are going to be removed from the inode. Wait
4747 * for dio in flight. Temporarily disable
4748 * dioread_nolock to prevent livelock.
4751 if (!ext4_should_journal_data(inode)) {
4752 ext4_inode_block_unlocked_dio(inode);
4753 inode_dio_wait(inode);
4754 ext4_inode_resume_unlocked_dio(inode);
4756 ext4_wait_for_tail_page_commit(inode);
4759 * Truncate pagecache after we've waited for commit
4760 * in data=journal mode to make pages freeable.
4762 truncate_pagecache(inode, oldsize, inode->i_size);
4764 ext4_truncate(inode);
4768 setattr_copy(inode, attr);
4769 mark_inode_dirty(inode);
4773 * If the call to ext4_truncate failed to get a transaction handle at
4774 * all, we need to clean up the in-core orphan list manually.
4776 if (orphan && inode->i_nlink)
4777 ext4_orphan_del(NULL, inode);
4779 if (!rc && (ia_valid & ATTR_MODE))
4780 rc = ext4_acl_chmod(inode);
4783 ext4_std_error(inode->i_sb, error);
4789 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4792 struct inode *inode;
4793 unsigned long delalloc_blocks;
4795 inode = dentry->d_inode;
4796 generic_fillattr(inode, stat);
4799 * We can't update i_blocks if the block allocation is delayed
4800 * otherwise in the case of system crash before the real block
4801 * allocation is done, we will have i_blocks inconsistent with
4802 * on-disk file blocks.
4803 * We always keep i_blocks updated together with real
4804 * allocation. But to not confuse with user, stat
4805 * will return the blocks that include the delayed allocation
4806 * blocks for this file.
4808 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4809 EXT4_I(inode)->i_reserved_data_blocks);
4811 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4815 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4817 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4818 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4819 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4823 * Account for index blocks, block groups bitmaps and block group
4824 * descriptor blocks if modify datablocks and index blocks
4825 * worse case, the indexs blocks spread over different block groups
4827 * If datablocks are discontiguous, they are possible to spread over
4828 * different block groups too. If they are contiguous, with flexbg,
4829 * they could still across block group boundary.
4831 * Also account for superblock, inode, quota and xattr blocks
4833 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4835 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4841 * How many index blocks need to touch to modify nrblocks?
4842 * The "Chunk" flag indicating whether the nrblocks is
4843 * physically contiguous on disk
4845 * For Direct IO and fallocate, they calls get_block to allocate
4846 * one single extent at a time, so they could set the "Chunk" flag
4848 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4853 * Now let's see how many group bitmaps and group descriptors need
4863 if (groups > ngroups)
4865 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4866 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4868 /* bitmaps and block group descriptor blocks */
4869 ret += groups + gdpblocks;
4871 /* Blocks for super block, inode, quota and xattr blocks */
4872 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4878 * Calculate the total number of credits to reserve to fit
4879 * the modification of a single pages into a single transaction,
4880 * which may include multiple chunks of block allocations.
4882 * This could be called via ext4_write_begin()
4884 * We need to consider the worse case, when
4885 * one new block per extent.
4887 int ext4_writepage_trans_blocks(struct inode *inode)
4889 int bpp = ext4_journal_blocks_per_page(inode);
4892 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4894 /* Account for data blocks for journalled mode */
4895 if (ext4_should_journal_data(inode))
4901 * Calculate the journal credits for a chunk of data modification.
4903 * This is called from DIO, fallocate or whoever calling
4904 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4906 * journal buffers for data blocks are not included here, as DIO
4907 * and fallocate do no need to journal data buffers.
4909 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4911 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4915 * The caller must have previously called ext4_reserve_inode_write().
4916 * Give this, we know that the caller already has write access to iloc->bh.
4918 int ext4_mark_iloc_dirty(handle_t *handle,
4919 struct inode *inode, struct ext4_iloc *iloc)
4923 if (IS_I_VERSION(inode))
4924 inode_inc_iversion(inode);
4926 /* the do_update_inode consumes one bh->b_count */
4929 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4930 err = ext4_do_update_inode(handle, inode, iloc);
4936 * On success, We end up with an outstanding reference count against
4937 * iloc->bh. This _must_ be cleaned up later.
4941 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4942 struct ext4_iloc *iloc)
4946 err = ext4_get_inode_loc(inode, iloc);
4948 BUFFER_TRACE(iloc->bh, "get_write_access");
4949 err = ext4_journal_get_write_access(handle, iloc->bh);
4955 ext4_std_error(inode->i_sb, err);
4960 * Expand an inode by new_extra_isize bytes.
4961 * Returns 0 on success or negative error number on failure.
4963 static int ext4_expand_extra_isize(struct inode *inode,
4964 unsigned int new_extra_isize,
4965 struct ext4_iloc iloc,
4968 struct ext4_inode *raw_inode;
4969 struct ext4_xattr_ibody_header *header;
4971 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4974 raw_inode = ext4_raw_inode(&iloc);
4976 header = IHDR(inode, raw_inode);
4978 /* No extended attributes present */
4979 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4980 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4981 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4983 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4987 /* try to expand with EAs present */
4988 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4993 * What we do here is to mark the in-core inode as clean with respect to inode
4994 * dirtiness (it may still be data-dirty).
4995 * This means that the in-core inode may be reaped by prune_icache
4996 * without having to perform any I/O. This is a very good thing,
4997 * because *any* task may call prune_icache - even ones which
4998 * have a transaction open against a different journal.
5000 * Is this cheating? Not really. Sure, we haven't written the
5001 * inode out, but prune_icache isn't a user-visible syncing function.
5002 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5003 * we start and wait on commits.
5005 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5007 struct ext4_iloc iloc;
5008 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5009 static unsigned int mnt_count;
5013 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5014 err = ext4_reserve_inode_write(handle, inode, &iloc);
5015 if (ext4_handle_valid(handle) &&
5016 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5017 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5019 * We need extra buffer credits since we may write into EA block
5020 * with this same handle. If journal_extend fails, then it will
5021 * only result in a minor loss of functionality for that inode.
5022 * If this is felt to be critical, then e2fsck should be run to
5023 * force a large enough s_min_extra_isize.
5025 if ((jbd2_journal_extend(handle,
5026 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5027 ret = ext4_expand_extra_isize(inode,
5028 sbi->s_want_extra_isize,
5031 ext4_set_inode_state(inode,
5032 EXT4_STATE_NO_EXPAND);
5034 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5035 ext4_warning(inode->i_sb,
5036 "Unable to expand inode %lu. Delete"
5037 " some EAs or run e2fsck.",
5040 le16_to_cpu(sbi->s_es->s_mnt_count);
5046 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5051 * ext4_dirty_inode() is called from __mark_inode_dirty()
5053 * We're really interested in the case where a file is being extended.
5054 * i_size has been changed by generic_commit_write() and we thus need
5055 * to include the updated inode in the current transaction.
5057 * Also, dquot_alloc_block() will always dirty the inode when blocks
5058 * are allocated to the file.
5060 * If the inode is marked synchronous, we don't honour that here - doing
5061 * so would cause a commit on atime updates, which we don't bother doing.
5062 * We handle synchronous inodes at the highest possible level.
5064 void ext4_dirty_inode(struct inode *inode, int flags)
5068 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5072 ext4_mark_inode_dirty(handle, inode);
5074 ext4_journal_stop(handle);
5081 * Bind an inode's backing buffer_head into this transaction, to prevent
5082 * it from being flushed to disk early. Unlike
5083 * ext4_reserve_inode_write, this leaves behind no bh reference and
5084 * returns no iloc structure, so the caller needs to repeat the iloc
5085 * lookup to mark the inode dirty later.
5087 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5089 struct ext4_iloc iloc;
5093 err = ext4_get_inode_loc(inode, &iloc);
5095 BUFFER_TRACE(iloc.bh, "get_write_access");
5096 err = jbd2_journal_get_write_access(handle, iloc.bh);
5098 err = ext4_handle_dirty_metadata(handle,
5104 ext4_std_error(inode->i_sb, err);
5109 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5116 * We have to be very careful here: changing a data block's
5117 * journaling status dynamically is dangerous. If we write a
5118 * data block to the journal, change the status and then delete
5119 * that block, we risk forgetting to revoke the old log record
5120 * from the journal and so a subsequent replay can corrupt data.
5121 * So, first we make sure that the journal is empty and that
5122 * nobody is changing anything.
5125 journal = EXT4_JOURNAL(inode);
5128 if (is_journal_aborted(journal))
5130 /* We have to allocate physical blocks for delalloc blocks
5131 * before flushing journal. otherwise delalloc blocks can not
5132 * be allocated any more. even more truncate on delalloc blocks
5133 * could trigger BUG by flushing delalloc blocks in journal.
5134 * There is no delalloc block in non-journal data mode.
5136 if (val && test_opt(inode->i_sb, DELALLOC)) {
5137 err = ext4_alloc_da_blocks(inode);
5142 /* Wait for all existing dio workers */
5143 ext4_inode_block_unlocked_dio(inode);
5144 inode_dio_wait(inode);
5146 jbd2_journal_lock_updates(journal);
5149 * OK, there are no updates running now, and all cached data is
5150 * synced to disk. We are now in a completely consistent state
5151 * which doesn't have anything in the journal, and we know that
5152 * no filesystem updates are running, so it is safe to modify
5153 * the inode's in-core data-journaling state flag now.
5157 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5159 jbd2_journal_flush(journal);
5160 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5162 ext4_set_aops(inode);
5164 jbd2_journal_unlock_updates(journal);
5165 ext4_inode_resume_unlocked_dio(inode);
5167 /* Finally we can mark the inode as dirty. */
5169 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5171 return PTR_ERR(handle);
5173 err = ext4_mark_inode_dirty(handle, inode);
5174 ext4_handle_sync(handle);
5175 ext4_journal_stop(handle);
5176 ext4_std_error(inode->i_sb, err);
5181 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5183 return !buffer_mapped(bh);
5186 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5188 struct page *page = vmf->page;
5192 struct file *file = vma->vm_file;
5193 struct inode *inode = file_inode(file);
5194 struct address_space *mapping = inode->i_mapping;
5196 get_block_t *get_block;
5199 sb_start_pagefault(inode->i_sb);
5200 file_update_time(vma->vm_file);
5201 /* Delalloc case is easy... */
5202 if (test_opt(inode->i_sb, DELALLOC) &&
5203 !ext4_should_journal_data(inode) &&
5204 !ext4_nonda_switch(inode->i_sb)) {
5206 ret = __block_page_mkwrite(vma, vmf,
5207 ext4_da_get_block_prep);
5208 } while (ret == -ENOSPC &&
5209 ext4_should_retry_alloc(inode->i_sb, &retries));
5214 size = i_size_read(inode);
5215 /* Page got truncated from under us? */
5216 if (page->mapping != mapping || page_offset(page) > size) {
5218 ret = VM_FAULT_NOPAGE;
5222 if (page->index == size >> PAGE_CACHE_SHIFT)
5223 len = size & ~PAGE_CACHE_MASK;
5225 len = PAGE_CACHE_SIZE;
5227 * Return if we have all the buffers mapped. This avoids the need to do
5228 * journal_start/journal_stop which can block and take a long time
5230 if (page_has_buffers(page)) {
5231 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5233 ext4_bh_unmapped)) {
5234 /* Wait so that we don't change page under IO */
5235 wait_for_stable_page(page);
5236 ret = VM_FAULT_LOCKED;
5241 /* OK, we need to fill the hole... */
5242 if (ext4_should_dioread_nolock(inode))
5243 get_block = ext4_get_block_write;
5245 get_block = ext4_get_block;
5247 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5248 ext4_writepage_trans_blocks(inode));
5249 if (IS_ERR(handle)) {
5250 ret = VM_FAULT_SIGBUS;
5253 ret = __block_page_mkwrite(vma, vmf, get_block);
5254 if (!ret && ext4_should_journal_data(inode)) {
5255 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5256 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5258 ret = VM_FAULT_SIGBUS;
5259 ext4_journal_stop(handle);
5262 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5264 ext4_journal_stop(handle);
5265 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5268 ret = block_page_mkwrite_return(ret);
5270 sb_end_pagefault(inode->i_sb);