2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/aio.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
60 csum_lo = le16_to_cpu(raw->i_checksum_lo);
61 raw->i_checksum_lo = 0;
62 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
63 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
64 csum_hi = le16_to_cpu(raw->i_checksum_hi);
65 raw->i_checksum_hi = 0;
68 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
69 EXT4_INODE_SIZE(inode->i_sb));
71 raw->i_checksum_lo = cpu_to_le16(csum_lo);
72 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
73 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
74 raw->i_checksum_hi = cpu_to_le16(csum_hi);
79 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
80 struct ext4_inode_info *ei)
82 __u32 provided, calculated;
84 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
85 cpu_to_le32(EXT4_OS_LINUX) ||
86 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
87 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
90 provided = le16_to_cpu(raw->i_checksum_lo);
91 calculated = ext4_inode_csum(inode, raw, ei);
92 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
93 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
94 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
98 return provided == calculated;
101 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
102 struct ext4_inode_info *ei)
106 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
107 cpu_to_le32(EXT4_OS_LINUX) ||
108 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
109 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
112 csum = ext4_inode_csum(inode, raw, ei);
113 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
114 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
115 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
116 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
119 static inline int ext4_begin_ordered_truncate(struct inode *inode,
122 trace_ext4_begin_ordered_truncate(inode, new_size);
124 * If jinode is zero, then we never opened the file for
125 * writing, so there's no need to call
126 * jbd2_journal_begin_ordered_truncate() since there's no
127 * outstanding writes we need to flush.
129 if (!EXT4_I(inode)->jinode)
131 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
132 EXT4_I(inode)->jinode,
136 static void ext4_invalidatepage(struct page *page, unsigned long offset);
137 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
138 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
139 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
140 struct inode *inode, struct page *page, loff_t from,
141 loff_t length, int flags);
144 * Test whether an inode is a fast symlink.
146 static int ext4_inode_is_fast_symlink(struct inode *inode)
148 int ea_blocks = EXT4_I(inode)->i_file_acl ?
149 (inode->i_sb->s_blocksize >> 9) : 0;
151 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
155 * Restart the transaction associated with *handle. This does a commit,
156 * so before we call here everything must be consistently dirtied against
159 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
165 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
166 * moment, get_block can be called only for blocks inside i_size since
167 * page cache has been already dropped and writes are blocked by
168 * i_mutex. So we can safely drop the i_data_sem here.
170 BUG_ON(EXT4_JOURNAL(inode) == NULL);
171 jbd_debug(2, "restarting handle %p\n", handle);
172 up_write(&EXT4_I(inode)->i_data_sem);
173 ret = ext4_journal_restart(handle, nblocks);
174 down_write(&EXT4_I(inode)->i_data_sem);
175 ext4_discard_preallocations(inode);
181 * Called at the last iput() if i_nlink is zero.
183 void ext4_evict_inode(struct inode *inode)
188 trace_ext4_evict_inode(inode);
190 if (inode->i_nlink) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode) &&
210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
211 inode->i_ino != EXT4_JOURNAL_INO) {
212 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
213 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
215 jbd2_complete_transaction(journal, commit_tid);
216 filemap_write_and_wait(&inode->i_data);
218 truncate_inode_pages(&inode->i_data, 0);
219 ext4_ioend_shutdown(inode);
223 if (!is_bad_inode(inode))
224 dquot_initialize(inode);
226 if (ext4_should_order_data(inode))
227 ext4_begin_ordered_truncate(inode, 0);
228 truncate_inode_pages(&inode->i_data, 0);
229 ext4_ioend_shutdown(inode);
231 if (is_bad_inode(inode))
235 * Protect us against freezing - iput() caller didn't have to have any
236 * protection against it
238 sb_start_intwrite(inode->i_sb);
239 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
240 ext4_blocks_for_truncate(inode)+3);
241 if (IS_ERR(handle)) {
242 ext4_std_error(inode->i_sb, PTR_ERR(handle));
244 * If we're going to skip the normal cleanup, we still need to
245 * make sure that the in-core orphan linked list is properly
248 ext4_orphan_del(NULL, inode);
249 sb_end_intwrite(inode->i_sb);
254 ext4_handle_sync(handle);
256 err = ext4_mark_inode_dirty(handle, inode);
258 ext4_warning(inode->i_sb,
259 "couldn't mark inode dirty (err %d)", err);
263 ext4_truncate(inode);
266 * ext4_ext_truncate() doesn't reserve any slop when it
267 * restarts journal transactions; therefore there may not be
268 * enough credits left in the handle to remove the inode from
269 * the orphan list and set the dtime field.
271 if (!ext4_handle_has_enough_credits(handle, 3)) {
272 err = ext4_journal_extend(handle, 3);
274 err = ext4_journal_restart(handle, 3);
276 ext4_warning(inode->i_sb,
277 "couldn't extend journal (err %d)", err);
279 ext4_journal_stop(handle);
280 ext4_orphan_del(NULL, inode);
281 sb_end_intwrite(inode->i_sb);
287 * Kill off the orphan record which ext4_truncate created.
288 * AKPM: I think this can be inside the above `if'.
289 * Note that ext4_orphan_del() has to be able to cope with the
290 * deletion of a non-existent orphan - this is because we don't
291 * know if ext4_truncate() actually created an orphan record.
292 * (Well, we could do this if we need to, but heck - it works)
294 ext4_orphan_del(handle, inode);
295 EXT4_I(inode)->i_dtime = get_seconds();
298 * One subtle ordering requirement: if anything has gone wrong
299 * (transaction abort, IO errors, whatever), then we can still
300 * do these next steps (the fs will already have been marked as
301 * having errors), but we can't free the inode if the mark_dirty
304 if (ext4_mark_inode_dirty(handle, inode))
305 /* If that failed, just do the required in-core inode clear. */
306 ext4_clear_inode(inode);
308 ext4_free_inode(handle, inode);
309 ext4_journal_stop(handle);
310 sb_end_intwrite(inode->i_sb);
313 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
317 qsize_t *ext4_get_reserved_space(struct inode *inode)
319 return &EXT4_I(inode)->i_reserved_quota;
324 * Calculate the number of metadata blocks need to reserve
325 * to allocate a block located at @lblock
327 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
329 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
330 return ext4_ext_calc_metadata_amount(inode, lblock);
332 return ext4_ind_calc_metadata_amount(inode, lblock);
336 * Called with i_data_sem down, which is important since we can call
337 * ext4_discard_preallocations() from here.
339 void ext4_da_update_reserve_space(struct inode *inode,
340 int used, int quota_claim)
342 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
343 struct ext4_inode_info *ei = EXT4_I(inode);
345 spin_lock(&ei->i_block_reservation_lock);
346 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
347 if (unlikely(used > ei->i_reserved_data_blocks)) {
348 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
349 "with only %d reserved data blocks",
350 __func__, inode->i_ino, used,
351 ei->i_reserved_data_blocks);
353 used = ei->i_reserved_data_blocks;
356 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
357 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
358 "with only %d reserved metadata blocks "
359 "(releasing %d blocks with reserved %d data blocks)",
360 inode->i_ino, ei->i_allocated_meta_blocks,
361 ei->i_reserved_meta_blocks, used,
362 ei->i_reserved_data_blocks);
364 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
367 /* Update per-inode reservations */
368 ei->i_reserved_data_blocks -= used;
369 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
370 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
371 used + ei->i_allocated_meta_blocks);
372 ei->i_allocated_meta_blocks = 0;
374 if (ei->i_reserved_data_blocks == 0) {
376 * We can release all of the reserved metadata blocks
377 * only when we have written all of the delayed
380 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
381 ei->i_reserved_meta_blocks);
382 ei->i_reserved_meta_blocks = 0;
383 ei->i_da_metadata_calc_len = 0;
385 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
387 /* Update quota subsystem for data blocks */
389 dquot_claim_block(inode, EXT4_C2B(sbi, used));
392 * We did fallocate with an offset that is already delayed
393 * allocated. So on delayed allocated writeback we should
394 * not re-claim the quota for fallocated blocks.
396 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
400 * If we have done all the pending block allocations and if
401 * there aren't any writers on the inode, we can discard the
402 * inode's preallocations.
404 if ((ei->i_reserved_data_blocks == 0) &&
405 (atomic_read(&inode->i_writecount) == 0))
406 ext4_discard_preallocations(inode);
409 static int __check_block_validity(struct inode *inode, const char *func,
411 struct ext4_map_blocks *map)
413 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
415 ext4_error_inode(inode, func, line, map->m_pblk,
416 "lblock %lu mapped to illegal pblock "
417 "(length %d)", (unsigned long) map->m_lblk,
424 #define check_block_validity(inode, map) \
425 __check_block_validity((inode), __func__, __LINE__, (map))
428 * Return the number of contiguous dirty pages in a given inode
429 * starting at page frame idx.
431 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
432 unsigned int max_pages)
434 struct address_space *mapping = inode->i_mapping;
438 int i, nr_pages, done = 0;
442 pagevec_init(&pvec, 0);
445 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
447 (pgoff_t)PAGEVEC_SIZE);
450 for (i = 0; i < nr_pages; i++) {
451 struct page *page = pvec.pages[i];
452 struct buffer_head *bh, *head;
455 if (unlikely(page->mapping != mapping) ||
457 PageWriteback(page) ||
458 page->index != idx) {
463 if (page_has_buffers(page)) {
464 bh = head = page_buffers(page);
466 if (!buffer_delay(bh) &&
467 !buffer_unwritten(bh))
469 bh = bh->b_this_page;
470 } while (!done && (bh != head));
477 if (num >= max_pages) {
482 pagevec_release(&pvec);
487 #ifdef ES_AGGRESSIVE_TEST
488 static void ext4_map_blocks_es_recheck(handle_t *handle,
490 struct ext4_map_blocks *es_map,
491 struct ext4_map_blocks *map,
498 * There is a race window that the result is not the same.
499 * e.g. xfstests #223 when dioread_nolock enables. The reason
500 * is that we lookup a block mapping in extent status tree with
501 * out taking i_data_sem. So at the time the unwritten extent
502 * could be converted.
504 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
505 down_read((&EXT4_I(inode)->i_data_sem));
506 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
507 retval = ext4_ext_map_blocks(handle, inode, map, flags &
508 EXT4_GET_BLOCKS_KEEP_SIZE);
510 retval = ext4_ind_map_blocks(handle, inode, map, flags &
511 EXT4_GET_BLOCKS_KEEP_SIZE);
513 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
514 up_read((&EXT4_I(inode)->i_data_sem));
516 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
517 * because it shouldn't be marked in es_map->m_flags.
519 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
522 * We don't check m_len because extent will be collpased in status
523 * tree. So the m_len might not equal.
525 if (es_map->m_lblk != map->m_lblk ||
526 es_map->m_flags != map->m_flags ||
527 es_map->m_pblk != map->m_pblk) {
528 printk("ES cache assertation failed for inode: %lu "
529 "es_cached ex [%d/%d/%llu/%x] != "
530 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
531 inode->i_ino, es_map->m_lblk, es_map->m_len,
532 es_map->m_pblk, es_map->m_flags, map->m_lblk,
533 map->m_len, map->m_pblk, map->m_flags,
537 #endif /* ES_AGGRESSIVE_TEST */
540 * The ext4_map_blocks() function tries to look up the requested blocks,
541 * and returns if the blocks are already mapped.
543 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
544 * and store the allocated blocks in the result buffer head and mark it
547 * If file type is extents based, it will call ext4_ext_map_blocks(),
548 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
551 * On success, it returns the number of blocks being mapped or allocate.
552 * if create==0 and the blocks are pre-allocated and uninitialized block,
553 * the result buffer head is unmapped. If the create ==1, it will make sure
554 * the buffer head is mapped.
556 * It returns 0 if plain look up failed (blocks have not been allocated), in
557 * that case, buffer head is unmapped
559 * It returns the error in case of allocation failure.
561 int ext4_map_blocks(handle_t *handle, struct inode *inode,
562 struct ext4_map_blocks *map, int flags)
564 struct extent_status es;
566 #ifdef ES_AGGRESSIVE_TEST
567 struct ext4_map_blocks orig_map;
569 memcpy(&orig_map, map, sizeof(*map));
573 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
574 "logical block %lu\n", inode->i_ino, flags, map->m_len,
575 (unsigned long) map->m_lblk);
577 /* Lookup extent status tree firstly */
578 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
579 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
580 map->m_pblk = ext4_es_pblock(&es) +
581 map->m_lblk - es.es_lblk;
582 map->m_flags |= ext4_es_is_written(&es) ?
583 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
584 retval = es.es_len - (map->m_lblk - es.es_lblk);
585 if (retval > map->m_len)
588 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
593 #ifdef ES_AGGRESSIVE_TEST
594 ext4_map_blocks_es_recheck(handle, inode, map,
601 * Try to see if we can get the block without requesting a new
604 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
605 down_read((&EXT4_I(inode)->i_data_sem));
606 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
607 retval = ext4_ext_map_blocks(handle, inode, map, flags &
608 EXT4_GET_BLOCKS_KEEP_SIZE);
610 retval = ext4_ind_map_blocks(handle, inode, map, flags &
611 EXT4_GET_BLOCKS_KEEP_SIZE);
615 unsigned long long status;
617 #ifdef ES_AGGRESSIVE_TEST
618 if (retval != map->m_len) {
619 printk("ES len assertation failed for inode: %lu "
620 "retval %d != map->m_len %d "
621 "in %s (lookup)\n", inode->i_ino, retval,
622 map->m_len, __func__);
626 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
627 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
628 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
629 ext4_find_delalloc_range(inode, map->m_lblk,
630 map->m_lblk + map->m_len - 1))
631 status |= EXTENT_STATUS_DELAYED;
632 ret = ext4_es_insert_extent(inode, map->m_lblk,
633 map->m_len, map->m_pblk, status);
637 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
638 up_read((&EXT4_I(inode)->i_data_sem));
641 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
642 int ret = check_block_validity(inode, map);
647 /* If it is only a block(s) look up */
648 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
652 * Returns if the blocks have already allocated
654 * Note that if blocks have been preallocated
655 * ext4_ext_get_block() returns the create = 0
656 * with buffer head unmapped.
658 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
662 * Here we clear m_flags because after allocating an new extent,
663 * it will be set again.
665 map->m_flags &= ~EXT4_MAP_FLAGS;
668 * New blocks allocate and/or writing to uninitialized extent
669 * will possibly result in updating i_data, so we take
670 * the write lock of i_data_sem, and call get_blocks()
671 * with create == 1 flag.
673 down_write((&EXT4_I(inode)->i_data_sem));
676 * if the caller is from delayed allocation writeout path
677 * we have already reserved fs blocks for allocation
678 * let the underlying get_block() function know to
679 * avoid double accounting
681 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
682 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
684 * We need to check for EXT4 here because migrate
685 * could have changed the inode type in between
687 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
688 retval = ext4_ext_map_blocks(handle, inode, map, flags);
690 retval = ext4_ind_map_blocks(handle, inode, map, flags);
692 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
694 * We allocated new blocks which will result in
695 * i_data's format changing. Force the migrate
696 * to fail by clearing migrate flags
698 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
702 * Update reserved blocks/metadata blocks after successful
703 * block allocation which had been deferred till now. We don't
704 * support fallocate for non extent files. So we can update
705 * reserve space here.
708 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
709 ext4_da_update_reserve_space(inode, retval, 1);
711 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
712 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
716 unsigned long long status;
718 #ifdef ES_AGGRESSIVE_TEST
719 if (retval != map->m_len) {
720 printk("ES len assertation failed for inode: %lu "
721 "retval %d != map->m_len %d "
722 "in %s (allocation)\n", inode->i_ino, retval,
723 map->m_len, __func__);
728 * If the extent has been zeroed out, we don't need to update
729 * extent status tree.
731 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
732 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
733 if (ext4_es_is_written(&es))
736 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
737 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
738 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
739 ext4_find_delalloc_range(inode, map->m_lblk,
740 map->m_lblk + map->m_len - 1))
741 status |= EXTENT_STATUS_DELAYED;
742 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
743 map->m_pblk, status);
749 up_write((&EXT4_I(inode)->i_data_sem));
750 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
751 int ret = check_block_validity(inode, map);
758 /* Maximum number of blocks we map for direct IO at once. */
759 #define DIO_MAX_BLOCKS 4096
761 static int _ext4_get_block(struct inode *inode, sector_t iblock,
762 struct buffer_head *bh, int flags)
764 handle_t *handle = ext4_journal_current_handle();
765 struct ext4_map_blocks map;
766 int ret = 0, started = 0;
769 if (ext4_has_inline_data(inode))
773 map.m_len = bh->b_size >> inode->i_blkbits;
775 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
776 /* Direct IO write... */
777 if (map.m_len > DIO_MAX_BLOCKS)
778 map.m_len = DIO_MAX_BLOCKS;
779 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
780 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
782 if (IS_ERR(handle)) {
783 ret = PTR_ERR(handle);
789 ret = ext4_map_blocks(handle, inode, &map, flags);
791 map_bh(bh, inode->i_sb, map.m_pblk);
792 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
793 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
797 ext4_journal_stop(handle);
801 int ext4_get_block(struct inode *inode, sector_t iblock,
802 struct buffer_head *bh, int create)
804 return _ext4_get_block(inode, iblock, bh,
805 create ? EXT4_GET_BLOCKS_CREATE : 0);
809 * `handle' can be NULL if create is zero
811 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
812 ext4_lblk_t block, int create, int *errp)
814 struct ext4_map_blocks map;
815 struct buffer_head *bh;
818 J_ASSERT(handle != NULL || create == 0);
822 err = ext4_map_blocks(handle, inode, &map,
823 create ? EXT4_GET_BLOCKS_CREATE : 0);
825 /* ensure we send some value back into *errp */
828 if (create && err == 0)
829 err = -ENOSPC; /* should never happen */
835 bh = sb_getblk(inode->i_sb, map.m_pblk);
840 if (map.m_flags & EXT4_MAP_NEW) {
841 J_ASSERT(create != 0);
842 J_ASSERT(handle != NULL);
845 * Now that we do not always journal data, we should
846 * keep in mind whether this should always journal the
847 * new buffer as metadata. For now, regular file
848 * writes use ext4_get_block instead, so it's not a
852 BUFFER_TRACE(bh, "call get_create_access");
853 fatal = ext4_journal_get_create_access(handle, bh);
854 if (!fatal && !buffer_uptodate(bh)) {
855 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
856 set_buffer_uptodate(bh);
859 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
860 err = ext4_handle_dirty_metadata(handle, inode, bh);
864 BUFFER_TRACE(bh, "not a new buffer");
874 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
875 ext4_lblk_t block, int create, int *err)
877 struct buffer_head *bh;
879 bh = ext4_getblk(handle, inode, block, create, err);
882 if (buffer_uptodate(bh))
884 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
886 if (buffer_uptodate(bh))
893 int ext4_walk_page_buffers(handle_t *handle,
894 struct buffer_head *head,
898 int (*fn)(handle_t *handle,
899 struct buffer_head *bh))
901 struct buffer_head *bh;
902 unsigned block_start, block_end;
903 unsigned blocksize = head->b_size;
905 struct buffer_head *next;
907 for (bh = head, block_start = 0;
908 ret == 0 && (bh != head || !block_start);
909 block_start = block_end, bh = next) {
910 next = bh->b_this_page;
911 block_end = block_start + blocksize;
912 if (block_end <= from || block_start >= to) {
913 if (partial && !buffer_uptodate(bh))
917 err = (*fn)(handle, bh);
925 * To preserve ordering, it is essential that the hole instantiation and
926 * the data write be encapsulated in a single transaction. We cannot
927 * close off a transaction and start a new one between the ext4_get_block()
928 * and the commit_write(). So doing the jbd2_journal_start at the start of
929 * prepare_write() is the right place.
931 * Also, this function can nest inside ext4_writepage(). In that case, we
932 * *know* that ext4_writepage() has generated enough buffer credits to do the
933 * whole page. So we won't block on the journal in that case, which is good,
934 * because the caller may be PF_MEMALLOC.
936 * By accident, ext4 can be reentered when a transaction is open via
937 * quota file writes. If we were to commit the transaction while thus
938 * reentered, there can be a deadlock - we would be holding a quota
939 * lock, and the commit would never complete if another thread had a
940 * transaction open and was blocking on the quota lock - a ranking
943 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
944 * will _not_ run commit under these circumstances because handle->h_ref
945 * is elevated. We'll still have enough credits for the tiny quotafile
948 int do_journal_get_write_access(handle_t *handle,
949 struct buffer_head *bh)
951 int dirty = buffer_dirty(bh);
954 if (!buffer_mapped(bh) || buffer_freed(bh))
957 * __block_write_begin() could have dirtied some buffers. Clean
958 * the dirty bit as jbd2_journal_get_write_access() could complain
959 * otherwise about fs integrity issues. Setting of the dirty bit
960 * by __block_write_begin() isn't a real problem here as we clear
961 * the bit before releasing a page lock and thus writeback cannot
962 * ever write the buffer.
965 clear_buffer_dirty(bh);
966 ret = ext4_journal_get_write_access(handle, bh);
968 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
972 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
973 struct buffer_head *bh_result, int create);
974 static int ext4_write_begin(struct file *file, struct address_space *mapping,
975 loff_t pos, unsigned len, unsigned flags,
976 struct page **pagep, void **fsdata)
978 struct inode *inode = mapping->host;
979 int ret, needed_blocks;
986 trace_ext4_write_begin(inode, pos, len, flags);
988 * Reserve one block more for addition to orphan list in case
989 * we allocate blocks but write fails for some reason
991 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
992 index = pos >> PAGE_CACHE_SHIFT;
993 from = pos & (PAGE_CACHE_SIZE - 1);
996 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
997 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1006 * grab_cache_page_write_begin() can take a long time if the
1007 * system is thrashing due to memory pressure, or if the page
1008 * is being written back. So grab it first before we start
1009 * the transaction handle. This also allows us to allocate
1010 * the page (if needed) without using GFP_NOFS.
1013 page = grab_cache_page_write_begin(mapping, index, flags);
1019 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1020 if (IS_ERR(handle)) {
1021 page_cache_release(page);
1022 return PTR_ERR(handle);
1026 if (page->mapping != mapping) {
1027 /* The page got truncated from under us */
1029 page_cache_release(page);
1030 ext4_journal_stop(handle);
1033 wait_on_page_writeback(page);
1035 if (ext4_should_dioread_nolock(inode))
1036 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1038 ret = __block_write_begin(page, pos, len, ext4_get_block);
1040 if (!ret && ext4_should_journal_data(inode)) {
1041 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1043 do_journal_get_write_access);
1049 * __block_write_begin may have instantiated a few blocks
1050 * outside i_size. Trim these off again. Don't need
1051 * i_size_read because we hold i_mutex.
1053 * Add inode to orphan list in case we crash before
1056 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1057 ext4_orphan_add(handle, inode);
1059 ext4_journal_stop(handle);
1060 if (pos + len > inode->i_size) {
1061 ext4_truncate_failed_write(inode);
1063 * If truncate failed early the inode might
1064 * still be on the orphan list; we need to
1065 * make sure the inode is removed from the
1066 * orphan list in that case.
1069 ext4_orphan_del(NULL, inode);
1072 if (ret == -ENOSPC &&
1073 ext4_should_retry_alloc(inode->i_sb, &retries))
1075 page_cache_release(page);
1082 /* For write_end() in data=journal mode */
1083 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1086 if (!buffer_mapped(bh) || buffer_freed(bh))
1088 set_buffer_uptodate(bh);
1089 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1090 clear_buffer_meta(bh);
1091 clear_buffer_prio(bh);
1096 * We need to pick up the new inode size which generic_commit_write gave us
1097 * `file' can be NULL - eg, when called from page_symlink().
1099 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1100 * buffers are managed internally.
1102 static int ext4_write_end(struct file *file,
1103 struct address_space *mapping,
1104 loff_t pos, unsigned len, unsigned copied,
1105 struct page *page, void *fsdata)
1107 handle_t *handle = ext4_journal_current_handle();
1108 struct inode *inode = mapping->host;
1110 int i_size_changed = 0;
1112 trace_ext4_write_end(inode, pos, len, copied);
1113 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1114 ret = ext4_jbd2_file_inode(handle, inode);
1117 page_cache_release(page);
1122 if (ext4_has_inline_data(inode)) {
1123 ret = ext4_write_inline_data_end(inode, pos, len,
1129 copied = block_write_end(file, mapping, pos,
1130 len, copied, page, fsdata);
1133 * No need to use i_size_read() here, the i_size
1134 * cannot change under us because we hole i_mutex.
1136 * But it's important to update i_size while still holding page lock:
1137 * page writeout could otherwise come in and zero beyond i_size.
1139 if (pos + copied > inode->i_size) {
1140 i_size_write(inode, pos + copied);
1144 if (pos + copied > EXT4_I(inode)->i_disksize) {
1145 /* We need to mark inode dirty even if
1146 * new_i_size is less that inode->i_size
1147 * but greater than i_disksize. (hint delalloc)
1149 ext4_update_i_disksize(inode, (pos + copied));
1153 page_cache_release(page);
1156 * Don't mark the inode dirty under page lock. First, it unnecessarily
1157 * makes the holding time of page lock longer. Second, it forces lock
1158 * ordering of page lock and transaction start for journaling
1162 ext4_mark_inode_dirty(handle, inode);
1166 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1167 /* if we have allocated more blocks and copied
1168 * less. We will have blocks allocated outside
1169 * inode->i_size. So truncate them
1171 ext4_orphan_add(handle, inode);
1173 ret2 = ext4_journal_stop(handle);
1177 if (pos + len > inode->i_size) {
1178 ext4_truncate_failed_write(inode);
1180 * If truncate failed early the inode might still be
1181 * on the orphan list; we need to make sure the inode
1182 * is removed from the orphan list in that case.
1185 ext4_orphan_del(NULL, inode);
1188 return ret ? ret : copied;
1191 static int ext4_journalled_write_end(struct file *file,
1192 struct address_space *mapping,
1193 loff_t pos, unsigned len, unsigned copied,
1194 struct page *page, void *fsdata)
1196 handle_t *handle = ext4_journal_current_handle();
1197 struct inode *inode = mapping->host;
1203 trace_ext4_journalled_write_end(inode, pos, len, copied);
1204 from = pos & (PAGE_CACHE_SIZE - 1);
1207 BUG_ON(!ext4_handle_valid(handle));
1209 if (ext4_has_inline_data(inode))
1210 copied = ext4_write_inline_data_end(inode, pos, len,
1214 if (!PageUptodate(page))
1216 page_zero_new_buffers(page, from+copied, to);
1219 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1220 to, &partial, write_end_fn);
1222 SetPageUptodate(page);
1224 new_i_size = pos + copied;
1225 if (new_i_size > inode->i_size)
1226 i_size_write(inode, pos+copied);
1227 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1228 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1229 if (new_i_size > EXT4_I(inode)->i_disksize) {
1230 ext4_update_i_disksize(inode, new_i_size);
1231 ret2 = ext4_mark_inode_dirty(handle, inode);
1237 page_cache_release(page);
1238 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1239 /* if we have allocated more blocks and copied
1240 * less. We will have blocks allocated outside
1241 * inode->i_size. So truncate them
1243 ext4_orphan_add(handle, inode);
1245 ret2 = ext4_journal_stop(handle);
1248 if (pos + len > inode->i_size) {
1249 ext4_truncate_failed_write(inode);
1251 * If truncate failed early the inode might still be
1252 * on the orphan list; we need to make sure the inode
1253 * is removed from the orphan list in that case.
1256 ext4_orphan_del(NULL, inode);
1259 return ret ? ret : copied;
1263 * Reserve a metadata for a single block located at lblock
1265 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1267 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1268 struct ext4_inode_info *ei = EXT4_I(inode);
1269 unsigned int md_needed;
1270 ext4_lblk_t save_last_lblock;
1274 * recalculate the amount of metadata blocks to reserve
1275 * in order to allocate nrblocks
1276 * worse case is one extent per block
1278 spin_lock(&ei->i_block_reservation_lock);
1280 * ext4_calc_metadata_amount() has side effects, which we have
1281 * to be prepared undo if we fail to claim space.
1283 save_len = ei->i_da_metadata_calc_len;
1284 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1285 md_needed = EXT4_NUM_B2C(sbi,
1286 ext4_calc_metadata_amount(inode, lblock));
1287 trace_ext4_da_reserve_space(inode, md_needed);
1290 * We do still charge estimated metadata to the sb though;
1291 * we cannot afford to run out of free blocks.
1293 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1294 ei->i_da_metadata_calc_len = save_len;
1295 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1296 spin_unlock(&ei->i_block_reservation_lock);
1299 ei->i_reserved_meta_blocks += md_needed;
1300 spin_unlock(&ei->i_block_reservation_lock);
1302 return 0; /* success */
1306 * Reserve a single cluster located at lblock
1308 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1310 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1311 struct ext4_inode_info *ei = EXT4_I(inode);
1312 unsigned int md_needed;
1314 ext4_lblk_t save_last_lblock;
1318 * We will charge metadata quota at writeout time; this saves
1319 * us from metadata over-estimation, though we may go over by
1320 * a small amount in the end. Here we just reserve for data.
1322 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1327 * recalculate the amount of metadata blocks to reserve
1328 * in order to allocate nrblocks
1329 * worse case is one extent per block
1331 spin_lock(&ei->i_block_reservation_lock);
1333 * ext4_calc_metadata_amount() has side effects, which we have
1334 * to be prepared undo if we fail to claim space.
1336 save_len = ei->i_da_metadata_calc_len;
1337 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1338 md_needed = EXT4_NUM_B2C(sbi,
1339 ext4_calc_metadata_amount(inode, lblock));
1340 trace_ext4_da_reserve_space(inode, md_needed);
1343 * We do still charge estimated metadata to the sb though;
1344 * we cannot afford to run out of free blocks.
1346 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1347 ei->i_da_metadata_calc_len = save_len;
1348 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1349 spin_unlock(&ei->i_block_reservation_lock);
1350 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1353 ei->i_reserved_data_blocks++;
1354 ei->i_reserved_meta_blocks += md_needed;
1355 spin_unlock(&ei->i_block_reservation_lock);
1357 return 0; /* success */
1360 static void ext4_da_release_space(struct inode *inode, int to_free)
1362 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1363 struct ext4_inode_info *ei = EXT4_I(inode);
1366 return; /* Nothing to release, exit */
1368 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1370 trace_ext4_da_release_space(inode, to_free);
1371 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1373 * if there aren't enough reserved blocks, then the
1374 * counter is messed up somewhere. Since this
1375 * function is called from invalidate page, it's
1376 * harmless to return without any action.
1378 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1379 "ino %lu, to_free %d with only %d reserved "
1380 "data blocks", inode->i_ino, to_free,
1381 ei->i_reserved_data_blocks);
1383 to_free = ei->i_reserved_data_blocks;
1385 ei->i_reserved_data_blocks -= to_free;
1387 if (ei->i_reserved_data_blocks == 0) {
1389 * We can release all of the reserved metadata blocks
1390 * only when we have written all of the delayed
1391 * allocation blocks.
1392 * Note that in case of bigalloc, i_reserved_meta_blocks,
1393 * i_reserved_data_blocks, etc. refer to number of clusters.
1395 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1396 ei->i_reserved_meta_blocks);
1397 ei->i_reserved_meta_blocks = 0;
1398 ei->i_da_metadata_calc_len = 0;
1401 /* update fs dirty data blocks counter */
1402 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1404 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1406 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1409 static void ext4_da_page_release_reservation(struct page *page,
1410 unsigned long offset)
1413 struct buffer_head *head, *bh;
1414 unsigned int curr_off = 0;
1415 struct inode *inode = page->mapping->host;
1416 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1420 head = page_buffers(page);
1423 unsigned int next_off = curr_off + bh->b_size;
1425 if ((offset <= curr_off) && (buffer_delay(bh))) {
1427 clear_buffer_delay(bh);
1429 curr_off = next_off;
1430 } while ((bh = bh->b_this_page) != head);
1433 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1434 ext4_es_remove_extent(inode, lblk, to_release);
1437 /* If we have released all the blocks belonging to a cluster, then we
1438 * need to release the reserved space for that cluster. */
1439 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1440 while (num_clusters > 0) {
1441 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1442 ((num_clusters - 1) << sbi->s_cluster_bits);
1443 if (sbi->s_cluster_ratio == 1 ||
1444 !ext4_find_delalloc_cluster(inode, lblk))
1445 ext4_da_release_space(inode, 1);
1452 * Delayed allocation stuff
1456 * mpage_da_submit_io - walks through extent of pages and try to write
1457 * them with writepage() call back
1459 * @mpd->inode: inode
1460 * @mpd->first_page: first page of the extent
1461 * @mpd->next_page: page after the last page of the extent
1463 * By the time mpage_da_submit_io() is called we expect all blocks
1464 * to be allocated. this may be wrong if allocation failed.
1466 * As pages are already locked by write_cache_pages(), we can't use it
1468 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1469 struct ext4_map_blocks *map)
1471 struct pagevec pvec;
1472 unsigned long index, end;
1473 int ret = 0, err, nr_pages, i;
1474 struct inode *inode = mpd->inode;
1475 struct address_space *mapping = inode->i_mapping;
1476 loff_t size = i_size_read(inode);
1477 unsigned int len, block_start;
1478 struct buffer_head *bh, *page_bufs = NULL;
1479 sector_t pblock = 0, cur_logical = 0;
1480 struct ext4_io_submit io_submit;
1482 BUG_ON(mpd->next_page <= mpd->first_page);
1483 memset(&io_submit, 0, sizeof(io_submit));
1485 * We need to start from the first_page to the next_page - 1
1486 * to make sure we also write the mapped dirty buffer_heads.
1487 * If we look at mpd->b_blocknr we would only be looking
1488 * at the currently mapped buffer_heads.
1490 index = mpd->first_page;
1491 end = mpd->next_page - 1;
1493 pagevec_init(&pvec, 0);
1494 while (index <= end) {
1495 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1498 for (i = 0; i < nr_pages; i++) {
1500 struct page *page = pvec.pages[i];
1502 index = page->index;
1506 if (index == size >> PAGE_CACHE_SHIFT)
1507 len = size & ~PAGE_CACHE_MASK;
1509 len = PAGE_CACHE_SIZE;
1511 cur_logical = index << (PAGE_CACHE_SHIFT -
1513 pblock = map->m_pblk + (cur_logical -
1518 BUG_ON(!PageLocked(page));
1519 BUG_ON(PageWriteback(page));
1521 bh = page_bufs = page_buffers(page);
1524 if (map && (cur_logical >= map->m_lblk) &&
1525 (cur_logical <= (map->m_lblk +
1526 (map->m_len - 1)))) {
1527 if (buffer_delay(bh)) {
1528 clear_buffer_delay(bh);
1529 bh->b_blocknr = pblock;
1531 if (buffer_unwritten(bh) ||
1533 BUG_ON(bh->b_blocknr != pblock);
1534 if (map->m_flags & EXT4_MAP_UNINIT)
1535 set_buffer_uninit(bh);
1536 clear_buffer_unwritten(bh);
1540 * skip page if block allocation undone and
1543 if (ext4_bh_delay_or_unwritten(NULL, bh))
1545 bh = bh->b_this_page;
1546 block_start += bh->b_size;
1549 } while (bh != page_bufs);
1556 clear_page_dirty_for_io(page);
1557 err = ext4_bio_write_page(&io_submit, page, len,
1560 mpd->pages_written++;
1562 * In error case, we have to continue because
1563 * remaining pages are still locked
1568 pagevec_release(&pvec);
1570 ext4_io_submit(&io_submit);
1574 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1578 struct pagevec pvec;
1579 struct inode *inode = mpd->inode;
1580 struct address_space *mapping = inode->i_mapping;
1581 ext4_lblk_t start, last;
1583 index = mpd->first_page;
1584 end = mpd->next_page - 1;
1586 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1587 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1588 ext4_es_remove_extent(inode, start, last - start + 1);
1590 pagevec_init(&pvec, 0);
1591 while (index <= end) {
1592 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1595 for (i = 0; i < nr_pages; i++) {
1596 struct page *page = pvec.pages[i];
1597 if (page->index > end)
1599 BUG_ON(!PageLocked(page));
1600 BUG_ON(PageWriteback(page));
1601 block_invalidatepage(page, 0);
1602 ClearPageUptodate(page);
1605 index = pvec.pages[nr_pages - 1]->index + 1;
1606 pagevec_release(&pvec);
1611 static void ext4_print_free_blocks(struct inode *inode)
1613 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1614 struct super_block *sb = inode->i_sb;
1615 struct ext4_inode_info *ei = EXT4_I(inode);
1617 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1618 EXT4_C2B(EXT4_SB(inode->i_sb),
1619 ext4_count_free_clusters(sb)));
1620 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1621 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1622 (long long) EXT4_C2B(EXT4_SB(sb),
1623 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1624 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1625 (long long) EXT4_C2B(EXT4_SB(sb),
1626 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1627 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1628 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1629 ei->i_reserved_data_blocks);
1630 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1631 ei->i_reserved_meta_blocks);
1632 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1633 ei->i_allocated_meta_blocks);
1638 * mpage_da_map_and_submit - go through given space, map them
1639 * if necessary, and then submit them for I/O
1641 * @mpd - bh describing space
1643 * The function skips space we know is already mapped to disk blocks.
1646 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1648 int err, blks, get_blocks_flags;
1649 struct ext4_map_blocks map, *mapp = NULL;
1650 sector_t next = mpd->b_blocknr;
1651 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1652 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1653 handle_t *handle = NULL;
1656 * If the blocks are mapped already, or we couldn't accumulate
1657 * any blocks, then proceed immediately to the submission stage.
1659 if ((mpd->b_size == 0) ||
1660 ((mpd->b_state & (1 << BH_Mapped)) &&
1661 !(mpd->b_state & (1 << BH_Delay)) &&
1662 !(mpd->b_state & (1 << BH_Unwritten))))
1665 handle = ext4_journal_current_handle();
1669 * Call ext4_map_blocks() to allocate any delayed allocation
1670 * blocks, or to convert an uninitialized extent to be
1671 * initialized (in the case where we have written into
1672 * one or more preallocated blocks).
1674 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1675 * indicate that we are on the delayed allocation path. This
1676 * affects functions in many different parts of the allocation
1677 * call path. This flag exists primarily because we don't
1678 * want to change *many* call functions, so ext4_map_blocks()
1679 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1680 * inode's allocation semaphore is taken.
1682 * If the blocks in questions were delalloc blocks, set
1683 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1684 * variables are updated after the blocks have been allocated.
1687 map.m_len = max_blocks;
1689 * We're in delalloc path and it is possible that we're going to
1690 * need more metadata blocks than previously reserved. However
1691 * we must not fail because we're in writeback and there is
1692 * nothing we can do about it so it might result in data loss.
1693 * So use reserved blocks to allocate metadata if possible.
1695 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
1696 EXT4_GET_BLOCKS_METADATA_NOFAIL;
1697 if (ext4_should_dioread_nolock(mpd->inode))
1698 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1699 if (mpd->b_state & (1 << BH_Delay))
1700 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1703 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1705 struct super_block *sb = mpd->inode->i_sb;
1709 * If get block returns EAGAIN or ENOSPC and there
1710 * appears to be free blocks we will just let
1711 * mpage_da_submit_io() unlock all of the pages.
1716 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1722 * get block failure will cause us to loop in
1723 * writepages, because a_ops->writepage won't be able
1724 * to make progress. The page will be redirtied by
1725 * writepage and writepages will again try to write
1728 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1729 ext4_msg(sb, KERN_CRIT,
1730 "delayed block allocation failed for inode %lu "
1731 "at logical offset %llu with max blocks %zd "
1732 "with error %d", mpd->inode->i_ino,
1733 (unsigned long long) next,
1734 mpd->b_size >> mpd->inode->i_blkbits, err);
1735 ext4_msg(sb, KERN_CRIT,
1736 "This should not happen!! Data will be lost");
1738 ext4_print_free_blocks(mpd->inode);
1740 /* invalidate all the pages */
1741 ext4_da_block_invalidatepages(mpd);
1743 /* Mark this page range as having been completed */
1750 if (map.m_flags & EXT4_MAP_NEW) {
1751 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1754 for (i = 0; i < map.m_len; i++)
1755 unmap_underlying_metadata(bdev, map.m_pblk + i);
1759 * Update on-disk size along with block allocation.
1761 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1762 if (disksize > i_size_read(mpd->inode))
1763 disksize = i_size_read(mpd->inode);
1764 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1765 ext4_update_i_disksize(mpd->inode, disksize);
1766 err = ext4_mark_inode_dirty(handle, mpd->inode);
1768 ext4_error(mpd->inode->i_sb,
1769 "Failed to mark inode %lu dirty",
1774 mpage_da_submit_io(mpd, mapp);
1778 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1779 (1 << BH_Delay) | (1 << BH_Unwritten))
1782 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1784 * @mpd->lbh - extent of blocks
1785 * @logical - logical number of the block in the file
1786 * @b_state - b_state of the buffer head added
1788 * the function is used to collect contig. blocks in same state
1790 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1791 unsigned long b_state)
1794 int blkbits = mpd->inode->i_blkbits;
1795 int nrblocks = mpd->b_size >> blkbits;
1798 * XXX Don't go larger than mballoc is willing to allocate
1799 * This is a stopgap solution. We eventually need to fold
1800 * mpage_da_submit_io() into this function and then call
1801 * ext4_map_blocks() multiple times in a loop
1803 if (nrblocks >= (8*1024*1024 >> blkbits))
1806 /* check if the reserved journal credits might overflow */
1807 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1808 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1810 * With non-extent format we are limited by the journal
1811 * credit available. Total credit needed to insert
1812 * nrblocks contiguous blocks is dependent on the
1813 * nrblocks. So limit nrblocks.
1819 * First block in the extent
1821 if (mpd->b_size == 0) {
1822 mpd->b_blocknr = logical;
1823 mpd->b_size = 1 << blkbits;
1824 mpd->b_state = b_state & BH_FLAGS;
1828 next = mpd->b_blocknr + nrblocks;
1830 * Can we merge the block to our big extent?
1832 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1833 mpd->b_size += 1 << blkbits;
1839 * We couldn't merge the block to our extent, so we
1840 * need to flush current extent and start new one
1842 mpage_da_map_and_submit(mpd);
1846 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1848 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1852 * This function is grabs code from the very beginning of
1853 * ext4_map_blocks, but assumes that the caller is from delayed write
1854 * time. This function looks up the requested blocks and sets the
1855 * buffer delay bit under the protection of i_data_sem.
1857 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1858 struct ext4_map_blocks *map,
1859 struct buffer_head *bh)
1861 struct extent_status es;
1863 sector_t invalid_block = ~((sector_t) 0xffff);
1864 #ifdef ES_AGGRESSIVE_TEST
1865 struct ext4_map_blocks orig_map;
1867 memcpy(&orig_map, map, sizeof(*map));
1870 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1874 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1875 "logical block %lu\n", inode->i_ino, map->m_len,
1876 (unsigned long) map->m_lblk);
1878 /* Lookup extent status tree firstly */
1879 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1881 if (ext4_es_is_hole(&es)) {
1883 down_read((&EXT4_I(inode)->i_data_sem));
1888 * Delayed extent could be allocated by fallocate.
1889 * So we need to check it.
1891 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1892 map_bh(bh, inode->i_sb, invalid_block);
1894 set_buffer_delay(bh);
1898 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1899 retval = es.es_len - (iblock - es.es_lblk);
1900 if (retval > map->m_len)
1901 retval = map->m_len;
1902 map->m_len = retval;
1903 if (ext4_es_is_written(&es))
1904 map->m_flags |= EXT4_MAP_MAPPED;
1905 else if (ext4_es_is_unwritten(&es))
1906 map->m_flags |= EXT4_MAP_UNWRITTEN;
1910 #ifdef ES_AGGRESSIVE_TEST
1911 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1917 * Try to see if we can get the block without requesting a new
1918 * file system block.
1920 down_read((&EXT4_I(inode)->i_data_sem));
1921 if (ext4_has_inline_data(inode)) {
1923 * We will soon create blocks for this page, and let
1924 * us pretend as if the blocks aren't allocated yet.
1925 * In case of clusters, we have to handle the work
1926 * of mapping from cluster so that the reserved space
1927 * is calculated properly.
1929 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1930 ext4_find_delalloc_cluster(inode, map->m_lblk))
1931 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1933 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1934 retval = ext4_ext_map_blocks(NULL, inode, map,
1935 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1937 retval = ext4_ind_map_blocks(NULL, inode, map,
1938 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1944 * XXX: __block_prepare_write() unmaps passed block,
1948 * If the block was allocated from previously allocated cluster,
1949 * then we don't need to reserve it again. However we still need
1950 * to reserve metadata for every block we're going to write.
1952 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1953 ret = ext4_da_reserve_space(inode, iblock);
1955 /* not enough space to reserve */
1960 ret = ext4_da_reserve_metadata(inode, iblock);
1962 /* not enough space to reserve */
1968 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1969 ~0, EXTENT_STATUS_DELAYED);
1975 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1976 * and it should not appear on the bh->b_state.
1978 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1980 map_bh(bh, inode->i_sb, invalid_block);
1982 set_buffer_delay(bh);
1983 } else if (retval > 0) {
1985 unsigned long long status;
1987 #ifdef ES_AGGRESSIVE_TEST
1988 if (retval != map->m_len) {
1989 printk("ES len assertation failed for inode: %lu "
1990 "retval %d != map->m_len %d "
1991 "in %s (lookup)\n", inode->i_ino, retval,
1992 map->m_len, __func__);
1996 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1997 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1998 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1999 map->m_pblk, status);
2005 up_read((&EXT4_I(inode)->i_data_sem));
2011 * This is a special get_blocks_t callback which is used by
2012 * ext4_da_write_begin(). It will either return mapped block or
2013 * reserve space for a single block.
2015 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2016 * We also have b_blocknr = -1 and b_bdev initialized properly
2018 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2019 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2020 * initialized properly.
2022 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2023 struct buffer_head *bh, int create)
2025 struct ext4_map_blocks map;
2028 BUG_ON(create == 0);
2029 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2031 map.m_lblk = iblock;
2035 * first, we need to know whether the block is allocated already
2036 * preallocated blocks are unmapped but should treated
2037 * the same as allocated blocks.
2039 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2043 map_bh(bh, inode->i_sb, map.m_pblk);
2044 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2046 if (buffer_unwritten(bh)) {
2047 /* A delayed write to unwritten bh should be marked
2048 * new and mapped. Mapped ensures that we don't do
2049 * get_block multiple times when we write to the same
2050 * offset and new ensures that we do proper zero out
2051 * for partial write.
2054 set_buffer_mapped(bh);
2059 static int bget_one(handle_t *handle, struct buffer_head *bh)
2065 static int bput_one(handle_t *handle, struct buffer_head *bh)
2071 static int __ext4_journalled_writepage(struct page *page,
2074 struct address_space *mapping = page->mapping;
2075 struct inode *inode = mapping->host;
2076 struct buffer_head *page_bufs = NULL;
2077 handle_t *handle = NULL;
2078 int ret = 0, err = 0;
2079 int inline_data = ext4_has_inline_data(inode);
2080 struct buffer_head *inode_bh = NULL;
2082 ClearPageChecked(page);
2085 BUG_ON(page->index != 0);
2086 BUG_ON(len > ext4_get_max_inline_size(inode));
2087 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2088 if (inode_bh == NULL)
2091 page_bufs = page_buffers(page);
2096 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2099 /* As soon as we unlock the page, it can go away, but we have
2100 * references to buffers so we are safe */
2103 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2104 ext4_writepage_trans_blocks(inode));
2105 if (IS_ERR(handle)) {
2106 ret = PTR_ERR(handle);
2110 BUG_ON(!ext4_handle_valid(handle));
2113 ret = ext4_journal_get_write_access(handle, inode_bh);
2115 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2118 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2119 do_journal_get_write_access);
2121 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2126 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2127 err = ext4_journal_stop(handle);
2131 if (!ext4_has_inline_data(inode))
2132 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2134 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2141 * Note that we don't need to start a transaction unless we're journaling data
2142 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2143 * need to file the inode to the transaction's list in ordered mode because if
2144 * we are writing back data added by write(), the inode is already there and if
2145 * we are writing back data modified via mmap(), no one guarantees in which
2146 * transaction the data will hit the disk. In case we are journaling data, we
2147 * cannot start transaction directly because transaction start ranks above page
2148 * lock so we have to do some magic.
2150 * This function can get called via...
2151 * - ext4_da_writepages after taking page lock (have journal handle)
2152 * - journal_submit_inode_data_buffers (no journal handle)
2153 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2154 * - grab_page_cache when doing write_begin (have journal handle)
2156 * We don't do any block allocation in this function. If we have page with
2157 * multiple blocks we need to write those buffer_heads that are mapped. This
2158 * is important for mmaped based write. So if we do with blocksize 1K
2159 * truncate(f, 1024);
2160 * a = mmap(f, 0, 4096);
2162 * truncate(f, 4096);
2163 * we have in the page first buffer_head mapped via page_mkwrite call back
2164 * but other buffer_heads would be unmapped but dirty (dirty done via the
2165 * do_wp_page). So writepage should write the first block. If we modify
2166 * the mmap area beyond 1024 we will again get a page_fault and the
2167 * page_mkwrite callback will do the block allocation and mark the
2168 * buffer_heads mapped.
2170 * We redirty the page if we have any buffer_heads that is either delay or
2171 * unwritten in the page.
2173 * We can get recursively called as show below.
2175 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2178 * But since we don't do any block allocation we should not deadlock.
2179 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2181 static int ext4_writepage(struct page *page,
2182 struct writeback_control *wbc)
2187 struct buffer_head *page_bufs = NULL;
2188 struct inode *inode = page->mapping->host;
2189 struct ext4_io_submit io_submit;
2191 trace_ext4_writepage(page);
2192 size = i_size_read(inode);
2193 if (page->index == size >> PAGE_CACHE_SHIFT)
2194 len = size & ~PAGE_CACHE_MASK;
2196 len = PAGE_CACHE_SIZE;
2198 page_bufs = page_buffers(page);
2200 * We cannot do block allocation or other extent handling in this
2201 * function. If there are buffers needing that, we have to redirty
2202 * the page. But we may reach here when we do a journal commit via
2203 * journal_submit_inode_data_buffers() and in that case we must write
2204 * allocated buffers to achieve data=ordered mode guarantees.
2206 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2207 ext4_bh_delay_or_unwritten)) {
2208 redirty_page_for_writepage(wbc, page);
2209 if (current->flags & PF_MEMALLOC) {
2211 * For memory cleaning there's no point in writing only
2212 * some buffers. So just bail out. Warn if we came here
2213 * from direct reclaim.
2215 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2222 if (PageChecked(page) && ext4_should_journal_data(inode))
2224 * It's mmapped pagecache. Add buffers and journal it. There
2225 * doesn't seem much point in redirtying the page here.
2227 return __ext4_journalled_writepage(page, len);
2229 memset(&io_submit, 0, sizeof(io_submit));
2230 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2231 ext4_io_submit(&io_submit);
2236 * This is called via ext4_da_writepages() to
2237 * calculate the total number of credits to reserve to fit
2238 * a single extent allocation into a single transaction,
2239 * ext4_da_writpeages() will loop calling this before
2240 * the block allocation.
2243 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2245 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2248 * With non-extent format the journal credit needed to
2249 * insert nrblocks contiguous block is dependent on
2250 * number of contiguous block. So we will limit
2251 * number of contiguous block to a sane value
2253 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2254 (max_blocks > EXT4_MAX_TRANS_DATA))
2255 max_blocks = EXT4_MAX_TRANS_DATA;
2257 return ext4_chunk_trans_blocks(inode, max_blocks);
2261 * write_cache_pages_da - walk the list of dirty pages of the given
2262 * address space and accumulate pages that need writing, and call
2263 * mpage_da_map_and_submit to map a single contiguous memory region
2264 * and then write them.
2266 static int write_cache_pages_da(handle_t *handle,
2267 struct address_space *mapping,
2268 struct writeback_control *wbc,
2269 struct mpage_da_data *mpd,
2270 pgoff_t *done_index)
2272 struct buffer_head *bh, *head;
2273 struct inode *inode = mapping->host;
2274 struct pagevec pvec;
2275 unsigned int nr_pages;
2278 long nr_to_write = wbc->nr_to_write;
2279 int i, tag, ret = 0;
2281 memset(mpd, 0, sizeof(struct mpage_da_data));
2284 pagevec_init(&pvec, 0);
2285 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2286 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2288 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2289 tag = PAGECACHE_TAG_TOWRITE;
2291 tag = PAGECACHE_TAG_DIRTY;
2293 *done_index = index;
2294 while (index <= end) {
2295 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2296 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2300 for (i = 0; i < nr_pages; i++) {
2301 struct page *page = pvec.pages[i];
2304 * At this point, the page may be truncated or
2305 * invalidated (changing page->mapping to NULL), or
2306 * even swizzled back from swapper_space to tmpfs file
2307 * mapping. However, page->index will not change
2308 * because we have a reference on the page.
2310 if (page->index > end)
2313 *done_index = page->index + 1;
2316 * If we can't merge this page, and we have
2317 * accumulated an contiguous region, write it
2319 if ((mpd->next_page != page->index) &&
2320 (mpd->next_page != mpd->first_page)) {
2321 mpage_da_map_and_submit(mpd);
2322 goto ret_extent_tail;
2328 * If the page is no longer dirty, or its
2329 * mapping no longer corresponds to inode we
2330 * are writing (which means it has been
2331 * truncated or invalidated), or the page is
2332 * already under writeback and we are not
2333 * doing a data integrity writeback, skip the page
2335 if (!PageDirty(page) ||
2336 (PageWriteback(page) &&
2337 (wbc->sync_mode == WB_SYNC_NONE)) ||
2338 unlikely(page->mapping != mapping)) {
2343 wait_on_page_writeback(page);
2344 BUG_ON(PageWriteback(page));
2347 * If we have inline data and arrive here, it means that
2348 * we will soon create the block for the 1st page, so
2349 * we'd better clear the inline data here.
2351 if (ext4_has_inline_data(inode)) {
2352 BUG_ON(ext4_test_inode_state(inode,
2353 EXT4_STATE_MAY_INLINE_DATA));
2354 ext4_destroy_inline_data(handle, inode);
2357 if (mpd->next_page != page->index)
2358 mpd->first_page = page->index;
2359 mpd->next_page = page->index + 1;
2360 logical = (sector_t) page->index <<
2361 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2363 /* Add all dirty buffers to mpd */
2364 head = page_buffers(page);
2367 BUG_ON(buffer_locked(bh));
2369 * We need to try to allocate unmapped blocks
2370 * in the same page. Otherwise we won't make
2371 * progress with the page in ext4_writepage
2373 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2374 mpage_add_bh_to_extent(mpd, logical,
2377 goto ret_extent_tail;
2378 } else if (buffer_dirty(bh) &&
2379 buffer_mapped(bh)) {
2381 * mapped dirty buffer. We need to
2382 * update the b_state because we look
2383 * at b_state in mpage_da_map_blocks.
2384 * We don't update b_size because if we
2385 * find an unmapped buffer_head later
2386 * we need to use the b_state flag of
2389 if (mpd->b_size == 0)
2391 bh->b_state & BH_FLAGS;
2394 } while ((bh = bh->b_this_page) != head);
2396 if (nr_to_write > 0) {
2398 if (nr_to_write == 0 &&
2399 wbc->sync_mode == WB_SYNC_NONE)
2401 * We stop writing back only if we are
2402 * not doing integrity sync. In case of
2403 * integrity sync we have to keep going
2404 * because someone may be concurrently
2405 * dirtying pages, and we might have
2406 * synced a lot of newly appeared dirty
2407 * pages, but have not synced all of the
2413 pagevec_release(&pvec);
2418 ret = MPAGE_DA_EXTENT_TAIL;
2420 pagevec_release(&pvec);
2426 static int ext4_da_writepages(struct address_space *mapping,
2427 struct writeback_control *wbc)
2430 int range_whole = 0;
2431 handle_t *handle = NULL;
2432 struct mpage_da_data mpd;
2433 struct inode *inode = mapping->host;
2434 int pages_written = 0;
2435 unsigned int max_pages;
2436 int range_cyclic, cycled = 1, io_done = 0;
2437 int needed_blocks, ret = 0;
2438 long desired_nr_to_write, nr_to_writebump = 0;
2439 loff_t range_start = wbc->range_start;
2440 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2441 pgoff_t done_index = 0;
2443 struct blk_plug plug;
2445 trace_ext4_da_writepages(inode, wbc);
2448 * No pages to write? This is mainly a kludge to avoid starting
2449 * a transaction for special inodes like journal inode on last iput()
2450 * because that could violate lock ordering on umount
2452 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2456 * If the filesystem has aborted, it is read-only, so return
2457 * right away instead of dumping stack traces later on that
2458 * will obscure the real source of the problem. We test
2459 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2460 * the latter could be true if the filesystem is mounted
2461 * read-only, and in that case, ext4_da_writepages should
2462 * *never* be called, so if that ever happens, we would want
2465 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2468 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2471 range_cyclic = wbc->range_cyclic;
2472 if (wbc->range_cyclic) {
2473 index = mapping->writeback_index;
2476 wbc->range_start = index << PAGE_CACHE_SHIFT;
2477 wbc->range_end = LLONG_MAX;
2478 wbc->range_cyclic = 0;
2481 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2482 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2486 * This works around two forms of stupidity. The first is in
2487 * the writeback code, which caps the maximum number of pages
2488 * written to be 1024 pages. This is wrong on multiple
2489 * levels; different architectues have a different page size,
2490 * which changes the maximum amount of data which gets
2491 * written. Secondly, 4 megabytes is way too small. XFS
2492 * forces this value to be 16 megabytes by multiplying
2493 * nr_to_write parameter by four, and then relies on its
2494 * allocator to allocate larger extents to make them
2495 * contiguous. Unfortunately this brings us to the second
2496 * stupidity, which is that ext4's mballoc code only allocates
2497 * at most 2048 blocks. So we force contiguous writes up to
2498 * the number of dirty blocks in the inode, or
2499 * sbi->max_writeback_mb_bump whichever is smaller.
2501 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2502 if (!range_cyclic && range_whole) {
2503 if (wbc->nr_to_write == LONG_MAX)
2504 desired_nr_to_write = wbc->nr_to_write;
2506 desired_nr_to_write = wbc->nr_to_write * 8;
2508 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2510 if (desired_nr_to_write > max_pages)
2511 desired_nr_to_write = max_pages;
2513 if (wbc->nr_to_write < desired_nr_to_write) {
2514 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2515 wbc->nr_to_write = desired_nr_to_write;
2519 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2520 tag_pages_for_writeback(mapping, index, end);
2522 blk_start_plug(&plug);
2523 while (!ret && wbc->nr_to_write > 0) {
2526 * we insert one extent at a time. So we need
2527 * credit needed for single extent allocation.
2528 * journalled mode is currently not supported
2531 BUG_ON(ext4_should_journal_data(inode));
2532 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2534 /* start a new transaction*/
2535 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2537 if (IS_ERR(handle)) {
2538 ret = PTR_ERR(handle);
2539 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2540 "%ld pages, ino %lu; err %d", __func__,
2541 wbc->nr_to_write, inode->i_ino, ret);
2542 blk_finish_plug(&plug);
2543 goto out_writepages;
2547 * Now call write_cache_pages_da() to find the next
2548 * contiguous region of logical blocks that need
2549 * blocks to be allocated by ext4 and submit them.
2551 ret = write_cache_pages_da(handle, mapping,
2552 wbc, &mpd, &done_index);
2554 * If we have a contiguous extent of pages and we
2555 * haven't done the I/O yet, map the blocks and submit
2558 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2559 mpage_da_map_and_submit(&mpd);
2560 ret = MPAGE_DA_EXTENT_TAIL;
2562 trace_ext4_da_write_pages(inode, &mpd);
2563 wbc->nr_to_write -= mpd.pages_written;
2565 ext4_journal_stop(handle);
2567 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2568 /* commit the transaction which would
2569 * free blocks released in the transaction
2572 jbd2_journal_force_commit_nested(sbi->s_journal);
2574 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2576 * Got one extent now try with rest of the pages.
2577 * If mpd.retval is set -EIO, journal is aborted.
2578 * So we don't need to write any more.
2580 pages_written += mpd.pages_written;
2583 } else if (wbc->nr_to_write)
2585 * There is no more writeout needed
2586 * or we requested for a noblocking writeout
2587 * and we found the device congested
2591 blk_finish_plug(&plug);
2592 if (!io_done && !cycled) {
2595 wbc->range_start = index << PAGE_CACHE_SHIFT;
2596 wbc->range_end = mapping->writeback_index - 1;
2601 wbc->range_cyclic = range_cyclic;
2602 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2604 * set the writeback_index so that range_cyclic
2605 * mode will write it back later
2607 mapping->writeback_index = done_index;
2610 wbc->nr_to_write -= nr_to_writebump;
2611 wbc->range_start = range_start;
2612 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2616 static int ext4_nonda_switch(struct super_block *sb)
2618 s64 free_clusters, dirty_clusters;
2619 struct ext4_sb_info *sbi = EXT4_SB(sb);
2622 * switch to non delalloc mode if we are running low
2623 * on free block. The free block accounting via percpu
2624 * counters can get slightly wrong with percpu_counter_batch getting
2625 * accumulated on each CPU without updating global counters
2626 * Delalloc need an accurate free block accounting. So switch
2627 * to non delalloc when we are near to error range.
2630 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2632 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2634 * Start pushing delalloc when 1/2 of free blocks are dirty.
2636 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2637 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2639 if (2 * free_clusters < 3 * dirty_clusters ||
2640 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2642 * free block count is less than 150% of dirty blocks
2643 * or free blocks is less than watermark
2650 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2651 loff_t pos, unsigned len, unsigned flags,
2652 struct page **pagep, void **fsdata)
2654 int ret, retries = 0;
2657 struct inode *inode = mapping->host;
2660 index = pos >> PAGE_CACHE_SHIFT;
2662 if (ext4_nonda_switch(inode->i_sb)) {
2663 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2664 return ext4_write_begin(file, mapping, pos,
2665 len, flags, pagep, fsdata);
2667 *fsdata = (void *)0;
2668 trace_ext4_da_write_begin(inode, pos, len, flags);
2670 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2671 ret = ext4_da_write_inline_data_begin(mapping, inode,
2681 * grab_cache_page_write_begin() can take a long time if the
2682 * system is thrashing due to memory pressure, or if the page
2683 * is being written back. So grab it first before we start
2684 * the transaction handle. This also allows us to allocate
2685 * the page (if needed) without using GFP_NOFS.
2688 page = grab_cache_page_write_begin(mapping, index, flags);
2694 * With delayed allocation, we don't log the i_disksize update
2695 * if there is delayed block allocation. But we still need
2696 * to journalling the i_disksize update if writes to the end
2697 * of file which has an already mapped buffer.
2700 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2701 if (IS_ERR(handle)) {
2702 page_cache_release(page);
2703 return PTR_ERR(handle);
2707 if (page->mapping != mapping) {
2708 /* The page got truncated from under us */
2710 page_cache_release(page);
2711 ext4_journal_stop(handle);
2714 /* In case writeback began while the page was unlocked */
2715 wait_on_page_writeback(page);
2717 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2720 ext4_journal_stop(handle);
2722 * block_write_begin may have instantiated a few blocks
2723 * outside i_size. Trim these off again. Don't need
2724 * i_size_read because we hold i_mutex.
2726 if (pos + len > inode->i_size)
2727 ext4_truncate_failed_write(inode);
2729 if (ret == -ENOSPC &&
2730 ext4_should_retry_alloc(inode->i_sb, &retries))
2733 page_cache_release(page);
2742 * Check if we should update i_disksize
2743 * when write to the end of file but not require block allocation
2745 static int ext4_da_should_update_i_disksize(struct page *page,
2746 unsigned long offset)
2748 struct buffer_head *bh;
2749 struct inode *inode = page->mapping->host;
2753 bh = page_buffers(page);
2754 idx = offset >> inode->i_blkbits;
2756 for (i = 0; i < idx; i++)
2757 bh = bh->b_this_page;
2759 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2764 static int ext4_da_write_end(struct file *file,
2765 struct address_space *mapping,
2766 loff_t pos, unsigned len, unsigned copied,
2767 struct page *page, void *fsdata)
2769 struct inode *inode = mapping->host;
2771 handle_t *handle = ext4_journal_current_handle();
2773 unsigned long start, end;
2774 int write_mode = (int)(unsigned long)fsdata;
2776 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2777 return ext4_write_end(file, mapping, pos,
2778 len, copied, page, fsdata);
2780 trace_ext4_da_write_end(inode, pos, len, copied);
2781 start = pos & (PAGE_CACHE_SIZE - 1);
2782 end = start + copied - 1;
2785 * generic_write_end() will run mark_inode_dirty() if i_size
2786 * changes. So let's piggyback the i_disksize mark_inode_dirty
2789 new_i_size = pos + copied;
2790 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2791 if (ext4_has_inline_data(inode) ||
2792 ext4_da_should_update_i_disksize(page, end)) {
2793 down_write(&EXT4_I(inode)->i_data_sem);
2794 if (new_i_size > EXT4_I(inode)->i_disksize)
2795 EXT4_I(inode)->i_disksize = new_i_size;
2796 up_write(&EXT4_I(inode)->i_data_sem);
2797 /* We need to mark inode dirty even if
2798 * new_i_size is less that inode->i_size
2799 * bu greater than i_disksize.(hint delalloc)
2801 ext4_mark_inode_dirty(handle, inode);
2805 if (write_mode != CONVERT_INLINE_DATA &&
2806 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2807 ext4_has_inline_data(inode))
2808 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2811 ret2 = generic_write_end(file, mapping, pos, len, copied,
2817 ret2 = ext4_journal_stop(handle);
2821 return ret ? ret : copied;
2824 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2827 * Drop reserved blocks
2829 BUG_ON(!PageLocked(page));
2830 if (!page_has_buffers(page))
2833 ext4_da_page_release_reservation(page, offset);
2836 ext4_invalidatepage(page, offset);
2842 * Force all delayed allocation blocks to be allocated for a given inode.
2844 int ext4_alloc_da_blocks(struct inode *inode)
2846 trace_ext4_alloc_da_blocks(inode);
2848 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2849 !EXT4_I(inode)->i_reserved_meta_blocks)
2853 * We do something simple for now. The filemap_flush() will
2854 * also start triggering a write of the data blocks, which is
2855 * not strictly speaking necessary (and for users of
2856 * laptop_mode, not even desirable). However, to do otherwise
2857 * would require replicating code paths in:
2859 * ext4_da_writepages() ->
2860 * write_cache_pages() ---> (via passed in callback function)
2861 * __mpage_da_writepage() -->
2862 * mpage_add_bh_to_extent()
2863 * mpage_da_map_blocks()
2865 * The problem is that write_cache_pages(), located in
2866 * mm/page-writeback.c, marks pages clean in preparation for
2867 * doing I/O, which is not desirable if we're not planning on
2870 * We could call write_cache_pages(), and then redirty all of
2871 * the pages by calling redirty_page_for_writepage() but that
2872 * would be ugly in the extreme. So instead we would need to
2873 * replicate parts of the code in the above functions,
2874 * simplifying them because we wouldn't actually intend to
2875 * write out the pages, but rather only collect contiguous
2876 * logical block extents, call the multi-block allocator, and
2877 * then update the buffer heads with the block allocations.
2879 * For now, though, we'll cheat by calling filemap_flush(),
2880 * which will map the blocks, and start the I/O, but not
2881 * actually wait for the I/O to complete.
2883 return filemap_flush(inode->i_mapping);
2887 * bmap() is special. It gets used by applications such as lilo and by
2888 * the swapper to find the on-disk block of a specific piece of data.
2890 * Naturally, this is dangerous if the block concerned is still in the
2891 * journal. If somebody makes a swapfile on an ext4 data-journaling
2892 * filesystem and enables swap, then they may get a nasty shock when the
2893 * data getting swapped to that swapfile suddenly gets overwritten by
2894 * the original zero's written out previously to the journal and
2895 * awaiting writeback in the kernel's buffer cache.
2897 * So, if we see any bmap calls here on a modified, data-journaled file,
2898 * take extra steps to flush any blocks which might be in the cache.
2900 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2902 struct inode *inode = mapping->host;
2907 * We can get here for an inline file via the FIBMAP ioctl
2909 if (ext4_has_inline_data(inode))
2912 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2913 test_opt(inode->i_sb, DELALLOC)) {
2915 * With delalloc we want to sync the file
2916 * so that we can make sure we allocate
2919 filemap_write_and_wait(mapping);
2922 if (EXT4_JOURNAL(inode) &&
2923 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2925 * This is a REALLY heavyweight approach, but the use of
2926 * bmap on dirty files is expected to be extremely rare:
2927 * only if we run lilo or swapon on a freshly made file
2928 * do we expect this to happen.
2930 * (bmap requires CAP_SYS_RAWIO so this does not
2931 * represent an unprivileged user DOS attack --- we'd be
2932 * in trouble if mortal users could trigger this path at
2935 * NB. EXT4_STATE_JDATA is not set on files other than
2936 * regular files. If somebody wants to bmap a directory
2937 * or symlink and gets confused because the buffer
2938 * hasn't yet been flushed to disk, they deserve
2939 * everything they get.
2942 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2943 journal = EXT4_JOURNAL(inode);
2944 jbd2_journal_lock_updates(journal);
2945 err = jbd2_journal_flush(journal);
2946 jbd2_journal_unlock_updates(journal);
2952 return generic_block_bmap(mapping, block, ext4_get_block);
2955 static int ext4_readpage(struct file *file, struct page *page)
2958 struct inode *inode = page->mapping->host;
2960 trace_ext4_readpage(page);
2962 if (ext4_has_inline_data(inode))
2963 ret = ext4_readpage_inline(inode, page);
2966 return mpage_readpage(page, ext4_get_block);
2972 ext4_readpages(struct file *file, struct address_space *mapping,
2973 struct list_head *pages, unsigned nr_pages)
2975 struct inode *inode = mapping->host;
2977 /* If the file has inline data, no need to do readpages. */
2978 if (ext4_has_inline_data(inode))
2981 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2984 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2986 trace_ext4_invalidatepage(page, offset);
2988 /* No journalling happens on data buffers when this function is used */
2989 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2991 block_invalidatepage(page, offset);
2994 static int __ext4_journalled_invalidatepage(struct page *page,
2995 unsigned long offset)
2997 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2999 trace_ext4_journalled_invalidatepage(page, offset);
3002 * If it's a full truncate we just forget about the pending dirtying
3005 ClearPageChecked(page);
3007 return jbd2_journal_invalidatepage(journal, page, offset);
3010 /* Wrapper for aops... */
3011 static void ext4_journalled_invalidatepage(struct page *page,
3012 unsigned long offset)
3014 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3017 static int ext4_releasepage(struct page *page, gfp_t wait)
3019 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3021 trace_ext4_releasepage(page);
3023 /* Page has dirty journalled data -> cannot release */
3024 if (PageChecked(page))
3027 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3029 return try_to_free_buffers(page);
3033 * ext4_get_block used when preparing for a DIO write or buffer write.
3034 * We allocate an uinitialized extent if blocks haven't been allocated.
3035 * The extent will be converted to initialized after the IO is complete.
3037 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3038 struct buffer_head *bh_result, int create)
3040 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3041 inode->i_ino, create);
3042 return _ext4_get_block(inode, iblock, bh_result,
3043 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3046 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3047 struct buffer_head *bh_result, int create)
3049 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3050 inode->i_ino, create);
3051 return _ext4_get_block(inode, iblock, bh_result,
3052 EXT4_GET_BLOCKS_NO_LOCK);
3055 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3056 ssize_t size, void *private, int ret,
3059 struct inode *inode = file_inode(iocb->ki_filp);
3060 ext4_io_end_t *io_end = iocb->private;
3062 /* if not async direct IO or dio with 0 bytes write, just return */
3063 if (!io_end || !size)
3066 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3067 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3068 iocb->private, io_end->inode->i_ino, iocb, offset,
3071 iocb->private = NULL;
3073 /* if not aio dio with unwritten extents, just free io and return */
3074 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3075 ext4_free_io_end(io_end);
3077 inode_dio_done(inode);
3079 aio_complete(iocb, ret, 0);
3083 io_end->offset = offset;
3084 io_end->size = size;
3086 io_end->iocb = iocb;
3087 io_end->result = ret;
3090 ext4_add_complete_io(io_end);
3094 * For ext4 extent files, ext4 will do direct-io write to holes,
3095 * preallocated extents, and those write extend the file, no need to
3096 * fall back to buffered IO.
3098 * For holes, we fallocate those blocks, mark them as uninitialized
3099 * If those blocks were preallocated, we mark sure they are split, but
3100 * still keep the range to write as uninitialized.
3102 * The unwritten extents will be converted to written when DIO is completed.
3103 * For async direct IO, since the IO may still pending when return, we
3104 * set up an end_io call back function, which will do the conversion
3105 * when async direct IO completed.
3107 * If the O_DIRECT write will extend the file then add this inode to the
3108 * orphan list. So recovery will truncate it back to the original size
3109 * if the machine crashes during the write.
3112 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3113 const struct iovec *iov, loff_t offset,
3114 unsigned long nr_segs)
3116 struct file *file = iocb->ki_filp;
3117 struct inode *inode = file->f_mapping->host;
3119 size_t count = iov_length(iov, nr_segs);
3121 get_block_t *get_block_func = NULL;
3123 loff_t final_size = offset + count;
3125 /* Use the old path for reads and writes beyond i_size. */
3126 if (rw != WRITE || final_size > inode->i_size)
3127 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3129 BUG_ON(iocb->private == NULL);
3131 /* If we do a overwrite dio, i_mutex locking can be released */
3132 overwrite = *((int *)iocb->private);
3135 atomic_inc(&inode->i_dio_count);
3136 down_read(&EXT4_I(inode)->i_data_sem);
3137 mutex_unlock(&inode->i_mutex);
3141 * We could direct write to holes and fallocate.
3143 * Allocated blocks to fill the hole are marked as
3144 * uninitialized to prevent parallel buffered read to expose
3145 * the stale data before DIO complete the data IO.
3147 * As to previously fallocated extents, ext4 get_block will
3148 * just simply mark the buffer mapped but still keep the
3149 * extents uninitialized.
3151 * For non AIO case, we will convert those unwritten extents
3152 * to written after return back from blockdev_direct_IO.
3154 * For async DIO, the conversion needs to be deferred when the
3155 * IO is completed. The ext4 end_io callback function will be
3156 * called to take care of the conversion work. Here for async
3157 * case, we allocate an io_end structure to hook to the iocb.
3159 iocb->private = NULL;
3160 ext4_inode_aio_set(inode, NULL);
3161 if (!is_sync_kiocb(iocb)) {
3162 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3167 io_end->flag |= EXT4_IO_END_DIRECT;
3168 iocb->private = io_end;
3170 * we save the io structure for current async direct
3171 * IO, so that later ext4_map_blocks() could flag the
3172 * io structure whether there is a unwritten extents
3173 * needs to be converted when IO is completed.
3175 ext4_inode_aio_set(inode, io_end);
3179 get_block_func = ext4_get_block_write_nolock;
3181 get_block_func = ext4_get_block_write;
3182 dio_flags = DIO_LOCKING;
3184 ret = __blockdev_direct_IO(rw, iocb, inode,
3185 inode->i_sb->s_bdev, iov,
3193 ext4_inode_aio_set(inode, NULL);
3195 * The io_end structure takes a reference to the inode, that
3196 * structure needs to be destroyed and the reference to the
3197 * inode need to be dropped, when IO is complete, even with 0
3198 * byte write, or failed.
3200 * In the successful AIO DIO case, the io_end structure will
3201 * be destroyed and the reference to the inode will be dropped
3202 * after the end_io call back function is called.
3204 * In the case there is 0 byte write, or error case, since VFS
3205 * direct IO won't invoke the end_io call back function, we
3206 * need to free the end_io structure here.
3208 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3209 ext4_free_io_end(iocb->private);
3210 iocb->private = NULL;
3211 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3212 EXT4_STATE_DIO_UNWRITTEN)) {
3215 * for non AIO case, since the IO is already
3216 * completed, we could do the conversion right here
3218 err = ext4_convert_unwritten_extents(inode,
3222 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3226 /* take i_mutex locking again if we do a ovewrite dio */
3228 inode_dio_done(inode);
3229 up_read(&EXT4_I(inode)->i_data_sem);
3230 mutex_lock(&inode->i_mutex);
3236 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3237 const struct iovec *iov, loff_t offset,
3238 unsigned long nr_segs)
3240 struct file *file = iocb->ki_filp;
3241 struct inode *inode = file->f_mapping->host;
3245 * If we are doing data journalling we don't support O_DIRECT
3247 if (ext4_should_journal_data(inode))
3250 /* Let buffer I/O handle the inline data case. */
3251 if (ext4_has_inline_data(inode))
3254 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3255 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3256 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3258 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3259 trace_ext4_direct_IO_exit(inode, offset,
3260 iov_length(iov, nr_segs), rw, ret);
3265 * Pages can be marked dirty completely asynchronously from ext4's journalling
3266 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3267 * much here because ->set_page_dirty is called under VFS locks. The page is
3268 * not necessarily locked.
3270 * We cannot just dirty the page and leave attached buffers clean, because the
3271 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3272 * or jbddirty because all the journalling code will explode.
3274 * So what we do is to mark the page "pending dirty" and next time writepage
3275 * is called, propagate that into the buffers appropriately.
3277 static int ext4_journalled_set_page_dirty(struct page *page)
3279 SetPageChecked(page);
3280 return __set_page_dirty_nobuffers(page);
3283 static const struct address_space_operations ext4_aops = {
3284 .readpage = ext4_readpage,
3285 .readpages = ext4_readpages,
3286 .writepage = ext4_writepage,
3287 .write_begin = ext4_write_begin,
3288 .write_end = ext4_write_end,
3290 .invalidatepage = ext4_invalidatepage,
3291 .releasepage = ext4_releasepage,
3292 .direct_IO = ext4_direct_IO,
3293 .migratepage = buffer_migrate_page,
3294 .is_partially_uptodate = block_is_partially_uptodate,
3295 .error_remove_page = generic_error_remove_page,
3298 static const struct address_space_operations ext4_journalled_aops = {
3299 .readpage = ext4_readpage,
3300 .readpages = ext4_readpages,
3301 .writepage = ext4_writepage,
3302 .write_begin = ext4_write_begin,
3303 .write_end = ext4_journalled_write_end,
3304 .set_page_dirty = ext4_journalled_set_page_dirty,
3306 .invalidatepage = ext4_journalled_invalidatepage,
3307 .releasepage = ext4_releasepage,
3308 .direct_IO = ext4_direct_IO,
3309 .is_partially_uptodate = block_is_partially_uptodate,
3310 .error_remove_page = generic_error_remove_page,
3313 static const struct address_space_operations ext4_da_aops = {
3314 .readpage = ext4_readpage,
3315 .readpages = ext4_readpages,
3316 .writepage = ext4_writepage,
3317 .writepages = ext4_da_writepages,
3318 .write_begin = ext4_da_write_begin,
3319 .write_end = ext4_da_write_end,
3321 .invalidatepage = ext4_da_invalidatepage,
3322 .releasepage = ext4_releasepage,
3323 .direct_IO = ext4_direct_IO,
3324 .migratepage = buffer_migrate_page,
3325 .is_partially_uptodate = block_is_partially_uptodate,
3326 .error_remove_page = generic_error_remove_page,
3329 void ext4_set_aops(struct inode *inode)
3331 switch (ext4_inode_journal_mode(inode)) {
3332 case EXT4_INODE_ORDERED_DATA_MODE:
3333 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3335 case EXT4_INODE_WRITEBACK_DATA_MODE:
3336 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3338 case EXT4_INODE_JOURNAL_DATA_MODE:
3339 inode->i_mapping->a_ops = &ext4_journalled_aops;
3344 if (test_opt(inode->i_sb, DELALLOC))
3345 inode->i_mapping->a_ops = &ext4_da_aops;
3347 inode->i_mapping->a_ops = &ext4_aops;
3352 * ext4_discard_partial_page_buffers()
3353 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3354 * This function finds and locks the page containing the offset
3355 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3356 * Calling functions that already have the page locked should call
3357 * ext4_discard_partial_page_buffers_no_lock directly.
3359 int ext4_discard_partial_page_buffers(handle_t *handle,
3360 struct address_space *mapping, loff_t from,
3361 loff_t length, int flags)
3363 struct inode *inode = mapping->host;
3367 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3368 mapping_gfp_mask(mapping) & ~__GFP_FS);
3372 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3373 from, length, flags);
3376 page_cache_release(page);
3381 * ext4_discard_partial_page_buffers_no_lock()
3382 * Zeros a page range of length 'length' starting from offset 'from'.
3383 * Buffer heads that correspond to the block aligned regions of the
3384 * zeroed range will be unmapped. Unblock aligned regions
3385 * will have the corresponding buffer head mapped if needed so that
3386 * that region of the page can be updated with the partial zero out.
3388 * This function assumes that the page has already been locked. The
3389 * The range to be discarded must be contained with in the given page.
3390 * If the specified range exceeds the end of the page it will be shortened
3391 * to the end of the page that corresponds to 'from'. This function is
3392 * appropriate for updating a page and it buffer heads to be unmapped and
3393 * zeroed for blocks that have been either released, or are going to be
3396 * handle: The journal handle
3397 * inode: The files inode
3398 * page: A locked page that contains the offset "from"
3399 * from: The starting byte offset (from the beginning of the file)
3400 * to begin discarding
3401 * len: The length of bytes to discard
3402 * flags: Optional flags that may be used:
3404 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3405 * Only zero the regions of the page whose buffer heads
3406 * have already been unmapped. This flag is appropriate
3407 * for updating the contents of a page whose blocks may
3408 * have already been released, and we only want to zero
3409 * out the regions that correspond to those released blocks.
3411 * Returns zero on success or negative on failure.
3413 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3414 struct inode *inode, struct page *page, loff_t from,
3415 loff_t length, int flags)
3417 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3418 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3419 unsigned int blocksize, max, pos;
3421 struct buffer_head *bh;
3424 blocksize = inode->i_sb->s_blocksize;
3425 max = PAGE_CACHE_SIZE - offset;
3427 if (index != page->index)
3431 * correct length if it does not fall between
3432 * 'from' and the end of the page
3434 if (length > max || length < 0)
3437 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3439 if (!page_has_buffers(page))
3440 create_empty_buffers(page, blocksize, 0);
3442 /* Find the buffer that contains "offset" */
3443 bh = page_buffers(page);
3445 while (offset >= pos) {
3446 bh = bh->b_this_page;
3452 while (pos < offset + length) {
3453 unsigned int end_of_block, range_to_discard;
3457 /* The length of space left to zero and unmap */
3458 range_to_discard = offset + length - pos;
3460 /* The length of space until the end of the block */
3461 end_of_block = blocksize - (pos & (blocksize-1));
3464 * Do not unmap or zero past end of block
3465 * for this buffer head
3467 if (range_to_discard > end_of_block)
3468 range_to_discard = end_of_block;
3472 * Skip this buffer head if we are only zeroing unampped
3473 * regions of the page
3475 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3479 /* If the range is block aligned, unmap */
3480 if (range_to_discard == blocksize) {
3481 clear_buffer_dirty(bh);
3483 clear_buffer_mapped(bh);
3484 clear_buffer_req(bh);
3485 clear_buffer_new(bh);
3486 clear_buffer_delay(bh);
3487 clear_buffer_unwritten(bh);
3488 clear_buffer_uptodate(bh);
3489 zero_user(page, pos, range_to_discard);
3490 BUFFER_TRACE(bh, "Buffer discarded");
3495 * If this block is not completely contained in the range
3496 * to be discarded, then it is not going to be released. Because
3497 * we need to keep this block, we need to make sure this part
3498 * of the page is uptodate before we modify it by writeing
3499 * partial zeros on it.
3501 if (!buffer_mapped(bh)) {
3503 * Buffer head must be mapped before we can read
3506 BUFFER_TRACE(bh, "unmapped");
3507 ext4_get_block(inode, iblock, bh, 0);
3508 /* unmapped? It's a hole - nothing to do */
3509 if (!buffer_mapped(bh)) {
3510 BUFFER_TRACE(bh, "still unmapped");
3515 /* Ok, it's mapped. Make sure it's up-to-date */
3516 if (PageUptodate(page))
3517 set_buffer_uptodate(bh);
3519 if (!buffer_uptodate(bh)) {
3521 ll_rw_block(READ, 1, &bh);
3523 /* Uhhuh. Read error. Complain and punt.*/
3524 if (!buffer_uptodate(bh))
3528 if (ext4_should_journal_data(inode)) {
3529 BUFFER_TRACE(bh, "get write access");
3530 err = ext4_journal_get_write_access(handle, bh);
3535 zero_user(page, pos, range_to_discard);
3538 if (ext4_should_journal_data(inode)) {
3539 err = ext4_handle_dirty_metadata(handle, inode, bh);
3541 mark_buffer_dirty(bh);
3543 BUFFER_TRACE(bh, "Partial buffer zeroed");
3545 bh = bh->b_this_page;
3547 pos += range_to_discard;
3553 int ext4_can_truncate(struct inode *inode)
3555 if (S_ISREG(inode->i_mode))
3557 if (S_ISDIR(inode->i_mode))
3559 if (S_ISLNK(inode->i_mode))
3560 return !ext4_inode_is_fast_symlink(inode);
3565 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3566 * associated with the given offset and length
3568 * @inode: File inode
3569 * @offset: The offset where the hole will begin
3570 * @len: The length of the hole
3572 * Returns: 0 on success or negative on failure
3575 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3577 struct inode *inode = file_inode(file);
3578 struct super_block *sb = inode->i_sb;
3579 ext4_lblk_t first_block, stop_block;
3580 struct address_space *mapping = inode->i_mapping;
3581 loff_t first_page, last_page, page_len;
3582 loff_t first_page_offset, last_page_offset;
3584 unsigned int credits;
3587 if (!S_ISREG(inode->i_mode))
3590 if (EXT4_SB(sb)->s_cluster_ratio > 1) {
3591 /* TODO: Add support for bigalloc file systems */
3595 trace_ext4_punch_hole(inode, offset, length);
3598 * Write out all dirty pages to avoid race conditions
3599 * Then release them.
3601 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3602 ret = filemap_write_and_wait_range(mapping, offset,
3603 offset + length - 1);
3608 mutex_lock(&inode->i_mutex);
3609 /* It's not possible punch hole on append only file */
3610 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
3614 if (IS_SWAPFILE(inode)) {
3619 /* No need to punch hole beyond i_size */
3620 if (offset >= inode->i_size)
3624 * If the hole extends beyond i_size, set the hole
3625 * to end after the page that contains i_size
3627 if (offset + length > inode->i_size) {
3628 length = inode->i_size +
3629 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3633 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
3634 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
3636 first_page_offset = first_page << PAGE_CACHE_SHIFT;
3637 last_page_offset = last_page << PAGE_CACHE_SHIFT;
3639 /* Now release the pages */
3640 if (last_page_offset > first_page_offset) {
3641 truncate_pagecache_range(inode, first_page_offset,
3642 last_page_offset - 1);
3645 /* Wait all existing dio workers, newcomers will block on i_mutex */
3646 ext4_inode_block_unlocked_dio(inode);
3647 ret = ext4_flush_unwritten_io(inode);
3650 inode_dio_wait(inode);
3652 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3653 credits = ext4_writepage_trans_blocks(inode);
3655 credits = ext4_blocks_for_truncate(inode);
3656 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3657 if (IS_ERR(handle)) {
3658 ret = PTR_ERR(handle);
3659 ext4_std_error(sb, ret);
3664 * Now we need to zero out the non-page-aligned data in the
3665 * pages at the start and tail of the hole, and unmap the
3666 * buffer heads for the block aligned regions of the page that
3667 * were completely zeroed.
3669 if (first_page > last_page) {
3671 * If the file space being truncated is contained
3672 * within a page just zero out and unmap the middle of
3675 ret = ext4_discard_partial_page_buffers(handle,
3676 mapping, offset, length, 0);
3682 * zero out and unmap the partial page that contains
3683 * the start of the hole
3685 page_len = first_page_offset - offset;
3687 ret = ext4_discard_partial_page_buffers(handle, mapping,
3688 offset, page_len, 0);
3694 * zero out and unmap the partial page that contains
3695 * the end of the hole
3697 page_len = offset + length - last_page_offset;
3699 ret = ext4_discard_partial_page_buffers(handle, mapping,
3700 last_page_offset, page_len, 0);
3707 * If i_size is contained in the last page, we need to
3708 * unmap and zero the partial page after i_size
3710 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
3711 inode->i_size % PAGE_CACHE_SIZE != 0) {
3712 page_len = PAGE_CACHE_SIZE -
3713 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3716 ret = ext4_discard_partial_page_buffers(handle,
3717 mapping, inode->i_size, page_len, 0);
3724 first_block = (offset + sb->s_blocksize - 1) >>
3725 EXT4_BLOCK_SIZE_BITS(sb);
3726 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3728 /* If there are no blocks to remove, return now */
3729 if (first_block >= stop_block)
3732 down_write(&EXT4_I(inode)->i_data_sem);
3733 ext4_discard_preallocations(inode);
3735 ret = ext4_es_remove_extent(inode, first_block,
3736 stop_block - first_block);
3738 up_write(&EXT4_I(inode)->i_data_sem);
3742 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3743 ret = ext4_ext_remove_space(inode, first_block,
3746 ret = ext4_free_hole_blocks(handle, inode, first_block,
3749 ext4_discard_preallocations(inode);
3750 up_write(&EXT4_I(inode)->i_data_sem);
3752 ext4_handle_sync(handle);
3753 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3754 ext4_mark_inode_dirty(handle, inode);
3756 ext4_journal_stop(handle);
3758 ext4_inode_resume_unlocked_dio(inode);
3760 mutex_unlock(&inode->i_mutex);
3767 * We block out ext4_get_block() block instantiations across the entire
3768 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3769 * simultaneously on behalf of the same inode.
3771 * As we work through the truncate and commit bits of it to the journal there
3772 * is one core, guiding principle: the file's tree must always be consistent on
3773 * disk. We must be able to restart the truncate after a crash.
3775 * The file's tree may be transiently inconsistent in memory (although it
3776 * probably isn't), but whenever we close off and commit a journal transaction,
3777 * the contents of (the filesystem + the journal) must be consistent and
3778 * restartable. It's pretty simple, really: bottom up, right to left (although
3779 * left-to-right works OK too).
3781 * Note that at recovery time, journal replay occurs *before* the restart of
3782 * truncate against the orphan inode list.
3784 * The committed inode has the new, desired i_size (which is the same as
3785 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3786 * that this inode's truncate did not complete and it will again call
3787 * ext4_truncate() to have another go. So there will be instantiated blocks
3788 * to the right of the truncation point in a crashed ext4 filesystem. But
3789 * that's fine - as long as they are linked from the inode, the post-crash
3790 * ext4_truncate() run will find them and release them.
3792 void ext4_truncate(struct inode *inode)
3794 struct ext4_inode_info *ei = EXT4_I(inode);
3795 unsigned int credits;
3797 struct address_space *mapping = inode->i_mapping;
3801 * There is a possibility that we're either freeing the inode
3802 * or it completely new indode. In those cases we might not
3803 * have i_mutex locked because it's not necessary.
3805 if (!(inode->i_state & (I_NEW|I_FREEING)))
3806 WARN_ON(!mutex_is_locked(&inode->i_mutex));
3807 trace_ext4_truncate_enter(inode);
3809 if (!ext4_can_truncate(inode))
3812 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3814 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3815 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3817 if (ext4_has_inline_data(inode)) {
3820 ext4_inline_data_truncate(inode, &has_inline);
3826 * finish any pending end_io work so we won't run the risk of
3827 * converting any truncated blocks to initialized later
3829 ext4_flush_unwritten_io(inode);
3831 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3832 credits = ext4_writepage_trans_blocks(inode);
3834 credits = ext4_blocks_for_truncate(inode);
3836 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3837 if (IS_ERR(handle)) {
3838 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3842 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3843 page_len = PAGE_CACHE_SIZE -
3844 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3846 if (ext4_discard_partial_page_buffers(handle,
3847 mapping, inode->i_size, page_len, 0))
3852 * We add the inode to the orphan list, so that if this
3853 * truncate spans multiple transactions, and we crash, we will
3854 * resume the truncate when the filesystem recovers. It also
3855 * marks the inode dirty, to catch the new size.
3857 * Implication: the file must always be in a sane, consistent
3858 * truncatable state while each transaction commits.
3860 if (ext4_orphan_add(handle, inode))
3863 down_write(&EXT4_I(inode)->i_data_sem);
3865 ext4_discard_preallocations(inode);
3867 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3868 ext4_ext_truncate(handle, inode);
3870 ext4_ind_truncate(handle, inode);
3872 up_write(&ei->i_data_sem);
3875 ext4_handle_sync(handle);
3879 * If this was a simple ftruncate() and the file will remain alive,
3880 * then we need to clear up the orphan record which we created above.
3881 * However, if this was a real unlink then we were called by
3882 * ext4_delete_inode(), and we allow that function to clean up the
3883 * orphan info for us.
3886 ext4_orphan_del(handle, inode);
3888 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3889 ext4_mark_inode_dirty(handle, inode);
3890 ext4_journal_stop(handle);
3892 trace_ext4_truncate_exit(inode);
3896 * ext4_get_inode_loc returns with an extra refcount against the inode's
3897 * underlying buffer_head on success. If 'in_mem' is true, we have all
3898 * data in memory that is needed to recreate the on-disk version of this
3901 static int __ext4_get_inode_loc(struct inode *inode,
3902 struct ext4_iloc *iloc, int in_mem)
3904 struct ext4_group_desc *gdp;
3905 struct buffer_head *bh;
3906 struct super_block *sb = inode->i_sb;
3908 int inodes_per_block, inode_offset;
3911 if (!ext4_valid_inum(sb, inode->i_ino))
3914 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3915 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3920 * Figure out the offset within the block group inode table
3922 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3923 inode_offset = ((inode->i_ino - 1) %
3924 EXT4_INODES_PER_GROUP(sb));
3925 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3926 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3928 bh = sb_getblk(sb, block);
3931 if (!buffer_uptodate(bh)) {
3935 * If the buffer has the write error flag, we have failed
3936 * to write out another inode in the same block. In this
3937 * case, we don't have to read the block because we may
3938 * read the old inode data successfully.
3940 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3941 set_buffer_uptodate(bh);
3943 if (buffer_uptodate(bh)) {
3944 /* someone brought it uptodate while we waited */
3950 * If we have all information of the inode in memory and this
3951 * is the only valid inode in the block, we need not read the
3955 struct buffer_head *bitmap_bh;
3958 start = inode_offset & ~(inodes_per_block - 1);
3960 /* Is the inode bitmap in cache? */
3961 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3962 if (unlikely(!bitmap_bh))
3966 * If the inode bitmap isn't in cache then the
3967 * optimisation may end up performing two reads instead
3968 * of one, so skip it.
3970 if (!buffer_uptodate(bitmap_bh)) {
3974 for (i = start; i < start + inodes_per_block; i++) {
3975 if (i == inode_offset)
3977 if (ext4_test_bit(i, bitmap_bh->b_data))
3981 if (i == start + inodes_per_block) {
3982 /* all other inodes are free, so skip I/O */
3983 memset(bh->b_data, 0, bh->b_size);
3984 set_buffer_uptodate(bh);
3992 * If we need to do any I/O, try to pre-readahead extra
3993 * blocks from the inode table.
3995 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3996 ext4_fsblk_t b, end, table;
3998 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4000 table = ext4_inode_table(sb, gdp);
4001 /* s_inode_readahead_blks is always a power of 2 */
4002 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4006 num = EXT4_INODES_PER_GROUP(sb);
4007 if (ext4_has_group_desc_csum(sb))
4008 num -= ext4_itable_unused_count(sb, gdp);
4009 table += num / inodes_per_block;
4013 sb_breadahead(sb, b++);
4017 * There are other valid inodes in the buffer, this inode
4018 * has in-inode xattrs, or we don't have this inode in memory.
4019 * Read the block from disk.
4021 trace_ext4_load_inode(inode);
4023 bh->b_end_io = end_buffer_read_sync;
4024 submit_bh(READ | REQ_META | REQ_PRIO, bh);
4026 if (!buffer_uptodate(bh)) {
4027 EXT4_ERROR_INODE_BLOCK(inode, block,
4028 "unable to read itable block");
4038 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4040 /* We have all inode data except xattrs in memory here. */
4041 return __ext4_get_inode_loc(inode, iloc,
4042 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4045 void ext4_set_inode_flags(struct inode *inode)
4047 unsigned int flags = EXT4_I(inode)->i_flags;
4048 unsigned int new_fl = 0;
4050 if (flags & EXT4_SYNC_FL)
4052 if (flags & EXT4_APPEND_FL)
4054 if (flags & EXT4_IMMUTABLE_FL)
4055 new_fl |= S_IMMUTABLE;
4056 if (flags & EXT4_NOATIME_FL)
4057 new_fl |= S_NOATIME;
4058 if (flags & EXT4_DIRSYNC_FL)
4059 new_fl |= S_DIRSYNC;
4060 set_mask_bits(&inode->i_flags,
4061 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC, new_fl);
4064 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4065 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4067 unsigned int vfs_fl;
4068 unsigned long old_fl, new_fl;
4071 vfs_fl = ei->vfs_inode.i_flags;
4072 old_fl = ei->i_flags;
4073 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4074 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4076 if (vfs_fl & S_SYNC)
4077 new_fl |= EXT4_SYNC_FL;
4078 if (vfs_fl & S_APPEND)
4079 new_fl |= EXT4_APPEND_FL;
4080 if (vfs_fl & S_IMMUTABLE)
4081 new_fl |= EXT4_IMMUTABLE_FL;
4082 if (vfs_fl & S_NOATIME)
4083 new_fl |= EXT4_NOATIME_FL;
4084 if (vfs_fl & S_DIRSYNC)
4085 new_fl |= EXT4_DIRSYNC_FL;
4086 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4089 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4090 struct ext4_inode_info *ei)
4093 struct inode *inode = &(ei->vfs_inode);
4094 struct super_block *sb = inode->i_sb;
4096 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4097 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4098 /* we are using combined 48 bit field */
4099 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4100 le32_to_cpu(raw_inode->i_blocks_lo);
4101 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4102 /* i_blocks represent file system block size */
4103 return i_blocks << (inode->i_blkbits - 9);
4108 return le32_to_cpu(raw_inode->i_blocks_lo);
4112 static inline void ext4_iget_extra_inode(struct inode *inode,
4113 struct ext4_inode *raw_inode,
4114 struct ext4_inode_info *ei)
4116 __le32 *magic = (void *)raw_inode +
4117 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4118 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4119 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4120 ext4_find_inline_data_nolock(inode);
4122 EXT4_I(inode)->i_inline_off = 0;
4125 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4127 struct ext4_iloc iloc;
4128 struct ext4_inode *raw_inode;
4129 struct ext4_inode_info *ei;
4130 struct inode *inode;
4131 journal_t *journal = EXT4_SB(sb)->s_journal;
4137 inode = iget_locked(sb, ino);
4139 return ERR_PTR(-ENOMEM);
4140 if (!(inode->i_state & I_NEW))
4146 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4149 raw_inode = ext4_raw_inode(&iloc);
4151 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4152 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4153 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4154 EXT4_INODE_SIZE(inode->i_sb)) {
4155 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4156 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4157 EXT4_INODE_SIZE(inode->i_sb));
4162 ei->i_extra_isize = 0;
4164 /* Precompute checksum seed for inode metadata */
4165 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4166 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4167 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4169 __le32 inum = cpu_to_le32(inode->i_ino);
4170 __le32 gen = raw_inode->i_generation;
4171 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4173 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4177 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4178 EXT4_ERROR_INODE(inode, "checksum invalid");
4183 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4184 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4185 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4186 if (!(test_opt(inode->i_sb, NO_UID32))) {
4187 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4188 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4190 i_uid_write(inode, i_uid);
4191 i_gid_write(inode, i_gid);
4192 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4194 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4195 ei->i_inline_off = 0;
4196 ei->i_dir_start_lookup = 0;
4197 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4198 /* We now have enough fields to check if the inode was active or not.
4199 * This is needed because nfsd might try to access dead inodes
4200 * the test is that same one that e2fsck uses
4201 * NeilBrown 1999oct15
4203 if (inode->i_nlink == 0) {
4204 if ((inode->i_mode == 0 ||
4205 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4206 ino != EXT4_BOOT_LOADER_INO) {
4207 /* this inode is deleted */
4211 /* The only unlinked inodes we let through here have
4212 * valid i_mode and are being read by the orphan
4213 * recovery code: that's fine, we're about to complete
4214 * the process of deleting those.
4215 * OR it is the EXT4_BOOT_LOADER_INO which is
4216 * not initialized on a new filesystem. */
4218 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4219 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4220 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4221 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4223 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4224 inode->i_size = ext4_isize(raw_inode);
4225 ei->i_disksize = inode->i_size;
4227 ei->i_reserved_quota = 0;
4229 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4230 ei->i_block_group = iloc.block_group;
4231 ei->i_last_alloc_group = ~0;
4233 * NOTE! The in-memory inode i_data array is in little-endian order
4234 * even on big-endian machines: we do NOT byteswap the block numbers!
4236 for (block = 0; block < EXT4_N_BLOCKS; block++)
4237 ei->i_data[block] = raw_inode->i_block[block];
4238 INIT_LIST_HEAD(&ei->i_orphan);
4241 * Set transaction id's of transactions that have to be committed
4242 * to finish f[data]sync. We set them to currently running transaction
4243 * as we cannot be sure that the inode or some of its metadata isn't
4244 * part of the transaction - the inode could have been reclaimed and
4245 * now it is reread from disk.
4248 transaction_t *transaction;
4251 read_lock(&journal->j_state_lock);
4252 if (journal->j_running_transaction)
4253 transaction = journal->j_running_transaction;
4255 transaction = journal->j_committing_transaction;
4257 tid = transaction->t_tid;
4259 tid = journal->j_commit_sequence;
4260 read_unlock(&journal->j_state_lock);
4261 ei->i_sync_tid = tid;
4262 ei->i_datasync_tid = tid;
4265 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4266 if (ei->i_extra_isize == 0) {
4267 /* The extra space is currently unused. Use it. */
4268 ei->i_extra_isize = sizeof(struct ext4_inode) -
4269 EXT4_GOOD_OLD_INODE_SIZE;
4271 ext4_iget_extra_inode(inode, raw_inode, ei);
4275 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4276 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4277 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4278 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4280 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4281 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4282 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4284 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4288 if (ei->i_file_acl &&
4289 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4290 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4294 } else if (!ext4_has_inline_data(inode)) {
4295 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4296 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4297 (S_ISLNK(inode->i_mode) &&
4298 !ext4_inode_is_fast_symlink(inode))))
4299 /* Validate extent which is part of inode */
4300 ret = ext4_ext_check_inode(inode);
4301 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4302 (S_ISLNK(inode->i_mode) &&
4303 !ext4_inode_is_fast_symlink(inode))) {
4304 /* Validate block references which are part of inode */
4305 ret = ext4_ind_check_inode(inode);
4311 if (S_ISREG(inode->i_mode)) {
4312 inode->i_op = &ext4_file_inode_operations;
4313 inode->i_fop = &ext4_file_operations;
4314 ext4_set_aops(inode);
4315 } else if (S_ISDIR(inode->i_mode)) {
4316 inode->i_op = &ext4_dir_inode_operations;
4317 inode->i_fop = &ext4_dir_operations;
4318 } else if (S_ISLNK(inode->i_mode)) {
4319 if (ext4_inode_is_fast_symlink(inode)) {
4320 inode->i_op = &ext4_fast_symlink_inode_operations;
4321 nd_terminate_link(ei->i_data, inode->i_size,
4322 sizeof(ei->i_data) - 1);
4324 inode->i_op = &ext4_symlink_inode_operations;
4325 ext4_set_aops(inode);
4327 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4328 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4329 inode->i_op = &ext4_special_inode_operations;
4330 if (raw_inode->i_block[0])
4331 init_special_inode(inode, inode->i_mode,
4332 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4334 init_special_inode(inode, inode->i_mode,
4335 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4336 } else if (ino == EXT4_BOOT_LOADER_INO) {
4337 make_bad_inode(inode);
4340 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4344 ext4_set_inode_flags(inode);
4345 unlock_new_inode(inode);
4351 return ERR_PTR(ret);
4354 static int ext4_inode_blocks_set(handle_t *handle,
4355 struct ext4_inode *raw_inode,
4356 struct ext4_inode_info *ei)
4358 struct inode *inode = &(ei->vfs_inode);
4359 u64 i_blocks = inode->i_blocks;
4360 struct super_block *sb = inode->i_sb;
4362 if (i_blocks <= ~0U) {
4364 * i_blocks can be represented in a 32 bit variable
4365 * as multiple of 512 bytes
4367 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4368 raw_inode->i_blocks_high = 0;
4369 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4372 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4375 if (i_blocks <= 0xffffffffffffULL) {
4377 * i_blocks can be represented in a 48 bit variable
4378 * as multiple of 512 bytes
4380 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4381 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4382 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4384 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4385 /* i_block is stored in file system block size */
4386 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4387 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4388 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4394 * Post the struct inode info into an on-disk inode location in the
4395 * buffer-cache. This gobbles the caller's reference to the
4396 * buffer_head in the inode location struct.
4398 * The caller must have write access to iloc->bh.
4400 static int ext4_do_update_inode(handle_t *handle,
4401 struct inode *inode,
4402 struct ext4_iloc *iloc)
4404 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4405 struct ext4_inode_info *ei = EXT4_I(inode);
4406 struct buffer_head *bh = iloc->bh;
4407 int err = 0, rc, block;
4408 int need_datasync = 0;
4412 /* For fields not not tracking in the in-memory inode,
4413 * initialise them to zero for new inodes. */
4414 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4415 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4417 ext4_get_inode_flags(ei);
4418 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4419 i_uid = i_uid_read(inode);
4420 i_gid = i_gid_read(inode);
4421 if (!(test_opt(inode->i_sb, NO_UID32))) {
4422 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4423 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4425 * Fix up interoperability with old kernels. Otherwise, old inodes get
4426 * re-used with the upper 16 bits of the uid/gid intact
4429 raw_inode->i_uid_high =
4430 cpu_to_le16(high_16_bits(i_uid));
4431 raw_inode->i_gid_high =
4432 cpu_to_le16(high_16_bits(i_gid));
4434 raw_inode->i_uid_high = 0;
4435 raw_inode->i_gid_high = 0;
4438 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4439 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4440 raw_inode->i_uid_high = 0;
4441 raw_inode->i_gid_high = 0;
4443 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4445 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4446 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4447 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4448 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4450 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4452 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4453 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4454 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4455 cpu_to_le32(EXT4_OS_HURD))
4456 raw_inode->i_file_acl_high =
4457 cpu_to_le16(ei->i_file_acl >> 32);
4458 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4459 if (ei->i_disksize != ext4_isize(raw_inode)) {
4460 ext4_isize_set(raw_inode, ei->i_disksize);
4463 if (ei->i_disksize > 0x7fffffffULL) {
4464 struct super_block *sb = inode->i_sb;
4465 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4466 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4467 EXT4_SB(sb)->s_es->s_rev_level ==
4468 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4469 /* If this is the first large file
4470 * created, add a flag to the superblock.
4472 err = ext4_journal_get_write_access(handle,
4473 EXT4_SB(sb)->s_sbh);
4476 ext4_update_dynamic_rev(sb);
4477 EXT4_SET_RO_COMPAT_FEATURE(sb,
4478 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4479 ext4_handle_sync(handle);
4480 err = ext4_handle_dirty_super(handle, sb);
4483 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4484 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4485 if (old_valid_dev(inode->i_rdev)) {
4486 raw_inode->i_block[0] =
4487 cpu_to_le32(old_encode_dev(inode->i_rdev));
4488 raw_inode->i_block[1] = 0;
4490 raw_inode->i_block[0] = 0;
4491 raw_inode->i_block[1] =
4492 cpu_to_le32(new_encode_dev(inode->i_rdev));
4493 raw_inode->i_block[2] = 0;
4495 } else if (!ext4_has_inline_data(inode)) {
4496 for (block = 0; block < EXT4_N_BLOCKS; block++)
4497 raw_inode->i_block[block] = ei->i_data[block];
4500 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4501 if (ei->i_extra_isize) {
4502 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4503 raw_inode->i_version_hi =
4504 cpu_to_le32(inode->i_version >> 32);
4505 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4508 ext4_inode_csum_set(inode, raw_inode, ei);
4510 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4511 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4514 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4516 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4519 ext4_std_error(inode->i_sb, err);
4524 * ext4_write_inode()
4526 * We are called from a few places:
4528 * - Within generic_file_write() for O_SYNC files.
4529 * Here, there will be no transaction running. We wait for any running
4530 * transaction to commit.
4532 * - Within sys_sync(), kupdate and such.
4533 * We wait on commit, if tol to.
4535 * - Within prune_icache() (PF_MEMALLOC == true)
4536 * Here we simply return. We can't afford to block kswapd on the
4539 * In all cases it is actually safe for us to return without doing anything,
4540 * because the inode has been copied into a raw inode buffer in
4541 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4544 * Note that we are absolutely dependent upon all inode dirtiers doing the
4545 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4546 * which we are interested.
4548 * It would be a bug for them to not do this. The code:
4550 * mark_inode_dirty(inode)
4552 * inode->i_size = expr;
4554 * is in error because a kswapd-driven write_inode() could occur while
4555 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4556 * will no longer be on the superblock's dirty inode list.
4558 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4562 if (current->flags & PF_MEMALLOC)
4565 if (EXT4_SB(inode->i_sb)->s_journal) {
4566 if (ext4_journal_current_handle()) {
4567 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4572 if (wbc->sync_mode != WB_SYNC_ALL)
4575 err = ext4_force_commit(inode->i_sb);
4577 struct ext4_iloc iloc;
4579 err = __ext4_get_inode_loc(inode, &iloc, 0);
4582 if (wbc->sync_mode == WB_SYNC_ALL)
4583 sync_dirty_buffer(iloc.bh);
4584 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4585 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4586 "IO error syncing inode");
4595 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4596 * buffers that are attached to a page stradding i_size and are undergoing
4597 * commit. In that case we have to wait for commit to finish and try again.
4599 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4603 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4604 tid_t commit_tid = 0;
4607 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4609 * All buffers in the last page remain valid? Then there's nothing to
4610 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4613 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4616 page = find_lock_page(inode->i_mapping,
4617 inode->i_size >> PAGE_CACHE_SHIFT);
4620 ret = __ext4_journalled_invalidatepage(page, offset);
4622 page_cache_release(page);
4626 read_lock(&journal->j_state_lock);
4627 if (journal->j_committing_transaction)
4628 commit_tid = journal->j_committing_transaction->t_tid;
4629 read_unlock(&journal->j_state_lock);
4631 jbd2_log_wait_commit(journal, commit_tid);
4638 * Called from notify_change.
4640 * We want to trap VFS attempts to truncate the file as soon as
4641 * possible. In particular, we want to make sure that when the VFS
4642 * shrinks i_size, we put the inode on the orphan list and modify
4643 * i_disksize immediately, so that during the subsequent flushing of
4644 * dirty pages and freeing of disk blocks, we can guarantee that any
4645 * commit will leave the blocks being flushed in an unused state on
4646 * disk. (On recovery, the inode will get truncated and the blocks will
4647 * be freed, so we have a strong guarantee that no future commit will
4648 * leave these blocks visible to the user.)
4650 * Another thing we have to assure is that if we are in ordered mode
4651 * and inode is still attached to the committing transaction, we must
4652 * we start writeout of all the dirty pages which are being truncated.
4653 * This way we are sure that all the data written in the previous
4654 * transaction are already on disk (truncate waits for pages under
4657 * Called with inode->i_mutex down.
4659 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4661 struct inode *inode = dentry->d_inode;
4664 const unsigned int ia_valid = attr->ia_valid;
4666 error = inode_change_ok(inode, attr);
4670 if (is_quota_modification(inode, attr))
4671 dquot_initialize(inode);
4672 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4673 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4676 /* (user+group)*(old+new) structure, inode write (sb,
4677 * inode block, ? - but truncate inode update has it) */
4678 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4679 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4680 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4681 if (IS_ERR(handle)) {
4682 error = PTR_ERR(handle);
4685 error = dquot_transfer(inode, attr);
4687 ext4_journal_stop(handle);
4690 /* Update corresponding info in inode so that everything is in
4691 * one transaction */
4692 if (attr->ia_valid & ATTR_UID)
4693 inode->i_uid = attr->ia_uid;
4694 if (attr->ia_valid & ATTR_GID)
4695 inode->i_gid = attr->ia_gid;
4696 error = ext4_mark_inode_dirty(handle, inode);
4697 ext4_journal_stop(handle);
4700 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4702 loff_t oldsize = inode->i_size;
4704 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4705 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4707 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4711 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4712 inode_inc_iversion(inode);
4714 if (S_ISREG(inode->i_mode) &&
4715 (attr->ia_size < inode->i_size)) {
4716 if (ext4_should_order_data(inode)) {
4717 error = ext4_begin_ordered_truncate(inode,
4722 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4723 if (IS_ERR(handle)) {
4724 error = PTR_ERR(handle);
4727 if (ext4_handle_valid(handle)) {
4728 error = ext4_orphan_add(handle, inode);
4731 EXT4_I(inode)->i_disksize = attr->ia_size;
4732 rc = ext4_mark_inode_dirty(handle, inode);
4735 ext4_journal_stop(handle);
4737 ext4_orphan_del(NULL, inode);
4742 i_size_write(inode, attr->ia_size);
4744 * Blocks are going to be removed from the inode. Wait
4745 * for dio in flight. Temporarily disable
4746 * dioread_nolock to prevent livelock.
4749 if (!ext4_should_journal_data(inode)) {
4750 ext4_inode_block_unlocked_dio(inode);
4751 inode_dio_wait(inode);
4752 ext4_inode_resume_unlocked_dio(inode);
4754 ext4_wait_for_tail_page_commit(inode);
4757 * Truncate pagecache after we've waited for commit
4758 * in data=journal mode to make pages freeable.
4760 truncate_pagecache(inode, oldsize, inode->i_size);
4763 * We want to call ext4_truncate() even if attr->ia_size ==
4764 * inode->i_size for cases like truncation of fallocated space
4766 if (attr->ia_valid & ATTR_SIZE)
4767 ext4_truncate(inode);
4770 setattr_copy(inode, attr);
4771 mark_inode_dirty(inode);
4775 * If the call to ext4_truncate failed to get a transaction handle at
4776 * all, we need to clean up the in-core orphan list manually.
4778 if (orphan && inode->i_nlink)
4779 ext4_orphan_del(NULL, inode);
4781 if (!rc && (ia_valid & ATTR_MODE))
4782 rc = ext4_acl_chmod(inode);
4785 ext4_std_error(inode->i_sb, error);
4791 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4794 struct inode *inode;
4795 unsigned long long delalloc_blocks;
4797 inode = dentry->d_inode;
4798 generic_fillattr(inode, stat);
4801 * We can't update i_blocks if the block allocation is delayed
4802 * otherwise in the case of system crash before the real block
4803 * allocation is done, we will have i_blocks inconsistent with
4804 * on-disk file blocks.
4805 * We always keep i_blocks updated together with real
4806 * allocation. But to not confuse with user, stat
4807 * will return the blocks that include the delayed allocation
4808 * blocks for this file.
4810 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4811 EXT4_I(inode)->i_reserved_data_blocks);
4813 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9);
4817 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4819 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4820 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4821 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4825 * Account for index blocks, block groups bitmaps and block group
4826 * descriptor blocks if modify datablocks and index blocks
4827 * worse case, the indexs blocks spread over different block groups
4829 * If datablocks are discontiguous, they are possible to spread over
4830 * different block groups too. If they are contiguous, with flexbg,
4831 * they could still across block group boundary.
4833 * Also account for superblock, inode, quota and xattr blocks
4835 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4837 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4843 * How many index blocks need to touch to modify nrblocks?
4844 * The "Chunk" flag indicating whether the nrblocks is
4845 * physically contiguous on disk
4847 * For Direct IO and fallocate, they calls get_block to allocate
4848 * one single extent at a time, so they could set the "Chunk" flag
4850 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4855 * Now let's see how many group bitmaps and group descriptors need
4865 if (groups > ngroups)
4867 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4868 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4870 /* bitmaps and block group descriptor blocks */
4871 ret += groups + gdpblocks;
4873 /* Blocks for super block, inode, quota and xattr blocks */
4874 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4880 * Calculate the total number of credits to reserve to fit
4881 * the modification of a single pages into a single transaction,
4882 * which may include multiple chunks of block allocations.
4884 * This could be called via ext4_write_begin()
4886 * We need to consider the worse case, when
4887 * one new block per extent.
4889 int ext4_writepage_trans_blocks(struct inode *inode)
4891 int bpp = ext4_journal_blocks_per_page(inode);
4894 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4896 /* Account for data blocks for journalled mode */
4897 if (ext4_should_journal_data(inode))
4903 * Calculate the journal credits for a chunk of data modification.
4905 * This is called from DIO, fallocate or whoever calling
4906 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4908 * journal buffers for data blocks are not included here, as DIO
4909 * and fallocate do no need to journal data buffers.
4911 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4913 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4917 * The caller must have previously called ext4_reserve_inode_write().
4918 * Give this, we know that the caller already has write access to iloc->bh.
4920 int ext4_mark_iloc_dirty(handle_t *handle,
4921 struct inode *inode, struct ext4_iloc *iloc)
4925 if (IS_I_VERSION(inode))
4926 inode_inc_iversion(inode);
4928 /* the do_update_inode consumes one bh->b_count */
4931 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4932 err = ext4_do_update_inode(handle, inode, iloc);
4938 * On success, We end up with an outstanding reference count against
4939 * iloc->bh. This _must_ be cleaned up later.
4943 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4944 struct ext4_iloc *iloc)
4948 err = ext4_get_inode_loc(inode, iloc);
4950 BUFFER_TRACE(iloc->bh, "get_write_access");
4951 err = ext4_journal_get_write_access(handle, iloc->bh);
4957 ext4_std_error(inode->i_sb, err);
4962 * Expand an inode by new_extra_isize bytes.
4963 * Returns 0 on success or negative error number on failure.
4965 static int ext4_expand_extra_isize(struct inode *inode,
4966 unsigned int new_extra_isize,
4967 struct ext4_iloc iloc,
4970 struct ext4_inode *raw_inode;
4971 struct ext4_xattr_ibody_header *header;
4973 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4976 raw_inode = ext4_raw_inode(&iloc);
4978 header = IHDR(inode, raw_inode);
4980 /* No extended attributes present */
4981 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4982 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4983 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4985 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4989 /* try to expand with EAs present */
4990 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4995 * What we do here is to mark the in-core inode as clean with respect to inode
4996 * dirtiness (it may still be data-dirty).
4997 * This means that the in-core inode may be reaped by prune_icache
4998 * without having to perform any I/O. This is a very good thing,
4999 * because *any* task may call prune_icache - even ones which
5000 * have a transaction open against a different journal.
5002 * Is this cheating? Not really. Sure, we haven't written the
5003 * inode out, but prune_icache isn't a user-visible syncing function.
5004 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5005 * we start and wait on commits.
5007 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5009 struct ext4_iloc iloc;
5010 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5011 static unsigned int mnt_count;
5015 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5016 err = ext4_reserve_inode_write(handle, inode, &iloc);
5017 if (ext4_handle_valid(handle) &&
5018 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5019 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5021 * We need extra buffer credits since we may write into EA block
5022 * with this same handle. If journal_extend fails, then it will
5023 * only result in a minor loss of functionality for that inode.
5024 * If this is felt to be critical, then e2fsck should be run to
5025 * force a large enough s_min_extra_isize.
5027 if ((jbd2_journal_extend(handle,
5028 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5029 ret = ext4_expand_extra_isize(inode,
5030 sbi->s_want_extra_isize,
5033 ext4_set_inode_state(inode,
5034 EXT4_STATE_NO_EXPAND);
5036 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5037 ext4_warning(inode->i_sb,
5038 "Unable to expand inode %lu. Delete"
5039 " some EAs or run e2fsck.",
5042 le16_to_cpu(sbi->s_es->s_mnt_count);
5048 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5053 * ext4_dirty_inode() is called from __mark_inode_dirty()
5055 * We're really interested in the case where a file is being extended.
5056 * i_size has been changed by generic_commit_write() and we thus need
5057 * to include the updated inode in the current transaction.
5059 * Also, dquot_alloc_block() will always dirty the inode when blocks
5060 * are allocated to the file.
5062 * If the inode is marked synchronous, we don't honour that here - doing
5063 * so would cause a commit on atime updates, which we don't bother doing.
5064 * We handle synchronous inodes at the highest possible level.
5066 void ext4_dirty_inode(struct inode *inode, int flags)
5070 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5074 ext4_mark_inode_dirty(handle, inode);
5076 ext4_journal_stop(handle);
5083 * Bind an inode's backing buffer_head into this transaction, to prevent
5084 * it from being flushed to disk early. Unlike
5085 * ext4_reserve_inode_write, this leaves behind no bh reference and
5086 * returns no iloc structure, so the caller needs to repeat the iloc
5087 * lookup to mark the inode dirty later.
5089 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5091 struct ext4_iloc iloc;
5095 err = ext4_get_inode_loc(inode, &iloc);
5097 BUFFER_TRACE(iloc.bh, "get_write_access");
5098 err = jbd2_journal_get_write_access(handle, iloc.bh);
5100 err = ext4_handle_dirty_metadata(handle,
5106 ext4_std_error(inode->i_sb, err);
5111 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5118 * We have to be very careful here: changing a data block's
5119 * journaling status dynamically is dangerous. If we write a
5120 * data block to the journal, change the status and then delete
5121 * that block, we risk forgetting to revoke the old log record
5122 * from the journal and so a subsequent replay can corrupt data.
5123 * So, first we make sure that the journal is empty and that
5124 * nobody is changing anything.
5127 journal = EXT4_JOURNAL(inode);
5130 if (is_journal_aborted(journal))
5132 /* We have to allocate physical blocks for delalloc blocks
5133 * before flushing journal. otherwise delalloc blocks can not
5134 * be allocated any more. even more truncate on delalloc blocks
5135 * could trigger BUG by flushing delalloc blocks in journal.
5136 * There is no delalloc block in non-journal data mode.
5138 if (val && test_opt(inode->i_sb, DELALLOC)) {
5139 err = ext4_alloc_da_blocks(inode);
5144 /* Wait for all existing dio workers */
5145 ext4_inode_block_unlocked_dio(inode);
5146 inode_dio_wait(inode);
5148 jbd2_journal_lock_updates(journal);
5151 * OK, there are no updates running now, and all cached data is
5152 * synced to disk. We are now in a completely consistent state
5153 * which doesn't have anything in the journal, and we know that
5154 * no filesystem updates are running, so it is safe to modify
5155 * the inode's in-core data-journaling state flag now.
5159 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5161 jbd2_journal_flush(journal);
5162 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5164 ext4_set_aops(inode);
5166 jbd2_journal_unlock_updates(journal);
5167 ext4_inode_resume_unlocked_dio(inode);
5169 /* Finally we can mark the inode as dirty. */
5171 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5173 return PTR_ERR(handle);
5175 err = ext4_mark_inode_dirty(handle, inode);
5176 ext4_handle_sync(handle);
5177 ext4_journal_stop(handle);
5178 ext4_std_error(inode->i_sb, err);
5183 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5185 return !buffer_mapped(bh);
5188 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5190 struct page *page = vmf->page;
5194 struct file *file = vma->vm_file;
5195 struct inode *inode = file_inode(file);
5196 struct address_space *mapping = inode->i_mapping;
5198 get_block_t *get_block;
5201 sb_start_pagefault(inode->i_sb);
5202 file_update_time(vma->vm_file);
5203 /* Delalloc case is easy... */
5204 if (test_opt(inode->i_sb, DELALLOC) &&
5205 !ext4_should_journal_data(inode) &&
5206 !ext4_nonda_switch(inode->i_sb)) {
5208 ret = __block_page_mkwrite(vma, vmf,
5209 ext4_da_get_block_prep);
5210 } while (ret == -ENOSPC &&
5211 ext4_should_retry_alloc(inode->i_sb, &retries));
5216 size = i_size_read(inode);
5217 /* Page got truncated from under us? */
5218 if (page->mapping != mapping || page_offset(page) > size) {
5220 ret = VM_FAULT_NOPAGE;
5224 if (page->index == size >> PAGE_CACHE_SHIFT)
5225 len = size & ~PAGE_CACHE_MASK;
5227 len = PAGE_CACHE_SIZE;
5229 * Return if we have all the buffers mapped. This avoids the need to do
5230 * journal_start/journal_stop which can block and take a long time
5232 if (page_has_buffers(page)) {
5233 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5235 ext4_bh_unmapped)) {
5236 /* Wait so that we don't change page under IO */
5237 wait_for_stable_page(page);
5238 ret = VM_FAULT_LOCKED;
5243 /* OK, we need to fill the hole... */
5244 if (ext4_should_dioread_nolock(inode))
5245 get_block = ext4_get_block_write;
5247 get_block = ext4_get_block;
5249 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5250 ext4_writepage_trans_blocks(inode));
5251 if (IS_ERR(handle)) {
5252 ret = VM_FAULT_SIGBUS;
5255 ret = __block_page_mkwrite(vma, vmf, get_block);
5256 if (!ret && ext4_should_journal_data(inode)) {
5257 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5258 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5260 ret = VM_FAULT_SIGBUS;
5261 ext4_journal_stop(handle);
5264 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5266 ext4_journal_stop(handle);
5267 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5270 ret = block_page_mkwrite_return(ret);
5272 sb_end_pagefault(inode->i_sb);