2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
80 __u32 provided, calculated;
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
96 return provided == calculated;
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 trace_ext4_begin_ordered_truncate(inode, new_size);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode)->jinode)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
134 static void ext4_invalidatepage(struct page *page, unsigned long offset);
135 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
136 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
137 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
138 struct inode *inode, struct page *page, loff_t from,
139 loff_t length, int flags);
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode *inode)
146 int ea_blocks = EXT4_I(inode)->i_file_acl ?
147 (inode->i_sb->s_blocksize >> 9) : 0;
149 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
157 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
168 BUG_ON(EXT4_JOURNAL(inode) == NULL);
169 jbd_debug(2, "restarting handle %p\n", handle);
170 up_write(&EXT4_I(inode)->i_data_sem);
171 ret = ext4_journal_restart(handle, nblocks);
172 down_write(&EXT4_I(inode)->i_data_sem);
173 ext4_discard_preallocations(inode);
179 * Called at the last iput() if i_nlink is zero.
181 void ext4_evict_inode(struct inode *inode)
186 trace_ext4_evict_inode(inode);
188 if (inode->i_nlink) {
190 * When journalling data dirty buffers are tracked only in the
191 * journal. So although mm thinks everything is clean and
192 * ready for reaping the inode might still have some pages to
193 * write in the running transaction or waiting to be
194 * checkpointed. Thus calling jbd2_journal_invalidatepage()
195 * (via truncate_inode_pages()) to discard these buffers can
196 * cause data loss. Also even if we did not discard these
197 * buffers, we would have no way to find them after the inode
198 * is reaped and thus user could see stale data if he tries to
199 * read them before the transaction is checkpointed. So be
200 * careful and force everything to disk here... We use
201 * ei->i_datasync_tid to store the newest transaction
202 * containing inode's data.
204 * Note that directories do not have this problem because they
205 * don't use page cache.
207 if (ext4_should_journal_data(inode) &&
208 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
209 inode->i_ino != EXT4_JOURNAL_INO) {
210 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
211 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
213 jbd2_log_start_commit(journal, commit_tid);
214 jbd2_log_wait_commit(journal, commit_tid);
215 filemap_write_and_wait(&inode->i_data);
217 truncate_inode_pages(&inode->i_data, 0);
218 ext4_ioend_shutdown(inode);
222 if (!is_bad_inode(inode))
223 dquot_initialize(inode);
225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
227 truncate_inode_pages(&inode->i_data, 0);
228 ext4_ioend_shutdown(inode);
230 if (is_bad_inode(inode))
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode->i_sb);
238 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
239 ext4_blocks_for_truncate(inode)+3);
240 if (IS_ERR(handle)) {
241 ext4_std_error(inode->i_sb, PTR_ERR(handle));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL, inode);
248 sb_end_intwrite(inode->i_sb);
253 ext4_handle_sync(handle);
255 err = ext4_mark_inode_dirty(handle, inode);
257 ext4_warning(inode->i_sb,
258 "couldn't mark inode dirty (err %d)", err);
262 ext4_truncate(inode);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle, 3)) {
271 err = ext4_journal_extend(handle, 3);
273 err = ext4_journal_restart(handle, 3);
275 ext4_warning(inode->i_sb,
276 "couldn't extend journal (err %d)", err);
278 ext4_journal_stop(handle);
279 ext4_orphan_del(NULL, inode);
280 sb_end_intwrite(inode->i_sb);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle, inode);
294 EXT4_I(inode)->i_dtime = get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle, inode))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode);
307 ext4_free_inode(handle, inode);
308 ext4_journal_stop(handle);
309 sb_end_intwrite(inode->i_sb);
312 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
316 qsize_t *ext4_get_reserved_space(struct inode *inode)
318 return &EXT4_I(inode)->i_reserved_quota;
323 * Calculate the number of metadata blocks need to reserve
324 * to allocate a block located at @lblock
326 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
328 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
329 return ext4_ext_calc_metadata_amount(inode, lblock);
331 return ext4_ind_calc_metadata_amount(inode, lblock);
335 * Called with i_data_sem down, which is important since we can call
336 * ext4_discard_preallocations() from here.
338 void ext4_da_update_reserve_space(struct inode *inode,
339 int used, int quota_claim)
341 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
342 struct ext4_inode_info *ei = EXT4_I(inode);
344 spin_lock(&ei->i_block_reservation_lock);
345 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
346 if (unlikely(used > ei->i_reserved_data_blocks)) {
347 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
348 "with only %d reserved data blocks",
349 __func__, inode->i_ino, used,
350 ei->i_reserved_data_blocks);
352 used = ei->i_reserved_data_blocks;
355 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
356 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
357 "with only %d reserved metadata blocks "
358 "(releasing %d blocks with reserved %d data blocks)",
359 inode->i_ino, ei->i_allocated_meta_blocks,
360 ei->i_reserved_meta_blocks, used,
361 ei->i_reserved_data_blocks);
363 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
366 /* Update per-inode reservations */
367 ei->i_reserved_data_blocks -= used;
368 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
369 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
370 used + ei->i_allocated_meta_blocks);
371 ei->i_allocated_meta_blocks = 0;
373 if (ei->i_reserved_data_blocks == 0) {
375 * We can release all of the reserved metadata blocks
376 * only when we have written all of the delayed
379 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
380 ei->i_reserved_meta_blocks);
381 ei->i_reserved_meta_blocks = 0;
382 ei->i_da_metadata_calc_len = 0;
384 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
386 /* Update quota subsystem for data blocks */
388 dquot_claim_block(inode, EXT4_C2B(sbi, used));
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
393 * not re-claim the quota for fallocated blocks.
395 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
403 if ((ei->i_reserved_data_blocks == 0) &&
404 (atomic_read(&inode->i_writecount) == 0))
405 ext4_discard_preallocations(inode);
408 static int __check_block_validity(struct inode *inode, const char *func,
410 struct ext4_map_blocks *map)
412 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
414 ext4_error_inode(inode, func, line, map->m_pblk,
415 "lblock %lu mapped to illegal pblock "
416 "(length %d)", (unsigned long) map->m_lblk,
423 #define check_block_validity(inode, map) \
424 __check_block_validity((inode), __func__, __LINE__, (map))
427 * Return the number of contiguous dirty pages in a given inode
428 * starting at page frame idx.
430 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
431 unsigned int max_pages)
433 struct address_space *mapping = inode->i_mapping;
437 int i, nr_pages, done = 0;
441 pagevec_init(&pvec, 0);
444 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
446 (pgoff_t)PAGEVEC_SIZE);
449 for (i = 0; i < nr_pages; i++) {
450 struct page *page = pvec.pages[i];
451 struct buffer_head *bh, *head;
454 if (unlikely(page->mapping != mapping) ||
456 PageWriteback(page) ||
457 page->index != idx) {
462 if (page_has_buffers(page)) {
463 bh = head = page_buffers(page);
465 if (!buffer_delay(bh) &&
466 !buffer_unwritten(bh))
468 bh = bh->b_this_page;
469 } while (!done && (bh != head));
476 if (num >= max_pages) {
481 pagevec_release(&pvec);
486 #ifdef ES_AGGRESSIVE_TEST
487 static void ext4_map_blocks_es_recheck(handle_t *handle,
489 struct ext4_map_blocks *es_map,
490 struct ext4_map_blocks *map,
497 * There is a race window that the result is not the same.
498 * e.g. xfstests #223 when dioread_nolock enables. The reason
499 * is that we lookup a block mapping in extent status tree with
500 * out taking i_data_sem. So at the time the unwritten extent
501 * could be converted.
503 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
504 down_read((&EXT4_I(inode)->i_data_sem));
505 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
506 retval = ext4_ext_map_blocks(handle, inode, map, flags &
507 EXT4_GET_BLOCKS_KEEP_SIZE);
509 retval = ext4_ind_map_blocks(handle, inode, map, flags &
510 EXT4_GET_BLOCKS_KEEP_SIZE);
512 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
513 up_read((&EXT4_I(inode)->i_data_sem));
515 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
516 * because it shouldn't be marked in es_map->m_flags.
518 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
521 * We don't check m_len because extent will be collpased in status
522 * tree. So the m_len might not equal.
524 if (es_map->m_lblk != map->m_lblk ||
525 es_map->m_flags != map->m_flags ||
526 es_map->m_pblk != map->m_pblk) {
527 printk("ES cache assertation failed for inode: %lu "
528 "es_cached ex [%d/%d/%llu/%x] != "
529 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
530 inode->i_ino, es_map->m_lblk, es_map->m_len,
531 es_map->m_pblk, es_map->m_flags, map->m_lblk,
532 map->m_len, map->m_pblk, map->m_flags,
536 #endif /* ES_AGGRESSIVE_TEST */
539 * The ext4_map_blocks() function tries to look up the requested blocks,
540 * and returns if the blocks are already mapped.
542 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
543 * and store the allocated blocks in the result buffer head and mark it
546 * If file type is extents based, it will call ext4_ext_map_blocks(),
547 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
550 * On success, it returns the number of blocks being mapped or allocate.
551 * if create==0 and the blocks are pre-allocated and uninitialized block,
552 * the result buffer head is unmapped. If the create ==1, it will make sure
553 * the buffer head is mapped.
555 * It returns 0 if plain look up failed (blocks have not been allocated), in
556 * that case, buffer head is unmapped
558 * It returns the error in case of allocation failure.
560 int ext4_map_blocks(handle_t *handle, struct inode *inode,
561 struct ext4_map_blocks *map, int flags)
563 struct extent_status es;
565 #ifdef ES_AGGRESSIVE_TEST
566 struct ext4_map_blocks orig_map;
568 memcpy(&orig_map, map, sizeof(*map));
572 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
573 "logical block %lu\n", inode->i_ino, flags, map->m_len,
574 (unsigned long) map->m_lblk);
576 /* Lookup extent status tree firstly */
577 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
578 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
579 map->m_pblk = ext4_es_pblock(&es) +
580 map->m_lblk - es.es_lblk;
581 map->m_flags |= ext4_es_is_written(&es) ?
582 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
583 retval = es.es_len - (map->m_lblk - es.es_lblk);
584 if (retval > map->m_len)
587 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
592 #ifdef ES_AGGRESSIVE_TEST
593 ext4_map_blocks_es_recheck(handle, inode, map,
600 * Try to see if we can get the block without requesting a new
603 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
604 down_read((&EXT4_I(inode)->i_data_sem));
605 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
606 retval = ext4_ext_map_blocks(handle, inode, map, flags &
607 EXT4_GET_BLOCKS_KEEP_SIZE);
609 retval = ext4_ind_map_blocks(handle, inode, map, flags &
610 EXT4_GET_BLOCKS_KEEP_SIZE);
614 unsigned long long status;
616 #ifdef ES_AGGRESSIVE_TEST
617 if (retval != map->m_len) {
618 printk("ES len assertation failed for inode: %lu "
619 "retval %d != map->m_len %d "
620 "in %s (lookup)\n", inode->i_ino, retval,
621 map->m_len, __func__);
625 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
626 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
627 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
628 ext4_find_delalloc_range(inode, map->m_lblk,
629 map->m_lblk + map->m_len - 1))
630 status |= EXTENT_STATUS_DELAYED;
631 ret = ext4_es_insert_extent(inode, map->m_lblk,
632 map->m_len, map->m_pblk, status);
636 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
637 up_read((&EXT4_I(inode)->i_data_sem));
640 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
641 int ret = check_block_validity(inode, map);
646 /* If it is only a block(s) look up */
647 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
651 * Returns if the blocks have already allocated
653 * Note that if blocks have been preallocated
654 * ext4_ext_get_block() returns the create = 0
655 * with buffer head unmapped.
657 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
661 * Here we clear m_flags because after allocating an new extent,
662 * it will be set again.
664 map->m_flags &= ~EXT4_MAP_FLAGS;
667 * New blocks allocate and/or writing to uninitialized extent
668 * will possibly result in updating i_data, so we take
669 * the write lock of i_data_sem, and call get_blocks()
670 * with create == 1 flag.
672 down_write((&EXT4_I(inode)->i_data_sem));
675 * if the caller is from delayed allocation writeout path
676 * we have already reserved fs blocks for allocation
677 * let the underlying get_block() function know to
678 * avoid double accounting
680 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
681 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
683 * We need to check for EXT4 here because migrate
684 * could have changed the inode type in between
686 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
687 retval = ext4_ext_map_blocks(handle, inode, map, flags);
689 retval = ext4_ind_map_blocks(handle, inode, map, flags);
691 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
693 * We allocated new blocks which will result in
694 * i_data's format changing. Force the migrate
695 * to fail by clearing migrate flags
697 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
701 * Update reserved blocks/metadata blocks after successful
702 * block allocation which had been deferred till now. We don't
703 * support fallocate for non extent files. So we can update
704 * reserve space here.
707 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
708 ext4_da_update_reserve_space(inode, retval, 1);
710 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
711 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
715 unsigned long long status;
717 #ifdef ES_AGGRESSIVE_TEST
718 if (retval != map->m_len) {
719 printk("ES len assertation failed for inode: %lu "
720 "retval %d != map->m_len %d "
721 "in %s (allocation)\n", inode->i_ino, retval,
722 map->m_len, __func__);
727 * If the extent has been zeroed out, we don't need to update
728 * extent status tree.
730 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
731 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
732 if (ext4_es_is_written(&es))
735 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
736 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
737 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
738 ext4_find_delalloc_range(inode, map->m_lblk,
739 map->m_lblk + map->m_len - 1))
740 status |= EXTENT_STATUS_DELAYED;
741 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
742 map->m_pblk, status);
748 up_write((&EXT4_I(inode)->i_data_sem));
749 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
750 int ret = check_block_validity(inode, map);
757 /* Maximum number of blocks we map for direct IO at once. */
758 #define DIO_MAX_BLOCKS 4096
760 static int _ext4_get_block(struct inode *inode, sector_t iblock,
761 struct buffer_head *bh, int flags)
763 handle_t *handle = ext4_journal_current_handle();
764 struct ext4_map_blocks map;
765 int ret = 0, started = 0;
768 if (ext4_has_inline_data(inode))
772 map.m_len = bh->b_size >> inode->i_blkbits;
774 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
775 /* Direct IO write... */
776 if (map.m_len > DIO_MAX_BLOCKS)
777 map.m_len = DIO_MAX_BLOCKS;
778 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
779 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
781 if (IS_ERR(handle)) {
782 ret = PTR_ERR(handle);
788 ret = ext4_map_blocks(handle, inode, &map, flags);
790 map_bh(bh, inode->i_sb, map.m_pblk);
791 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
792 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
796 ext4_journal_stop(handle);
800 int ext4_get_block(struct inode *inode, sector_t iblock,
801 struct buffer_head *bh, int create)
803 return _ext4_get_block(inode, iblock, bh,
804 create ? EXT4_GET_BLOCKS_CREATE : 0);
808 * `handle' can be NULL if create is zero
810 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
811 ext4_lblk_t block, int create, int *errp)
813 struct ext4_map_blocks map;
814 struct buffer_head *bh;
817 J_ASSERT(handle != NULL || create == 0);
821 err = ext4_map_blocks(handle, inode, &map,
822 create ? EXT4_GET_BLOCKS_CREATE : 0);
824 /* ensure we send some value back into *errp */
827 if (create && err == 0)
828 err = -ENOSPC; /* should never happen */
834 bh = sb_getblk(inode->i_sb, map.m_pblk);
839 if (map.m_flags & EXT4_MAP_NEW) {
840 J_ASSERT(create != 0);
841 J_ASSERT(handle != NULL);
844 * Now that we do not always journal data, we should
845 * keep in mind whether this should always journal the
846 * new buffer as metadata. For now, regular file
847 * writes use ext4_get_block instead, so it's not a
851 BUFFER_TRACE(bh, "call get_create_access");
852 fatal = ext4_journal_get_create_access(handle, bh);
853 if (!fatal && !buffer_uptodate(bh)) {
854 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
855 set_buffer_uptodate(bh);
858 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
859 err = ext4_handle_dirty_metadata(handle, inode, bh);
863 BUFFER_TRACE(bh, "not a new buffer");
873 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
874 ext4_lblk_t block, int create, int *err)
876 struct buffer_head *bh;
878 bh = ext4_getblk(handle, inode, block, create, err);
881 if (buffer_uptodate(bh))
883 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
885 if (buffer_uptodate(bh))
892 int ext4_walk_page_buffers(handle_t *handle,
893 struct buffer_head *head,
897 int (*fn)(handle_t *handle,
898 struct buffer_head *bh))
900 struct buffer_head *bh;
901 unsigned block_start, block_end;
902 unsigned blocksize = head->b_size;
904 struct buffer_head *next;
906 for (bh = head, block_start = 0;
907 ret == 0 && (bh != head || !block_start);
908 block_start = block_end, bh = next) {
909 next = bh->b_this_page;
910 block_end = block_start + blocksize;
911 if (block_end <= from || block_start >= to) {
912 if (partial && !buffer_uptodate(bh))
916 err = (*fn)(handle, bh);
924 * To preserve ordering, it is essential that the hole instantiation and
925 * the data write be encapsulated in a single transaction. We cannot
926 * close off a transaction and start a new one between the ext4_get_block()
927 * and the commit_write(). So doing the jbd2_journal_start at the start of
928 * prepare_write() is the right place.
930 * Also, this function can nest inside ext4_writepage(). In that case, we
931 * *know* that ext4_writepage() has generated enough buffer credits to do the
932 * whole page. So we won't block on the journal in that case, which is good,
933 * because the caller may be PF_MEMALLOC.
935 * By accident, ext4 can be reentered when a transaction is open via
936 * quota file writes. If we were to commit the transaction while thus
937 * reentered, there can be a deadlock - we would be holding a quota
938 * lock, and the commit would never complete if another thread had a
939 * transaction open and was blocking on the quota lock - a ranking
942 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
943 * will _not_ run commit under these circumstances because handle->h_ref
944 * is elevated. We'll still have enough credits for the tiny quotafile
947 int do_journal_get_write_access(handle_t *handle,
948 struct buffer_head *bh)
950 int dirty = buffer_dirty(bh);
953 if (!buffer_mapped(bh) || buffer_freed(bh))
956 * __block_write_begin() could have dirtied some buffers. Clean
957 * the dirty bit as jbd2_journal_get_write_access() could complain
958 * otherwise about fs integrity issues. Setting of the dirty bit
959 * by __block_write_begin() isn't a real problem here as we clear
960 * the bit before releasing a page lock and thus writeback cannot
961 * ever write the buffer.
964 clear_buffer_dirty(bh);
965 ret = ext4_journal_get_write_access(handle, bh);
967 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
971 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
972 struct buffer_head *bh_result, int create);
973 static int ext4_write_begin(struct file *file, struct address_space *mapping,
974 loff_t pos, unsigned len, unsigned flags,
975 struct page **pagep, void **fsdata)
977 struct inode *inode = mapping->host;
978 int ret, needed_blocks;
985 trace_ext4_write_begin(inode, pos, len, flags);
987 * Reserve one block more for addition to orphan list in case
988 * we allocate blocks but write fails for some reason
990 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
991 index = pos >> PAGE_CACHE_SHIFT;
992 from = pos & (PAGE_CACHE_SIZE - 1);
995 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
996 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1005 * grab_cache_page_write_begin() can take a long time if the
1006 * system is thrashing due to memory pressure, or if the page
1007 * is being written back. So grab it first before we start
1008 * the transaction handle. This also allows us to allocate
1009 * the page (if needed) without using GFP_NOFS.
1012 page = grab_cache_page_write_begin(mapping, index, flags);
1018 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1019 if (IS_ERR(handle)) {
1020 page_cache_release(page);
1021 return PTR_ERR(handle);
1025 if (page->mapping != mapping) {
1026 /* The page got truncated from under us */
1028 page_cache_release(page);
1029 ext4_journal_stop(handle);
1032 wait_on_page_writeback(page);
1034 if (ext4_should_dioread_nolock(inode))
1035 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1037 ret = __block_write_begin(page, pos, len, ext4_get_block);
1039 if (!ret && ext4_should_journal_data(inode)) {
1040 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1042 do_journal_get_write_access);
1048 * __block_write_begin may have instantiated a few blocks
1049 * outside i_size. Trim these off again. Don't need
1050 * i_size_read because we hold i_mutex.
1052 * Add inode to orphan list in case we crash before
1055 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1056 ext4_orphan_add(handle, inode);
1058 ext4_journal_stop(handle);
1059 if (pos + len > inode->i_size) {
1060 ext4_truncate_failed_write(inode);
1062 * If truncate failed early the inode might
1063 * still be on the orphan list; we need to
1064 * make sure the inode is removed from the
1065 * orphan list in that case.
1068 ext4_orphan_del(NULL, inode);
1071 if (ret == -ENOSPC &&
1072 ext4_should_retry_alloc(inode->i_sb, &retries))
1074 page_cache_release(page);
1081 /* For write_end() in data=journal mode */
1082 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1084 if (!buffer_mapped(bh) || buffer_freed(bh))
1086 set_buffer_uptodate(bh);
1087 return ext4_handle_dirty_metadata(handle, NULL, bh);
1090 static int ext4_generic_write_end(struct file *file,
1091 struct address_space *mapping,
1092 loff_t pos, unsigned len, unsigned copied,
1093 struct page *page, void *fsdata)
1095 int i_size_changed = 0;
1096 struct inode *inode = mapping->host;
1097 handle_t *handle = ext4_journal_current_handle();
1099 if (ext4_has_inline_data(inode))
1100 copied = ext4_write_inline_data_end(inode, pos, len,
1103 copied = block_write_end(file, mapping, pos,
1104 len, copied, page, fsdata);
1107 * No need to use i_size_read() here, the i_size
1108 * cannot change under us because we hold i_mutex.
1110 * But it's important to update i_size while still holding page lock:
1111 * page writeout could otherwise come in and zero beyond i_size.
1113 if (pos + copied > inode->i_size) {
1114 i_size_write(inode, pos + copied);
1118 if (pos + copied > EXT4_I(inode)->i_disksize) {
1119 /* We need to mark inode dirty even if
1120 * new_i_size is less that inode->i_size
1121 * bu greater than i_disksize.(hint delalloc)
1123 ext4_update_i_disksize(inode, (pos + copied));
1127 page_cache_release(page);
1130 * Don't mark the inode dirty under page lock. First, it unnecessarily
1131 * makes the holding time of page lock longer. Second, it forces lock
1132 * ordering of page lock and transaction start for journaling
1136 ext4_mark_inode_dirty(handle, inode);
1142 * We need to pick up the new inode size which generic_commit_write gave us
1143 * `file' can be NULL - eg, when called from page_symlink().
1145 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1146 * buffers are managed internally.
1148 static int ext4_ordered_write_end(struct file *file,
1149 struct address_space *mapping,
1150 loff_t pos, unsigned len, unsigned copied,
1151 struct page *page, void *fsdata)
1153 handle_t *handle = ext4_journal_current_handle();
1154 struct inode *inode = mapping->host;
1157 trace_ext4_ordered_write_end(inode, pos, len, copied);
1158 ret = ext4_jbd2_file_inode(handle, inode);
1161 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1164 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1165 /* if we have allocated more blocks and copied
1166 * less. We will have blocks allocated outside
1167 * inode->i_size. So truncate them
1169 ext4_orphan_add(handle, inode);
1174 page_cache_release(page);
1177 ret2 = ext4_journal_stop(handle);
1181 if (pos + len > inode->i_size) {
1182 ext4_truncate_failed_write(inode);
1184 * If truncate failed early the inode might still be
1185 * on the orphan list; we need to make sure the inode
1186 * is removed from the orphan list in that case.
1189 ext4_orphan_del(NULL, inode);
1193 return ret ? ret : copied;
1196 static int ext4_writeback_write_end(struct file *file,
1197 struct address_space *mapping,
1198 loff_t pos, unsigned len, unsigned copied,
1199 struct page *page, void *fsdata)
1201 handle_t *handle = ext4_journal_current_handle();
1202 struct inode *inode = mapping->host;
1205 trace_ext4_writeback_write_end(inode, pos, len, copied);
1206 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1209 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1210 /* if we have allocated more blocks and copied
1211 * less. We will have blocks allocated outside
1212 * inode->i_size. So truncate them
1214 ext4_orphan_add(handle, inode);
1219 ret2 = ext4_journal_stop(handle);
1223 if (pos + len > inode->i_size) {
1224 ext4_truncate_failed_write(inode);
1226 * If truncate failed early the inode might still be
1227 * on the orphan list; we need to make sure the inode
1228 * is removed from the orphan list in that case.
1231 ext4_orphan_del(NULL, inode);
1234 return ret ? ret : copied;
1237 static int ext4_journalled_write_end(struct file *file,
1238 struct address_space *mapping,
1239 loff_t pos, unsigned len, unsigned copied,
1240 struct page *page, void *fsdata)
1242 handle_t *handle = ext4_journal_current_handle();
1243 struct inode *inode = mapping->host;
1249 trace_ext4_journalled_write_end(inode, pos, len, copied);
1250 from = pos & (PAGE_CACHE_SIZE - 1);
1253 BUG_ON(!ext4_handle_valid(handle));
1255 if (ext4_has_inline_data(inode))
1256 copied = ext4_write_inline_data_end(inode, pos, len,
1260 if (!PageUptodate(page))
1262 page_zero_new_buffers(page, from+copied, to);
1265 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1266 to, &partial, write_end_fn);
1268 SetPageUptodate(page);
1270 new_i_size = pos + copied;
1271 if (new_i_size > inode->i_size)
1272 i_size_write(inode, pos+copied);
1273 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1274 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1275 if (new_i_size > EXT4_I(inode)->i_disksize) {
1276 ext4_update_i_disksize(inode, new_i_size);
1277 ret2 = ext4_mark_inode_dirty(handle, inode);
1283 page_cache_release(page);
1284 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1285 /* if we have allocated more blocks and copied
1286 * less. We will have blocks allocated outside
1287 * inode->i_size. So truncate them
1289 ext4_orphan_add(handle, inode);
1291 ret2 = ext4_journal_stop(handle);
1294 if (pos + len > inode->i_size) {
1295 ext4_truncate_failed_write(inode);
1297 * If truncate failed early the inode might still be
1298 * on the orphan list; we need to make sure the inode
1299 * is removed from the orphan list in that case.
1302 ext4_orphan_del(NULL, inode);
1305 return ret ? ret : copied;
1309 * Reserve a metadata for a single block located at lblock
1311 static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1314 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1315 struct ext4_inode_info *ei = EXT4_I(inode);
1316 unsigned int md_needed;
1317 ext4_lblk_t save_last_lblock;
1321 * recalculate the amount of metadata blocks to reserve
1322 * in order to allocate nrblocks
1323 * worse case is one extent per block
1326 spin_lock(&ei->i_block_reservation_lock);
1328 * ext4_calc_metadata_amount() has side effects, which we have
1329 * to be prepared undo if we fail to claim space.
1331 save_len = ei->i_da_metadata_calc_len;
1332 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1333 md_needed = EXT4_NUM_B2C(sbi,
1334 ext4_calc_metadata_amount(inode, lblock));
1335 trace_ext4_da_reserve_space(inode, md_needed);
1338 * We do still charge estimated metadata to the sb though;
1339 * we cannot afford to run out of free blocks.
1341 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1342 ei->i_da_metadata_calc_len = save_len;
1343 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1344 spin_unlock(&ei->i_block_reservation_lock);
1345 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1351 ei->i_reserved_meta_blocks += md_needed;
1352 spin_unlock(&ei->i_block_reservation_lock);
1354 return 0; /* success */
1358 * Reserve a single cluster located at lblock
1360 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1363 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1364 struct ext4_inode_info *ei = EXT4_I(inode);
1365 unsigned int md_needed;
1367 ext4_lblk_t save_last_lblock;
1371 * We will charge metadata quota at writeout time; this saves
1372 * us from metadata over-estimation, though we may go over by
1373 * a small amount in the end. Here we just reserve for data.
1375 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1380 * recalculate the amount of metadata blocks to reserve
1381 * in order to allocate nrblocks
1382 * worse case is one extent per block
1385 spin_lock(&ei->i_block_reservation_lock);
1387 * ext4_calc_metadata_amount() has side effects, which we have
1388 * to be prepared undo if we fail to claim space.
1390 save_len = ei->i_da_metadata_calc_len;
1391 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1392 md_needed = EXT4_NUM_B2C(sbi,
1393 ext4_calc_metadata_amount(inode, lblock));
1394 trace_ext4_da_reserve_space(inode, md_needed);
1397 * We do still charge estimated metadata to the sb though;
1398 * we cannot afford to run out of free blocks.
1400 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1401 ei->i_da_metadata_calc_len = save_len;
1402 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1403 spin_unlock(&ei->i_block_reservation_lock);
1404 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1408 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1411 ei->i_reserved_data_blocks++;
1412 ei->i_reserved_meta_blocks += md_needed;
1413 spin_unlock(&ei->i_block_reservation_lock);
1415 return 0; /* success */
1418 static void ext4_da_release_space(struct inode *inode, int to_free)
1420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1421 struct ext4_inode_info *ei = EXT4_I(inode);
1424 return; /* Nothing to release, exit */
1426 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1428 trace_ext4_da_release_space(inode, to_free);
1429 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1431 * if there aren't enough reserved blocks, then the
1432 * counter is messed up somewhere. Since this
1433 * function is called from invalidate page, it's
1434 * harmless to return without any action.
1436 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1437 "ino %lu, to_free %d with only %d reserved "
1438 "data blocks", inode->i_ino, to_free,
1439 ei->i_reserved_data_blocks);
1441 to_free = ei->i_reserved_data_blocks;
1443 ei->i_reserved_data_blocks -= to_free;
1445 if (ei->i_reserved_data_blocks == 0) {
1447 * We can release all of the reserved metadata blocks
1448 * only when we have written all of the delayed
1449 * allocation blocks.
1450 * Note that in case of bigalloc, i_reserved_meta_blocks,
1451 * i_reserved_data_blocks, etc. refer to number of clusters.
1453 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1454 ei->i_reserved_meta_blocks);
1455 ei->i_reserved_meta_blocks = 0;
1456 ei->i_da_metadata_calc_len = 0;
1459 /* update fs dirty data blocks counter */
1460 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1462 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1464 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1467 static void ext4_da_page_release_reservation(struct page *page,
1468 unsigned long offset)
1471 struct buffer_head *head, *bh;
1472 unsigned int curr_off = 0;
1473 struct inode *inode = page->mapping->host;
1474 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1478 head = page_buffers(page);
1481 unsigned int next_off = curr_off + bh->b_size;
1483 if ((offset <= curr_off) && (buffer_delay(bh))) {
1485 clear_buffer_delay(bh);
1487 curr_off = next_off;
1488 } while ((bh = bh->b_this_page) != head);
1491 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1492 ext4_es_remove_extent(inode, lblk, to_release);
1495 /* If we have released all the blocks belonging to a cluster, then we
1496 * need to release the reserved space for that cluster. */
1497 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1498 while (num_clusters > 0) {
1499 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1500 ((num_clusters - 1) << sbi->s_cluster_bits);
1501 if (sbi->s_cluster_ratio == 1 ||
1502 !ext4_find_delalloc_cluster(inode, lblk))
1503 ext4_da_release_space(inode, 1);
1510 * Delayed allocation stuff
1514 * mpage_da_submit_io - walks through extent of pages and try to write
1515 * them with writepage() call back
1517 * @mpd->inode: inode
1518 * @mpd->first_page: first page of the extent
1519 * @mpd->next_page: page after the last page of the extent
1521 * By the time mpage_da_submit_io() is called we expect all blocks
1522 * to be allocated. this may be wrong if allocation failed.
1524 * As pages are already locked by write_cache_pages(), we can't use it
1526 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1527 struct ext4_map_blocks *map)
1529 struct pagevec pvec;
1530 unsigned long index, end;
1531 int ret = 0, err, nr_pages, i;
1532 struct inode *inode = mpd->inode;
1533 struct address_space *mapping = inode->i_mapping;
1534 loff_t size = i_size_read(inode);
1535 unsigned int len, block_start;
1536 struct buffer_head *bh, *page_bufs = NULL;
1537 sector_t pblock = 0, cur_logical = 0;
1538 struct ext4_io_submit io_submit;
1540 BUG_ON(mpd->next_page <= mpd->first_page);
1541 memset(&io_submit, 0, sizeof(io_submit));
1543 * We need to start from the first_page to the next_page - 1
1544 * to make sure we also write the mapped dirty buffer_heads.
1545 * If we look at mpd->b_blocknr we would only be looking
1546 * at the currently mapped buffer_heads.
1548 index = mpd->first_page;
1549 end = mpd->next_page - 1;
1551 pagevec_init(&pvec, 0);
1552 while (index <= end) {
1553 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1556 for (i = 0; i < nr_pages; i++) {
1558 struct page *page = pvec.pages[i];
1560 index = page->index;
1564 if (index == size >> PAGE_CACHE_SHIFT)
1565 len = size & ~PAGE_CACHE_MASK;
1567 len = PAGE_CACHE_SIZE;
1569 cur_logical = index << (PAGE_CACHE_SHIFT -
1571 pblock = map->m_pblk + (cur_logical -
1576 BUG_ON(!PageLocked(page));
1577 BUG_ON(PageWriteback(page));
1579 bh = page_bufs = page_buffers(page);
1582 if (map && (cur_logical >= map->m_lblk) &&
1583 (cur_logical <= (map->m_lblk +
1584 (map->m_len - 1)))) {
1585 if (buffer_delay(bh)) {
1586 clear_buffer_delay(bh);
1587 bh->b_blocknr = pblock;
1589 if (buffer_unwritten(bh) ||
1591 BUG_ON(bh->b_blocknr != pblock);
1592 if (map->m_flags & EXT4_MAP_UNINIT)
1593 set_buffer_uninit(bh);
1594 clear_buffer_unwritten(bh);
1598 * skip page if block allocation undone and
1601 if (ext4_bh_delay_or_unwritten(NULL, bh))
1603 bh = bh->b_this_page;
1604 block_start += bh->b_size;
1607 } while (bh != page_bufs);
1614 clear_page_dirty_for_io(page);
1615 err = ext4_bio_write_page(&io_submit, page, len,
1618 mpd->pages_written++;
1620 * In error case, we have to continue because
1621 * remaining pages are still locked
1626 pagevec_release(&pvec);
1628 ext4_io_submit(&io_submit);
1632 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1636 struct pagevec pvec;
1637 struct inode *inode = mpd->inode;
1638 struct address_space *mapping = inode->i_mapping;
1639 ext4_lblk_t start, last;
1641 index = mpd->first_page;
1642 end = mpd->next_page - 1;
1644 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1645 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1646 ext4_es_remove_extent(inode, start, last - start + 1);
1648 pagevec_init(&pvec, 0);
1649 while (index <= end) {
1650 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1653 for (i = 0; i < nr_pages; i++) {
1654 struct page *page = pvec.pages[i];
1655 if (page->index > end)
1657 BUG_ON(!PageLocked(page));
1658 BUG_ON(PageWriteback(page));
1659 block_invalidatepage(page, 0);
1660 ClearPageUptodate(page);
1663 index = pvec.pages[nr_pages - 1]->index + 1;
1664 pagevec_release(&pvec);
1669 static void ext4_print_free_blocks(struct inode *inode)
1671 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1672 struct super_block *sb = inode->i_sb;
1674 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1675 EXT4_C2B(EXT4_SB(inode->i_sb),
1676 ext4_count_free_clusters(inode->i_sb)));
1677 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1678 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1679 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1680 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1681 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1682 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1683 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1684 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1685 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1686 EXT4_I(inode)->i_reserved_data_blocks);
1687 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1688 EXT4_I(inode)->i_reserved_meta_blocks);
1693 * mpage_da_map_and_submit - go through given space, map them
1694 * if necessary, and then submit them for I/O
1696 * @mpd - bh describing space
1698 * The function skips space we know is already mapped to disk blocks.
1701 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1703 int err, blks, get_blocks_flags;
1704 struct ext4_map_blocks map, *mapp = NULL;
1705 sector_t next = mpd->b_blocknr;
1706 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1707 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1708 handle_t *handle = NULL;
1711 * If the blocks are mapped already, or we couldn't accumulate
1712 * any blocks, then proceed immediately to the submission stage.
1714 if ((mpd->b_size == 0) ||
1715 ((mpd->b_state & (1 << BH_Mapped)) &&
1716 !(mpd->b_state & (1 << BH_Delay)) &&
1717 !(mpd->b_state & (1 << BH_Unwritten))))
1720 handle = ext4_journal_current_handle();
1724 * Call ext4_map_blocks() to allocate any delayed allocation
1725 * blocks, or to convert an uninitialized extent to be
1726 * initialized (in the case where we have written into
1727 * one or more preallocated blocks).
1729 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1730 * indicate that we are on the delayed allocation path. This
1731 * affects functions in many different parts of the allocation
1732 * call path. This flag exists primarily because we don't
1733 * want to change *many* call functions, so ext4_map_blocks()
1734 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1735 * inode's allocation semaphore is taken.
1737 * If the blocks in questions were delalloc blocks, set
1738 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1739 * variables are updated after the blocks have been allocated.
1742 map.m_len = max_blocks;
1743 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1744 if (ext4_should_dioread_nolock(mpd->inode))
1745 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1746 if (mpd->b_state & (1 << BH_Delay))
1747 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1749 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1751 struct super_block *sb = mpd->inode->i_sb;
1755 * If get block returns EAGAIN or ENOSPC and there
1756 * appears to be free blocks we will just let
1757 * mpage_da_submit_io() unlock all of the pages.
1762 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1768 * get block failure will cause us to loop in
1769 * writepages, because a_ops->writepage won't be able
1770 * to make progress. The page will be redirtied by
1771 * writepage and writepages will again try to write
1774 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1775 ext4_msg(sb, KERN_CRIT,
1776 "delayed block allocation failed for inode %lu "
1777 "at logical offset %llu with max blocks %zd "
1778 "with error %d", mpd->inode->i_ino,
1779 (unsigned long long) next,
1780 mpd->b_size >> mpd->inode->i_blkbits, err);
1781 ext4_msg(sb, KERN_CRIT,
1782 "This should not happen!! Data will be lost");
1784 ext4_print_free_blocks(mpd->inode);
1786 /* invalidate all the pages */
1787 ext4_da_block_invalidatepages(mpd);
1789 /* Mark this page range as having been completed */
1796 if (map.m_flags & EXT4_MAP_NEW) {
1797 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1800 for (i = 0; i < map.m_len; i++)
1801 unmap_underlying_metadata(bdev, map.m_pblk + i);
1805 * Update on-disk size along with block allocation.
1807 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1808 if (disksize > i_size_read(mpd->inode))
1809 disksize = i_size_read(mpd->inode);
1810 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1811 ext4_update_i_disksize(mpd->inode, disksize);
1812 err = ext4_mark_inode_dirty(handle, mpd->inode);
1814 ext4_error(mpd->inode->i_sb,
1815 "Failed to mark inode %lu dirty",
1820 mpage_da_submit_io(mpd, mapp);
1824 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1825 (1 << BH_Delay) | (1 << BH_Unwritten))
1828 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1830 * @mpd->lbh - extent of blocks
1831 * @logical - logical number of the block in the file
1832 * @b_state - b_state of the buffer head added
1834 * the function is used to collect contig. blocks in same state
1836 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical,
1837 unsigned long b_state)
1840 int blkbits = mpd->inode->i_blkbits;
1841 int nrblocks = mpd->b_size >> blkbits;
1844 * XXX Don't go larger than mballoc is willing to allocate
1845 * This is a stopgap solution. We eventually need to fold
1846 * mpage_da_submit_io() into this function and then call
1847 * ext4_map_blocks() multiple times in a loop
1849 if (nrblocks >= (8*1024*1024 >> blkbits))
1852 /* check if the reserved journal credits might overflow */
1853 if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) {
1854 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1856 * With non-extent format we are limited by the journal
1857 * credit available. Total credit needed to insert
1858 * nrblocks contiguous blocks is dependent on the
1859 * nrblocks. So limit nrblocks.
1865 * First block in the extent
1867 if (mpd->b_size == 0) {
1868 mpd->b_blocknr = logical;
1869 mpd->b_size = 1 << blkbits;
1870 mpd->b_state = b_state & BH_FLAGS;
1874 next = mpd->b_blocknr + nrblocks;
1876 * Can we merge the block to our big extent?
1878 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1879 mpd->b_size += 1 << blkbits;
1885 * We couldn't merge the block to our extent, so we
1886 * need to flush current extent and start new one
1888 mpage_da_map_and_submit(mpd);
1892 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1894 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1898 * This function is grabs code from the very beginning of
1899 * ext4_map_blocks, but assumes that the caller is from delayed write
1900 * time. This function looks up the requested blocks and sets the
1901 * buffer delay bit under the protection of i_data_sem.
1903 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1904 struct ext4_map_blocks *map,
1905 struct buffer_head *bh)
1907 struct extent_status es;
1909 sector_t invalid_block = ~((sector_t) 0xffff);
1910 #ifdef ES_AGGRESSIVE_TEST
1911 struct ext4_map_blocks orig_map;
1913 memcpy(&orig_map, map, sizeof(*map));
1916 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1920 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1921 "logical block %lu\n", inode->i_ino, map->m_len,
1922 (unsigned long) map->m_lblk);
1924 /* Lookup extent status tree firstly */
1925 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1927 if (ext4_es_is_hole(&es)) {
1929 down_read((&EXT4_I(inode)->i_data_sem));
1934 * Delayed extent could be allocated by fallocate.
1935 * So we need to check it.
1937 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1938 map_bh(bh, inode->i_sb, invalid_block);
1940 set_buffer_delay(bh);
1944 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1945 retval = es.es_len - (iblock - es.es_lblk);
1946 if (retval > map->m_len)
1947 retval = map->m_len;
1948 map->m_len = retval;
1949 if (ext4_es_is_written(&es))
1950 map->m_flags |= EXT4_MAP_MAPPED;
1951 else if (ext4_es_is_unwritten(&es))
1952 map->m_flags |= EXT4_MAP_UNWRITTEN;
1956 #ifdef ES_AGGRESSIVE_TEST
1957 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1963 * Try to see if we can get the block without requesting a new
1964 * file system block.
1966 down_read((&EXT4_I(inode)->i_data_sem));
1967 if (ext4_has_inline_data(inode)) {
1969 * We will soon create blocks for this page, and let
1970 * us pretend as if the blocks aren't allocated yet.
1971 * In case of clusters, we have to handle the work
1972 * of mapping from cluster so that the reserved space
1973 * is calculated properly.
1975 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1976 ext4_find_delalloc_cluster(inode, map->m_lblk))
1977 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1979 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1980 retval = ext4_ext_map_blocks(NULL, inode, map,
1981 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1983 retval = ext4_ind_map_blocks(NULL, inode, map,
1984 EXT4_GET_BLOCKS_NO_PUT_HOLE);
1990 * XXX: __block_prepare_write() unmaps passed block,
1994 * If the block was allocated from previously allocated cluster,
1995 * then we don't need to reserve it again. However we still need
1996 * to reserve metadata for every block we're going to write.
1998 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1999 ret = ext4_da_reserve_space(inode, iblock);
2001 /* not enough space to reserve */
2006 ret = ext4_da_reserve_metadata(inode, iblock);
2008 /* not enough space to reserve */
2014 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2015 ~0, EXTENT_STATUS_DELAYED);
2021 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
2022 * and it should not appear on the bh->b_state.
2024 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
2026 map_bh(bh, inode->i_sb, invalid_block);
2028 set_buffer_delay(bh);
2029 } else if (retval > 0) {
2031 unsigned long long status;
2033 #ifdef ES_AGGRESSIVE_TEST
2034 if (retval != map->m_len) {
2035 printk("ES len assertation failed for inode: %lu "
2036 "retval %d != map->m_len %d "
2037 "in %s (lookup)\n", inode->i_ino, retval,
2038 map->m_len, __func__);
2042 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
2043 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
2044 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
2045 map->m_pblk, status);
2051 up_read((&EXT4_I(inode)->i_data_sem));
2057 * This is a special get_blocks_t callback which is used by
2058 * ext4_da_write_begin(). It will either return mapped block or
2059 * reserve space for a single block.
2061 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2062 * We also have b_blocknr = -1 and b_bdev initialized properly
2064 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2065 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2066 * initialized properly.
2068 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2069 struct buffer_head *bh, int create)
2071 struct ext4_map_blocks map;
2074 BUG_ON(create == 0);
2075 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2077 map.m_lblk = iblock;
2081 * first, we need to know whether the block is allocated already
2082 * preallocated blocks are unmapped but should treated
2083 * the same as allocated blocks.
2085 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
2089 map_bh(bh, inode->i_sb, map.m_pblk);
2090 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2092 if (buffer_unwritten(bh)) {
2093 /* A delayed write to unwritten bh should be marked
2094 * new and mapped. Mapped ensures that we don't do
2095 * get_block multiple times when we write to the same
2096 * offset and new ensures that we do proper zero out
2097 * for partial write.
2100 set_buffer_mapped(bh);
2105 static int bget_one(handle_t *handle, struct buffer_head *bh)
2111 static int bput_one(handle_t *handle, struct buffer_head *bh)
2117 static int __ext4_journalled_writepage(struct page *page,
2120 struct address_space *mapping = page->mapping;
2121 struct inode *inode = mapping->host;
2122 struct buffer_head *page_bufs = NULL;
2123 handle_t *handle = NULL;
2124 int ret = 0, err = 0;
2125 int inline_data = ext4_has_inline_data(inode);
2126 struct buffer_head *inode_bh = NULL;
2128 ClearPageChecked(page);
2131 BUG_ON(page->index != 0);
2132 BUG_ON(len > ext4_get_max_inline_size(inode));
2133 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2134 if (inode_bh == NULL)
2137 page_bufs = page_buffers(page);
2142 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2145 /* As soon as we unlock the page, it can go away, but we have
2146 * references to buffers so we are safe */
2149 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2150 ext4_writepage_trans_blocks(inode));
2151 if (IS_ERR(handle)) {
2152 ret = PTR_ERR(handle);
2156 BUG_ON(!ext4_handle_valid(handle));
2159 ret = ext4_journal_get_write_access(handle, inode_bh);
2161 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
2164 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2165 do_journal_get_write_access);
2167 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2172 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2173 err = ext4_journal_stop(handle);
2177 if (!ext4_has_inline_data(inode))
2178 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2180 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2187 * Note that we don't need to start a transaction unless we're journaling data
2188 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2189 * need to file the inode to the transaction's list in ordered mode because if
2190 * we are writing back data added by write(), the inode is already there and if
2191 * we are writing back data modified via mmap(), no one guarantees in which
2192 * transaction the data will hit the disk. In case we are journaling data, we
2193 * cannot start transaction directly because transaction start ranks above page
2194 * lock so we have to do some magic.
2196 * This function can get called via...
2197 * - ext4_da_writepages after taking page lock (have journal handle)
2198 * - journal_submit_inode_data_buffers (no journal handle)
2199 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2200 * - grab_page_cache when doing write_begin (have journal handle)
2202 * We don't do any block allocation in this function. If we have page with
2203 * multiple blocks we need to write those buffer_heads that are mapped. This
2204 * is important for mmaped based write. So if we do with blocksize 1K
2205 * truncate(f, 1024);
2206 * a = mmap(f, 0, 4096);
2208 * truncate(f, 4096);
2209 * we have in the page first buffer_head mapped via page_mkwrite call back
2210 * but other buffer_heads would be unmapped but dirty (dirty done via the
2211 * do_wp_page). So writepage should write the first block. If we modify
2212 * the mmap area beyond 1024 we will again get a page_fault and the
2213 * page_mkwrite callback will do the block allocation and mark the
2214 * buffer_heads mapped.
2216 * We redirty the page if we have any buffer_heads that is either delay or
2217 * unwritten in the page.
2219 * We can get recursively called as show below.
2221 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2224 * But since we don't do any block allocation we should not deadlock.
2225 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2227 static int ext4_writepage(struct page *page,
2228 struct writeback_control *wbc)
2233 struct buffer_head *page_bufs = NULL;
2234 struct inode *inode = page->mapping->host;
2235 struct ext4_io_submit io_submit;
2237 trace_ext4_writepage(page);
2238 size = i_size_read(inode);
2239 if (page->index == size >> PAGE_CACHE_SHIFT)
2240 len = size & ~PAGE_CACHE_MASK;
2242 len = PAGE_CACHE_SIZE;
2244 page_bufs = page_buffers(page);
2246 * We cannot do block allocation or other extent handling in this
2247 * function. If there are buffers needing that, we have to redirty
2248 * the page. But we may reach here when we do a journal commit via
2249 * journal_submit_inode_data_buffers() and in that case we must write
2250 * allocated buffers to achieve data=ordered mode guarantees.
2252 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2253 ext4_bh_delay_or_unwritten)) {
2254 redirty_page_for_writepage(wbc, page);
2255 if (current->flags & PF_MEMALLOC) {
2257 * For memory cleaning there's no point in writing only
2258 * some buffers. So just bail out. Warn if we came here
2259 * from direct reclaim.
2261 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2268 if (PageChecked(page) && ext4_should_journal_data(inode))
2270 * It's mmapped pagecache. Add buffers and journal it. There
2271 * doesn't seem much point in redirtying the page here.
2273 return __ext4_journalled_writepage(page, len);
2275 memset(&io_submit, 0, sizeof(io_submit));
2276 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
2277 ext4_io_submit(&io_submit);
2282 * This is called via ext4_da_writepages() to
2283 * calculate the total number of credits to reserve to fit
2284 * a single extent allocation into a single transaction,
2285 * ext4_da_writpeages() will loop calling this before
2286 * the block allocation.
2289 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2291 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2294 * With non-extent format the journal credit needed to
2295 * insert nrblocks contiguous block is dependent on
2296 * number of contiguous block. So we will limit
2297 * number of contiguous block to a sane value
2299 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2300 (max_blocks > EXT4_MAX_TRANS_DATA))
2301 max_blocks = EXT4_MAX_TRANS_DATA;
2303 return ext4_chunk_trans_blocks(inode, max_blocks);
2307 * write_cache_pages_da - walk the list of dirty pages of the given
2308 * address space and accumulate pages that need writing, and call
2309 * mpage_da_map_and_submit to map a single contiguous memory region
2310 * and then write them.
2312 static int write_cache_pages_da(handle_t *handle,
2313 struct address_space *mapping,
2314 struct writeback_control *wbc,
2315 struct mpage_da_data *mpd,
2316 pgoff_t *done_index)
2318 struct buffer_head *bh, *head;
2319 struct inode *inode = mapping->host;
2320 struct pagevec pvec;
2321 unsigned int nr_pages;
2324 long nr_to_write = wbc->nr_to_write;
2325 int i, tag, ret = 0;
2327 memset(mpd, 0, sizeof(struct mpage_da_data));
2330 pagevec_init(&pvec, 0);
2331 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2332 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2334 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2335 tag = PAGECACHE_TAG_TOWRITE;
2337 tag = PAGECACHE_TAG_DIRTY;
2339 *done_index = index;
2340 while (index <= end) {
2341 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2342 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2346 for (i = 0; i < nr_pages; i++) {
2347 struct page *page = pvec.pages[i];
2350 * At this point, the page may be truncated or
2351 * invalidated (changing page->mapping to NULL), or
2352 * even swizzled back from swapper_space to tmpfs file
2353 * mapping. However, page->index will not change
2354 * because we have a reference on the page.
2356 if (page->index > end)
2359 *done_index = page->index + 1;
2362 * If we can't merge this page, and we have
2363 * accumulated an contiguous region, write it
2365 if ((mpd->next_page != page->index) &&
2366 (mpd->next_page != mpd->first_page)) {
2367 mpage_da_map_and_submit(mpd);
2368 goto ret_extent_tail;
2374 * If the page is no longer dirty, or its
2375 * mapping no longer corresponds to inode we
2376 * are writing (which means it has been
2377 * truncated or invalidated), or the page is
2378 * already under writeback and we are not
2379 * doing a data integrity writeback, skip the page
2381 if (!PageDirty(page) ||
2382 (PageWriteback(page) &&
2383 (wbc->sync_mode == WB_SYNC_NONE)) ||
2384 unlikely(page->mapping != mapping)) {
2389 wait_on_page_writeback(page);
2390 BUG_ON(PageWriteback(page));
2393 * If we have inline data and arrive here, it means that
2394 * we will soon create the block for the 1st page, so
2395 * we'd better clear the inline data here.
2397 if (ext4_has_inline_data(inode)) {
2398 BUG_ON(ext4_test_inode_state(inode,
2399 EXT4_STATE_MAY_INLINE_DATA));
2400 ext4_destroy_inline_data(handle, inode);
2403 if (mpd->next_page != page->index)
2404 mpd->first_page = page->index;
2405 mpd->next_page = page->index + 1;
2406 logical = (sector_t) page->index <<
2407 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2409 /* Add all dirty buffers to mpd */
2410 head = page_buffers(page);
2413 BUG_ON(buffer_locked(bh));
2415 * We need to try to allocate unmapped blocks
2416 * in the same page. Otherwise we won't make
2417 * progress with the page in ext4_writepage
2419 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2420 mpage_add_bh_to_extent(mpd, logical,
2423 goto ret_extent_tail;
2424 } else if (buffer_dirty(bh) &&
2425 buffer_mapped(bh)) {
2427 * mapped dirty buffer. We need to
2428 * update the b_state because we look
2429 * at b_state in mpage_da_map_blocks.
2430 * We don't update b_size because if we
2431 * find an unmapped buffer_head later
2432 * we need to use the b_state flag of
2435 if (mpd->b_size == 0)
2437 bh->b_state & BH_FLAGS;
2440 } while ((bh = bh->b_this_page) != head);
2442 if (nr_to_write > 0) {
2444 if (nr_to_write == 0 &&
2445 wbc->sync_mode == WB_SYNC_NONE)
2447 * We stop writing back only if we are
2448 * not doing integrity sync. In case of
2449 * integrity sync we have to keep going
2450 * because someone may be concurrently
2451 * dirtying pages, and we might have
2452 * synced a lot of newly appeared dirty
2453 * pages, but have not synced all of the
2459 pagevec_release(&pvec);
2464 ret = MPAGE_DA_EXTENT_TAIL;
2466 pagevec_release(&pvec);
2472 static int ext4_da_writepages(struct address_space *mapping,
2473 struct writeback_control *wbc)
2476 int range_whole = 0;
2477 handle_t *handle = NULL;
2478 struct mpage_da_data mpd;
2479 struct inode *inode = mapping->host;
2480 int pages_written = 0;
2481 unsigned int max_pages;
2482 int range_cyclic, cycled = 1, io_done = 0;
2483 int needed_blocks, ret = 0;
2484 long desired_nr_to_write, nr_to_writebump = 0;
2485 loff_t range_start = wbc->range_start;
2486 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2487 pgoff_t done_index = 0;
2489 struct blk_plug plug;
2491 trace_ext4_da_writepages(inode, wbc);
2494 * No pages to write? This is mainly a kludge to avoid starting
2495 * a transaction for special inodes like journal inode on last iput()
2496 * because that could violate lock ordering on umount
2498 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2502 * If the filesystem has aborted, it is read-only, so return
2503 * right away instead of dumping stack traces later on that
2504 * will obscure the real source of the problem. We test
2505 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2506 * the latter could be true if the filesystem is mounted
2507 * read-only, and in that case, ext4_da_writepages should
2508 * *never* be called, so if that ever happens, we would want
2511 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2514 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2517 range_cyclic = wbc->range_cyclic;
2518 if (wbc->range_cyclic) {
2519 index = mapping->writeback_index;
2522 wbc->range_start = index << PAGE_CACHE_SHIFT;
2523 wbc->range_end = LLONG_MAX;
2524 wbc->range_cyclic = 0;
2527 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2528 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2532 * This works around two forms of stupidity. The first is in
2533 * the writeback code, which caps the maximum number of pages
2534 * written to be 1024 pages. This is wrong on multiple
2535 * levels; different architectues have a different page size,
2536 * which changes the maximum amount of data which gets
2537 * written. Secondly, 4 megabytes is way too small. XFS
2538 * forces this value to be 16 megabytes by multiplying
2539 * nr_to_write parameter by four, and then relies on its
2540 * allocator to allocate larger extents to make them
2541 * contiguous. Unfortunately this brings us to the second
2542 * stupidity, which is that ext4's mballoc code only allocates
2543 * at most 2048 blocks. So we force contiguous writes up to
2544 * the number of dirty blocks in the inode, or
2545 * sbi->max_writeback_mb_bump whichever is smaller.
2547 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2548 if (!range_cyclic && range_whole) {
2549 if (wbc->nr_to_write == LONG_MAX)
2550 desired_nr_to_write = wbc->nr_to_write;
2552 desired_nr_to_write = wbc->nr_to_write * 8;
2554 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2556 if (desired_nr_to_write > max_pages)
2557 desired_nr_to_write = max_pages;
2559 if (wbc->nr_to_write < desired_nr_to_write) {
2560 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2561 wbc->nr_to_write = desired_nr_to_write;
2565 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2566 tag_pages_for_writeback(mapping, index, end);
2568 blk_start_plug(&plug);
2569 while (!ret && wbc->nr_to_write > 0) {
2572 * we insert one extent at a time. So we need
2573 * credit needed for single extent allocation.
2574 * journalled mode is currently not supported
2577 BUG_ON(ext4_should_journal_data(inode));
2578 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2580 /* start a new transaction*/
2581 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2583 if (IS_ERR(handle)) {
2584 ret = PTR_ERR(handle);
2585 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2586 "%ld pages, ino %lu; err %d", __func__,
2587 wbc->nr_to_write, inode->i_ino, ret);
2588 blk_finish_plug(&plug);
2589 goto out_writepages;
2593 * Now call write_cache_pages_da() to find the next
2594 * contiguous region of logical blocks that need
2595 * blocks to be allocated by ext4 and submit them.
2597 ret = write_cache_pages_da(handle, mapping,
2598 wbc, &mpd, &done_index);
2600 * If we have a contiguous extent of pages and we
2601 * haven't done the I/O yet, map the blocks and submit
2604 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2605 mpage_da_map_and_submit(&mpd);
2606 ret = MPAGE_DA_EXTENT_TAIL;
2608 trace_ext4_da_write_pages(inode, &mpd);
2609 wbc->nr_to_write -= mpd.pages_written;
2611 ext4_journal_stop(handle);
2613 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2614 /* commit the transaction which would
2615 * free blocks released in the transaction
2618 jbd2_journal_force_commit_nested(sbi->s_journal);
2620 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2622 * Got one extent now try with rest of the pages.
2623 * If mpd.retval is set -EIO, journal is aborted.
2624 * So we don't need to write any more.
2626 pages_written += mpd.pages_written;
2629 } else if (wbc->nr_to_write)
2631 * There is no more writeout needed
2632 * or we requested for a noblocking writeout
2633 * and we found the device congested
2637 blk_finish_plug(&plug);
2638 if (!io_done && !cycled) {
2641 wbc->range_start = index << PAGE_CACHE_SHIFT;
2642 wbc->range_end = mapping->writeback_index - 1;
2647 wbc->range_cyclic = range_cyclic;
2648 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2650 * set the writeback_index so that range_cyclic
2651 * mode will write it back later
2653 mapping->writeback_index = done_index;
2656 wbc->nr_to_write -= nr_to_writebump;
2657 wbc->range_start = range_start;
2658 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2662 static int ext4_nonda_switch(struct super_block *sb)
2664 s64 free_blocks, dirty_blocks;
2665 struct ext4_sb_info *sbi = EXT4_SB(sb);
2668 * switch to non delalloc mode if we are running low
2669 * on free block. The free block accounting via percpu
2670 * counters can get slightly wrong with percpu_counter_batch getting
2671 * accumulated on each CPU without updating global counters
2672 * Delalloc need an accurate free block accounting. So switch
2673 * to non delalloc when we are near to error range.
2675 free_blocks = EXT4_C2B(sbi,
2676 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2677 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2679 * Start pushing delalloc when 1/2 of free blocks are dirty.
2681 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2682 !writeback_in_progress(sb->s_bdi) &&
2683 down_read_trylock(&sb->s_umount)) {
2684 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2685 up_read(&sb->s_umount);
2688 if (2 * free_blocks < 3 * dirty_blocks ||
2689 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2691 * free block count is less than 150% of dirty blocks
2692 * or free blocks is less than watermark
2699 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2700 loff_t pos, unsigned len, unsigned flags,
2701 struct page **pagep, void **fsdata)
2703 int ret, retries = 0;
2706 struct inode *inode = mapping->host;
2709 index = pos >> PAGE_CACHE_SHIFT;
2711 if (ext4_nonda_switch(inode->i_sb)) {
2712 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2713 return ext4_write_begin(file, mapping, pos,
2714 len, flags, pagep, fsdata);
2716 *fsdata = (void *)0;
2717 trace_ext4_da_write_begin(inode, pos, len, flags);
2719 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2720 ret = ext4_da_write_inline_data_begin(mapping, inode,
2730 * grab_cache_page_write_begin() can take a long time if the
2731 * system is thrashing due to memory pressure, or if the page
2732 * is being written back. So grab it first before we start
2733 * the transaction handle. This also allows us to allocate
2734 * the page (if needed) without using GFP_NOFS.
2737 page = grab_cache_page_write_begin(mapping, index, flags);
2743 * With delayed allocation, we don't log the i_disksize update
2744 * if there is delayed block allocation. But we still need
2745 * to journalling the i_disksize update if writes to the end
2746 * of file which has an already mapped buffer.
2749 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
2750 if (IS_ERR(handle)) {
2751 page_cache_release(page);
2752 return PTR_ERR(handle);
2756 if (page->mapping != mapping) {
2757 /* The page got truncated from under us */
2759 page_cache_release(page);
2760 ext4_journal_stop(handle);
2763 /* In case writeback began while the page was unlocked */
2764 wait_on_page_writeback(page);
2766 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2769 ext4_journal_stop(handle);
2771 * block_write_begin may have instantiated a few blocks
2772 * outside i_size. Trim these off again. Don't need
2773 * i_size_read because we hold i_mutex.
2775 if (pos + len > inode->i_size)
2776 ext4_truncate_failed_write(inode);
2778 if (ret == -ENOSPC &&
2779 ext4_should_retry_alloc(inode->i_sb, &retries))
2782 page_cache_release(page);
2791 * Check if we should update i_disksize
2792 * when write to the end of file but not require block allocation
2794 static int ext4_da_should_update_i_disksize(struct page *page,
2795 unsigned long offset)
2797 struct buffer_head *bh;
2798 struct inode *inode = page->mapping->host;
2802 bh = page_buffers(page);
2803 idx = offset >> inode->i_blkbits;
2805 for (i = 0; i < idx; i++)
2806 bh = bh->b_this_page;
2808 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2813 static int ext4_da_write_end(struct file *file,
2814 struct address_space *mapping,
2815 loff_t pos, unsigned len, unsigned copied,
2816 struct page *page, void *fsdata)
2818 struct inode *inode = mapping->host;
2820 handle_t *handle = ext4_journal_current_handle();
2822 unsigned long start, end;
2823 int write_mode = (int)(unsigned long)fsdata;
2825 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2826 switch (ext4_inode_journal_mode(inode)) {
2827 case EXT4_INODE_ORDERED_DATA_MODE:
2828 return ext4_ordered_write_end(file, mapping, pos,
2829 len, copied, page, fsdata);
2830 case EXT4_INODE_WRITEBACK_DATA_MODE:
2831 return ext4_writeback_write_end(file, mapping, pos,
2832 len, copied, page, fsdata);
2838 trace_ext4_da_write_end(inode, pos, len, copied);
2839 start = pos & (PAGE_CACHE_SIZE - 1);
2840 end = start + copied - 1;
2843 * generic_write_end() will run mark_inode_dirty() if i_size
2844 * changes. So let's piggyback the i_disksize mark_inode_dirty
2847 new_i_size = pos + copied;
2848 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2849 if (ext4_has_inline_data(inode) ||
2850 ext4_da_should_update_i_disksize(page, end)) {
2851 down_write(&EXT4_I(inode)->i_data_sem);
2852 if (new_i_size > EXT4_I(inode)->i_disksize)
2853 EXT4_I(inode)->i_disksize = new_i_size;
2854 up_write(&EXT4_I(inode)->i_data_sem);
2855 /* We need to mark inode dirty even if
2856 * new_i_size is less that inode->i_size
2857 * bu greater than i_disksize.(hint delalloc)
2859 ext4_mark_inode_dirty(handle, inode);
2863 if (write_mode != CONVERT_INLINE_DATA &&
2864 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2865 ext4_has_inline_data(inode))
2866 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2869 ret2 = generic_write_end(file, mapping, pos, len, copied,
2875 ret2 = ext4_journal_stop(handle);
2879 return ret ? ret : copied;
2882 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2885 * Drop reserved blocks
2887 BUG_ON(!PageLocked(page));
2888 if (!page_has_buffers(page))
2891 ext4_da_page_release_reservation(page, offset);
2894 ext4_invalidatepage(page, offset);
2900 * Force all delayed allocation blocks to be allocated for a given inode.
2902 int ext4_alloc_da_blocks(struct inode *inode)
2904 trace_ext4_alloc_da_blocks(inode);
2906 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2907 !EXT4_I(inode)->i_reserved_meta_blocks)
2911 * We do something simple for now. The filemap_flush() will
2912 * also start triggering a write of the data blocks, which is
2913 * not strictly speaking necessary (and for users of
2914 * laptop_mode, not even desirable). However, to do otherwise
2915 * would require replicating code paths in:
2917 * ext4_da_writepages() ->
2918 * write_cache_pages() ---> (via passed in callback function)
2919 * __mpage_da_writepage() -->
2920 * mpage_add_bh_to_extent()
2921 * mpage_da_map_blocks()
2923 * The problem is that write_cache_pages(), located in
2924 * mm/page-writeback.c, marks pages clean in preparation for
2925 * doing I/O, which is not desirable if we're not planning on
2928 * We could call write_cache_pages(), and then redirty all of
2929 * the pages by calling redirty_page_for_writepage() but that
2930 * would be ugly in the extreme. So instead we would need to
2931 * replicate parts of the code in the above functions,
2932 * simplifying them because we wouldn't actually intend to
2933 * write out the pages, but rather only collect contiguous
2934 * logical block extents, call the multi-block allocator, and
2935 * then update the buffer heads with the block allocations.
2937 * For now, though, we'll cheat by calling filemap_flush(),
2938 * which will map the blocks, and start the I/O, but not
2939 * actually wait for the I/O to complete.
2941 return filemap_flush(inode->i_mapping);
2945 * bmap() is special. It gets used by applications such as lilo and by
2946 * the swapper to find the on-disk block of a specific piece of data.
2948 * Naturally, this is dangerous if the block concerned is still in the
2949 * journal. If somebody makes a swapfile on an ext4 data-journaling
2950 * filesystem and enables swap, then they may get a nasty shock when the
2951 * data getting swapped to that swapfile suddenly gets overwritten by
2952 * the original zero's written out previously to the journal and
2953 * awaiting writeback in the kernel's buffer cache.
2955 * So, if we see any bmap calls here on a modified, data-journaled file,
2956 * take extra steps to flush any blocks which might be in the cache.
2958 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2960 struct inode *inode = mapping->host;
2965 * We can get here for an inline file via the FIBMAP ioctl
2967 if (ext4_has_inline_data(inode))
2970 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2971 test_opt(inode->i_sb, DELALLOC)) {
2973 * With delalloc we want to sync the file
2974 * so that we can make sure we allocate
2977 filemap_write_and_wait(mapping);
2980 if (EXT4_JOURNAL(inode) &&
2981 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2983 * This is a REALLY heavyweight approach, but the use of
2984 * bmap on dirty files is expected to be extremely rare:
2985 * only if we run lilo or swapon on a freshly made file
2986 * do we expect this to happen.
2988 * (bmap requires CAP_SYS_RAWIO so this does not
2989 * represent an unprivileged user DOS attack --- we'd be
2990 * in trouble if mortal users could trigger this path at
2993 * NB. EXT4_STATE_JDATA is not set on files other than
2994 * regular files. If somebody wants to bmap a directory
2995 * or symlink and gets confused because the buffer
2996 * hasn't yet been flushed to disk, they deserve
2997 * everything they get.
3000 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3001 journal = EXT4_JOURNAL(inode);
3002 jbd2_journal_lock_updates(journal);
3003 err = jbd2_journal_flush(journal);
3004 jbd2_journal_unlock_updates(journal);
3010 return generic_block_bmap(mapping, block, ext4_get_block);
3013 static int ext4_readpage(struct file *file, struct page *page)
3016 struct inode *inode = page->mapping->host;
3018 trace_ext4_readpage(page);
3020 if (ext4_has_inline_data(inode))
3021 ret = ext4_readpage_inline(inode, page);
3024 return mpage_readpage(page, ext4_get_block);
3030 ext4_readpages(struct file *file, struct address_space *mapping,
3031 struct list_head *pages, unsigned nr_pages)
3033 struct inode *inode = mapping->host;
3035 /* If the file has inline data, no need to do readpages. */
3036 if (ext4_has_inline_data(inode))
3039 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3042 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3044 trace_ext4_invalidatepage(page, offset);
3046 /* No journalling happens on data buffers when this function is used */
3047 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3049 block_invalidatepage(page, offset);
3052 static int __ext4_journalled_invalidatepage(struct page *page,
3053 unsigned long offset)
3055 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3057 trace_ext4_journalled_invalidatepage(page, offset);
3060 * If it's a full truncate we just forget about the pending dirtying
3063 ClearPageChecked(page);
3065 return jbd2_journal_invalidatepage(journal, page, offset);
3068 /* Wrapper for aops... */
3069 static void ext4_journalled_invalidatepage(struct page *page,
3070 unsigned long offset)
3072 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
3075 static int ext4_releasepage(struct page *page, gfp_t wait)
3077 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3079 trace_ext4_releasepage(page);
3081 /* Page has dirty journalled data -> cannot release */
3082 if (PageChecked(page))
3085 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3087 return try_to_free_buffers(page);
3091 * ext4_get_block used when preparing for a DIO write or buffer write.
3092 * We allocate an uinitialized extent if blocks haven't been allocated.
3093 * The extent will be converted to initialized after the IO is complete.
3095 int ext4_get_block_write(struct inode *inode, sector_t iblock,
3096 struct buffer_head *bh_result, int create)
3098 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3099 inode->i_ino, create);
3100 return _ext4_get_block(inode, iblock, bh_result,
3101 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3104 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
3105 struct buffer_head *bh_result, int create)
3107 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3108 inode->i_ino, create);
3109 return _ext4_get_block(inode, iblock, bh_result,
3110 EXT4_GET_BLOCKS_NO_LOCK);
3113 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3114 ssize_t size, void *private, int ret,
3117 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
3118 ext4_io_end_t *io_end = iocb->private;
3120 /* if not async direct IO or dio with 0 bytes write, just return */
3121 if (!io_end || !size)
3124 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3125 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3126 iocb->private, io_end->inode->i_ino, iocb, offset,
3129 iocb->private = NULL;
3131 /* if not aio dio with unwritten extents, just free io and return */
3132 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3133 ext4_free_io_end(io_end);
3135 inode_dio_done(inode);
3137 aio_complete(iocb, ret, 0);
3141 io_end->offset = offset;
3142 io_end->size = size;
3144 io_end->iocb = iocb;
3145 io_end->result = ret;
3148 ext4_add_complete_io(io_end);
3152 * For ext4 extent files, ext4 will do direct-io write to holes,
3153 * preallocated extents, and those write extend the file, no need to
3154 * fall back to buffered IO.
3156 * For holes, we fallocate those blocks, mark them as uninitialized
3157 * If those blocks were preallocated, we mark sure they are split, but
3158 * still keep the range to write as uninitialized.
3160 * The unwritten extents will be converted to written when DIO is completed.
3161 * For async direct IO, since the IO may still pending when return, we
3162 * set up an end_io call back function, which will do the conversion
3163 * when async direct IO completed.
3165 * If the O_DIRECT write will extend the file then add this inode to the
3166 * orphan list. So recovery will truncate it back to the original size
3167 * if the machine crashes during the write.
3170 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3171 const struct iovec *iov, loff_t offset,
3172 unsigned long nr_segs)
3174 struct file *file = iocb->ki_filp;
3175 struct inode *inode = file->f_mapping->host;
3177 size_t count = iov_length(iov, nr_segs);
3179 get_block_t *get_block_func = NULL;
3181 loff_t final_size = offset + count;
3183 /* Use the old path for reads and writes beyond i_size. */
3184 if (rw != WRITE || final_size > inode->i_size)
3185 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3187 BUG_ON(iocb->private == NULL);
3189 /* If we do a overwrite dio, i_mutex locking can be released */
3190 overwrite = *((int *)iocb->private);
3193 atomic_inc(&inode->i_dio_count);
3194 down_read(&EXT4_I(inode)->i_data_sem);
3195 mutex_unlock(&inode->i_mutex);
3199 * We could direct write to holes and fallocate.
3201 * Allocated blocks to fill the hole are marked as
3202 * uninitialized to prevent parallel buffered read to expose
3203 * the stale data before DIO complete the data IO.
3205 * As to previously fallocated extents, ext4 get_block will
3206 * just simply mark the buffer mapped but still keep the
3207 * extents uninitialized.
3209 * For non AIO case, we will convert those unwritten extents
3210 * to written after return back from blockdev_direct_IO.
3212 * For async DIO, the conversion needs to be deferred when the
3213 * IO is completed. The ext4 end_io callback function will be
3214 * called to take care of the conversion work. Here for async
3215 * case, we allocate an io_end structure to hook to the iocb.
3217 iocb->private = NULL;
3218 ext4_inode_aio_set(inode, NULL);
3219 if (!is_sync_kiocb(iocb)) {
3220 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3225 io_end->flag |= EXT4_IO_END_DIRECT;
3226 iocb->private = io_end;
3228 * we save the io structure for current async direct
3229 * IO, so that later ext4_map_blocks() could flag the
3230 * io structure whether there is a unwritten extents
3231 * needs to be converted when IO is completed.
3233 ext4_inode_aio_set(inode, io_end);
3237 get_block_func = ext4_get_block_write_nolock;
3239 get_block_func = ext4_get_block_write;
3240 dio_flags = DIO_LOCKING;
3242 ret = __blockdev_direct_IO(rw, iocb, inode,
3243 inode->i_sb->s_bdev, iov,
3251 ext4_inode_aio_set(inode, NULL);
3253 * The io_end structure takes a reference to the inode, that
3254 * structure needs to be destroyed and the reference to the
3255 * inode need to be dropped, when IO is complete, even with 0
3256 * byte write, or failed.
3258 * In the successful AIO DIO case, the io_end structure will
3259 * be destroyed and the reference to the inode will be dropped
3260 * after the end_io call back function is called.
3262 * In the case there is 0 byte write, or error case, since VFS
3263 * direct IO won't invoke the end_io call back function, we
3264 * need to free the end_io structure here.
3266 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3267 ext4_free_io_end(iocb->private);
3268 iocb->private = NULL;
3269 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3270 EXT4_STATE_DIO_UNWRITTEN)) {
3273 * for non AIO case, since the IO is already
3274 * completed, we could do the conversion right here
3276 err = ext4_convert_unwritten_extents(inode,
3280 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3284 /* take i_mutex locking again if we do a ovewrite dio */
3286 inode_dio_done(inode);
3287 up_read(&EXT4_I(inode)->i_data_sem);
3288 mutex_lock(&inode->i_mutex);
3294 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3295 const struct iovec *iov, loff_t offset,
3296 unsigned long nr_segs)
3298 struct file *file = iocb->ki_filp;
3299 struct inode *inode = file->f_mapping->host;
3303 * If we are doing data journalling we don't support O_DIRECT
3305 if (ext4_should_journal_data(inode))
3308 /* Let buffer I/O handle the inline data case. */
3309 if (ext4_has_inline_data(inode))
3312 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3313 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3314 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3316 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3317 trace_ext4_direct_IO_exit(inode, offset,
3318 iov_length(iov, nr_segs), rw, ret);
3323 * Pages can be marked dirty completely asynchronously from ext4's journalling
3324 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3325 * much here because ->set_page_dirty is called under VFS locks. The page is
3326 * not necessarily locked.
3328 * We cannot just dirty the page and leave attached buffers clean, because the
3329 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3330 * or jbddirty because all the journalling code will explode.
3332 * So what we do is to mark the page "pending dirty" and next time writepage
3333 * is called, propagate that into the buffers appropriately.
3335 static int ext4_journalled_set_page_dirty(struct page *page)
3337 SetPageChecked(page);
3338 return __set_page_dirty_nobuffers(page);
3341 static const struct address_space_operations ext4_ordered_aops = {
3342 .readpage = ext4_readpage,
3343 .readpages = ext4_readpages,
3344 .writepage = ext4_writepage,
3345 .write_begin = ext4_write_begin,
3346 .write_end = ext4_ordered_write_end,
3348 .invalidatepage = ext4_invalidatepage,
3349 .releasepage = ext4_releasepage,
3350 .direct_IO = ext4_direct_IO,
3351 .migratepage = buffer_migrate_page,
3352 .is_partially_uptodate = block_is_partially_uptodate,
3353 .error_remove_page = generic_error_remove_page,
3356 static const struct address_space_operations ext4_writeback_aops = {
3357 .readpage = ext4_readpage,
3358 .readpages = ext4_readpages,
3359 .writepage = ext4_writepage,
3360 .write_begin = ext4_write_begin,
3361 .write_end = ext4_writeback_write_end,
3363 .invalidatepage = ext4_invalidatepage,
3364 .releasepage = ext4_releasepage,
3365 .direct_IO = ext4_direct_IO,
3366 .migratepage = buffer_migrate_page,
3367 .is_partially_uptodate = block_is_partially_uptodate,
3368 .error_remove_page = generic_error_remove_page,
3371 static const struct address_space_operations ext4_journalled_aops = {
3372 .readpage = ext4_readpage,
3373 .readpages = ext4_readpages,
3374 .writepage = ext4_writepage,
3375 .write_begin = ext4_write_begin,
3376 .write_end = ext4_journalled_write_end,
3377 .set_page_dirty = ext4_journalled_set_page_dirty,
3379 .invalidatepage = ext4_journalled_invalidatepage,
3380 .releasepage = ext4_releasepage,
3381 .direct_IO = ext4_direct_IO,
3382 .is_partially_uptodate = block_is_partially_uptodate,
3383 .error_remove_page = generic_error_remove_page,
3386 static const struct address_space_operations ext4_da_aops = {
3387 .readpage = ext4_readpage,
3388 .readpages = ext4_readpages,
3389 .writepage = ext4_writepage,
3390 .writepages = ext4_da_writepages,
3391 .write_begin = ext4_da_write_begin,
3392 .write_end = ext4_da_write_end,
3394 .invalidatepage = ext4_da_invalidatepage,
3395 .releasepage = ext4_releasepage,
3396 .direct_IO = ext4_direct_IO,
3397 .migratepage = buffer_migrate_page,
3398 .is_partially_uptodate = block_is_partially_uptodate,
3399 .error_remove_page = generic_error_remove_page,
3402 void ext4_set_aops(struct inode *inode)
3404 switch (ext4_inode_journal_mode(inode)) {
3405 case EXT4_INODE_ORDERED_DATA_MODE:
3406 if (test_opt(inode->i_sb, DELALLOC))
3407 inode->i_mapping->a_ops = &ext4_da_aops;
3409 inode->i_mapping->a_ops = &ext4_ordered_aops;
3411 case EXT4_INODE_WRITEBACK_DATA_MODE:
3412 if (test_opt(inode->i_sb, DELALLOC))
3413 inode->i_mapping->a_ops = &ext4_da_aops;
3415 inode->i_mapping->a_ops = &ext4_writeback_aops;
3417 case EXT4_INODE_JOURNAL_DATA_MODE:
3418 inode->i_mapping->a_ops = &ext4_journalled_aops;
3427 * ext4_discard_partial_page_buffers()
3428 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3429 * This function finds and locks the page containing the offset
3430 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3431 * Calling functions that already have the page locked should call
3432 * ext4_discard_partial_page_buffers_no_lock directly.
3434 int ext4_discard_partial_page_buffers(handle_t *handle,
3435 struct address_space *mapping, loff_t from,
3436 loff_t length, int flags)
3438 struct inode *inode = mapping->host;
3442 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3443 mapping_gfp_mask(mapping) & ~__GFP_FS);
3447 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3448 from, length, flags);
3451 page_cache_release(page);
3456 * ext4_discard_partial_page_buffers_no_lock()
3457 * Zeros a page range of length 'length' starting from offset 'from'.
3458 * Buffer heads that correspond to the block aligned regions of the
3459 * zeroed range will be unmapped. Unblock aligned regions
3460 * will have the corresponding buffer head mapped if needed so that
3461 * that region of the page can be updated with the partial zero out.
3463 * This function assumes that the page has already been locked. The
3464 * The range to be discarded must be contained with in the given page.
3465 * If the specified range exceeds the end of the page it will be shortened
3466 * to the end of the page that corresponds to 'from'. This function is
3467 * appropriate for updating a page and it buffer heads to be unmapped and
3468 * zeroed for blocks that have been either released, or are going to be
3471 * handle: The journal handle
3472 * inode: The files inode
3473 * page: A locked page that contains the offset "from"
3474 * from: The starting byte offset (from the beginning of the file)
3475 * to begin discarding
3476 * len: The length of bytes to discard
3477 * flags: Optional flags that may be used:
3479 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3480 * Only zero the regions of the page whose buffer heads
3481 * have already been unmapped. This flag is appropriate
3482 * for updating the contents of a page whose blocks may
3483 * have already been released, and we only want to zero
3484 * out the regions that correspond to those released blocks.
3486 * Returns zero on success or negative on failure.
3488 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3489 struct inode *inode, struct page *page, loff_t from,
3490 loff_t length, int flags)
3492 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3493 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3494 unsigned int blocksize, max, pos;
3496 struct buffer_head *bh;
3499 blocksize = inode->i_sb->s_blocksize;
3500 max = PAGE_CACHE_SIZE - offset;
3502 if (index != page->index)
3506 * correct length if it does not fall between
3507 * 'from' and the end of the page
3509 if (length > max || length < 0)
3512 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3514 if (!page_has_buffers(page))
3515 create_empty_buffers(page, blocksize, 0);
3517 /* Find the buffer that contains "offset" */
3518 bh = page_buffers(page);
3520 while (offset >= pos) {
3521 bh = bh->b_this_page;
3527 while (pos < offset + length) {
3528 unsigned int end_of_block, range_to_discard;
3532 /* The length of space left to zero and unmap */
3533 range_to_discard = offset + length - pos;
3535 /* The length of space until the end of the block */
3536 end_of_block = blocksize - (pos & (blocksize-1));
3539 * Do not unmap or zero past end of block
3540 * for this buffer head
3542 if (range_to_discard > end_of_block)
3543 range_to_discard = end_of_block;
3547 * Skip this buffer head if we are only zeroing unampped
3548 * regions of the page
3550 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3554 /* If the range is block aligned, unmap */
3555 if (range_to_discard == blocksize) {
3556 clear_buffer_dirty(bh);
3558 clear_buffer_mapped(bh);
3559 clear_buffer_req(bh);
3560 clear_buffer_new(bh);
3561 clear_buffer_delay(bh);
3562 clear_buffer_unwritten(bh);
3563 clear_buffer_uptodate(bh);
3564 zero_user(page, pos, range_to_discard);
3565 BUFFER_TRACE(bh, "Buffer discarded");
3570 * If this block is not completely contained in the range
3571 * to be discarded, then it is not going to be released. Because
3572 * we need to keep this block, we need to make sure this part
3573 * of the page is uptodate before we modify it by writeing
3574 * partial zeros on it.
3576 if (!buffer_mapped(bh)) {
3578 * Buffer head must be mapped before we can read
3581 BUFFER_TRACE(bh, "unmapped");
3582 ext4_get_block(inode, iblock, bh, 0);
3583 /* unmapped? It's a hole - nothing to do */
3584 if (!buffer_mapped(bh)) {
3585 BUFFER_TRACE(bh, "still unmapped");
3590 /* Ok, it's mapped. Make sure it's up-to-date */
3591 if (PageUptodate(page))
3592 set_buffer_uptodate(bh);
3594 if (!buffer_uptodate(bh)) {
3596 ll_rw_block(READ, 1, &bh);
3598 /* Uhhuh. Read error. Complain and punt.*/
3599 if (!buffer_uptodate(bh))
3603 if (ext4_should_journal_data(inode)) {
3604 BUFFER_TRACE(bh, "get write access");
3605 err = ext4_journal_get_write_access(handle, bh);
3610 zero_user(page, pos, range_to_discard);
3613 if (ext4_should_journal_data(inode)) {
3614 err = ext4_handle_dirty_metadata(handle, inode, bh);
3616 mark_buffer_dirty(bh);
3618 BUFFER_TRACE(bh, "Partial buffer zeroed");
3620 bh = bh->b_this_page;
3622 pos += range_to_discard;
3628 int ext4_can_truncate(struct inode *inode)
3630 if (S_ISREG(inode->i_mode))
3632 if (S_ISDIR(inode->i_mode))
3634 if (S_ISLNK(inode->i_mode))
3635 return !ext4_inode_is_fast_symlink(inode);
3640 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3641 * associated with the given offset and length
3643 * @inode: File inode
3644 * @offset: The offset where the hole will begin
3645 * @len: The length of the hole
3647 * Returns: 0 on success or negative on failure
3650 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3652 struct inode *inode = file->f_path.dentry->d_inode;
3653 if (!S_ISREG(inode->i_mode))
3656 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3657 return ext4_ind_punch_hole(file, offset, length);
3659 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3660 /* TODO: Add support for bigalloc file systems */
3664 trace_ext4_punch_hole(inode, offset, length);
3666 return ext4_ext_punch_hole(file, offset, length);
3672 * We block out ext4_get_block() block instantiations across the entire
3673 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3674 * simultaneously on behalf of the same inode.
3676 * As we work through the truncate and commit bits of it to the journal there
3677 * is one core, guiding principle: the file's tree must always be consistent on
3678 * disk. We must be able to restart the truncate after a crash.
3680 * The file's tree may be transiently inconsistent in memory (although it
3681 * probably isn't), but whenever we close off and commit a journal transaction,
3682 * the contents of (the filesystem + the journal) must be consistent and
3683 * restartable. It's pretty simple, really: bottom up, right to left (although
3684 * left-to-right works OK too).
3686 * Note that at recovery time, journal replay occurs *before* the restart of
3687 * truncate against the orphan inode list.
3689 * The committed inode has the new, desired i_size (which is the same as
3690 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3691 * that this inode's truncate did not complete and it will again call
3692 * ext4_truncate() to have another go. So there will be instantiated blocks
3693 * to the right of the truncation point in a crashed ext4 filesystem. But
3694 * that's fine - as long as they are linked from the inode, the post-crash
3695 * ext4_truncate() run will find them and release them.
3697 void ext4_truncate(struct inode *inode)
3699 trace_ext4_truncate_enter(inode);
3701 if (!ext4_can_truncate(inode))
3704 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3706 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3707 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3709 if (ext4_has_inline_data(inode)) {
3712 ext4_inline_data_truncate(inode, &has_inline);
3717 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3718 ext4_ext_truncate(inode);
3720 ext4_ind_truncate(inode);
3722 trace_ext4_truncate_exit(inode);
3726 * ext4_get_inode_loc returns with an extra refcount against the inode's
3727 * underlying buffer_head on success. If 'in_mem' is true, we have all
3728 * data in memory that is needed to recreate the on-disk version of this
3731 static int __ext4_get_inode_loc(struct inode *inode,
3732 struct ext4_iloc *iloc, int in_mem)
3734 struct ext4_group_desc *gdp;
3735 struct buffer_head *bh;
3736 struct super_block *sb = inode->i_sb;
3738 int inodes_per_block, inode_offset;
3741 if (!ext4_valid_inum(sb, inode->i_ino))
3744 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3745 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3750 * Figure out the offset within the block group inode table
3752 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3753 inode_offset = ((inode->i_ino - 1) %
3754 EXT4_INODES_PER_GROUP(sb));
3755 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3756 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3758 bh = sb_getblk(sb, block);
3761 if (!buffer_uptodate(bh)) {
3765 * If the buffer has the write error flag, we have failed
3766 * to write out another inode in the same block. In this
3767 * case, we don't have to read the block because we may
3768 * read the old inode data successfully.
3770 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3771 set_buffer_uptodate(bh);
3773 if (buffer_uptodate(bh)) {
3774 /* someone brought it uptodate while we waited */
3780 * If we have all information of the inode in memory and this
3781 * is the only valid inode in the block, we need not read the
3785 struct buffer_head *bitmap_bh;
3788 start = inode_offset & ~(inodes_per_block - 1);
3790 /* Is the inode bitmap in cache? */
3791 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3792 if (unlikely(!bitmap_bh))
3796 * If the inode bitmap isn't in cache then the
3797 * optimisation may end up performing two reads instead
3798 * of one, so skip it.
3800 if (!buffer_uptodate(bitmap_bh)) {
3804 for (i = start; i < start + inodes_per_block; i++) {
3805 if (i == inode_offset)
3807 if (ext4_test_bit(i, bitmap_bh->b_data))
3811 if (i == start + inodes_per_block) {
3812 /* all other inodes are free, so skip I/O */
3813 memset(bh->b_data, 0, bh->b_size);
3814 set_buffer_uptodate(bh);
3822 * If we need to do any I/O, try to pre-readahead extra
3823 * blocks from the inode table.
3825 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3826 ext4_fsblk_t b, end, table;
3829 table = ext4_inode_table(sb, gdp);
3830 /* s_inode_readahead_blks is always a power of 2 */
3831 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3834 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3835 num = EXT4_INODES_PER_GROUP(sb);
3836 if (ext4_has_group_desc_csum(sb))
3837 num -= ext4_itable_unused_count(sb, gdp);
3838 table += num / inodes_per_block;
3842 sb_breadahead(sb, b++);
3846 * There are other valid inodes in the buffer, this inode
3847 * has in-inode xattrs, or we don't have this inode in memory.
3848 * Read the block from disk.
3850 trace_ext4_load_inode(inode);
3852 bh->b_end_io = end_buffer_read_sync;
3853 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3855 if (!buffer_uptodate(bh)) {
3856 EXT4_ERROR_INODE_BLOCK(inode, block,
3857 "unable to read itable block");
3867 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3869 /* We have all inode data except xattrs in memory here. */
3870 return __ext4_get_inode_loc(inode, iloc,
3871 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3874 void ext4_set_inode_flags(struct inode *inode)
3876 unsigned int flags = EXT4_I(inode)->i_flags;
3878 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3879 if (flags & EXT4_SYNC_FL)
3880 inode->i_flags |= S_SYNC;
3881 if (flags & EXT4_APPEND_FL)
3882 inode->i_flags |= S_APPEND;
3883 if (flags & EXT4_IMMUTABLE_FL)
3884 inode->i_flags |= S_IMMUTABLE;
3885 if (flags & EXT4_NOATIME_FL)
3886 inode->i_flags |= S_NOATIME;
3887 if (flags & EXT4_DIRSYNC_FL)
3888 inode->i_flags |= S_DIRSYNC;
3891 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3892 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3894 unsigned int vfs_fl;
3895 unsigned long old_fl, new_fl;
3898 vfs_fl = ei->vfs_inode.i_flags;
3899 old_fl = ei->i_flags;
3900 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3901 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3903 if (vfs_fl & S_SYNC)
3904 new_fl |= EXT4_SYNC_FL;
3905 if (vfs_fl & S_APPEND)
3906 new_fl |= EXT4_APPEND_FL;
3907 if (vfs_fl & S_IMMUTABLE)
3908 new_fl |= EXT4_IMMUTABLE_FL;
3909 if (vfs_fl & S_NOATIME)
3910 new_fl |= EXT4_NOATIME_FL;
3911 if (vfs_fl & S_DIRSYNC)
3912 new_fl |= EXT4_DIRSYNC_FL;
3913 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3916 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3917 struct ext4_inode_info *ei)
3920 struct inode *inode = &(ei->vfs_inode);
3921 struct super_block *sb = inode->i_sb;
3923 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3924 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3925 /* we are using combined 48 bit field */
3926 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3927 le32_to_cpu(raw_inode->i_blocks_lo);
3928 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3929 /* i_blocks represent file system block size */
3930 return i_blocks << (inode->i_blkbits - 9);
3935 return le32_to_cpu(raw_inode->i_blocks_lo);
3939 static inline void ext4_iget_extra_inode(struct inode *inode,
3940 struct ext4_inode *raw_inode,
3941 struct ext4_inode_info *ei)
3943 __le32 *magic = (void *)raw_inode +
3944 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
3945 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
3946 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3947 ext4_find_inline_data_nolock(inode);
3949 EXT4_I(inode)->i_inline_off = 0;
3952 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3954 struct ext4_iloc iloc;
3955 struct ext4_inode *raw_inode;
3956 struct ext4_inode_info *ei;
3957 struct inode *inode;
3958 journal_t *journal = EXT4_SB(sb)->s_journal;
3964 inode = iget_locked(sb, ino);
3966 return ERR_PTR(-ENOMEM);
3967 if (!(inode->i_state & I_NEW))
3973 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3976 raw_inode = ext4_raw_inode(&iloc);
3978 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3979 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3980 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3981 EXT4_INODE_SIZE(inode->i_sb)) {
3982 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3983 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3984 EXT4_INODE_SIZE(inode->i_sb));
3989 ei->i_extra_isize = 0;
3991 /* Precompute checksum seed for inode metadata */
3992 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3993 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3994 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3996 __le32 inum = cpu_to_le32(inode->i_ino);
3997 __le32 gen = raw_inode->i_generation;
3998 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4000 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4004 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4005 EXT4_ERROR_INODE(inode, "checksum invalid");
4010 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4011 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4012 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4013 if (!(test_opt(inode->i_sb, NO_UID32))) {
4014 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4015 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4017 i_uid_write(inode, i_uid);
4018 i_gid_write(inode, i_gid);
4019 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4021 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4022 ei->i_inline_off = 0;
4023 ei->i_dir_start_lookup = 0;
4024 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4025 /* We now have enough fields to check if the inode was active or not.
4026 * This is needed because nfsd might try to access dead inodes
4027 * the test is that same one that e2fsck uses
4028 * NeilBrown 1999oct15
4030 if (inode->i_nlink == 0) {
4031 if (inode->i_mode == 0 ||
4032 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4033 /* this inode is deleted */
4037 /* The only unlinked inodes we let through here have
4038 * valid i_mode and are being read by the orphan
4039 * recovery code: that's fine, we're about to complete
4040 * the process of deleting those. */
4042 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4043 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4044 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4045 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4047 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4048 inode->i_size = ext4_isize(raw_inode);
4049 ei->i_disksize = inode->i_size;
4051 ei->i_reserved_quota = 0;
4053 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4054 ei->i_block_group = iloc.block_group;
4055 ei->i_last_alloc_group = ~0;
4057 * NOTE! The in-memory inode i_data array is in little-endian order
4058 * even on big-endian machines: we do NOT byteswap the block numbers!
4060 for (block = 0; block < EXT4_N_BLOCKS; block++)
4061 ei->i_data[block] = raw_inode->i_block[block];
4062 INIT_LIST_HEAD(&ei->i_orphan);
4065 * Set transaction id's of transactions that have to be committed
4066 * to finish f[data]sync. We set them to currently running transaction
4067 * as we cannot be sure that the inode or some of its metadata isn't
4068 * part of the transaction - the inode could have been reclaimed and
4069 * now it is reread from disk.
4072 transaction_t *transaction;
4075 read_lock(&journal->j_state_lock);
4076 if (journal->j_running_transaction)
4077 transaction = journal->j_running_transaction;
4079 transaction = journal->j_committing_transaction;
4081 tid = transaction->t_tid;
4083 tid = journal->j_commit_sequence;
4084 read_unlock(&journal->j_state_lock);
4085 ei->i_sync_tid = tid;
4086 ei->i_datasync_tid = tid;
4089 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4090 if (ei->i_extra_isize == 0) {
4091 /* The extra space is currently unused. Use it. */
4092 ei->i_extra_isize = sizeof(struct ext4_inode) -
4093 EXT4_GOOD_OLD_INODE_SIZE;
4095 ext4_iget_extra_inode(inode, raw_inode, ei);
4099 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4100 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4101 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4102 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4104 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4105 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4106 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4108 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4112 if (ei->i_file_acl &&
4113 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4114 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4118 } else if (!ext4_has_inline_data(inode)) {
4119 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4120 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4121 (S_ISLNK(inode->i_mode) &&
4122 !ext4_inode_is_fast_symlink(inode))))
4123 /* Validate extent which is part of inode */
4124 ret = ext4_ext_check_inode(inode);
4125 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4126 (S_ISLNK(inode->i_mode) &&
4127 !ext4_inode_is_fast_symlink(inode))) {
4128 /* Validate block references which are part of inode */
4129 ret = ext4_ind_check_inode(inode);
4135 if (S_ISREG(inode->i_mode)) {
4136 inode->i_op = &ext4_file_inode_operations;
4137 inode->i_fop = &ext4_file_operations;
4138 ext4_set_aops(inode);
4139 } else if (S_ISDIR(inode->i_mode)) {
4140 inode->i_op = &ext4_dir_inode_operations;
4141 inode->i_fop = &ext4_dir_operations;
4142 } else if (S_ISLNK(inode->i_mode)) {
4143 if (ext4_inode_is_fast_symlink(inode)) {
4144 inode->i_op = &ext4_fast_symlink_inode_operations;
4145 nd_terminate_link(ei->i_data, inode->i_size,
4146 sizeof(ei->i_data) - 1);
4148 inode->i_op = &ext4_symlink_inode_operations;
4149 ext4_set_aops(inode);
4151 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4152 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4153 inode->i_op = &ext4_special_inode_operations;
4154 if (raw_inode->i_block[0])
4155 init_special_inode(inode, inode->i_mode,
4156 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4158 init_special_inode(inode, inode->i_mode,
4159 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4162 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4166 ext4_set_inode_flags(inode);
4167 unlock_new_inode(inode);
4173 return ERR_PTR(ret);
4176 static int ext4_inode_blocks_set(handle_t *handle,
4177 struct ext4_inode *raw_inode,
4178 struct ext4_inode_info *ei)
4180 struct inode *inode = &(ei->vfs_inode);
4181 u64 i_blocks = inode->i_blocks;
4182 struct super_block *sb = inode->i_sb;
4184 if (i_blocks <= ~0U) {
4186 * i_blocks can be represented in a 32 bit variable
4187 * as multiple of 512 bytes
4189 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4190 raw_inode->i_blocks_high = 0;
4191 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4194 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4197 if (i_blocks <= 0xffffffffffffULL) {
4199 * i_blocks can be represented in a 48 bit variable
4200 * as multiple of 512 bytes
4202 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4203 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4204 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4206 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4207 /* i_block is stored in file system block size */
4208 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4209 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4210 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4216 * Post the struct inode info into an on-disk inode location in the
4217 * buffer-cache. This gobbles the caller's reference to the
4218 * buffer_head in the inode location struct.
4220 * The caller must have write access to iloc->bh.
4222 static int ext4_do_update_inode(handle_t *handle,
4223 struct inode *inode,
4224 struct ext4_iloc *iloc)
4226 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4227 struct ext4_inode_info *ei = EXT4_I(inode);
4228 struct buffer_head *bh = iloc->bh;
4229 int err = 0, rc, block;
4230 int need_datasync = 0;
4234 /* For fields not not tracking in the in-memory inode,
4235 * initialise them to zero for new inodes. */
4236 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4237 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4239 ext4_get_inode_flags(ei);
4240 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4241 i_uid = i_uid_read(inode);
4242 i_gid = i_gid_read(inode);
4243 if (!(test_opt(inode->i_sb, NO_UID32))) {
4244 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4245 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4247 * Fix up interoperability with old kernels. Otherwise, old inodes get
4248 * re-used with the upper 16 bits of the uid/gid intact
4251 raw_inode->i_uid_high =
4252 cpu_to_le16(high_16_bits(i_uid));
4253 raw_inode->i_gid_high =
4254 cpu_to_le16(high_16_bits(i_gid));
4256 raw_inode->i_uid_high = 0;
4257 raw_inode->i_gid_high = 0;
4260 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4261 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4262 raw_inode->i_uid_high = 0;
4263 raw_inode->i_gid_high = 0;
4265 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4267 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4268 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4269 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4270 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4272 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4274 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4275 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4276 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4277 cpu_to_le32(EXT4_OS_HURD))
4278 raw_inode->i_file_acl_high =
4279 cpu_to_le16(ei->i_file_acl >> 32);
4280 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4281 if (ei->i_disksize != ext4_isize(raw_inode)) {
4282 ext4_isize_set(raw_inode, ei->i_disksize);
4285 if (ei->i_disksize > 0x7fffffffULL) {
4286 struct super_block *sb = inode->i_sb;
4287 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4288 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4289 EXT4_SB(sb)->s_es->s_rev_level ==
4290 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4291 /* If this is the first large file
4292 * created, add a flag to the superblock.
4294 err = ext4_journal_get_write_access(handle,
4295 EXT4_SB(sb)->s_sbh);
4298 ext4_update_dynamic_rev(sb);
4299 EXT4_SET_RO_COMPAT_FEATURE(sb,
4300 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4301 ext4_handle_sync(handle);
4302 err = ext4_handle_dirty_super(handle, sb);
4305 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4306 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4307 if (old_valid_dev(inode->i_rdev)) {
4308 raw_inode->i_block[0] =
4309 cpu_to_le32(old_encode_dev(inode->i_rdev));
4310 raw_inode->i_block[1] = 0;
4312 raw_inode->i_block[0] = 0;
4313 raw_inode->i_block[1] =
4314 cpu_to_le32(new_encode_dev(inode->i_rdev));
4315 raw_inode->i_block[2] = 0;
4317 } else if (!ext4_has_inline_data(inode)) {
4318 for (block = 0; block < EXT4_N_BLOCKS; block++)
4319 raw_inode->i_block[block] = ei->i_data[block];
4322 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4323 if (ei->i_extra_isize) {
4324 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4325 raw_inode->i_version_hi =
4326 cpu_to_le32(inode->i_version >> 32);
4327 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4330 ext4_inode_csum_set(inode, raw_inode, ei);
4332 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4333 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4336 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4338 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4341 ext4_std_error(inode->i_sb, err);
4346 * ext4_write_inode()
4348 * We are called from a few places:
4350 * - Within generic_file_write() for O_SYNC files.
4351 * Here, there will be no transaction running. We wait for any running
4352 * transaction to commit.
4354 * - Within sys_sync(), kupdate and such.
4355 * We wait on commit, if tol to.
4357 * - Within prune_icache() (PF_MEMALLOC == true)
4358 * Here we simply return. We can't afford to block kswapd on the
4361 * In all cases it is actually safe for us to return without doing anything,
4362 * because the inode has been copied into a raw inode buffer in
4363 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4366 * Note that we are absolutely dependent upon all inode dirtiers doing the
4367 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4368 * which we are interested.
4370 * It would be a bug for them to not do this. The code:
4372 * mark_inode_dirty(inode)
4374 * inode->i_size = expr;
4376 * is in error because a kswapd-driven write_inode() could occur while
4377 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4378 * will no longer be on the superblock's dirty inode list.
4380 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4384 if (current->flags & PF_MEMALLOC)
4387 if (EXT4_SB(inode->i_sb)->s_journal) {
4388 if (ext4_journal_current_handle()) {
4389 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4394 if (wbc->sync_mode != WB_SYNC_ALL)
4397 err = ext4_force_commit(inode->i_sb);
4399 struct ext4_iloc iloc;
4401 err = __ext4_get_inode_loc(inode, &iloc, 0);
4404 if (wbc->sync_mode == WB_SYNC_ALL)
4405 sync_dirty_buffer(iloc.bh);
4406 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4407 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4408 "IO error syncing inode");
4417 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4418 * buffers that are attached to a page stradding i_size and are undergoing
4419 * commit. In that case we have to wait for commit to finish and try again.
4421 static void ext4_wait_for_tail_page_commit(struct inode *inode)
4425 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4426 tid_t commit_tid = 0;
4429 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4431 * All buffers in the last page remain valid? Then there's nothing to
4432 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4435 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4438 page = find_lock_page(inode->i_mapping,
4439 inode->i_size >> PAGE_CACHE_SHIFT);
4442 ret = __ext4_journalled_invalidatepage(page, offset);
4444 page_cache_release(page);
4448 read_lock(&journal->j_state_lock);
4449 if (journal->j_committing_transaction)
4450 commit_tid = journal->j_committing_transaction->t_tid;
4451 read_unlock(&journal->j_state_lock);
4453 jbd2_log_wait_commit(journal, commit_tid);
4460 * Called from notify_change.
4462 * We want to trap VFS attempts to truncate the file as soon as
4463 * possible. In particular, we want to make sure that when the VFS
4464 * shrinks i_size, we put the inode on the orphan list and modify
4465 * i_disksize immediately, so that during the subsequent flushing of
4466 * dirty pages and freeing of disk blocks, we can guarantee that any
4467 * commit will leave the blocks being flushed in an unused state on
4468 * disk. (On recovery, the inode will get truncated and the blocks will
4469 * be freed, so we have a strong guarantee that no future commit will
4470 * leave these blocks visible to the user.)
4472 * Another thing we have to assure is that if we are in ordered mode
4473 * and inode is still attached to the committing transaction, we must
4474 * we start writeout of all the dirty pages which are being truncated.
4475 * This way we are sure that all the data written in the previous
4476 * transaction are already on disk (truncate waits for pages under
4479 * Called with inode->i_mutex down.
4481 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4483 struct inode *inode = dentry->d_inode;
4486 const unsigned int ia_valid = attr->ia_valid;
4488 error = inode_change_ok(inode, attr);
4492 if (is_quota_modification(inode, attr))
4493 dquot_initialize(inode);
4494 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4495 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4498 /* (user+group)*(old+new) structure, inode write (sb,
4499 * inode block, ? - but truncate inode update has it) */
4500 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4501 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4502 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
4503 if (IS_ERR(handle)) {
4504 error = PTR_ERR(handle);
4507 error = dquot_transfer(inode, attr);
4509 ext4_journal_stop(handle);
4512 /* Update corresponding info in inode so that everything is in
4513 * one transaction */
4514 if (attr->ia_valid & ATTR_UID)
4515 inode->i_uid = attr->ia_uid;
4516 if (attr->ia_valid & ATTR_GID)
4517 inode->i_gid = attr->ia_gid;
4518 error = ext4_mark_inode_dirty(handle, inode);
4519 ext4_journal_stop(handle);
4522 if (attr->ia_valid & ATTR_SIZE) {
4524 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4525 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4527 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4532 if (S_ISREG(inode->i_mode) &&
4533 attr->ia_valid & ATTR_SIZE &&
4534 (attr->ia_size < inode->i_size)) {
4537 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4538 if (IS_ERR(handle)) {
4539 error = PTR_ERR(handle);
4542 if (ext4_handle_valid(handle)) {
4543 error = ext4_orphan_add(handle, inode);
4546 EXT4_I(inode)->i_disksize = attr->ia_size;
4547 rc = ext4_mark_inode_dirty(handle, inode);
4550 ext4_journal_stop(handle);
4552 if (ext4_should_order_data(inode)) {
4553 error = ext4_begin_ordered_truncate(inode,
4556 /* Do as much error cleanup as possible */
4557 handle = ext4_journal_start(inode,
4559 if (IS_ERR(handle)) {
4560 ext4_orphan_del(NULL, inode);
4563 ext4_orphan_del(handle, inode);
4565 ext4_journal_stop(handle);
4571 if (attr->ia_valid & ATTR_SIZE) {
4572 if (attr->ia_size != inode->i_size) {
4573 loff_t oldsize = inode->i_size;
4575 i_size_write(inode, attr->ia_size);
4577 * Blocks are going to be removed from the inode. Wait
4578 * for dio in flight. Temporarily disable
4579 * dioread_nolock to prevent livelock.
4582 if (!ext4_should_journal_data(inode)) {
4583 ext4_inode_block_unlocked_dio(inode);
4584 inode_dio_wait(inode);
4585 ext4_inode_resume_unlocked_dio(inode);
4587 ext4_wait_for_tail_page_commit(inode);
4590 * Truncate pagecache after we've waited for commit
4591 * in data=journal mode to make pages freeable.
4593 truncate_pagecache(inode, oldsize, inode->i_size);
4595 ext4_truncate(inode);
4599 setattr_copy(inode, attr);
4600 mark_inode_dirty(inode);
4604 * If the call to ext4_truncate failed to get a transaction handle at
4605 * all, we need to clean up the in-core orphan list manually.
4607 if (orphan && inode->i_nlink)
4608 ext4_orphan_del(NULL, inode);
4610 if (!rc && (ia_valid & ATTR_MODE))
4611 rc = ext4_acl_chmod(inode);
4614 ext4_std_error(inode->i_sb, error);
4620 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4623 struct inode *inode;
4624 unsigned long delalloc_blocks;
4626 inode = dentry->d_inode;
4627 generic_fillattr(inode, stat);
4630 * We can't update i_blocks if the block allocation is delayed
4631 * otherwise in the case of system crash before the real block
4632 * allocation is done, we will have i_blocks inconsistent with
4633 * on-disk file blocks.
4634 * We always keep i_blocks updated together with real
4635 * allocation. But to not confuse with user, stat
4636 * will return the blocks that include the delayed allocation
4637 * blocks for this file.
4639 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4640 EXT4_I(inode)->i_reserved_data_blocks);
4642 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4646 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4648 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4649 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4650 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4654 * Account for index blocks, block groups bitmaps and block group
4655 * descriptor blocks if modify datablocks and index blocks
4656 * worse case, the indexs blocks spread over different block groups
4658 * If datablocks are discontiguous, they are possible to spread over
4659 * different block groups too. If they are contiguous, with flexbg,
4660 * they could still across block group boundary.
4662 * Also account for superblock, inode, quota and xattr blocks
4664 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4666 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4672 * How many index blocks need to touch to modify nrblocks?
4673 * The "Chunk" flag indicating whether the nrblocks is
4674 * physically contiguous on disk
4676 * For Direct IO and fallocate, they calls get_block to allocate
4677 * one single extent at a time, so they could set the "Chunk" flag
4679 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4684 * Now let's see how many group bitmaps and group descriptors need
4694 if (groups > ngroups)
4696 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4697 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4699 /* bitmaps and block group descriptor blocks */
4700 ret += groups + gdpblocks;
4702 /* Blocks for super block, inode, quota and xattr blocks */
4703 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4709 * Calculate the total number of credits to reserve to fit
4710 * the modification of a single pages into a single transaction,
4711 * which may include multiple chunks of block allocations.
4713 * This could be called via ext4_write_begin()
4715 * We need to consider the worse case, when
4716 * one new block per extent.
4718 int ext4_writepage_trans_blocks(struct inode *inode)
4720 int bpp = ext4_journal_blocks_per_page(inode);
4723 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4725 /* Account for data blocks for journalled mode */
4726 if (ext4_should_journal_data(inode))
4732 * Calculate the journal credits for a chunk of data modification.
4734 * This is called from DIO, fallocate or whoever calling
4735 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4737 * journal buffers for data blocks are not included here, as DIO
4738 * and fallocate do no need to journal data buffers.
4740 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4742 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4746 * The caller must have previously called ext4_reserve_inode_write().
4747 * Give this, we know that the caller already has write access to iloc->bh.
4749 int ext4_mark_iloc_dirty(handle_t *handle,
4750 struct inode *inode, struct ext4_iloc *iloc)
4754 if (IS_I_VERSION(inode))
4755 inode_inc_iversion(inode);
4757 /* the do_update_inode consumes one bh->b_count */
4760 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4761 err = ext4_do_update_inode(handle, inode, iloc);
4767 * On success, We end up with an outstanding reference count against
4768 * iloc->bh. This _must_ be cleaned up later.
4772 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4773 struct ext4_iloc *iloc)
4777 err = ext4_get_inode_loc(inode, iloc);
4779 BUFFER_TRACE(iloc->bh, "get_write_access");
4780 err = ext4_journal_get_write_access(handle, iloc->bh);
4786 ext4_std_error(inode->i_sb, err);
4791 * Expand an inode by new_extra_isize bytes.
4792 * Returns 0 on success or negative error number on failure.
4794 static int ext4_expand_extra_isize(struct inode *inode,
4795 unsigned int new_extra_isize,
4796 struct ext4_iloc iloc,
4799 struct ext4_inode *raw_inode;
4800 struct ext4_xattr_ibody_header *header;
4802 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4805 raw_inode = ext4_raw_inode(&iloc);
4807 header = IHDR(inode, raw_inode);
4809 /* No extended attributes present */
4810 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4811 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4812 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4814 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4818 /* try to expand with EAs present */
4819 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4824 * What we do here is to mark the in-core inode as clean with respect to inode
4825 * dirtiness (it may still be data-dirty).
4826 * This means that the in-core inode may be reaped by prune_icache
4827 * without having to perform any I/O. This is a very good thing,
4828 * because *any* task may call prune_icache - even ones which
4829 * have a transaction open against a different journal.
4831 * Is this cheating? Not really. Sure, we haven't written the
4832 * inode out, but prune_icache isn't a user-visible syncing function.
4833 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4834 * we start and wait on commits.
4836 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4838 struct ext4_iloc iloc;
4839 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4840 static unsigned int mnt_count;
4844 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4845 err = ext4_reserve_inode_write(handle, inode, &iloc);
4846 if (ext4_handle_valid(handle) &&
4847 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4848 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4850 * We need extra buffer credits since we may write into EA block
4851 * with this same handle. If journal_extend fails, then it will
4852 * only result in a minor loss of functionality for that inode.
4853 * If this is felt to be critical, then e2fsck should be run to
4854 * force a large enough s_min_extra_isize.
4856 if ((jbd2_journal_extend(handle,
4857 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4858 ret = ext4_expand_extra_isize(inode,
4859 sbi->s_want_extra_isize,
4862 ext4_set_inode_state(inode,
4863 EXT4_STATE_NO_EXPAND);
4865 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4866 ext4_warning(inode->i_sb,
4867 "Unable to expand inode %lu. Delete"
4868 " some EAs or run e2fsck.",
4871 le16_to_cpu(sbi->s_es->s_mnt_count);
4877 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4882 * ext4_dirty_inode() is called from __mark_inode_dirty()
4884 * We're really interested in the case where a file is being extended.
4885 * i_size has been changed by generic_commit_write() and we thus need
4886 * to include the updated inode in the current transaction.
4888 * Also, dquot_alloc_block() will always dirty the inode when blocks
4889 * are allocated to the file.
4891 * If the inode is marked synchronous, we don't honour that here - doing
4892 * so would cause a commit on atime updates, which we don't bother doing.
4893 * We handle synchronous inodes at the highest possible level.
4895 void ext4_dirty_inode(struct inode *inode, int flags)
4899 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4903 ext4_mark_inode_dirty(handle, inode);
4905 ext4_journal_stop(handle);
4912 * Bind an inode's backing buffer_head into this transaction, to prevent
4913 * it from being flushed to disk early. Unlike
4914 * ext4_reserve_inode_write, this leaves behind no bh reference and
4915 * returns no iloc structure, so the caller needs to repeat the iloc
4916 * lookup to mark the inode dirty later.
4918 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4920 struct ext4_iloc iloc;
4924 err = ext4_get_inode_loc(inode, &iloc);
4926 BUFFER_TRACE(iloc.bh, "get_write_access");
4927 err = jbd2_journal_get_write_access(handle, iloc.bh);
4929 err = ext4_handle_dirty_metadata(handle,
4935 ext4_std_error(inode->i_sb, err);
4940 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4947 * We have to be very careful here: changing a data block's
4948 * journaling status dynamically is dangerous. If we write a
4949 * data block to the journal, change the status and then delete
4950 * that block, we risk forgetting to revoke the old log record
4951 * from the journal and so a subsequent replay can corrupt data.
4952 * So, first we make sure that the journal is empty and that
4953 * nobody is changing anything.
4956 journal = EXT4_JOURNAL(inode);
4959 if (is_journal_aborted(journal))
4961 /* We have to allocate physical blocks for delalloc blocks
4962 * before flushing journal. otherwise delalloc blocks can not
4963 * be allocated any more. even more truncate on delalloc blocks
4964 * could trigger BUG by flushing delalloc blocks in journal.
4965 * There is no delalloc block in non-journal data mode.
4967 if (val && test_opt(inode->i_sb, DELALLOC)) {
4968 err = ext4_alloc_da_blocks(inode);
4973 /* Wait for all existing dio workers */
4974 ext4_inode_block_unlocked_dio(inode);
4975 inode_dio_wait(inode);
4977 jbd2_journal_lock_updates(journal);
4980 * OK, there are no updates running now, and all cached data is
4981 * synced to disk. We are now in a completely consistent state
4982 * which doesn't have anything in the journal, and we know that
4983 * no filesystem updates are running, so it is safe to modify
4984 * the inode's in-core data-journaling state flag now.
4988 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4990 jbd2_journal_flush(journal);
4991 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4993 ext4_set_aops(inode);
4995 jbd2_journal_unlock_updates(journal);
4996 ext4_inode_resume_unlocked_dio(inode);
4998 /* Finally we can mark the inode as dirty. */
5000 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5002 return PTR_ERR(handle);
5004 err = ext4_mark_inode_dirty(handle, inode);
5005 ext4_handle_sync(handle);
5006 ext4_journal_stop(handle);
5007 ext4_std_error(inode->i_sb, err);
5012 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5014 return !buffer_mapped(bh);
5017 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5019 struct page *page = vmf->page;
5023 struct file *file = vma->vm_file;
5024 struct inode *inode = file->f_path.dentry->d_inode;
5025 struct address_space *mapping = inode->i_mapping;
5027 get_block_t *get_block;
5030 sb_start_pagefault(inode->i_sb);
5031 file_update_time(vma->vm_file);
5032 /* Delalloc case is easy... */
5033 if (test_opt(inode->i_sb, DELALLOC) &&
5034 !ext4_should_journal_data(inode) &&
5035 !ext4_nonda_switch(inode->i_sb)) {
5037 ret = __block_page_mkwrite(vma, vmf,
5038 ext4_da_get_block_prep);
5039 } while (ret == -ENOSPC &&
5040 ext4_should_retry_alloc(inode->i_sb, &retries));
5045 size = i_size_read(inode);
5046 /* Page got truncated from under us? */
5047 if (page->mapping != mapping || page_offset(page) > size) {
5049 ret = VM_FAULT_NOPAGE;
5053 if (page->index == size >> PAGE_CACHE_SHIFT)
5054 len = size & ~PAGE_CACHE_MASK;
5056 len = PAGE_CACHE_SIZE;
5058 * Return if we have all the buffers mapped. This avoids the need to do
5059 * journal_start/journal_stop which can block and take a long time
5061 if (page_has_buffers(page)) {
5062 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5064 ext4_bh_unmapped)) {
5065 /* Wait so that we don't change page under IO */
5066 wait_on_page_writeback(page);
5067 ret = VM_FAULT_LOCKED;
5072 /* OK, we need to fill the hole... */
5073 if (ext4_should_dioread_nolock(inode))
5074 get_block = ext4_get_block_write;
5076 get_block = ext4_get_block;
5078 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5079 ext4_writepage_trans_blocks(inode));
5080 if (IS_ERR(handle)) {
5081 ret = VM_FAULT_SIGBUS;
5084 ret = __block_page_mkwrite(vma, vmf, get_block);
5085 if (!ret && ext4_should_journal_data(inode)) {
5086 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5087 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5089 ret = VM_FAULT_SIGBUS;
5090 ext4_journal_stop(handle);
5093 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5095 ext4_journal_stop(handle);
5096 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5099 ret = block_page_mkwrite_return(ret);
5101 sb_end_pagefault(inode->i_sb);