2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
32 #include <linux/module.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
50 * combine low and high parts of physical block number into ext4_fsblk_t
52 ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
56 block = le32_to_cpu(ex->ee_start_lo);
57 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
63 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
65 ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
69 block = le32_to_cpu(ix->ei_leaf_lo);
70 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
75 * ext4_ext_store_pblock:
76 * stores a large physical block number into an extent struct,
77 * breaking it into parts
79 void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
81 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
82 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
86 * ext4_idx_store_pblock:
87 * stores a large physical block number into an index struct,
88 * breaking it into parts
90 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
92 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
93 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
96 static int ext4_ext_truncate_extend_restart(handle_t *handle,
102 if (!ext4_handle_valid(handle))
104 if (handle->h_buffer_credits > needed)
106 err = ext4_journal_extend(handle, needed);
109 err = ext4_truncate_restart_trans(handle, inode, needed);
111 * We have dropped i_data_sem so someone might have cached again
112 * an extent we are going to truncate.
114 ext4_ext_invalidate_cache(inode);
124 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
125 struct ext4_ext_path *path)
128 /* path points to block */
129 return ext4_journal_get_write_access(handle, path->p_bh);
131 /* path points to leaf/index in inode body */
132 /* we use in-core data, no need to protect them */
142 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
143 struct ext4_ext_path *path)
147 /* path points to block */
148 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
150 /* path points to leaf/index in inode body */
151 err = ext4_mark_inode_dirty(handle, inode);
156 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
157 struct ext4_ext_path *path,
160 struct ext4_inode_info *ei = EXT4_I(inode);
161 ext4_fsblk_t bg_start;
162 ext4_fsblk_t last_block;
163 ext4_grpblk_t colour;
164 ext4_group_t block_group;
165 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
169 struct ext4_extent *ex;
170 depth = path->p_depth;
172 /* try to predict block placement */
173 ex = path[depth].p_ext;
175 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
177 /* it looks like index is empty;
178 * try to find starting block from index itself */
179 if (path[depth].p_bh)
180 return path[depth].p_bh->b_blocknr;
183 /* OK. use inode's group */
184 block_group = ei->i_block_group;
185 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
187 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
188 * block groups per flexgroup, reserve the first block
189 * group for directories and special files. Regular
190 * files will start at the second block group. This
191 * tends to speed up directory access and improves
194 block_group &= ~(flex_size-1);
195 if (S_ISREG(inode->i_mode))
198 bg_start = (block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
199 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
200 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
203 * If we are doing delayed allocation, we don't need take
204 * colour into account.
206 if (test_opt(inode->i_sb, DELALLOC))
209 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
210 colour = (current->pid % 16) *
211 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
213 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
214 return bg_start + colour + block;
218 * Allocation for a meta data block
221 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
222 struct ext4_ext_path *path,
223 struct ext4_extent *ex, int *err)
225 ext4_fsblk_t goal, newblock;
227 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
228 newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
232 static inline int ext4_ext_space_block(struct inode *inode, int check)
236 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
237 / sizeof(struct ext4_extent);
239 #ifdef AGGRESSIVE_TEST
247 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
251 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
252 / sizeof(struct ext4_extent_idx);
254 #ifdef AGGRESSIVE_TEST
262 static inline int ext4_ext_space_root(struct inode *inode, int check)
266 size = sizeof(EXT4_I(inode)->i_data);
267 size -= sizeof(struct ext4_extent_header);
268 size /= sizeof(struct ext4_extent);
270 #ifdef AGGRESSIVE_TEST
278 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
282 size = sizeof(EXT4_I(inode)->i_data);
283 size -= sizeof(struct ext4_extent_header);
284 size /= sizeof(struct ext4_extent_idx);
286 #ifdef AGGRESSIVE_TEST
295 * Calculate the number of metadata blocks needed
296 * to allocate @blocks
297 * Worse case is one block per extent
299 int ext4_ext_calc_metadata_amount(struct inode *inode, sector_t lblock)
301 struct ext4_inode_info *ei = EXT4_I(inode);
304 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
305 / sizeof(struct ext4_extent_idx));
308 * If the new delayed allocation block is contiguous with the
309 * previous da block, it can share index blocks with the
310 * previous block, so we only need to allocate a new index
311 * block every idxs leaf blocks. At ldxs**2 blocks, we need
312 * an additional index block, and at ldxs**3 blocks, yet
313 * another index blocks.
315 if (ei->i_da_metadata_calc_len &&
316 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
317 if ((ei->i_da_metadata_calc_len % idxs) == 0)
319 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
321 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
323 ei->i_da_metadata_calc_len = 0;
325 ei->i_da_metadata_calc_len++;
326 ei->i_da_metadata_calc_last_lblock++;
331 * In the worst case we need a new set of index blocks at
332 * every level of the inode's extent tree.
334 ei->i_da_metadata_calc_len = 1;
335 ei->i_da_metadata_calc_last_lblock = lblock;
336 return ext_depth(inode) + 1;
340 ext4_ext_max_entries(struct inode *inode, int depth)
344 if (depth == ext_depth(inode)) {
346 max = ext4_ext_space_root(inode, 1);
348 max = ext4_ext_space_root_idx(inode, 1);
351 max = ext4_ext_space_block(inode, 1);
353 max = ext4_ext_space_block_idx(inode, 1);
359 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
361 ext4_fsblk_t block = ext_pblock(ext);
362 int len = ext4_ext_get_actual_len(ext);
364 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
367 static int ext4_valid_extent_idx(struct inode *inode,
368 struct ext4_extent_idx *ext_idx)
370 ext4_fsblk_t block = idx_pblock(ext_idx);
372 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
375 static int ext4_valid_extent_entries(struct inode *inode,
376 struct ext4_extent_header *eh,
379 struct ext4_extent *ext;
380 struct ext4_extent_idx *ext_idx;
381 unsigned short entries;
382 if (eh->eh_entries == 0)
385 entries = le16_to_cpu(eh->eh_entries);
389 ext = EXT_FIRST_EXTENT(eh);
391 if (!ext4_valid_extent(inode, ext))
397 ext_idx = EXT_FIRST_INDEX(eh);
399 if (!ext4_valid_extent_idx(inode, ext_idx))
408 static int __ext4_ext_check(const char *function, struct inode *inode,
409 struct ext4_extent_header *eh,
412 const char *error_msg;
415 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
416 error_msg = "invalid magic";
419 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
420 error_msg = "unexpected eh_depth";
423 if (unlikely(eh->eh_max == 0)) {
424 error_msg = "invalid eh_max";
427 max = ext4_ext_max_entries(inode, depth);
428 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
429 error_msg = "too large eh_max";
432 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
433 error_msg = "invalid eh_entries";
436 if (!ext4_valid_extent_entries(inode, eh, depth)) {
437 error_msg = "invalid extent entries";
443 __ext4_error(inode->i_sb, function,
444 "bad header/extent in inode #%lu: %s - magic %x, "
445 "entries %u, max %u(%u), depth %u(%u)",
446 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
447 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
448 max, le16_to_cpu(eh->eh_depth), depth);
453 #define ext4_ext_check(inode, eh, depth) \
454 __ext4_ext_check(__func__, inode, eh, depth)
456 int ext4_ext_check_inode(struct inode *inode)
458 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
462 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
464 int k, l = path->p_depth;
467 for (k = 0; k <= l; k++, path++) {
469 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
470 idx_pblock(path->p_idx));
471 } else if (path->p_ext) {
472 ext_debug(" %d:[%d]%d:%llu ",
473 le32_to_cpu(path->p_ext->ee_block),
474 ext4_ext_is_uninitialized(path->p_ext),
475 ext4_ext_get_actual_len(path->p_ext),
476 ext_pblock(path->p_ext));
483 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
485 int depth = ext_depth(inode);
486 struct ext4_extent_header *eh;
487 struct ext4_extent *ex;
493 eh = path[depth].p_hdr;
494 ex = EXT_FIRST_EXTENT(eh);
496 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
498 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
499 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
500 ext4_ext_is_uninitialized(ex),
501 ext4_ext_get_actual_len(ex), ext_pblock(ex));
506 #define ext4_ext_show_path(inode, path)
507 #define ext4_ext_show_leaf(inode, path)
510 void ext4_ext_drop_refs(struct ext4_ext_path *path)
512 int depth = path->p_depth;
515 for (i = 0; i <= depth; i++, path++)
523 * ext4_ext_binsearch_idx:
524 * binary search for the closest index of the given block
525 * the header must be checked before calling this
528 ext4_ext_binsearch_idx(struct inode *inode,
529 struct ext4_ext_path *path, ext4_lblk_t block)
531 struct ext4_extent_header *eh = path->p_hdr;
532 struct ext4_extent_idx *r, *l, *m;
535 ext_debug("binsearch for %u(idx): ", block);
537 l = EXT_FIRST_INDEX(eh) + 1;
538 r = EXT_LAST_INDEX(eh);
541 if (block < le32_to_cpu(m->ei_block))
545 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
546 m, le32_to_cpu(m->ei_block),
547 r, le32_to_cpu(r->ei_block));
551 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
552 idx_pblock(path->p_idx));
554 #ifdef CHECK_BINSEARCH
556 struct ext4_extent_idx *chix, *ix;
559 chix = ix = EXT_FIRST_INDEX(eh);
560 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
562 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
563 printk(KERN_DEBUG "k=%d, ix=0x%p, "
565 ix, EXT_FIRST_INDEX(eh));
566 printk(KERN_DEBUG "%u <= %u\n",
567 le32_to_cpu(ix->ei_block),
568 le32_to_cpu(ix[-1].ei_block));
570 BUG_ON(k && le32_to_cpu(ix->ei_block)
571 <= le32_to_cpu(ix[-1].ei_block));
572 if (block < le32_to_cpu(ix->ei_block))
576 BUG_ON(chix != path->p_idx);
583 * ext4_ext_binsearch:
584 * binary search for closest extent of the given block
585 * the header must be checked before calling this
588 ext4_ext_binsearch(struct inode *inode,
589 struct ext4_ext_path *path, ext4_lblk_t block)
591 struct ext4_extent_header *eh = path->p_hdr;
592 struct ext4_extent *r, *l, *m;
594 if (eh->eh_entries == 0) {
596 * this leaf is empty:
597 * we get such a leaf in split/add case
602 ext_debug("binsearch for %u: ", block);
604 l = EXT_FIRST_EXTENT(eh) + 1;
605 r = EXT_LAST_EXTENT(eh);
609 if (block < le32_to_cpu(m->ee_block))
613 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
614 m, le32_to_cpu(m->ee_block),
615 r, le32_to_cpu(r->ee_block));
619 ext_debug(" -> %d:%llu:[%d]%d ",
620 le32_to_cpu(path->p_ext->ee_block),
621 ext_pblock(path->p_ext),
622 ext4_ext_is_uninitialized(path->p_ext),
623 ext4_ext_get_actual_len(path->p_ext));
625 #ifdef CHECK_BINSEARCH
627 struct ext4_extent *chex, *ex;
630 chex = ex = EXT_FIRST_EXTENT(eh);
631 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
632 BUG_ON(k && le32_to_cpu(ex->ee_block)
633 <= le32_to_cpu(ex[-1].ee_block));
634 if (block < le32_to_cpu(ex->ee_block))
638 BUG_ON(chex != path->p_ext);
644 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
646 struct ext4_extent_header *eh;
648 eh = ext_inode_hdr(inode);
651 eh->eh_magic = EXT4_EXT_MAGIC;
652 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
653 ext4_mark_inode_dirty(handle, inode);
654 ext4_ext_invalidate_cache(inode);
658 struct ext4_ext_path *
659 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
660 struct ext4_ext_path *path)
662 struct ext4_extent_header *eh;
663 struct buffer_head *bh;
664 short int depth, i, ppos = 0, alloc = 0;
666 eh = ext_inode_hdr(inode);
667 depth = ext_depth(inode);
669 /* account possible depth increase */
671 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
674 return ERR_PTR(-ENOMEM);
681 /* walk through the tree */
683 int need_to_validate = 0;
685 ext_debug("depth %d: num %d, max %d\n",
686 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
688 ext4_ext_binsearch_idx(inode, path + ppos, block);
689 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
690 path[ppos].p_depth = i;
691 path[ppos].p_ext = NULL;
693 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
696 if (!bh_uptodate_or_lock(bh)) {
697 if (bh_submit_read(bh) < 0) {
701 /* validate the extent entries */
702 need_to_validate = 1;
704 eh = ext_block_hdr(bh);
706 if (unlikely(ppos > depth)) {
708 EXT4_ERROR_INODE(inode,
709 "ppos %d > depth %d", ppos, depth);
712 path[ppos].p_bh = bh;
713 path[ppos].p_hdr = eh;
716 if (need_to_validate && ext4_ext_check(inode, eh, i))
720 path[ppos].p_depth = i;
721 path[ppos].p_ext = NULL;
722 path[ppos].p_idx = NULL;
725 ext4_ext_binsearch(inode, path + ppos, block);
726 /* if not an empty leaf */
727 if (path[ppos].p_ext)
728 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
730 ext4_ext_show_path(inode, path);
735 ext4_ext_drop_refs(path);
738 return ERR_PTR(-EIO);
742 * ext4_ext_insert_index:
743 * insert new index [@logical;@ptr] into the block at @curp;
744 * check where to insert: before @curp or after @curp
746 int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
747 struct ext4_ext_path *curp,
748 int logical, ext4_fsblk_t ptr)
750 struct ext4_extent_idx *ix;
753 err = ext4_ext_get_access(handle, inode, curp);
757 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
758 EXT4_ERROR_INODE(inode,
759 "logical %d == ei_block %d!",
760 logical, le32_to_cpu(curp->p_idx->ei_block));
763 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
764 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
766 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
767 len = (len - 1) * sizeof(struct ext4_extent_idx);
768 len = len < 0 ? 0 : len;
769 ext_debug("insert new index %d after: %llu. "
770 "move %d from 0x%p to 0x%p\n",
772 (curp->p_idx + 1), (curp->p_idx + 2));
773 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
775 ix = curp->p_idx + 1;
778 len = len * sizeof(struct ext4_extent_idx);
779 len = len < 0 ? 0 : len;
780 ext_debug("insert new index %d before: %llu. "
781 "move %d from 0x%p to 0x%p\n",
783 curp->p_idx, (curp->p_idx + 1));
784 memmove(curp->p_idx + 1, curp->p_idx, len);
788 ix->ei_block = cpu_to_le32(logical);
789 ext4_idx_store_pblock(ix, ptr);
790 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
792 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
793 > le16_to_cpu(curp->p_hdr->eh_max))) {
794 EXT4_ERROR_INODE(inode,
795 "logical %d == ei_block %d!",
796 logical, le32_to_cpu(curp->p_idx->ei_block));
799 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
800 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
804 err = ext4_ext_dirty(handle, inode, curp);
805 ext4_std_error(inode->i_sb, err);
812 * inserts new subtree into the path, using free index entry
814 * - allocates all needed blocks (new leaf and all intermediate index blocks)
815 * - makes decision where to split
816 * - moves remaining extents and index entries (right to the split point)
817 * into the newly allocated blocks
818 * - initializes subtree
820 static int ext4_ext_split(handle_t *handle, struct inode *inode,
821 struct ext4_ext_path *path,
822 struct ext4_extent *newext, int at)
824 struct buffer_head *bh = NULL;
825 int depth = ext_depth(inode);
826 struct ext4_extent_header *neh;
827 struct ext4_extent_idx *fidx;
828 struct ext4_extent *ex;
830 ext4_fsblk_t newblock, oldblock;
832 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
835 /* make decision: where to split? */
836 /* FIXME: now decision is simplest: at current extent */
838 /* if current leaf will be split, then we should use
839 * border from split point */
840 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
841 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
844 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
845 border = path[depth].p_ext[1].ee_block;
846 ext_debug("leaf will be split."
847 " next leaf starts at %d\n",
848 le32_to_cpu(border));
850 border = newext->ee_block;
851 ext_debug("leaf will be added."
852 " next leaf starts at %d\n",
853 le32_to_cpu(border));
857 * If error occurs, then we break processing
858 * and mark filesystem read-only. index won't
859 * be inserted and tree will be in consistent
860 * state. Next mount will repair buffers too.
864 * Get array to track all allocated blocks.
865 * We need this to handle errors and free blocks
868 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
872 /* allocate all needed blocks */
873 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
874 for (a = 0; a < depth - at; a++) {
875 newblock = ext4_ext_new_meta_block(handle, inode, path,
879 ablocks[a] = newblock;
882 /* initialize new leaf */
883 newblock = ablocks[--a];
884 if (unlikely(newblock == 0)) {
885 EXT4_ERROR_INODE(inode, "newblock == 0!");
889 bh = sb_getblk(inode->i_sb, newblock);
896 err = ext4_journal_get_create_access(handle, bh);
900 neh = ext_block_hdr(bh);
902 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
903 neh->eh_magic = EXT4_EXT_MAGIC;
905 ex = EXT_FIRST_EXTENT(neh);
907 /* move remainder of path[depth] to the new leaf */
908 if (unlikely(path[depth].p_hdr->eh_entries !=
909 path[depth].p_hdr->eh_max)) {
910 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
911 path[depth].p_hdr->eh_entries,
912 path[depth].p_hdr->eh_max);
916 /* start copy from next extent */
917 /* TODO: we could do it by single memmove */
920 while (path[depth].p_ext <=
921 EXT_MAX_EXTENT(path[depth].p_hdr)) {
922 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
923 le32_to_cpu(path[depth].p_ext->ee_block),
924 ext_pblock(path[depth].p_ext),
925 ext4_ext_is_uninitialized(path[depth].p_ext),
926 ext4_ext_get_actual_len(path[depth].p_ext),
928 /*memmove(ex++, path[depth].p_ext++,
929 sizeof(struct ext4_extent));
935 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
936 le16_add_cpu(&neh->eh_entries, m);
939 set_buffer_uptodate(bh);
942 err = ext4_handle_dirty_metadata(handle, inode, bh);
948 /* correct old leaf */
950 err = ext4_ext_get_access(handle, inode, path + depth);
953 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
954 err = ext4_ext_dirty(handle, inode, path + depth);
960 /* create intermediate indexes */
962 if (unlikely(k < 0)) {
963 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
968 ext_debug("create %d intermediate indices\n", k);
969 /* insert new index into current index block */
970 /* current depth stored in i var */
974 newblock = ablocks[--a];
975 bh = sb_getblk(inode->i_sb, newblock);
982 err = ext4_journal_get_create_access(handle, bh);
986 neh = ext_block_hdr(bh);
987 neh->eh_entries = cpu_to_le16(1);
988 neh->eh_magic = EXT4_EXT_MAGIC;
989 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
990 neh->eh_depth = cpu_to_le16(depth - i);
991 fidx = EXT_FIRST_INDEX(neh);
992 fidx->ei_block = border;
993 ext4_idx_store_pblock(fidx, oldblock);
995 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
996 i, newblock, le32_to_cpu(border), oldblock);
1001 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1002 EXT_MAX_INDEX(path[i].p_hdr));
1003 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1004 EXT_LAST_INDEX(path[i].p_hdr))) {
1005 EXT4_ERROR_INODE(inode,
1006 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1007 le32_to_cpu(path[i].p_ext->ee_block));
1011 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
1012 ext_debug("%d: move %d:%llu in new index %llu\n", i,
1013 le32_to_cpu(path[i].p_idx->ei_block),
1014 idx_pblock(path[i].p_idx),
1016 /*memmove(++fidx, path[i].p_idx++,
1017 sizeof(struct ext4_extent_idx));
1019 BUG_ON(neh->eh_entries > neh->eh_max);*/
1024 memmove(++fidx, path[i].p_idx - m,
1025 sizeof(struct ext4_extent_idx) * m);
1026 le16_add_cpu(&neh->eh_entries, m);
1028 set_buffer_uptodate(bh);
1031 err = ext4_handle_dirty_metadata(handle, inode, bh);
1037 /* correct old index */
1039 err = ext4_ext_get_access(handle, inode, path + i);
1042 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1043 err = ext4_ext_dirty(handle, inode, path + i);
1051 /* insert new index */
1052 err = ext4_ext_insert_index(handle, inode, path + at,
1053 le32_to_cpu(border), newblock);
1057 if (buffer_locked(bh))
1063 /* free all allocated blocks in error case */
1064 for (i = 0; i < depth; i++) {
1067 ext4_free_blocks(handle, inode, 0, ablocks[i], 1,
1068 EXT4_FREE_BLOCKS_METADATA);
1077 * ext4_ext_grow_indepth:
1078 * implements tree growing procedure:
1079 * - allocates new block
1080 * - moves top-level data (index block or leaf) into the new block
1081 * - initializes new top-level, creating index that points to the
1082 * just created block
1084 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1085 struct ext4_ext_path *path,
1086 struct ext4_extent *newext)
1088 struct ext4_ext_path *curp = path;
1089 struct ext4_extent_header *neh;
1090 struct ext4_extent_idx *fidx;
1091 struct buffer_head *bh;
1092 ext4_fsblk_t newblock;
1095 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1099 bh = sb_getblk(inode->i_sb, newblock);
1102 ext4_std_error(inode->i_sb, err);
1107 err = ext4_journal_get_create_access(handle, bh);
1113 /* move top-level index/leaf into new block */
1114 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1116 /* set size of new block */
1117 neh = ext_block_hdr(bh);
1118 /* old root could have indexes or leaves
1119 * so calculate e_max right way */
1120 if (ext_depth(inode))
1121 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1123 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1124 neh->eh_magic = EXT4_EXT_MAGIC;
1125 set_buffer_uptodate(bh);
1128 err = ext4_handle_dirty_metadata(handle, inode, bh);
1132 /* create index in new top-level index: num,max,pointer */
1133 err = ext4_ext_get_access(handle, inode, curp);
1137 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1138 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1139 curp->p_hdr->eh_entries = cpu_to_le16(1);
1140 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1142 if (path[0].p_hdr->eh_depth)
1143 curp->p_idx->ei_block =
1144 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1146 curp->p_idx->ei_block =
1147 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1148 ext4_idx_store_pblock(curp->p_idx, newblock);
1150 neh = ext_inode_hdr(inode);
1151 fidx = EXT_FIRST_INDEX(neh);
1152 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1153 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1154 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1156 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1157 err = ext4_ext_dirty(handle, inode, curp);
1165 * ext4_ext_create_new_leaf:
1166 * finds empty index and adds new leaf.
1167 * if no free index is found, then it requests in-depth growing.
1169 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1170 struct ext4_ext_path *path,
1171 struct ext4_extent *newext)
1173 struct ext4_ext_path *curp;
1174 int depth, i, err = 0;
1177 i = depth = ext_depth(inode);
1179 /* walk up to the tree and look for free index entry */
1180 curp = path + depth;
1181 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1186 /* we use already allocated block for index block,
1187 * so subsequent data blocks should be contiguous */
1188 if (EXT_HAS_FREE_INDEX(curp)) {
1189 /* if we found index with free entry, then use that
1190 * entry: create all needed subtree and add new leaf */
1191 err = ext4_ext_split(handle, inode, path, newext, i);
1196 ext4_ext_drop_refs(path);
1197 path = ext4_ext_find_extent(inode,
1198 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1201 err = PTR_ERR(path);
1203 /* tree is full, time to grow in depth */
1204 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1209 ext4_ext_drop_refs(path);
1210 path = ext4_ext_find_extent(inode,
1211 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1214 err = PTR_ERR(path);
1219 * only first (depth 0 -> 1) produces free space;
1220 * in all other cases we have to split the grown tree
1222 depth = ext_depth(inode);
1223 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1224 /* now we need to split */
1234 * search the closest allocated block to the left for *logical
1235 * and returns it at @logical + it's physical address at @phys
1236 * if *logical is the smallest allocated block, the function
1237 * returns 0 at @phys
1238 * return value contains 0 (success) or error code
1241 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1242 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1244 struct ext4_extent_idx *ix;
1245 struct ext4_extent *ex;
1248 if (unlikely(path == NULL)) {
1249 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1252 depth = path->p_depth;
1255 if (depth == 0 && path->p_ext == NULL)
1258 /* usually extent in the path covers blocks smaller
1259 * then *logical, but it can be that extent is the
1260 * first one in the file */
1262 ex = path[depth].p_ext;
1263 ee_len = ext4_ext_get_actual_len(ex);
1264 if (*logical < le32_to_cpu(ex->ee_block)) {
1265 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1266 EXT4_ERROR_INODE(inode,
1267 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1268 *logical, le32_to_cpu(ex->ee_block));
1271 while (--depth >= 0) {
1272 ix = path[depth].p_idx;
1273 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1274 EXT4_ERROR_INODE(inode,
1275 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1276 ix != NULL ? ix->ei_block : 0,
1277 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1278 EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1286 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1287 EXT4_ERROR_INODE(inode,
1288 "logical %d < ee_block %d + ee_len %d!",
1289 *logical, le32_to_cpu(ex->ee_block), ee_len);
1293 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1294 *phys = ext_pblock(ex) + ee_len - 1;
1299 * search the closest allocated block to the right for *logical
1300 * and returns it at @logical + it's physical address at @phys
1301 * if *logical is the smallest allocated block, the function
1302 * returns 0 at @phys
1303 * return value contains 0 (success) or error code
1306 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1307 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1309 struct buffer_head *bh = NULL;
1310 struct ext4_extent_header *eh;
1311 struct ext4_extent_idx *ix;
1312 struct ext4_extent *ex;
1314 int depth; /* Note, NOT eh_depth; depth from top of tree */
1317 if (unlikely(path == NULL)) {
1318 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1321 depth = path->p_depth;
1324 if (depth == 0 && path->p_ext == NULL)
1327 /* usually extent in the path covers blocks smaller
1328 * then *logical, but it can be that extent is the
1329 * first one in the file */
1331 ex = path[depth].p_ext;
1332 ee_len = ext4_ext_get_actual_len(ex);
1333 if (*logical < le32_to_cpu(ex->ee_block)) {
1334 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1335 EXT4_ERROR_INODE(inode,
1336 "first_extent(path[%d].p_hdr) != ex",
1340 while (--depth >= 0) {
1341 ix = path[depth].p_idx;
1342 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1343 EXT4_ERROR_INODE(inode,
1344 "ix != EXT_FIRST_INDEX *logical %d!",
1349 *logical = le32_to_cpu(ex->ee_block);
1350 *phys = ext_pblock(ex);
1354 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1355 EXT4_ERROR_INODE(inode,
1356 "logical %d < ee_block %d + ee_len %d!",
1357 *logical, le32_to_cpu(ex->ee_block), ee_len);
1361 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1362 /* next allocated block in this leaf */
1364 *logical = le32_to_cpu(ex->ee_block);
1365 *phys = ext_pblock(ex);
1369 /* go up and search for index to the right */
1370 while (--depth >= 0) {
1371 ix = path[depth].p_idx;
1372 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1376 /* we've gone up to the root and found no index to the right */
1380 /* we've found index to the right, let's
1381 * follow it and find the closest allocated
1382 * block to the right */
1384 block = idx_pblock(ix);
1385 while (++depth < path->p_depth) {
1386 bh = sb_bread(inode->i_sb, block);
1389 eh = ext_block_hdr(bh);
1390 /* subtract from p_depth to get proper eh_depth */
1391 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1395 ix = EXT_FIRST_INDEX(eh);
1396 block = idx_pblock(ix);
1400 bh = sb_bread(inode->i_sb, block);
1403 eh = ext_block_hdr(bh);
1404 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1408 ex = EXT_FIRST_EXTENT(eh);
1409 *logical = le32_to_cpu(ex->ee_block);
1410 *phys = ext_pblock(ex);
1416 * ext4_ext_next_allocated_block:
1417 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1418 * NOTE: it considers block number from index entry as
1419 * allocated block. Thus, index entries have to be consistent
1423 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1427 BUG_ON(path == NULL);
1428 depth = path->p_depth;
1430 if (depth == 0 && path->p_ext == NULL)
1431 return EXT_MAX_BLOCK;
1433 while (depth >= 0) {
1434 if (depth == path->p_depth) {
1436 if (path[depth].p_ext !=
1437 EXT_LAST_EXTENT(path[depth].p_hdr))
1438 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1441 if (path[depth].p_idx !=
1442 EXT_LAST_INDEX(path[depth].p_hdr))
1443 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1448 return EXT_MAX_BLOCK;
1452 * ext4_ext_next_leaf_block:
1453 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1455 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1456 struct ext4_ext_path *path)
1460 BUG_ON(path == NULL);
1461 depth = path->p_depth;
1463 /* zero-tree has no leaf blocks at all */
1465 return EXT_MAX_BLOCK;
1467 /* go to index block */
1470 while (depth >= 0) {
1471 if (path[depth].p_idx !=
1472 EXT_LAST_INDEX(path[depth].p_hdr))
1473 return (ext4_lblk_t)
1474 le32_to_cpu(path[depth].p_idx[1].ei_block);
1478 return EXT_MAX_BLOCK;
1482 * ext4_ext_correct_indexes:
1483 * if leaf gets modified and modified extent is first in the leaf,
1484 * then we have to correct all indexes above.
1485 * TODO: do we need to correct tree in all cases?
1487 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1488 struct ext4_ext_path *path)
1490 struct ext4_extent_header *eh;
1491 int depth = ext_depth(inode);
1492 struct ext4_extent *ex;
1496 eh = path[depth].p_hdr;
1497 ex = path[depth].p_ext;
1499 if (unlikely(ex == NULL || eh == NULL)) {
1500 EXT4_ERROR_INODE(inode,
1501 "ex %p == NULL or eh %p == NULL", ex, eh);
1506 /* there is no tree at all */
1510 if (ex != EXT_FIRST_EXTENT(eh)) {
1511 /* we correct tree if first leaf got modified only */
1516 * TODO: we need correction if border is smaller than current one
1519 border = path[depth].p_ext->ee_block;
1520 err = ext4_ext_get_access(handle, inode, path + k);
1523 path[k].p_idx->ei_block = border;
1524 err = ext4_ext_dirty(handle, inode, path + k);
1529 /* change all left-side indexes */
1530 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1532 err = ext4_ext_get_access(handle, inode, path + k);
1535 path[k].p_idx->ei_block = border;
1536 err = ext4_ext_dirty(handle, inode, path + k);
1545 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1546 struct ext4_extent *ex2)
1548 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1551 * Make sure that either both extents are uninitialized, or
1554 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1557 if (ext4_ext_is_uninitialized(ex1))
1558 max_len = EXT_UNINIT_MAX_LEN;
1560 max_len = EXT_INIT_MAX_LEN;
1562 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1563 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1565 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1566 le32_to_cpu(ex2->ee_block))
1570 * To allow future support for preallocated extents to be added
1571 * as an RO_COMPAT feature, refuse to merge to extents if
1572 * this can result in the top bit of ee_len being set.
1574 if (ext1_ee_len + ext2_ee_len > max_len)
1576 #ifdef AGGRESSIVE_TEST
1577 if (ext1_ee_len >= 4)
1581 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1587 * This function tries to merge the "ex" extent to the next extent in the tree.
1588 * It always tries to merge towards right. If you want to merge towards
1589 * left, pass "ex - 1" as argument instead of "ex".
1590 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1591 * 1 if they got merged.
1593 int ext4_ext_try_to_merge(struct inode *inode,
1594 struct ext4_ext_path *path,
1595 struct ext4_extent *ex)
1597 struct ext4_extent_header *eh;
1598 unsigned int depth, len;
1600 int uninitialized = 0;
1602 depth = ext_depth(inode);
1603 BUG_ON(path[depth].p_hdr == NULL);
1604 eh = path[depth].p_hdr;
1606 while (ex < EXT_LAST_EXTENT(eh)) {
1607 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1609 /* merge with next extent! */
1610 if (ext4_ext_is_uninitialized(ex))
1612 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1613 + ext4_ext_get_actual_len(ex + 1));
1615 ext4_ext_mark_uninitialized(ex);
1617 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1618 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1619 * sizeof(struct ext4_extent);
1620 memmove(ex + 1, ex + 2, len);
1622 le16_add_cpu(&eh->eh_entries, -1);
1624 WARN_ON(eh->eh_entries == 0);
1625 if (!eh->eh_entries)
1626 ext4_error(inode->i_sb,
1627 "inode#%lu, eh->eh_entries = 0!",
1635 * check if a portion of the "newext" extent overlaps with an
1638 * If there is an overlap discovered, it updates the length of the newext
1639 * such that there will be no overlap, and then returns 1.
1640 * If there is no overlap found, it returns 0.
1642 unsigned int ext4_ext_check_overlap(struct inode *inode,
1643 struct ext4_extent *newext,
1644 struct ext4_ext_path *path)
1647 unsigned int depth, len1;
1648 unsigned int ret = 0;
1650 b1 = le32_to_cpu(newext->ee_block);
1651 len1 = ext4_ext_get_actual_len(newext);
1652 depth = ext_depth(inode);
1653 if (!path[depth].p_ext)
1655 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1658 * get the next allocated block if the extent in the path
1659 * is before the requested block(s)
1662 b2 = ext4_ext_next_allocated_block(path);
1663 if (b2 == EXT_MAX_BLOCK)
1667 /* check for wrap through zero on extent logical start block*/
1668 if (b1 + len1 < b1) {
1669 len1 = EXT_MAX_BLOCK - b1;
1670 newext->ee_len = cpu_to_le16(len1);
1674 /* check for overlap */
1675 if (b1 + len1 > b2) {
1676 newext->ee_len = cpu_to_le16(b2 - b1);
1684 * ext4_ext_insert_extent:
1685 * tries to merge requsted extent into the existing extent or
1686 * inserts requested extent as new one into the tree,
1687 * creating new leaf in the no-space case.
1689 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1690 struct ext4_ext_path *path,
1691 struct ext4_extent *newext, int flag)
1693 struct ext4_extent_header *eh;
1694 struct ext4_extent *ex, *fex;
1695 struct ext4_extent *nearex; /* nearest extent */
1696 struct ext4_ext_path *npath = NULL;
1697 int depth, len, err;
1699 unsigned uninitialized = 0;
1701 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1702 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1705 depth = ext_depth(inode);
1706 ex = path[depth].p_ext;
1707 if (unlikely(path[depth].p_hdr == NULL)) {
1708 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1712 /* try to insert block into found extent and return */
1713 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1714 && ext4_can_extents_be_merged(inode, ex, newext)) {
1715 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1716 ext4_ext_is_uninitialized(newext),
1717 ext4_ext_get_actual_len(newext),
1718 le32_to_cpu(ex->ee_block),
1719 ext4_ext_is_uninitialized(ex),
1720 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1721 err = ext4_ext_get_access(handle, inode, path + depth);
1726 * ext4_can_extents_be_merged should have checked that either
1727 * both extents are uninitialized, or both aren't. Thus we
1728 * need to check only one of them here.
1730 if (ext4_ext_is_uninitialized(ex))
1732 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1733 + ext4_ext_get_actual_len(newext));
1735 ext4_ext_mark_uninitialized(ex);
1736 eh = path[depth].p_hdr;
1742 depth = ext_depth(inode);
1743 eh = path[depth].p_hdr;
1744 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1747 /* probably next leaf has space for us? */
1748 fex = EXT_LAST_EXTENT(eh);
1749 next = ext4_ext_next_leaf_block(inode, path);
1750 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1751 && next != EXT_MAX_BLOCK) {
1752 ext_debug("next leaf block - %d\n", next);
1753 BUG_ON(npath != NULL);
1754 npath = ext4_ext_find_extent(inode, next, NULL);
1756 return PTR_ERR(npath);
1757 BUG_ON(npath->p_depth != path->p_depth);
1758 eh = npath[depth].p_hdr;
1759 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1760 ext_debug("next leaf isnt full(%d)\n",
1761 le16_to_cpu(eh->eh_entries));
1765 ext_debug("next leaf has no free space(%d,%d)\n",
1766 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1770 * There is no free space in the found leaf.
1771 * We're gonna add a new leaf in the tree.
1773 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1776 depth = ext_depth(inode);
1777 eh = path[depth].p_hdr;
1780 nearex = path[depth].p_ext;
1782 err = ext4_ext_get_access(handle, inode, path + depth);
1787 /* there is no extent in this leaf, create first one */
1788 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1789 le32_to_cpu(newext->ee_block),
1791 ext4_ext_is_uninitialized(newext),
1792 ext4_ext_get_actual_len(newext));
1793 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1794 } else if (le32_to_cpu(newext->ee_block)
1795 > le32_to_cpu(nearex->ee_block)) {
1796 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1797 if (nearex != EXT_LAST_EXTENT(eh)) {
1798 len = EXT_MAX_EXTENT(eh) - nearex;
1799 len = (len - 1) * sizeof(struct ext4_extent);
1800 len = len < 0 ? 0 : len;
1801 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1802 "move %d from 0x%p to 0x%p\n",
1803 le32_to_cpu(newext->ee_block),
1805 ext4_ext_is_uninitialized(newext),
1806 ext4_ext_get_actual_len(newext),
1807 nearex, len, nearex + 1, nearex + 2);
1808 memmove(nearex + 2, nearex + 1, len);
1810 path[depth].p_ext = nearex + 1;
1812 BUG_ON(newext->ee_block == nearex->ee_block);
1813 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1814 len = len < 0 ? 0 : len;
1815 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1816 "move %d from 0x%p to 0x%p\n",
1817 le32_to_cpu(newext->ee_block),
1819 ext4_ext_is_uninitialized(newext),
1820 ext4_ext_get_actual_len(newext),
1821 nearex, len, nearex + 1, nearex + 2);
1822 memmove(nearex + 1, nearex, len);
1823 path[depth].p_ext = nearex;
1826 le16_add_cpu(&eh->eh_entries, 1);
1827 nearex = path[depth].p_ext;
1828 nearex->ee_block = newext->ee_block;
1829 ext4_ext_store_pblock(nearex, ext_pblock(newext));
1830 nearex->ee_len = newext->ee_len;
1833 /* try to merge extents to the right */
1834 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1835 ext4_ext_try_to_merge(inode, path, nearex);
1837 /* try to merge extents to the left */
1839 /* time to correct all indexes above */
1840 err = ext4_ext_correct_indexes(handle, inode, path);
1844 err = ext4_ext_dirty(handle, inode, path + depth);
1848 ext4_ext_drop_refs(npath);
1851 ext4_ext_invalidate_cache(inode);
1855 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1856 ext4_lblk_t num, ext_prepare_callback func,
1859 struct ext4_ext_path *path = NULL;
1860 struct ext4_ext_cache cbex;
1861 struct ext4_extent *ex;
1862 ext4_lblk_t next, start = 0, end = 0;
1863 ext4_lblk_t last = block + num;
1864 int depth, exists, err = 0;
1866 BUG_ON(func == NULL);
1867 BUG_ON(inode == NULL);
1869 while (block < last && block != EXT_MAX_BLOCK) {
1871 /* find extent for this block */
1872 down_read(&EXT4_I(inode)->i_data_sem);
1873 path = ext4_ext_find_extent(inode, block, path);
1874 up_read(&EXT4_I(inode)->i_data_sem);
1876 err = PTR_ERR(path);
1881 depth = ext_depth(inode);
1882 if (unlikely(path[depth].p_hdr == NULL)) {
1883 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1887 ex = path[depth].p_ext;
1888 next = ext4_ext_next_allocated_block(path);
1892 /* there is no extent yet, so try to allocate
1893 * all requested space */
1896 } else if (le32_to_cpu(ex->ee_block) > block) {
1897 /* need to allocate space before found extent */
1899 end = le32_to_cpu(ex->ee_block);
1900 if (block + num < end)
1902 } else if (block >= le32_to_cpu(ex->ee_block)
1903 + ext4_ext_get_actual_len(ex)) {
1904 /* need to allocate space after found extent */
1909 } else if (block >= le32_to_cpu(ex->ee_block)) {
1911 * some part of requested space is covered
1915 end = le32_to_cpu(ex->ee_block)
1916 + ext4_ext_get_actual_len(ex);
1917 if (block + num < end)
1923 BUG_ON(end <= start);
1926 cbex.ec_block = start;
1927 cbex.ec_len = end - start;
1929 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1931 cbex.ec_block = le32_to_cpu(ex->ee_block);
1932 cbex.ec_len = ext4_ext_get_actual_len(ex);
1933 cbex.ec_start = ext_pblock(ex);
1934 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1937 if (unlikely(cbex.ec_len == 0)) {
1938 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1942 err = func(inode, path, &cbex, ex, cbdata);
1943 ext4_ext_drop_refs(path);
1948 if (err == EXT_REPEAT)
1950 else if (err == EXT_BREAK) {
1955 if (ext_depth(inode) != depth) {
1956 /* depth was changed. we have to realloc path */
1961 block = cbex.ec_block + cbex.ec_len;
1965 ext4_ext_drop_refs(path);
1973 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1974 __u32 len, ext4_fsblk_t start, int type)
1976 struct ext4_ext_cache *cex;
1978 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1979 cex = &EXT4_I(inode)->i_cached_extent;
1980 cex->ec_type = type;
1981 cex->ec_block = block;
1983 cex->ec_start = start;
1984 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1988 * ext4_ext_put_gap_in_cache:
1989 * calculate boundaries of the gap that the requested block fits into
1990 * and cache this gap
1993 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1996 int depth = ext_depth(inode);
1999 struct ext4_extent *ex;
2001 ex = path[depth].p_ext;
2003 /* there is no extent yet, so gap is [0;-] */
2005 len = EXT_MAX_BLOCK;
2006 ext_debug("cache gap(whole file):");
2007 } else if (block < le32_to_cpu(ex->ee_block)) {
2009 len = le32_to_cpu(ex->ee_block) - block;
2010 ext_debug("cache gap(before): %u [%u:%u]",
2012 le32_to_cpu(ex->ee_block),
2013 ext4_ext_get_actual_len(ex));
2014 } else if (block >= le32_to_cpu(ex->ee_block)
2015 + ext4_ext_get_actual_len(ex)) {
2017 lblock = le32_to_cpu(ex->ee_block)
2018 + ext4_ext_get_actual_len(ex);
2020 next = ext4_ext_next_allocated_block(path);
2021 ext_debug("cache gap(after): [%u:%u] %u",
2022 le32_to_cpu(ex->ee_block),
2023 ext4_ext_get_actual_len(ex),
2025 BUG_ON(next == lblock);
2026 len = next - lblock;
2032 ext_debug(" -> %u:%lu\n", lblock, len);
2033 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
2037 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2038 struct ext4_extent *ex)
2040 struct ext4_ext_cache *cex;
2041 int ret = EXT4_EXT_CACHE_NO;
2044 * We borrow i_block_reservation_lock to protect i_cached_extent
2046 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2047 cex = &EXT4_I(inode)->i_cached_extent;
2049 /* has cache valid data? */
2050 if (cex->ec_type == EXT4_EXT_CACHE_NO)
2053 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
2054 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
2055 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
2056 ex->ee_block = cpu_to_le32(cex->ec_block);
2057 ext4_ext_store_pblock(ex, cex->ec_start);
2058 ex->ee_len = cpu_to_le16(cex->ec_len);
2059 ext_debug("%u cached by %u:%u:%llu\n",
2061 cex->ec_block, cex->ec_len, cex->ec_start);
2065 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2071 * removes index from the index block.
2072 * It's used in truncate case only, thus all requests are for
2073 * last index in the block only.
2075 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2076 struct ext4_ext_path *path)
2081 /* free index block */
2083 leaf = idx_pblock(path->p_idx);
2084 if (unlikely(path->p_hdr->eh_entries == 0)) {
2085 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2088 err = ext4_ext_get_access(handle, inode, path);
2091 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2092 err = ext4_ext_dirty(handle, inode, path);
2095 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2096 ext4_free_blocks(handle, inode, 0, leaf, 1,
2097 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2102 * ext4_ext_calc_credits_for_single_extent:
2103 * This routine returns max. credits that needed to insert an extent
2104 * to the extent tree.
2105 * When pass the actual path, the caller should calculate credits
2108 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2109 struct ext4_ext_path *path)
2112 int depth = ext_depth(inode);
2115 /* probably there is space in leaf? */
2116 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2117 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2120 * There are some space in the leaf tree, no
2121 * need to account for leaf block credit
2123 * bitmaps and block group descriptor blocks
2124 * and other metadat blocks still need to be
2127 /* 1 bitmap, 1 block group descriptor */
2128 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2133 return ext4_chunk_trans_blocks(inode, nrblocks);
2137 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2139 * if nrblocks are fit in a single extent (chunk flag is 1), then
2140 * in the worse case, each tree level index/leaf need to be changed
2141 * if the tree split due to insert a new extent, then the old tree
2142 * index/leaf need to be updated too
2144 * If the nrblocks are discontiguous, they could cause
2145 * the whole tree split more than once, but this is really rare.
2147 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2150 int depth = ext_depth(inode);
2160 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2161 struct ext4_extent *ex,
2162 ext4_lblk_t from, ext4_lblk_t to)
2164 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2165 int flags = EXT4_FREE_BLOCKS_FORGET;
2167 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2168 flags |= EXT4_FREE_BLOCKS_METADATA;
2169 #ifdef EXTENTS_STATS
2171 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2172 spin_lock(&sbi->s_ext_stats_lock);
2173 sbi->s_ext_blocks += ee_len;
2174 sbi->s_ext_extents++;
2175 if (ee_len < sbi->s_ext_min)
2176 sbi->s_ext_min = ee_len;
2177 if (ee_len > sbi->s_ext_max)
2178 sbi->s_ext_max = ee_len;
2179 if (ext_depth(inode) > sbi->s_depth_max)
2180 sbi->s_depth_max = ext_depth(inode);
2181 spin_unlock(&sbi->s_ext_stats_lock);
2184 if (from >= le32_to_cpu(ex->ee_block)
2185 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2190 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2191 start = ext_pblock(ex) + ee_len - num;
2192 ext_debug("free last %u blocks starting %llu\n", num, start);
2193 ext4_free_blocks(handle, inode, 0, start, num, flags);
2194 } else if (from == le32_to_cpu(ex->ee_block)
2195 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2196 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2197 from, to, le32_to_cpu(ex->ee_block), ee_len);
2199 printk(KERN_INFO "strange request: removal(2) "
2200 "%u-%u from %u:%u\n",
2201 from, to, le32_to_cpu(ex->ee_block), ee_len);
2207 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2208 struct ext4_ext_path *path, ext4_lblk_t start)
2210 int err = 0, correct_index = 0;
2211 int depth = ext_depth(inode), credits;
2212 struct ext4_extent_header *eh;
2213 ext4_lblk_t a, b, block;
2215 ext4_lblk_t ex_ee_block;
2216 unsigned short ex_ee_len;
2217 unsigned uninitialized = 0;
2218 struct ext4_extent *ex;
2220 /* the header must be checked already in ext4_ext_remove_space() */
2221 ext_debug("truncate since %u in leaf\n", start);
2222 if (!path[depth].p_hdr)
2223 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2224 eh = path[depth].p_hdr;
2225 if (unlikely(path[depth].p_hdr == NULL)) {
2226 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2229 /* find where to start removing */
2230 ex = EXT_LAST_EXTENT(eh);
2232 ex_ee_block = le32_to_cpu(ex->ee_block);
2233 ex_ee_len = ext4_ext_get_actual_len(ex);
2235 while (ex >= EXT_FIRST_EXTENT(eh) &&
2236 ex_ee_block + ex_ee_len > start) {
2238 if (ext4_ext_is_uninitialized(ex))
2243 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2244 uninitialized, ex_ee_len);
2245 path[depth].p_ext = ex;
2247 a = ex_ee_block > start ? ex_ee_block : start;
2248 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2249 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2251 ext_debug(" border %u:%u\n", a, b);
2253 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2257 } else if (a != ex_ee_block) {
2258 /* remove tail of the extent */
2259 block = ex_ee_block;
2261 } else if (b != ex_ee_block + ex_ee_len - 1) {
2262 /* remove head of the extent */
2265 /* there is no "make a hole" API yet */
2268 /* remove whole extent: excellent! */
2269 block = ex_ee_block;
2271 BUG_ON(a != ex_ee_block);
2272 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2276 * 3 for leaf, sb, and inode plus 2 (bmap and group
2277 * descriptor) for each block group; assume two block
2278 * groups plus ex_ee_len/blocks_per_block_group for
2281 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2282 if (ex == EXT_FIRST_EXTENT(eh)) {
2284 credits += (ext_depth(inode)) + 1;
2286 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2288 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2292 err = ext4_ext_get_access(handle, inode, path + depth);
2296 err = ext4_remove_blocks(handle, inode, ex, a, b);
2301 /* this extent is removed; mark slot entirely unused */
2302 ext4_ext_store_pblock(ex, 0);
2303 le16_add_cpu(&eh->eh_entries, -1);
2306 ex->ee_block = cpu_to_le32(block);
2307 ex->ee_len = cpu_to_le16(num);
2309 * Do not mark uninitialized if all the blocks in the
2310 * extent have been removed.
2312 if (uninitialized && num)
2313 ext4_ext_mark_uninitialized(ex);
2315 err = ext4_ext_dirty(handle, inode, path + depth);
2319 ext_debug("new extent: %u:%u:%llu\n", block, num,
2322 ex_ee_block = le32_to_cpu(ex->ee_block);
2323 ex_ee_len = ext4_ext_get_actual_len(ex);
2326 if (correct_index && eh->eh_entries)
2327 err = ext4_ext_correct_indexes(handle, inode, path);
2329 /* if this leaf is free, then we should
2330 * remove it from index block above */
2331 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2332 err = ext4_ext_rm_idx(handle, inode, path + depth);
2339 * ext4_ext_more_to_rm:
2340 * returns 1 if current index has to be freed (even partial)
2343 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2345 BUG_ON(path->p_idx == NULL);
2347 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2351 * if truncate on deeper level happened, it wasn't partial,
2352 * so we have to consider current index for truncation
2354 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2359 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2361 struct super_block *sb = inode->i_sb;
2362 int depth = ext_depth(inode);
2363 struct ext4_ext_path *path;
2367 ext_debug("truncate since %u\n", start);
2369 /* probably first extent we're gonna free will be last in block */
2370 handle = ext4_journal_start(inode, depth + 1);
2372 return PTR_ERR(handle);
2374 ext4_ext_invalidate_cache(inode);
2377 * We start scanning from right side, freeing all the blocks
2378 * after i_size and walking into the tree depth-wise.
2380 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2382 ext4_journal_stop(handle);
2385 path[0].p_hdr = ext_inode_hdr(inode);
2386 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2390 path[0].p_depth = depth;
2392 while (i >= 0 && err == 0) {
2394 /* this is leaf block */
2395 err = ext4_ext_rm_leaf(handle, inode, path, start);
2396 /* root level has p_bh == NULL, brelse() eats this */
2397 brelse(path[i].p_bh);
2398 path[i].p_bh = NULL;
2403 /* this is index block */
2404 if (!path[i].p_hdr) {
2405 ext_debug("initialize header\n");
2406 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2409 if (!path[i].p_idx) {
2410 /* this level hasn't been touched yet */
2411 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2412 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2413 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2415 le16_to_cpu(path[i].p_hdr->eh_entries));
2417 /* we were already here, see at next index */
2421 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2422 i, EXT_FIRST_INDEX(path[i].p_hdr),
2424 if (ext4_ext_more_to_rm(path + i)) {
2425 struct buffer_head *bh;
2426 /* go to the next level */
2427 ext_debug("move to level %d (block %llu)\n",
2428 i + 1, idx_pblock(path[i].p_idx));
2429 memset(path + i + 1, 0, sizeof(*path));
2430 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2432 /* should we reset i_size? */
2436 if (WARN_ON(i + 1 > depth)) {
2440 if (ext4_ext_check(inode, ext_block_hdr(bh),
2445 path[i + 1].p_bh = bh;
2447 /* save actual number of indexes since this
2448 * number is changed at the next iteration */
2449 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2452 /* we finished processing this index, go up */
2453 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2454 /* index is empty, remove it;
2455 * handle must be already prepared by the
2456 * truncatei_leaf() */
2457 err = ext4_ext_rm_idx(handle, inode, path + i);
2459 /* root level has p_bh == NULL, brelse() eats this */
2460 brelse(path[i].p_bh);
2461 path[i].p_bh = NULL;
2463 ext_debug("return to level %d\n", i);
2467 /* TODO: flexible tree reduction should be here */
2468 if (path->p_hdr->eh_entries == 0) {
2470 * truncate to zero freed all the tree,
2471 * so we need to correct eh_depth
2473 err = ext4_ext_get_access(handle, inode, path);
2475 ext_inode_hdr(inode)->eh_depth = 0;
2476 ext_inode_hdr(inode)->eh_max =
2477 cpu_to_le16(ext4_ext_space_root(inode, 0));
2478 err = ext4_ext_dirty(handle, inode, path);
2482 ext4_ext_drop_refs(path);
2484 ext4_journal_stop(handle);
2490 * called at mount time
2492 void ext4_ext_init(struct super_block *sb)
2495 * possible initialization would be here
2498 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2499 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2500 printk(KERN_INFO "EXT4-fs: file extents enabled");
2501 #ifdef AGGRESSIVE_TEST
2502 printk(", aggressive tests");
2504 #ifdef CHECK_BINSEARCH
2505 printk(", check binsearch");
2507 #ifdef EXTENTS_STATS
2512 #ifdef EXTENTS_STATS
2513 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2514 EXT4_SB(sb)->s_ext_min = 1 << 30;
2515 EXT4_SB(sb)->s_ext_max = 0;
2521 * called at umount time
2523 void ext4_ext_release(struct super_block *sb)
2525 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2528 #ifdef EXTENTS_STATS
2529 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2530 struct ext4_sb_info *sbi = EXT4_SB(sb);
2531 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2532 sbi->s_ext_blocks, sbi->s_ext_extents,
2533 sbi->s_ext_blocks / sbi->s_ext_extents);
2534 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2535 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2540 static void bi_complete(struct bio *bio, int error)
2542 complete((struct completion *)bio->bi_private);
2545 /* FIXME!! we need to try to merge to left or right after zero-out */
2546 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2550 int blkbits, blocksize;
2552 struct completion event;
2553 unsigned int ee_len, len, done, offset;
2556 blkbits = inode->i_blkbits;
2557 blocksize = inode->i_sb->s_blocksize;
2558 ee_len = ext4_ext_get_actual_len(ex);
2559 ee_pblock = ext_pblock(ex);
2561 /* convert ee_pblock to 512 byte sectors */
2562 ee_pblock = ee_pblock << (blkbits - 9);
2564 while (ee_len > 0) {
2566 if (ee_len > BIO_MAX_PAGES)
2567 len = BIO_MAX_PAGES;
2571 bio = bio_alloc(GFP_NOIO, len);
2572 bio->bi_sector = ee_pblock;
2573 bio->bi_bdev = inode->i_sb->s_bdev;
2577 while (done < len) {
2578 ret = bio_add_page(bio, ZERO_PAGE(0),
2580 if (ret != blocksize) {
2582 * We can't add any more pages because of
2583 * hardware limitations. Start a new bio.
2588 offset += blocksize;
2589 if (offset >= PAGE_CACHE_SIZE)
2593 init_completion(&event);
2594 bio->bi_private = &event;
2595 bio->bi_end_io = bi_complete;
2596 submit_bio(WRITE, bio);
2597 wait_for_completion(&event);
2599 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2607 ee_pblock += done << (blkbits - 9);
2612 #define EXT4_EXT_ZERO_LEN 7
2614 * This function is called by ext4_ext_get_blocks() if someone tries to write
2615 * to an uninitialized extent. It may result in splitting the uninitialized
2616 * extent into multiple extents (upto three - one initialized and two
2618 * There are three possibilities:
2619 * a> There is no split required: Entire extent should be initialized
2620 * b> Splits in two extents: Write is happening at either end of the extent
2621 * c> Splits in three extents: Somone is writing in middle of the extent
2623 static int ext4_ext_convert_to_initialized(handle_t *handle,
2624 struct inode *inode,
2625 struct ext4_ext_path *path,
2627 unsigned int max_blocks)
2629 struct ext4_extent *ex, newex, orig_ex;
2630 struct ext4_extent *ex1 = NULL;
2631 struct ext4_extent *ex2 = NULL;
2632 struct ext4_extent *ex3 = NULL;
2633 struct ext4_extent_header *eh;
2634 ext4_lblk_t ee_block;
2635 unsigned int allocated, ee_len, depth;
2636 ext4_fsblk_t newblock;
2640 depth = ext_depth(inode);
2641 eh = path[depth].p_hdr;
2642 ex = path[depth].p_ext;
2643 ee_block = le32_to_cpu(ex->ee_block);
2644 ee_len = ext4_ext_get_actual_len(ex);
2645 allocated = ee_len - (iblock - ee_block);
2646 newblock = iblock - ee_block + ext_pblock(ex);
2648 orig_ex.ee_block = ex->ee_block;
2649 orig_ex.ee_len = cpu_to_le16(ee_len);
2650 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2652 err = ext4_ext_get_access(handle, inode, path + depth);
2655 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2656 if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2657 err = ext4_ext_zeroout(inode, &orig_ex);
2659 goto fix_extent_len;
2660 /* update the extent length and mark as initialized */
2661 ex->ee_block = orig_ex.ee_block;
2662 ex->ee_len = orig_ex.ee_len;
2663 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2664 ext4_ext_dirty(handle, inode, path + depth);
2665 /* zeroed the full extent */
2669 /* ex1: ee_block to iblock - 1 : uninitialized */
2670 if (iblock > ee_block) {
2672 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2673 ext4_ext_mark_uninitialized(ex1);
2677 * for sanity, update the length of the ex2 extent before
2678 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2679 * overlap of blocks.
2681 if (!ex1 && allocated > max_blocks)
2682 ex2->ee_len = cpu_to_le16(max_blocks);
2683 /* ex3: to ee_block + ee_len : uninitialised */
2684 if (allocated > max_blocks) {
2685 unsigned int newdepth;
2686 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2687 if (allocated <= EXT4_EXT_ZERO_LEN) {
2689 * iblock == ee_block is handled by the zerouout
2691 * Mark first half uninitialized.
2692 * Mark second half initialized and zero out the
2693 * initialized extent
2695 ex->ee_block = orig_ex.ee_block;
2696 ex->ee_len = cpu_to_le16(ee_len - allocated);
2697 ext4_ext_mark_uninitialized(ex);
2698 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2699 ext4_ext_dirty(handle, inode, path + depth);
2702 ex3->ee_block = cpu_to_le32(iblock);
2703 ext4_ext_store_pblock(ex3, newblock);
2704 ex3->ee_len = cpu_to_le16(allocated);
2705 err = ext4_ext_insert_extent(handle, inode, path,
2707 if (err == -ENOSPC) {
2708 err = ext4_ext_zeroout(inode, &orig_ex);
2710 goto fix_extent_len;
2711 ex->ee_block = orig_ex.ee_block;
2712 ex->ee_len = orig_ex.ee_len;
2713 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2714 ext4_ext_dirty(handle, inode, path + depth);
2715 /* blocks available from iblock */
2719 goto fix_extent_len;
2722 * We need to zero out the second half because
2723 * an fallocate request can update file size and
2724 * converting the second half to initialized extent
2725 * implies that we can leak some junk data to user
2728 err = ext4_ext_zeroout(inode, ex3);
2731 * We should actually mark the
2732 * second half as uninit and return error
2733 * Insert would have changed the extent
2735 depth = ext_depth(inode);
2736 ext4_ext_drop_refs(path);
2737 path = ext4_ext_find_extent(inode,
2740 err = PTR_ERR(path);
2743 /* get the second half extent details */
2744 ex = path[depth].p_ext;
2745 err = ext4_ext_get_access(handle, inode,
2749 ext4_ext_mark_uninitialized(ex);
2750 ext4_ext_dirty(handle, inode, path + depth);
2754 /* zeroed the second half */
2758 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2759 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2760 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2761 ext4_ext_mark_uninitialized(ex3);
2762 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2763 if (err == -ENOSPC) {
2764 err = ext4_ext_zeroout(inode, &orig_ex);
2766 goto fix_extent_len;
2767 /* update the extent length and mark as initialized */
2768 ex->ee_block = orig_ex.ee_block;
2769 ex->ee_len = orig_ex.ee_len;
2770 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2771 ext4_ext_dirty(handle, inode, path + depth);
2772 /* zeroed the full extent */
2773 /* blocks available from iblock */
2777 goto fix_extent_len;
2779 * The depth, and hence eh & ex might change
2780 * as part of the insert above.
2782 newdepth = ext_depth(inode);
2784 * update the extent length after successful insert of the
2787 orig_ex.ee_len = cpu_to_le16(ee_len -
2788 ext4_ext_get_actual_len(ex3));
2790 ext4_ext_drop_refs(path);
2791 path = ext4_ext_find_extent(inode, iblock, path);
2793 err = PTR_ERR(path);
2796 eh = path[depth].p_hdr;
2797 ex = path[depth].p_ext;
2801 err = ext4_ext_get_access(handle, inode, path + depth);
2805 allocated = max_blocks;
2807 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2808 * to insert a extent in the middle zerout directly
2809 * otherwise give the extent a chance to merge to left
2811 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2812 iblock != ee_block) {
2813 err = ext4_ext_zeroout(inode, &orig_ex);
2815 goto fix_extent_len;
2816 /* update the extent length and mark as initialized */
2817 ex->ee_block = orig_ex.ee_block;
2818 ex->ee_len = orig_ex.ee_len;
2819 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2820 ext4_ext_dirty(handle, inode, path + depth);
2821 /* zero out the first half */
2822 /* blocks available from iblock */
2827 * If there was a change of depth as part of the
2828 * insertion of ex3 above, we need to update the length
2829 * of the ex1 extent again here
2831 if (ex1 && ex1 != ex) {
2833 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2834 ext4_ext_mark_uninitialized(ex1);
2837 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2838 ex2->ee_block = cpu_to_le32(iblock);
2839 ext4_ext_store_pblock(ex2, newblock);
2840 ex2->ee_len = cpu_to_le16(allocated);
2844 * New (initialized) extent starts from the first block
2845 * in the current extent. i.e., ex2 == ex
2846 * We have to see if it can be merged with the extent
2849 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2851 * To merge left, pass "ex2 - 1" to try_to_merge(),
2852 * since it merges towards right _only_.
2854 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2856 err = ext4_ext_correct_indexes(handle, inode, path);
2859 depth = ext_depth(inode);
2864 * Try to Merge towards right. This might be required
2865 * only when the whole extent is being written to.
2866 * i.e. ex2 == ex and ex3 == NULL.
2869 ret = ext4_ext_try_to_merge(inode, path, ex2);
2871 err = ext4_ext_correct_indexes(handle, inode, path);
2876 /* Mark modified extent as dirty */
2877 err = ext4_ext_dirty(handle, inode, path + depth);
2880 err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2881 if (err == -ENOSPC) {
2882 err = ext4_ext_zeroout(inode, &orig_ex);
2884 goto fix_extent_len;
2885 /* update the extent length and mark as initialized */
2886 ex->ee_block = orig_ex.ee_block;
2887 ex->ee_len = orig_ex.ee_len;
2888 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2889 ext4_ext_dirty(handle, inode, path + depth);
2890 /* zero out the first half */
2893 goto fix_extent_len;
2895 ext4_ext_show_leaf(inode, path);
2896 return err ? err : allocated;
2899 ex->ee_block = orig_ex.ee_block;
2900 ex->ee_len = orig_ex.ee_len;
2901 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2902 ext4_ext_mark_uninitialized(ex);
2903 ext4_ext_dirty(handle, inode, path + depth);
2908 * This function is called by ext4_ext_get_blocks() from
2909 * ext4_get_blocks_dio_write() when DIO to write
2910 * to an uninitialized extent.
2912 * Writing to an uninitized extent may result in splitting the uninitialized
2913 * extent into multiple /intialized unintialized extents (up to three)
2914 * There are three possibilities:
2915 * a> There is no split required: Entire extent should be uninitialized
2916 * b> Splits in two extents: Write is happening at either end of the extent
2917 * c> Splits in three extents: Somone is writing in middle of the extent
2919 * One of more index blocks maybe needed if the extent tree grow after
2920 * the unintialized extent split. To prevent ENOSPC occur at the IO
2921 * complete, we need to split the uninitialized extent before DIO submit
2922 * the IO. The uninitilized extent called at this time will be split
2923 * into three uninitialized extent(at most). After IO complete, the part
2924 * being filled will be convert to initialized by the end_io callback function
2925 * via ext4_convert_unwritten_extents().
2927 * Returns the size of uninitialized extent to be written on success.
2929 static int ext4_split_unwritten_extents(handle_t *handle,
2930 struct inode *inode,
2931 struct ext4_ext_path *path,
2933 unsigned int max_blocks,
2936 struct ext4_extent *ex, newex, orig_ex;
2937 struct ext4_extent *ex1 = NULL;
2938 struct ext4_extent *ex2 = NULL;
2939 struct ext4_extent *ex3 = NULL;
2940 struct ext4_extent_header *eh;
2941 ext4_lblk_t ee_block;
2942 unsigned int allocated, ee_len, depth;
2943 ext4_fsblk_t newblock;
2946 ext_debug("ext4_split_unwritten_extents: inode %lu,"
2947 "iblock %llu, max_blocks %u\n", inode->i_ino,
2948 (unsigned long long)iblock, max_blocks);
2949 depth = ext_depth(inode);
2950 eh = path[depth].p_hdr;
2951 ex = path[depth].p_ext;
2952 ee_block = le32_to_cpu(ex->ee_block);
2953 ee_len = ext4_ext_get_actual_len(ex);
2954 allocated = ee_len - (iblock - ee_block);
2955 newblock = iblock - ee_block + ext_pblock(ex);
2957 orig_ex.ee_block = ex->ee_block;
2958 orig_ex.ee_len = cpu_to_le16(ee_len);
2959 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2962 * If the uninitialized extent begins at the same logical
2963 * block where the write begins, and the write completely
2964 * covers the extent, then we don't need to split it.
2966 if ((iblock == ee_block) && (allocated <= max_blocks))
2969 err = ext4_ext_get_access(handle, inode, path + depth);
2972 /* ex1: ee_block to iblock - 1 : uninitialized */
2973 if (iblock > ee_block) {
2975 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2976 ext4_ext_mark_uninitialized(ex1);
2980 * for sanity, update the length of the ex2 extent before
2981 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2982 * overlap of blocks.
2984 if (!ex1 && allocated > max_blocks)
2985 ex2->ee_len = cpu_to_le16(max_blocks);
2986 /* ex3: to ee_block + ee_len : uninitialised */
2987 if (allocated > max_blocks) {
2988 unsigned int newdepth;
2990 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2991 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2992 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2993 ext4_ext_mark_uninitialized(ex3);
2994 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2995 if (err == -ENOSPC) {
2996 err = ext4_ext_zeroout(inode, &orig_ex);
2998 goto fix_extent_len;
2999 /* update the extent length and mark as initialized */
3000 ex->ee_block = orig_ex.ee_block;
3001 ex->ee_len = orig_ex.ee_len;
3002 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3003 ext4_ext_dirty(handle, inode, path + depth);
3004 /* zeroed the full extent */
3005 /* blocks available from iblock */
3009 goto fix_extent_len;
3011 * The depth, and hence eh & ex might change
3012 * as part of the insert above.
3014 newdepth = ext_depth(inode);
3016 * update the extent length after successful insert of the
3019 orig_ex.ee_len = cpu_to_le16(ee_len -
3020 ext4_ext_get_actual_len(ex3));
3022 ext4_ext_drop_refs(path);
3023 path = ext4_ext_find_extent(inode, iblock, path);
3025 err = PTR_ERR(path);
3028 eh = path[depth].p_hdr;
3029 ex = path[depth].p_ext;
3033 err = ext4_ext_get_access(handle, inode, path + depth);
3037 allocated = max_blocks;
3040 * If there was a change of depth as part of the
3041 * insertion of ex3 above, we need to update the length
3042 * of the ex1 extent again here
3044 if (ex1 && ex1 != ex) {
3046 ex1->ee_len = cpu_to_le16(iblock - ee_block);
3047 ext4_ext_mark_uninitialized(ex1);
3051 * ex2: iblock to iblock + maxblocks-1 : to be direct IO written,
3052 * uninitialised still.
3054 ex2->ee_block = cpu_to_le32(iblock);
3055 ext4_ext_store_pblock(ex2, newblock);
3056 ex2->ee_len = cpu_to_le16(allocated);
3057 ext4_ext_mark_uninitialized(ex2);
3060 /* Mark modified extent as dirty */
3061 err = ext4_ext_dirty(handle, inode, path + depth);
3062 ext_debug("out here\n");
3065 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3066 if (err == -ENOSPC) {
3067 err = ext4_ext_zeroout(inode, &orig_ex);
3069 goto fix_extent_len;
3070 /* update the extent length and mark as initialized */
3071 ex->ee_block = orig_ex.ee_block;
3072 ex->ee_len = orig_ex.ee_len;
3073 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3074 ext4_ext_dirty(handle, inode, path + depth);
3075 /* zero out the first half */
3078 goto fix_extent_len;
3080 ext4_ext_show_leaf(inode, path);
3081 return err ? err : allocated;
3084 ex->ee_block = orig_ex.ee_block;
3085 ex->ee_len = orig_ex.ee_len;
3086 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3087 ext4_ext_mark_uninitialized(ex);
3088 ext4_ext_dirty(handle, inode, path + depth);
3091 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3092 struct inode *inode,
3093 struct ext4_ext_path *path)
3095 struct ext4_extent *ex;
3096 struct ext4_extent_header *eh;
3101 depth = ext_depth(inode);
3102 eh = path[depth].p_hdr;
3103 ex = path[depth].p_ext;
3105 err = ext4_ext_get_access(handle, inode, path + depth);
3108 /* first mark the extent as initialized */
3109 ext4_ext_mark_initialized(ex);
3112 * We have to see if it can be merged with the extent
3115 if (ex > EXT_FIRST_EXTENT(eh)) {
3117 * To merge left, pass "ex - 1" to try_to_merge(),
3118 * since it merges towards right _only_.
3120 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3122 err = ext4_ext_correct_indexes(handle, inode, path);
3125 depth = ext_depth(inode);
3130 * Try to Merge towards right.
3132 ret = ext4_ext_try_to_merge(inode, path, ex);
3134 err = ext4_ext_correct_indexes(handle, inode, path);
3137 depth = ext_depth(inode);
3139 /* Mark modified extent as dirty */
3140 err = ext4_ext_dirty(handle, inode, path + depth);
3142 ext4_ext_show_leaf(inode, path);
3146 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3147 sector_t block, int count)
3150 for (i = 0; i < count; i++)
3151 unmap_underlying_metadata(bdev, block + i);
3155 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3156 ext4_lblk_t iblock, unsigned int max_blocks,
3157 struct ext4_ext_path *path, int flags,
3158 unsigned int allocated, struct buffer_head *bh_result,
3159 ext4_fsblk_t newblock)
3163 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3165 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3166 "block %llu, max_blocks %u, flags %d, allocated %u",
3167 inode->i_ino, (unsigned long long)iblock, max_blocks,
3169 ext4_ext_show_leaf(inode, path);
3171 /* get_block() before submit the IO, split the extent */
3172 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3173 ret = ext4_split_unwritten_extents(handle,
3174 inode, path, iblock,
3177 * Flag the inode(non aio case) or end_io struct (aio case)
3178 * that this IO needs to convertion to written when IO is
3182 io->flag = EXT4_IO_UNWRITTEN;
3184 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3185 if (ext4_should_dioread_nolock(inode))
3186 set_buffer_uninit(bh_result);
3189 /* IO end_io complete, convert the filled extent to written */
3190 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3191 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3194 ext4_update_inode_fsync_trans(handle, inode, 1);
3197 /* buffered IO case */
3199 * repeat fallocate creation request
3200 * we already have an unwritten extent
3202 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3205 /* buffered READ or buffered write_begin() lookup */
3206 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3208 * We have blocks reserved already. We
3209 * return allocated blocks so that delalloc
3210 * won't do block reservation for us. But
3211 * the buffer head will be unmapped so that
3212 * a read from the block returns 0s.
3214 set_buffer_unwritten(bh_result);
3218 /* buffered write, writepage time, convert*/
3219 ret = ext4_ext_convert_to_initialized(handle, inode,
3223 ext4_update_inode_fsync_trans(handle, inode, 1);
3230 set_buffer_new(bh_result);
3232 * if we allocated more blocks than requested
3233 * we need to make sure we unmap the extra block
3234 * allocated. The actual needed block will get
3235 * unmapped later when we find the buffer_head marked
3238 if (allocated > max_blocks) {
3239 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3240 newblock + max_blocks,
3241 allocated - max_blocks);
3242 allocated = max_blocks;
3246 * If we have done fallocate with the offset that is already
3247 * delayed allocated, we would have block reservation
3248 * and quota reservation done in the delayed write path.
3249 * But fallocate would have already updated quota and block
3250 * count for this offset. So cancel these reservation
3252 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3253 ext4_da_update_reserve_space(inode, allocated, 0);
3256 set_buffer_mapped(bh_result);
3258 if (allocated > max_blocks)
3259 allocated = max_blocks;
3260 ext4_ext_show_leaf(inode, path);
3261 bh_result->b_bdev = inode->i_sb->s_bdev;
3262 bh_result->b_blocknr = newblock;
3265 ext4_ext_drop_refs(path);
3268 return err ? err : allocated;
3271 * Block allocation/map/preallocation routine for extents based files
3274 * Need to be called with
3275 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3276 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3278 * return > 0, number of of blocks already mapped/allocated
3279 * if create == 0 and these are pre-allocated blocks
3280 * buffer head is unmapped
3281 * otherwise blocks are mapped
3283 * return = 0, if plain look up failed (blocks have not been allocated)
3284 * buffer head is unmapped
3286 * return < 0, error case.
3288 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3290 unsigned int max_blocks, struct buffer_head *bh_result,
3293 struct ext4_ext_path *path = NULL;
3294 struct ext4_extent_header *eh;
3295 struct ext4_extent newex, *ex, *last_ex;
3296 ext4_fsblk_t newblock;
3297 int err = 0, depth, ret, cache_type;
3298 unsigned int allocated = 0;
3299 struct ext4_allocation_request ar;
3300 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3302 __clear_bit(BH_New, &bh_result->b_state);
3303 ext_debug("blocks %u/%u requested for inode %lu\n",
3304 iblock, max_blocks, inode->i_ino);
3306 /* check in cache */
3307 cache_type = ext4_ext_in_cache(inode, iblock, &newex);
3309 if (cache_type == EXT4_EXT_CACHE_GAP) {
3310 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3312 * block isn't allocated yet and
3313 * user doesn't want to allocate it
3317 /* we should allocate requested block */
3318 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3319 /* block is already allocated */
3321 - le32_to_cpu(newex.ee_block)
3322 + ext_pblock(&newex);
3323 /* number of remaining blocks in the extent */
3324 allocated = ext4_ext_get_actual_len(&newex) -
3325 (iblock - le32_to_cpu(newex.ee_block));
3332 /* find extent for this block */
3333 path = ext4_ext_find_extent(inode, iblock, NULL);
3335 err = PTR_ERR(path);
3340 depth = ext_depth(inode);
3343 * consistent leaf must not be empty;
3344 * this situation is possible, though, _during_ tree modification;
3345 * this is why assert can't be put in ext4_ext_find_extent()
3347 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3348 EXT4_ERROR_INODE(inode, "bad extent address "
3349 "iblock: %d, depth: %d pblock %lld",
3350 iblock, depth, path[depth].p_block);
3354 eh = path[depth].p_hdr;
3356 ex = path[depth].p_ext;
3358 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3359 ext4_fsblk_t ee_start = ext_pblock(ex);
3360 unsigned short ee_len;
3363 * Uninitialized extents are treated as holes, except that
3364 * we split out initialized portions during a write.
3366 ee_len = ext4_ext_get_actual_len(ex);
3367 /* if found extent covers block, simply return it */
3368 if (iblock >= ee_block && iblock < ee_block + ee_len) {
3369 newblock = iblock - ee_block + ee_start;
3370 /* number of remaining blocks in the extent */
3371 allocated = ee_len - (iblock - ee_block);
3372 ext_debug("%u fit into %u:%d -> %llu\n", iblock,
3373 ee_block, ee_len, newblock);
3375 /* Do not put uninitialized extent in the cache */
3376 if (!ext4_ext_is_uninitialized(ex)) {
3377 ext4_ext_put_in_cache(inode, ee_block,
3379 EXT4_EXT_CACHE_EXTENT);
3382 ret = ext4_ext_handle_uninitialized_extents(handle,
3383 inode, iblock, max_blocks, path,
3384 flags, allocated, bh_result, newblock);
3390 * requested block isn't allocated yet;
3391 * we couldn't try to create block if create flag is zero
3393 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3395 * put just found gap into cache to speed up
3396 * subsequent requests
3398 ext4_ext_put_gap_in_cache(inode, path, iblock);
3402 * Okay, we need to do block allocation.
3405 /* find neighbour allocated blocks */
3407 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3411 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3416 * See if request is beyond maximum number of blocks we can have in
3417 * a single extent. For an initialized extent this limit is
3418 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3419 * EXT_UNINIT_MAX_LEN.
3421 if (max_blocks > EXT_INIT_MAX_LEN &&
3422 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3423 max_blocks = EXT_INIT_MAX_LEN;
3424 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
3425 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3426 max_blocks = EXT_UNINIT_MAX_LEN;
3428 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
3429 newex.ee_block = cpu_to_le32(iblock);
3430 newex.ee_len = cpu_to_le16(max_blocks);
3431 err = ext4_ext_check_overlap(inode, &newex, path);
3433 allocated = ext4_ext_get_actual_len(&newex);
3435 allocated = max_blocks;
3437 /* allocate new block */
3439 ar.goal = ext4_ext_find_goal(inode, path, iblock);
3440 ar.logical = iblock;
3442 if (S_ISREG(inode->i_mode))
3443 ar.flags = EXT4_MB_HINT_DATA;
3445 /* disable in-core preallocation for non-regular files */
3447 newblock = ext4_mb_new_blocks(handle, &ar, &err);
3450 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3451 ar.goal, newblock, allocated);
3453 /* try to insert new extent into found leaf and return */
3454 ext4_ext_store_pblock(&newex, newblock);
3455 newex.ee_len = cpu_to_le16(ar.len);
3456 /* Mark uninitialized */
3457 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3458 ext4_ext_mark_uninitialized(&newex);
3460 * io_end structure was created for every IO write to an
3461 * uninitialized extent. To avoid unecessary conversion,
3462 * here we flag the IO that really needs the conversion.
3463 * For non asycn direct IO case, flag the inode state
3464 * that we need to perform convertion when IO is done.
3466 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3468 io->flag = EXT4_IO_UNWRITTEN;
3470 ext4_set_inode_state(inode,
3471 EXT4_STATE_DIO_UNWRITTEN);
3473 if (ext4_should_dioread_nolock(inode))
3474 set_buffer_uninit(bh_result);
3477 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL)) {
3478 if (unlikely(!eh->eh_entries)) {
3479 EXT4_ERROR_INODE(inode,
3480 "eh->eh_entries == 0 ee_block %d",
3485 last_ex = EXT_LAST_EXTENT(eh);
3486 if (iblock + ar.len > le32_to_cpu(last_ex->ee_block)
3487 + ext4_ext_get_actual_len(last_ex))
3488 EXT4_I(inode)->i_flags &= ~EXT4_EOFBLOCKS_FL;
3490 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3492 /* free data blocks we just allocated */
3493 /* not a good idea to call discard here directly,
3494 * but otherwise we'd need to call it every free() */
3495 ext4_discard_preallocations(inode);
3496 ext4_free_blocks(handle, inode, 0, ext_pblock(&newex),
3497 ext4_ext_get_actual_len(&newex), 0);
3501 /* previous routine could use block we allocated */
3502 newblock = ext_pblock(&newex);
3503 allocated = ext4_ext_get_actual_len(&newex);
3504 if (allocated > max_blocks)
3505 allocated = max_blocks;
3506 set_buffer_new(bh_result);
3509 * Update reserved blocks/metadata blocks after successful
3510 * block allocation which had been deferred till now.
3512 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3513 ext4_da_update_reserve_space(inode, allocated, 1);
3516 * Cache the extent and update transaction to commit on fdatasync only
3517 * when it is _not_ an uninitialized extent.
3519 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3520 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
3521 EXT4_EXT_CACHE_EXTENT);
3522 ext4_update_inode_fsync_trans(handle, inode, 1);
3524 ext4_update_inode_fsync_trans(handle, inode, 0);
3526 if (allocated > max_blocks)
3527 allocated = max_blocks;
3528 ext4_ext_show_leaf(inode, path);
3529 set_buffer_mapped(bh_result);
3530 bh_result->b_bdev = inode->i_sb->s_bdev;
3531 bh_result->b_blocknr = newblock;
3534 ext4_ext_drop_refs(path);
3537 return err ? err : allocated;
3540 void ext4_ext_truncate(struct inode *inode)
3542 struct address_space *mapping = inode->i_mapping;
3543 struct super_block *sb = inode->i_sb;
3544 ext4_lblk_t last_block;
3549 * probably first extent we're gonna free will be last in block
3551 err = ext4_writepage_trans_blocks(inode);
3552 handle = ext4_journal_start(inode, err);
3556 if (inode->i_size & (sb->s_blocksize - 1))
3557 ext4_block_truncate_page(handle, mapping, inode->i_size);
3559 if (ext4_orphan_add(handle, inode))
3562 down_write(&EXT4_I(inode)->i_data_sem);
3563 ext4_ext_invalidate_cache(inode);
3565 ext4_discard_preallocations(inode);
3568 * TODO: optimization is possible here.
3569 * Probably we need not scan at all,
3570 * because page truncation is enough.
3573 /* we have to know where to truncate from in crash case */
3574 EXT4_I(inode)->i_disksize = inode->i_size;
3575 ext4_mark_inode_dirty(handle, inode);
3577 last_block = (inode->i_size + sb->s_blocksize - 1)
3578 >> EXT4_BLOCK_SIZE_BITS(sb);
3579 err = ext4_ext_remove_space(inode, last_block);
3581 /* In a multi-transaction truncate, we only make the final
3582 * transaction synchronous.
3585 ext4_handle_sync(handle);
3588 up_write(&EXT4_I(inode)->i_data_sem);
3590 * If this was a simple ftruncate() and the file will remain alive,
3591 * then we need to clear up the orphan record which we created above.
3592 * However, if this was a real unlink then we were called by
3593 * ext4_delete_inode(), and we allow that function to clean up the
3594 * orphan info for us.
3597 ext4_orphan_del(handle, inode);
3599 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3600 ext4_mark_inode_dirty(handle, inode);
3601 ext4_journal_stop(handle);
3604 static void ext4_falloc_update_inode(struct inode *inode,
3605 int mode, loff_t new_size, int update_ctime)
3607 struct timespec now;
3610 now = current_fs_time(inode->i_sb);
3611 if (!timespec_equal(&inode->i_ctime, &now))
3612 inode->i_ctime = now;
3615 * Update only when preallocation was requested beyond
3618 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3619 if (new_size > i_size_read(inode))
3620 i_size_write(inode, new_size);
3621 if (new_size > EXT4_I(inode)->i_disksize)
3622 ext4_update_i_disksize(inode, new_size);
3625 * Mark that we allocate beyond EOF so the subsequent truncate
3626 * can proceed even if the new size is the same as i_size.
3628 if (new_size > i_size_read(inode))
3629 EXT4_I(inode)->i_flags |= EXT4_EOFBLOCKS_FL;
3635 * preallocate space for a file. This implements ext4's fallocate inode
3636 * operation, which gets called from sys_fallocate system call.
3637 * For block-mapped files, posix_fallocate should fall back to the method
3638 * of writing zeroes to the required new blocks (the same behavior which is
3639 * expected for file systems which do not support fallocate() system call).
3641 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3646 unsigned int max_blocks;
3650 struct buffer_head map_bh;
3651 unsigned int credits, blkbits = inode->i_blkbits;
3654 * currently supporting (pre)allocate mode for extent-based
3657 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3660 /* preallocation to directories is currently not supported */
3661 if (S_ISDIR(inode->i_mode))
3664 block = offset >> blkbits;
3666 * We can't just convert len to max_blocks because
3667 * If blocksize = 4096 offset = 3072 and len = 2048
3669 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3672 * credits to insert 1 extent into extent tree
3674 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3675 mutex_lock(&inode->i_mutex);
3677 while (ret >= 0 && ret < max_blocks) {
3678 block = block + ret;
3679 max_blocks = max_blocks - ret;
3680 handle = ext4_journal_start(inode, credits);
3681 if (IS_ERR(handle)) {
3682 ret = PTR_ERR(handle);
3686 ret = ext4_get_blocks(handle, inode, block,
3687 max_blocks, &map_bh,
3688 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3692 printk(KERN_ERR "%s: ext4_ext_get_blocks "
3693 "returned error inode#%lu, block=%u, "
3694 "max_blocks=%u", __func__,
3695 inode->i_ino, block, max_blocks);
3697 ext4_mark_inode_dirty(handle, inode);
3698 ret2 = ext4_journal_stop(handle);
3701 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3702 blkbits) >> blkbits))
3703 new_size = offset + len;
3705 new_size = (block + ret) << blkbits;
3707 ext4_falloc_update_inode(inode, mode, new_size,
3708 buffer_new(&map_bh));
3709 ext4_mark_inode_dirty(handle, inode);
3710 ret2 = ext4_journal_stop(handle);
3714 if (ret == -ENOSPC &&
3715 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3719 mutex_unlock(&inode->i_mutex);
3720 return ret > 0 ? ret2 : ret;
3724 * This function convert a range of blocks to written extents
3725 * The caller of this function will pass the start offset and the size.
3726 * all unwritten extents within this range will be converted to
3729 * This function is called from the direct IO end io call back
3730 * function, to convert the fallocated extents after IO is completed.
3731 * Returns 0 on success.
3733 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3738 unsigned int max_blocks;
3741 struct buffer_head map_bh;
3742 unsigned int credits, blkbits = inode->i_blkbits;
3744 block = offset >> blkbits;
3746 * We can't just convert len to max_blocks because
3747 * If blocksize = 4096 offset = 3072 and len = 2048
3749 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3752 * credits to insert 1 extent into extent tree
3754 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3755 while (ret >= 0 && ret < max_blocks) {
3756 block = block + ret;
3757 max_blocks = max_blocks - ret;
3758 handle = ext4_journal_start(inode, credits);
3759 if (IS_ERR(handle)) {
3760 ret = PTR_ERR(handle);
3764 ret = ext4_get_blocks(handle, inode, block,
3765 max_blocks, &map_bh,
3766 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3769 printk(KERN_ERR "%s: ext4_ext_get_blocks "
3770 "returned error inode#%lu, block=%u, "
3771 "max_blocks=%u", __func__,
3772 inode->i_ino, block, max_blocks);
3774 ext4_mark_inode_dirty(handle, inode);
3775 ret2 = ext4_journal_stop(handle);
3776 if (ret <= 0 || ret2 )
3779 return ret > 0 ? ret2 : ret;
3782 * Callback function called for each extent to gather FIEMAP information.
3784 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3785 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3788 struct fiemap_extent_info *fieinfo = data;
3789 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3796 logical = (__u64)newex->ec_block << blksize_bits;
3798 if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3801 struct buffer_head *bh = NULL;
3803 offset = logical >> PAGE_SHIFT;
3804 page = find_get_page(inode->i_mapping, offset);
3805 if (!page || !page_has_buffers(page))
3806 return EXT_CONTINUE;
3808 bh = page_buffers(page);
3811 return EXT_CONTINUE;
3813 if (buffer_delay(bh)) {
3814 flags |= FIEMAP_EXTENT_DELALLOC;
3815 page_cache_release(page);
3817 page_cache_release(page);
3818 return EXT_CONTINUE;
3822 physical = (__u64)newex->ec_start << blksize_bits;
3823 length = (__u64)newex->ec_len << blksize_bits;
3825 if (ex && ext4_ext_is_uninitialized(ex))
3826 flags |= FIEMAP_EXTENT_UNWRITTEN;
3829 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3831 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3832 * this also indicates no more allocated blocks.
3834 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3836 if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3837 newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3838 loff_t size = i_size_read(inode);
3839 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3841 flags |= FIEMAP_EXTENT_LAST;
3842 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3843 logical+length > size)
3844 length = (size - logical + bs - 1) & ~(bs-1);
3847 error = fiemap_fill_next_extent(fieinfo, logical, physical,
3854 return EXT_CONTINUE;
3857 /* fiemap flags we can handle specified here */
3858 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3860 static int ext4_xattr_fiemap(struct inode *inode,
3861 struct fiemap_extent_info *fieinfo)
3865 __u32 flags = FIEMAP_EXTENT_LAST;
3866 int blockbits = inode->i_sb->s_blocksize_bits;
3870 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3871 struct ext4_iloc iloc;
3872 int offset; /* offset of xattr in inode */
3874 error = ext4_get_inode_loc(inode, &iloc);
3877 physical = iloc.bh->b_blocknr << blockbits;
3878 offset = EXT4_GOOD_OLD_INODE_SIZE +
3879 EXT4_I(inode)->i_extra_isize;
3881 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3882 flags |= FIEMAP_EXTENT_DATA_INLINE;
3883 } else { /* external block */
3884 physical = EXT4_I(inode)->i_file_acl << blockbits;
3885 length = inode->i_sb->s_blocksize;
3889 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3891 return (error < 0 ? error : 0);
3894 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3895 __u64 start, __u64 len)
3897 ext4_lblk_t start_blk;
3900 /* fallback to generic here if not in extents fmt */
3901 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3902 return generic_block_fiemap(inode, fieinfo, start, len,
3905 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3908 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3909 error = ext4_xattr_fiemap(inode, fieinfo);
3911 ext4_lblk_t len_blks;
3914 start_blk = start >> inode->i_sb->s_blocksize_bits;
3915 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
3916 if (last_blk >= EXT_MAX_BLOCK)
3917 last_blk = EXT_MAX_BLOCK-1;
3918 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
3921 * Walk the extent tree gathering extent information.
3922 * ext4_ext_fiemap_cb will push extents back to user.
3924 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3925 ext4_ext_fiemap_cb, fieinfo);