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 static 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_journal_restart(handle_t *handle, int needed)
100 if (!ext4_handle_valid(handle))
102 if (handle->h_buffer_credits > needed)
104 err = ext4_journal_extend(handle, needed);
107 return ext4_journal_restart(handle, needed);
115 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
116 struct ext4_ext_path *path)
119 /* path points to block */
120 return ext4_journal_get_write_access(handle, path->p_bh);
122 /* path points to leaf/index in inode body */
123 /* we use in-core data, no need to protect them */
133 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
134 struct ext4_ext_path *path)
138 /* path points to block */
139 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
141 /* path points to leaf/index in inode body */
142 err = ext4_mark_inode_dirty(handle, inode);
147 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
148 struct ext4_ext_path *path,
151 struct ext4_inode_info *ei = EXT4_I(inode);
152 ext4_fsblk_t bg_start;
153 ext4_fsblk_t last_block;
154 ext4_grpblk_t colour;
155 ext4_group_t block_group;
156 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
160 struct ext4_extent *ex;
161 depth = path->p_depth;
163 /* try to predict block placement */
164 ex = path[depth].p_ext;
166 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
168 /* it looks like index is empty;
169 * try to find starting block from index itself */
170 if (path[depth].p_bh)
171 return path[depth].p_bh->b_blocknr;
174 /* OK. use inode's group */
175 block_group = ei->i_block_group;
176 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
178 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
179 * block groups per flexgroup, reserve the first block
180 * group for directories and special files. Regular
181 * files will start at the second block group. This
182 * tends to speed up directory access and improves
185 block_group &= ~(flex_size-1);
186 if (S_ISREG(inode->i_mode))
189 bg_start = (block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
190 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
191 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
194 * If we are doing delayed allocation, we don't need take
195 * colour into account.
197 if (test_opt(inode->i_sb, DELALLOC))
200 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
201 colour = (current->pid % 16) *
202 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
204 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
205 return bg_start + colour + block;
209 * Allocation for a meta data block
212 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
213 struct ext4_ext_path *path,
214 struct ext4_extent *ex, int *err)
216 ext4_fsblk_t goal, newblock;
218 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
219 newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
223 static int ext4_ext_space_block(struct inode *inode)
227 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
228 / sizeof(struct ext4_extent);
229 #ifdef AGGRESSIVE_TEST
236 static int ext4_ext_space_block_idx(struct inode *inode)
240 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
241 / sizeof(struct ext4_extent_idx);
242 #ifdef AGGRESSIVE_TEST
249 static int ext4_ext_space_root(struct inode *inode)
253 size = sizeof(EXT4_I(inode)->i_data);
254 size -= sizeof(struct ext4_extent_header);
255 size /= sizeof(struct ext4_extent);
256 #ifdef AGGRESSIVE_TEST
263 static int ext4_ext_space_root_idx(struct inode *inode)
267 size = sizeof(EXT4_I(inode)->i_data);
268 size -= sizeof(struct ext4_extent_header);
269 size /= sizeof(struct ext4_extent_idx);
270 #ifdef AGGRESSIVE_TEST
278 * Calculate the number of metadata blocks needed
279 * to allocate @blocks
280 * Worse case is one block per extent
282 int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
284 int lcap, icap, rcap, leafs, idxs, num;
285 int newextents = blocks;
287 rcap = ext4_ext_space_root_idx(inode);
288 lcap = ext4_ext_space_block(inode);
289 icap = ext4_ext_space_block_idx(inode);
291 /* number of new leaf blocks needed */
292 num = leafs = (newextents + lcap - 1) / lcap;
295 * Worse case, we need separate index block(s)
296 * to link all new leaf blocks
298 idxs = (leafs + icap - 1) / icap;
301 idxs = (idxs + icap - 1) / icap;
302 } while (idxs > rcap);
308 ext4_ext_max_entries(struct inode *inode, int depth)
312 if (depth == ext_depth(inode)) {
314 max = ext4_ext_space_root(inode);
316 max = ext4_ext_space_root_idx(inode);
319 max = ext4_ext_space_block(inode);
321 max = ext4_ext_space_block_idx(inode);
327 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
329 ext4_fsblk_t block = ext_pblock(ext);
330 int len = ext4_ext_get_actual_len(ext);
332 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
335 static int ext4_valid_extent_idx(struct inode *inode,
336 struct ext4_extent_idx *ext_idx)
338 ext4_fsblk_t block = idx_pblock(ext_idx);
340 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
343 static int ext4_valid_extent_entries(struct inode *inode,
344 struct ext4_extent_header *eh,
347 struct ext4_extent *ext;
348 struct ext4_extent_idx *ext_idx;
349 unsigned short entries;
350 if (eh->eh_entries == 0)
353 entries = le16_to_cpu(eh->eh_entries);
357 ext = EXT_FIRST_EXTENT(eh);
359 if (!ext4_valid_extent(inode, ext))
365 ext_idx = EXT_FIRST_INDEX(eh);
367 if (!ext4_valid_extent_idx(inode, ext_idx))
376 static int __ext4_ext_check(const char *function, struct inode *inode,
377 struct ext4_extent_header *eh,
380 const char *error_msg;
383 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
384 error_msg = "invalid magic";
387 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
388 error_msg = "unexpected eh_depth";
391 if (unlikely(eh->eh_max == 0)) {
392 error_msg = "invalid eh_max";
395 max = ext4_ext_max_entries(inode, depth);
396 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
397 error_msg = "too large eh_max";
400 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
401 error_msg = "invalid eh_entries";
404 if (!ext4_valid_extent_entries(inode, eh, depth)) {
405 error_msg = "invalid extent entries";
411 ext4_error(inode->i_sb, function,
412 "bad header/extent in inode #%lu: %s - magic %x, "
413 "entries %u, max %u(%u), depth %u(%u)",
414 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
415 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
416 max, le16_to_cpu(eh->eh_depth), depth);
421 #define ext4_ext_check(inode, eh, depth) \
422 __ext4_ext_check(__func__, inode, eh, depth)
424 int ext4_ext_check_inode(struct inode *inode)
426 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
430 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
432 int k, l = path->p_depth;
435 for (k = 0; k <= l; k++, path++) {
437 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
438 idx_pblock(path->p_idx));
439 } else if (path->p_ext) {
440 ext_debug(" %d:%d:%llu ",
441 le32_to_cpu(path->p_ext->ee_block),
442 ext4_ext_get_actual_len(path->p_ext),
443 ext_pblock(path->p_ext));
450 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
452 int depth = ext_depth(inode);
453 struct ext4_extent_header *eh;
454 struct ext4_extent *ex;
460 eh = path[depth].p_hdr;
461 ex = EXT_FIRST_EXTENT(eh);
463 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
464 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
465 ext4_ext_get_actual_len(ex), ext_pblock(ex));
470 #define ext4_ext_show_path(inode, path)
471 #define ext4_ext_show_leaf(inode, path)
474 void ext4_ext_drop_refs(struct ext4_ext_path *path)
476 int depth = path->p_depth;
479 for (i = 0; i <= depth; i++, path++)
487 * ext4_ext_binsearch_idx:
488 * binary search for the closest index of the given block
489 * the header must be checked before calling this
492 ext4_ext_binsearch_idx(struct inode *inode,
493 struct ext4_ext_path *path, ext4_lblk_t block)
495 struct ext4_extent_header *eh = path->p_hdr;
496 struct ext4_extent_idx *r, *l, *m;
499 ext_debug("binsearch for %u(idx): ", block);
501 l = EXT_FIRST_INDEX(eh) + 1;
502 r = EXT_LAST_INDEX(eh);
505 if (block < le32_to_cpu(m->ei_block))
509 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
510 m, le32_to_cpu(m->ei_block),
511 r, le32_to_cpu(r->ei_block));
515 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
516 idx_pblock(path->p_idx));
518 #ifdef CHECK_BINSEARCH
520 struct ext4_extent_idx *chix, *ix;
523 chix = ix = EXT_FIRST_INDEX(eh);
524 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
526 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
527 printk(KERN_DEBUG "k=%d, ix=0x%p, "
529 ix, EXT_FIRST_INDEX(eh));
530 printk(KERN_DEBUG "%u <= %u\n",
531 le32_to_cpu(ix->ei_block),
532 le32_to_cpu(ix[-1].ei_block));
534 BUG_ON(k && le32_to_cpu(ix->ei_block)
535 <= le32_to_cpu(ix[-1].ei_block));
536 if (block < le32_to_cpu(ix->ei_block))
540 BUG_ON(chix != path->p_idx);
547 * ext4_ext_binsearch:
548 * binary search for closest extent of the given block
549 * the header must be checked before calling this
552 ext4_ext_binsearch(struct inode *inode,
553 struct ext4_ext_path *path, ext4_lblk_t block)
555 struct ext4_extent_header *eh = path->p_hdr;
556 struct ext4_extent *r, *l, *m;
558 if (eh->eh_entries == 0) {
560 * this leaf is empty:
561 * we get such a leaf in split/add case
566 ext_debug("binsearch for %u: ", block);
568 l = EXT_FIRST_EXTENT(eh) + 1;
569 r = EXT_LAST_EXTENT(eh);
573 if (block < le32_to_cpu(m->ee_block))
577 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
578 m, le32_to_cpu(m->ee_block),
579 r, le32_to_cpu(r->ee_block));
583 ext_debug(" -> %d:%llu:%d ",
584 le32_to_cpu(path->p_ext->ee_block),
585 ext_pblock(path->p_ext),
586 ext4_ext_get_actual_len(path->p_ext));
588 #ifdef CHECK_BINSEARCH
590 struct ext4_extent *chex, *ex;
593 chex = ex = EXT_FIRST_EXTENT(eh);
594 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
595 BUG_ON(k && le32_to_cpu(ex->ee_block)
596 <= le32_to_cpu(ex[-1].ee_block));
597 if (block < le32_to_cpu(ex->ee_block))
601 BUG_ON(chex != path->p_ext);
607 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
609 struct ext4_extent_header *eh;
611 eh = ext_inode_hdr(inode);
614 eh->eh_magic = EXT4_EXT_MAGIC;
615 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
616 ext4_mark_inode_dirty(handle, inode);
617 ext4_ext_invalidate_cache(inode);
621 struct ext4_ext_path *
622 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
623 struct ext4_ext_path *path)
625 struct ext4_extent_header *eh;
626 struct buffer_head *bh;
627 short int depth, i, ppos = 0, alloc = 0;
629 eh = ext_inode_hdr(inode);
630 depth = ext_depth(inode);
632 /* account possible depth increase */
634 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
637 return ERR_PTR(-ENOMEM);
644 /* walk through the tree */
646 int need_to_validate = 0;
648 ext_debug("depth %d: num %d, max %d\n",
649 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
651 ext4_ext_binsearch_idx(inode, path + ppos, block);
652 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
653 path[ppos].p_depth = i;
654 path[ppos].p_ext = NULL;
656 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
659 if (!bh_uptodate_or_lock(bh)) {
660 if (bh_submit_read(bh) < 0) {
664 /* validate the extent entries */
665 need_to_validate = 1;
667 eh = ext_block_hdr(bh);
669 BUG_ON(ppos > depth);
670 path[ppos].p_bh = bh;
671 path[ppos].p_hdr = eh;
674 if (need_to_validate && ext4_ext_check(inode, eh, i))
678 path[ppos].p_depth = i;
679 path[ppos].p_ext = NULL;
680 path[ppos].p_idx = NULL;
683 ext4_ext_binsearch(inode, path + ppos, block);
684 /* if not an empty leaf */
685 if (path[ppos].p_ext)
686 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
688 ext4_ext_show_path(inode, path);
693 ext4_ext_drop_refs(path);
696 return ERR_PTR(-EIO);
700 * ext4_ext_insert_index:
701 * insert new index [@logical;@ptr] into the block at @curp;
702 * check where to insert: before @curp or after @curp
704 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
705 struct ext4_ext_path *curp,
706 int logical, ext4_fsblk_t ptr)
708 struct ext4_extent_idx *ix;
711 err = ext4_ext_get_access(handle, inode, curp);
715 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
716 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
717 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
719 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
720 len = (len - 1) * sizeof(struct ext4_extent_idx);
721 len = len < 0 ? 0 : len;
722 ext_debug("insert new index %d after: %llu. "
723 "move %d from 0x%p to 0x%p\n",
725 (curp->p_idx + 1), (curp->p_idx + 2));
726 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
728 ix = curp->p_idx + 1;
731 len = len * sizeof(struct ext4_extent_idx);
732 len = len < 0 ? 0 : len;
733 ext_debug("insert new index %d before: %llu. "
734 "move %d from 0x%p to 0x%p\n",
736 curp->p_idx, (curp->p_idx + 1));
737 memmove(curp->p_idx + 1, curp->p_idx, len);
741 ix->ei_block = cpu_to_le32(logical);
742 ext4_idx_store_pblock(ix, ptr);
743 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
745 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
746 > le16_to_cpu(curp->p_hdr->eh_max));
747 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
749 err = ext4_ext_dirty(handle, inode, curp);
750 ext4_std_error(inode->i_sb, err);
757 * inserts new subtree into the path, using free index entry
759 * - allocates all needed blocks (new leaf and all intermediate index blocks)
760 * - makes decision where to split
761 * - moves remaining extents and index entries (right to the split point)
762 * into the newly allocated blocks
763 * - initializes subtree
765 static int ext4_ext_split(handle_t *handle, struct inode *inode,
766 struct ext4_ext_path *path,
767 struct ext4_extent *newext, int at)
769 struct buffer_head *bh = NULL;
770 int depth = ext_depth(inode);
771 struct ext4_extent_header *neh;
772 struct ext4_extent_idx *fidx;
773 struct ext4_extent *ex;
775 ext4_fsblk_t newblock, oldblock;
777 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
780 /* make decision: where to split? */
781 /* FIXME: now decision is simplest: at current extent */
783 /* if current leaf will be split, then we should use
784 * border from split point */
785 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
786 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
787 border = path[depth].p_ext[1].ee_block;
788 ext_debug("leaf will be split."
789 " next leaf starts at %d\n",
790 le32_to_cpu(border));
792 border = newext->ee_block;
793 ext_debug("leaf will be added."
794 " next leaf starts at %d\n",
795 le32_to_cpu(border));
799 * If error occurs, then we break processing
800 * and mark filesystem read-only. index won't
801 * be inserted and tree will be in consistent
802 * state. Next mount will repair buffers too.
806 * Get array to track all allocated blocks.
807 * We need this to handle errors and free blocks
810 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
814 /* allocate all needed blocks */
815 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
816 for (a = 0; a < depth - at; a++) {
817 newblock = ext4_ext_new_meta_block(handle, inode, path,
821 ablocks[a] = newblock;
824 /* initialize new leaf */
825 newblock = ablocks[--a];
826 BUG_ON(newblock == 0);
827 bh = sb_getblk(inode->i_sb, newblock);
834 err = ext4_journal_get_create_access(handle, bh);
838 neh = ext_block_hdr(bh);
840 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
841 neh->eh_magic = EXT4_EXT_MAGIC;
843 ex = EXT_FIRST_EXTENT(neh);
845 /* move remainder of path[depth] to the new leaf */
846 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
847 /* start copy from next extent */
848 /* TODO: we could do it by single memmove */
851 while (path[depth].p_ext <=
852 EXT_MAX_EXTENT(path[depth].p_hdr)) {
853 ext_debug("move %d:%llu:%d in new leaf %llu\n",
854 le32_to_cpu(path[depth].p_ext->ee_block),
855 ext_pblock(path[depth].p_ext),
856 ext4_ext_get_actual_len(path[depth].p_ext),
858 /*memmove(ex++, path[depth].p_ext++,
859 sizeof(struct ext4_extent));
865 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
866 le16_add_cpu(&neh->eh_entries, m);
869 set_buffer_uptodate(bh);
872 err = ext4_handle_dirty_metadata(handle, inode, bh);
878 /* correct old leaf */
880 err = ext4_ext_get_access(handle, inode, path + depth);
883 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
884 err = ext4_ext_dirty(handle, inode, path + depth);
890 /* create intermediate indexes */
894 ext_debug("create %d intermediate indices\n", k);
895 /* insert new index into current index block */
896 /* current depth stored in i var */
900 newblock = ablocks[--a];
901 bh = sb_getblk(inode->i_sb, newblock);
908 err = ext4_journal_get_create_access(handle, bh);
912 neh = ext_block_hdr(bh);
913 neh->eh_entries = cpu_to_le16(1);
914 neh->eh_magic = EXT4_EXT_MAGIC;
915 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
916 neh->eh_depth = cpu_to_le16(depth - i);
917 fidx = EXT_FIRST_INDEX(neh);
918 fidx->ei_block = border;
919 ext4_idx_store_pblock(fidx, oldblock);
921 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
922 i, newblock, le32_to_cpu(border), oldblock);
927 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
928 EXT_MAX_INDEX(path[i].p_hdr));
929 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
930 EXT_LAST_INDEX(path[i].p_hdr));
931 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
932 ext_debug("%d: move %d:%llu in new index %llu\n", i,
933 le32_to_cpu(path[i].p_idx->ei_block),
934 idx_pblock(path[i].p_idx),
936 /*memmove(++fidx, path[i].p_idx++,
937 sizeof(struct ext4_extent_idx));
939 BUG_ON(neh->eh_entries > neh->eh_max);*/
944 memmove(++fidx, path[i].p_idx - m,
945 sizeof(struct ext4_extent_idx) * m);
946 le16_add_cpu(&neh->eh_entries, m);
948 set_buffer_uptodate(bh);
951 err = ext4_handle_dirty_metadata(handle, inode, bh);
957 /* correct old index */
959 err = ext4_ext_get_access(handle, inode, path + i);
962 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
963 err = ext4_ext_dirty(handle, inode, path + i);
971 /* insert new index */
972 err = ext4_ext_insert_index(handle, inode, path + at,
973 le32_to_cpu(border), newblock);
977 if (buffer_locked(bh))
983 /* free all allocated blocks in error case */
984 for (i = 0; i < depth; i++) {
987 ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
996 * ext4_ext_grow_indepth:
997 * implements tree growing procedure:
998 * - allocates new block
999 * - moves top-level data (index block or leaf) into the new block
1000 * - initializes new top-level, creating index that points to the
1001 * just created block
1003 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1004 struct ext4_ext_path *path,
1005 struct ext4_extent *newext)
1007 struct ext4_ext_path *curp = path;
1008 struct ext4_extent_header *neh;
1009 struct ext4_extent_idx *fidx;
1010 struct buffer_head *bh;
1011 ext4_fsblk_t newblock;
1014 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1018 bh = sb_getblk(inode->i_sb, newblock);
1021 ext4_std_error(inode->i_sb, err);
1026 err = ext4_journal_get_create_access(handle, bh);
1032 /* move top-level index/leaf into new block */
1033 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1035 /* set size of new block */
1036 neh = ext_block_hdr(bh);
1037 /* old root could have indexes or leaves
1038 * so calculate e_max right way */
1039 if (ext_depth(inode))
1040 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
1042 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
1043 neh->eh_magic = EXT4_EXT_MAGIC;
1044 set_buffer_uptodate(bh);
1047 err = ext4_handle_dirty_metadata(handle, inode, bh);
1051 /* create index in new top-level index: num,max,pointer */
1052 err = ext4_ext_get_access(handle, inode, curp);
1056 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1057 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
1058 curp->p_hdr->eh_entries = cpu_to_le16(1);
1059 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1061 if (path[0].p_hdr->eh_depth)
1062 curp->p_idx->ei_block =
1063 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1065 curp->p_idx->ei_block =
1066 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1067 ext4_idx_store_pblock(curp->p_idx, newblock);
1069 neh = ext_inode_hdr(inode);
1070 fidx = EXT_FIRST_INDEX(neh);
1071 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1072 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1073 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1075 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1076 err = ext4_ext_dirty(handle, inode, curp);
1084 * ext4_ext_create_new_leaf:
1085 * finds empty index and adds new leaf.
1086 * if no free index is found, then it requests in-depth growing.
1088 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1089 struct ext4_ext_path *path,
1090 struct ext4_extent *newext)
1092 struct ext4_ext_path *curp;
1093 int depth, i, err = 0;
1096 i = depth = ext_depth(inode);
1098 /* walk up to the tree and look for free index entry */
1099 curp = path + depth;
1100 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1105 /* we use already allocated block for index block,
1106 * so subsequent data blocks should be contiguous */
1107 if (EXT_HAS_FREE_INDEX(curp)) {
1108 /* if we found index with free entry, then use that
1109 * entry: create all needed subtree and add new leaf */
1110 err = ext4_ext_split(handle, inode, path, newext, i);
1115 ext4_ext_drop_refs(path);
1116 path = ext4_ext_find_extent(inode,
1117 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1120 err = PTR_ERR(path);
1122 /* tree is full, time to grow in depth */
1123 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1128 ext4_ext_drop_refs(path);
1129 path = ext4_ext_find_extent(inode,
1130 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1133 err = PTR_ERR(path);
1138 * only first (depth 0 -> 1) produces free space;
1139 * in all other cases we have to split the grown tree
1141 depth = ext_depth(inode);
1142 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1143 /* now we need to split */
1153 * search the closest allocated block to the left for *logical
1154 * and returns it at @logical + it's physical address at @phys
1155 * if *logical is the smallest allocated block, the function
1156 * returns 0 at @phys
1157 * return value contains 0 (success) or error code
1160 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1161 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1163 struct ext4_extent_idx *ix;
1164 struct ext4_extent *ex;
1167 BUG_ON(path == NULL);
1168 depth = path->p_depth;
1171 if (depth == 0 && path->p_ext == NULL)
1174 /* usually extent in the path covers blocks smaller
1175 * then *logical, but it can be that extent is the
1176 * first one in the file */
1178 ex = path[depth].p_ext;
1179 ee_len = ext4_ext_get_actual_len(ex);
1180 if (*logical < le32_to_cpu(ex->ee_block)) {
1181 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1182 while (--depth >= 0) {
1183 ix = path[depth].p_idx;
1184 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1189 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1191 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1192 *phys = ext_pblock(ex) + ee_len - 1;
1197 * search the closest allocated block to the right for *logical
1198 * and returns it at @logical + it's physical address at @phys
1199 * if *logical is the smallest allocated block, the function
1200 * returns 0 at @phys
1201 * return value contains 0 (success) or error code
1204 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1205 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1207 struct buffer_head *bh = NULL;
1208 struct ext4_extent_header *eh;
1209 struct ext4_extent_idx *ix;
1210 struct ext4_extent *ex;
1212 int depth; /* Note, NOT eh_depth; depth from top of tree */
1215 BUG_ON(path == NULL);
1216 depth = path->p_depth;
1219 if (depth == 0 && path->p_ext == NULL)
1222 /* usually extent in the path covers blocks smaller
1223 * then *logical, but it can be that extent is the
1224 * first one in the file */
1226 ex = path[depth].p_ext;
1227 ee_len = ext4_ext_get_actual_len(ex);
1228 if (*logical < le32_to_cpu(ex->ee_block)) {
1229 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1230 while (--depth >= 0) {
1231 ix = path[depth].p_idx;
1232 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1234 *logical = le32_to_cpu(ex->ee_block);
1235 *phys = ext_pblock(ex);
1239 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1241 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1242 /* next allocated block in this leaf */
1244 *logical = le32_to_cpu(ex->ee_block);
1245 *phys = ext_pblock(ex);
1249 /* go up and search for index to the right */
1250 while (--depth >= 0) {
1251 ix = path[depth].p_idx;
1252 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1256 /* we've gone up to the root and found no index to the right */
1260 /* we've found index to the right, let's
1261 * follow it and find the closest allocated
1262 * block to the right */
1264 block = idx_pblock(ix);
1265 while (++depth < path->p_depth) {
1266 bh = sb_bread(inode->i_sb, block);
1269 eh = ext_block_hdr(bh);
1270 /* subtract from p_depth to get proper eh_depth */
1271 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1275 ix = EXT_FIRST_INDEX(eh);
1276 block = idx_pblock(ix);
1280 bh = sb_bread(inode->i_sb, block);
1283 eh = ext_block_hdr(bh);
1284 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1288 ex = EXT_FIRST_EXTENT(eh);
1289 *logical = le32_to_cpu(ex->ee_block);
1290 *phys = ext_pblock(ex);
1296 * ext4_ext_next_allocated_block:
1297 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1298 * NOTE: it considers block number from index entry as
1299 * allocated block. Thus, index entries have to be consistent
1303 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1307 BUG_ON(path == NULL);
1308 depth = path->p_depth;
1310 if (depth == 0 && path->p_ext == NULL)
1311 return EXT_MAX_BLOCK;
1313 while (depth >= 0) {
1314 if (depth == path->p_depth) {
1316 if (path[depth].p_ext !=
1317 EXT_LAST_EXTENT(path[depth].p_hdr))
1318 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1321 if (path[depth].p_idx !=
1322 EXT_LAST_INDEX(path[depth].p_hdr))
1323 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1328 return EXT_MAX_BLOCK;
1332 * ext4_ext_next_leaf_block:
1333 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1335 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1336 struct ext4_ext_path *path)
1340 BUG_ON(path == NULL);
1341 depth = path->p_depth;
1343 /* zero-tree has no leaf blocks at all */
1345 return EXT_MAX_BLOCK;
1347 /* go to index block */
1350 while (depth >= 0) {
1351 if (path[depth].p_idx !=
1352 EXT_LAST_INDEX(path[depth].p_hdr))
1353 return (ext4_lblk_t)
1354 le32_to_cpu(path[depth].p_idx[1].ei_block);
1358 return EXT_MAX_BLOCK;
1362 * ext4_ext_correct_indexes:
1363 * if leaf gets modified and modified extent is first in the leaf,
1364 * then we have to correct all indexes above.
1365 * TODO: do we need to correct tree in all cases?
1367 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1368 struct ext4_ext_path *path)
1370 struct ext4_extent_header *eh;
1371 int depth = ext_depth(inode);
1372 struct ext4_extent *ex;
1376 eh = path[depth].p_hdr;
1377 ex = path[depth].p_ext;
1382 /* there is no tree at all */
1386 if (ex != EXT_FIRST_EXTENT(eh)) {
1387 /* we correct tree if first leaf got modified only */
1392 * TODO: we need correction if border is smaller than current one
1395 border = path[depth].p_ext->ee_block;
1396 err = ext4_ext_get_access(handle, inode, path + k);
1399 path[k].p_idx->ei_block = border;
1400 err = ext4_ext_dirty(handle, inode, path + k);
1405 /* change all left-side indexes */
1406 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1408 err = ext4_ext_get_access(handle, inode, path + k);
1411 path[k].p_idx->ei_block = border;
1412 err = ext4_ext_dirty(handle, inode, path + k);
1421 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1422 struct ext4_extent *ex2)
1424 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1427 * Make sure that either both extents are uninitialized, or
1430 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1433 if (ext4_ext_is_uninitialized(ex1))
1434 max_len = EXT_UNINIT_MAX_LEN;
1436 max_len = EXT_INIT_MAX_LEN;
1438 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1439 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1441 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1442 le32_to_cpu(ex2->ee_block))
1446 * To allow future support for preallocated extents to be added
1447 * as an RO_COMPAT feature, refuse to merge to extents if
1448 * this can result in the top bit of ee_len being set.
1450 if (ext1_ee_len + ext2_ee_len > max_len)
1452 #ifdef AGGRESSIVE_TEST
1453 if (ext1_ee_len >= 4)
1457 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1463 * This function tries to merge the "ex" extent to the next extent in the tree.
1464 * It always tries to merge towards right. If you want to merge towards
1465 * left, pass "ex - 1" as argument instead of "ex".
1466 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1467 * 1 if they got merged.
1469 int ext4_ext_try_to_merge(struct inode *inode,
1470 struct ext4_ext_path *path,
1471 struct ext4_extent *ex)
1473 struct ext4_extent_header *eh;
1474 unsigned int depth, len;
1476 int uninitialized = 0;
1478 depth = ext_depth(inode);
1479 BUG_ON(path[depth].p_hdr == NULL);
1480 eh = path[depth].p_hdr;
1482 while (ex < EXT_LAST_EXTENT(eh)) {
1483 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1485 /* merge with next extent! */
1486 if (ext4_ext_is_uninitialized(ex))
1488 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1489 + ext4_ext_get_actual_len(ex + 1));
1491 ext4_ext_mark_uninitialized(ex);
1493 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1494 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1495 * sizeof(struct ext4_extent);
1496 memmove(ex + 1, ex + 2, len);
1498 le16_add_cpu(&eh->eh_entries, -1);
1500 WARN_ON(eh->eh_entries == 0);
1501 if (!eh->eh_entries)
1502 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1503 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1510 * check if a portion of the "newext" extent overlaps with an
1513 * If there is an overlap discovered, it updates the length of the newext
1514 * such that there will be no overlap, and then returns 1.
1515 * If there is no overlap found, it returns 0.
1517 unsigned int ext4_ext_check_overlap(struct inode *inode,
1518 struct ext4_extent *newext,
1519 struct ext4_ext_path *path)
1522 unsigned int depth, len1;
1523 unsigned int ret = 0;
1525 b1 = le32_to_cpu(newext->ee_block);
1526 len1 = ext4_ext_get_actual_len(newext);
1527 depth = ext_depth(inode);
1528 if (!path[depth].p_ext)
1530 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1533 * get the next allocated block if the extent in the path
1534 * is before the requested block(s)
1537 b2 = ext4_ext_next_allocated_block(path);
1538 if (b2 == EXT_MAX_BLOCK)
1542 /* check for wrap through zero on extent logical start block*/
1543 if (b1 + len1 < b1) {
1544 len1 = EXT_MAX_BLOCK - b1;
1545 newext->ee_len = cpu_to_le16(len1);
1549 /* check for overlap */
1550 if (b1 + len1 > b2) {
1551 newext->ee_len = cpu_to_le16(b2 - b1);
1559 * ext4_ext_insert_extent:
1560 * tries to merge requsted extent into the existing extent or
1561 * inserts requested extent as new one into the tree,
1562 * creating new leaf in the no-space case.
1564 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1565 struct ext4_ext_path *path,
1566 struct ext4_extent *newext)
1568 struct ext4_extent_header *eh;
1569 struct ext4_extent *ex, *fex;
1570 struct ext4_extent *nearex; /* nearest extent */
1571 struct ext4_ext_path *npath = NULL;
1572 int depth, len, err;
1574 unsigned uninitialized = 0;
1576 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1577 depth = ext_depth(inode);
1578 ex = path[depth].p_ext;
1579 BUG_ON(path[depth].p_hdr == NULL);
1581 /* try to insert block into found extent and return */
1582 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1583 ext_debug("append %d block to %d:%d (from %llu)\n",
1584 ext4_ext_get_actual_len(newext),
1585 le32_to_cpu(ex->ee_block),
1586 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1587 err = ext4_ext_get_access(handle, inode, path + depth);
1592 * ext4_can_extents_be_merged should have checked that either
1593 * both extents are uninitialized, or both aren't. Thus we
1594 * need to check only one of them here.
1596 if (ext4_ext_is_uninitialized(ex))
1598 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1599 + ext4_ext_get_actual_len(newext));
1601 ext4_ext_mark_uninitialized(ex);
1602 eh = path[depth].p_hdr;
1608 depth = ext_depth(inode);
1609 eh = path[depth].p_hdr;
1610 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1613 /* probably next leaf has space for us? */
1614 fex = EXT_LAST_EXTENT(eh);
1615 next = ext4_ext_next_leaf_block(inode, path);
1616 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1617 && next != EXT_MAX_BLOCK) {
1618 ext_debug("next leaf block - %d\n", next);
1619 BUG_ON(npath != NULL);
1620 npath = ext4_ext_find_extent(inode, next, NULL);
1622 return PTR_ERR(npath);
1623 BUG_ON(npath->p_depth != path->p_depth);
1624 eh = npath[depth].p_hdr;
1625 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1626 ext_debug("next leaf isnt full(%d)\n",
1627 le16_to_cpu(eh->eh_entries));
1631 ext_debug("next leaf has no free space(%d,%d)\n",
1632 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1636 * There is no free space in the found leaf.
1637 * We're gonna add a new leaf in the tree.
1639 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1642 depth = ext_depth(inode);
1643 eh = path[depth].p_hdr;
1646 nearex = path[depth].p_ext;
1648 err = ext4_ext_get_access(handle, inode, path + depth);
1653 /* there is no extent in this leaf, create first one */
1654 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1655 le32_to_cpu(newext->ee_block),
1657 ext4_ext_get_actual_len(newext));
1658 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1659 } else if (le32_to_cpu(newext->ee_block)
1660 > le32_to_cpu(nearex->ee_block)) {
1661 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1662 if (nearex != EXT_LAST_EXTENT(eh)) {
1663 len = EXT_MAX_EXTENT(eh) - nearex;
1664 len = (len - 1) * sizeof(struct ext4_extent);
1665 len = len < 0 ? 0 : len;
1666 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1667 "move %d from 0x%p to 0x%p\n",
1668 le32_to_cpu(newext->ee_block),
1670 ext4_ext_get_actual_len(newext),
1671 nearex, len, nearex + 1, nearex + 2);
1672 memmove(nearex + 2, nearex + 1, len);
1674 path[depth].p_ext = nearex + 1;
1676 BUG_ON(newext->ee_block == nearex->ee_block);
1677 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1678 len = len < 0 ? 0 : len;
1679 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1680 "move %d from 0x%p to 0x%p\n",
1681 le32_to_cpu(newext->ee_block),
1683 ext4_ext_get_actual_len(newext),
1684 nearex, len, nearex + 1, nearex + 2);
1685 memmove(nearex + 1, nearex, len);
1686 path[depth].p_ext = nearex;
1689 le16_add_cpu(&eh->eh_entries, 1);
1690 nearex = path[depth].p_ext;
1691 nearex->ee_block = newext->ee_block;
1692 ext4_ext_store_pblock(nearex, ext_pblock(newext));
1693 nearex->ee_len = newext->ee_len;
1696 /* try to merge extents to the right */
1697 ext4_ext_try_to_merge(inode, path, nearex);
1699 /* try to merge extents to the left */
1701 /* time to correct all indexes above */
1702 err = ext4_ext_correct_indexes(handle, inode, path);
1706 err = ext4_ext_dirty(handle, inode, path + depth);
1710 ext4_ext_drop_refs(npath);
1713 ext4_ext_invalidate_cache(inode);
1717 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1718 ext4_lblk_t num, ext_prepare_callback func,
1721 struct ext4_ext_path *path = NULL;
1722 struct ext4_ext_cache cbex;
1723 struct ext4_extent *ex;
1724 ext4_lblk_t next, start = 0, end = 0;
1725 ext4_lblk_t last = block + num;
1726 int depth, exists, err = 0;
1728 BUG_ON(func == NULL);
1729 BUG_ON(inode == NULL);
1731 while (block < last && block != EXT_MAX_BLOCK) {
1733 /* find extent for this block */
1734 path = ext4_ext_find_extent(inode, block, path);
1736 err = PTR_ERR(path);
1741 depth = ext_depth(inode);
1742 BUG_ON(path[depth].p_hdr == NULL);
1743 ex = path[depth].p_ext;
1744 next = ext4_ext_next_allocated_block(path);
1748 /* there is no extent yet, so try to allocate
1749 * all requested space */
1752 } else if (le32_to_cpu(ex->ee_block) > block) {
1753 /* need to allocate space before found extent */
1755 end = le32_to_cpu(ex->ee_block);
1756 if (block + num < end)
1758 } else if (block >= le32_to_cpu(ex->ee_block)
1759 + ext4_ext_get_actual_len(ex)) {
1760 /* need to allocate space after found extent */
1765 } else if (block >= le32_to_cpu(ex->ee_block)) {
1767 * some part of requested space is covered
1771 end = le32_to_cpu(ex->ee_block)
1772 + ext4_ext_get_actual_len(ex);
1773 if (block + num < end)
1779 BUG_ON(end <= start);
1782 cbex.ec_block = start;
1783 cbex.ec_len = end - start;
1785 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1787 cbex.ec_block = le32_to_cpu(ex->ee_block);
1788 cbex.ec_len = ext4_ext_get_actual_len(ex);
1789 cbex.ec_start = ext_pblock(ex);
1790 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1793 BUG_ON(cbex.ec_len == 0);
1794 err = func(inode, path, &cbex, ex, cbdata);
1795 ext4_ext_drop_refs(path);
1800 if (err == EXT_REPEAT)
1802 else if (err == EXT_BREAK) {
1807 if (ext_depth(inode) != depth) {
1808 /* depth was changed. we have to realloc path */
1813 block = cbex.ec_block + cbex.ec_len;
1817 ext4_ext_drop_refs(path);
1825 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1826 __u32 len, ext4_fsblk_t start, int type)
1828 struct ext4_ext_cache *cex;
1830 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1831 cex = &EXT4_I(inode)->i_cached_extent;
1832 cex->ec_type = type;
1833 cex->ec_block = block;
1835 cex->ec_start = start;
1836 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1840 * ext4_ext_put_gap_in_cache:
1841 * calculate boundaries of the gap that the requested block fits into
1842 * and cache this gap
1845 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1848 int depth = ext_depth(inode);
1851 struct ext4_extent *ex;
1853 ex = path[depth].p_ext;
1855 /* there is no extent yet, so gap is [0;-] */
1857 len = EXT_MAX_BLOCK;
1858 ext_debug("cache gap(whole file):");
1859 } else if (block < le32_to_cpu(ex->ee_block)) {
1861 len = le32_to_cpu(ex->ee_block) - block;
1862 ext_debug("cache gap(before): %u [%u:%u]",
1864 le32_to_cpu(ex->ee_block),
1865 ext4_ext_get_actual_len(ex));
1866 } else if (block >= le32_to_cpu(ex->ee_block)
1867 + ext4_ext_get_actual_len(ex)) {
1869 lblock = le32_to_cpu(ex->ee_block)
1870 + ext4_ext_get_actual_len(ex);
1872 next = ext4_ext_next_allocated_block(path);
1873 ext_debug("cache gap(after): [%u:%u] %u",
1874 le32_to_cpu(ex->ee_block),
1875 ext4_ext_get_actual_len(ex),
1877 BUG_ON(next == lblock);
1878 len = next - lblock;
1884 ext_debug(" -> %u:%lu\n", lblock, len);
1885 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1889 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1890 struct ext4_extent *ex)
1892 struct ext4_ext_cache *cex;
1893 int ret = EXT4_EXT_CACHE_NO;
1896 * We borrow i_block_reservation_lock to protect i_cached_extent
1898 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1899 cex = &EXT4_I(inode)->i_cached_extent;
1901 /* has cache valid data? */
1902 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1905 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1906 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1907 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1908 ex->ee_block = cpu_to_le32(cex->ec_block);
1909 ext4_ext_store_pblock(ex, cex->ec_start);
1910 ex->ee_len = cpu_to_le16(cex->ec_len);
1911 ext_debug("%u cached by %u:%u:%llu\n",
1913 cex->ec_block, cex->ec_len, cex->ec_start);
1917 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1923 * removes index from the index block.
1924 * It's used in truncate case only, thus all requests are for
1925 * last index in the block only.
1927 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1928 struct ext4_ext_path *path)
1930 struct buffer_head *bh;
1934 /* free index block */
1936 leaf = idx_pblock(path->p_idx);
1937 BUG_ON(path->p_hdr->eh_entries == 0);
1938 err = ext4_ext_get_access(handle, inode, path);
1941 le16_add_cpu(&path->p_hdr->eh_entries, -1);
1942 err = ext4_ext_dirty(handle, inode, path);
1945 ext_debug("index is empty, remove it, free block %llu\n", leaf);
1946 bh = sb_find_get_block(inode->i_sb, leaf);
1947 ext4_forget(handle, 1, inode, bh, leaf);
1948 ext4_free_blocks(handle, inode, leaf, 1, 1);
1953 * ext4_ext_calc_credits_for_single_extent:
1954 * This routine returns max. credits that needed to insert an extent
1955 * to the extent tree.
1956 * When pass the actual path, the caller should calculate credits
1959 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
1960 struct ext4_ext_path *path)
1963 int depth = ext_depth(inode);
1966 /* probably there is space in leaf? */
1967 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1968 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
1971 * There are some space in the leaf tree, no
1972 * need to account for leaf block credit
1974 * bitmaps and block group descriptor blocks
1975 * and other metadat blocks still need to be
1978 /* 1 bitmap, 1 block group descriptor */
1979 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
1983 return ext4_chunk_trans_blocks(inode, nrblocks);
1987 * How many index/leaf blocks need to change/allocate to modify nrblocks?
1989 * if nrblocks are fit in a single extent (chunk flag is 1), then
1990 * in the worse case, each tree level index/leaf need to be changed
1991 * if the tree split due to insert a new extent, then the old tree
1992 * index/leaf need to be updated too
1994 * If the nrblocks are discontiguous, they could cause
1995 * the whole tree split more than once, but this is really rare.
1997 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2000 int depth = ext_depth(inode);
2010 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2011 struct ext4_extent *ex,
2012 ext4_lblk_t from, ext4_lblk_t to)
2014 struct buffer_head *bh;
2015 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2016 int i, metadata = 0;
2018 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2020 #ifdef EXTENTS_STATS
2022 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2023 spin_lock(&sbi->s_ext_stats_lock);
2024 sbi->s_ext_blocks += ee_len;
2025 sbi->s_ext_extents++;
2026 if (ee_len < sbi->s_ext_min)
2027 sbi->s_ext_min = ee_len;
2028 if (ee_len > sbi->s_ext_max)
2029 sbi->s_ext_max = ee_len;
2030 if (ext_depth(inode) > sbi->s_depth_max)
2031 sbi->s_depth_max = ext_depth(inode);
2032 spin_unlock(&sbi->s_ext_stats_lock);
2035 if (from >= le32_to_cpu(ex->ee_block)
2036 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2041 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2042 start = ext_pblock(ex) + ee_len - num;
2043 ext_debug("free last %u blocks starting %llu\n", num, start);
2044 for (i = 0; i < num; i++) {
2045 bh = sb_find_get_block(inode->i_sb, start + i);
2046 ext4_forget(handle, 0, inode, bh, start + i);
2048 ext4_free_blocks(handle, inode, start, num, metadata);
2049 } else if (from == le32_to_cpu(ex->ee_block)
2050 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2051 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2052 from, to, le32_to_cpu(ex->ee_block), ee_len);
2054 printk(KERN_INFO "strange request: removal(2) "
2055 "%u-%u from %u:%u\n",
2056 from, to, le32_to_cpu(ex->ee_block), ee_len);
2062 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2063 struct ext4_ext_path *path, ext4_lblk_t start)
2065 int err = 0, correct_index = 0;
2066 int depth = ext_depth(inode), credits;
2067 struct ext4_extent_header *eh;
2068 ext4_lblk_t a, b, block;
2070 ext4_lblk_t ex_ee_block;
2071 unsigned short ex_ee_len;
2072 unsigned uninitialized = 0;
2073 struct ext4_extent *ex;
2075 /* the header must be checked already in ext4_ext_remove_space() */
2076 ext_debug("truncate since %u in leaf\n", start);
2077 if (!path[depth].p_hdr)
2078 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2079 eh = path[depth].p_hdr;
2082 /* find where to start removing */
2083 ex = EXT_LAST_EXTENT(eh);
2085 ex_ee_block = le32_to_cpu(ex->ee_block);
2086 if (ext4_ext_is_uninitialized(ex))
2088 ex_ee_len = ext4_ext_get_actual_len(ex);
2090 while (ex >= EXT_FIRST_EXTENT(eh) &&
2091 ex_ee_block + ex_ee_len > start) {
2092 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
2093 path[depth].p_ext = ex;
2095 a = ex_ee_block > start ? ex_ee_block : start;
2096 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2097 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2099 ext_debug(" border %u:%u\n", a, b);
2101 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2105 } else if (a != ex_ee_block) {
2106 /* remove tail of the extent */
2107 block = ex_ee_block;
2109 } else if (b != ex_ee_block + ex_ee_len - 1) {
2110 /* remove head of the extent */
2113 /* there is no "make a hole" API yet */
2116 /* remove whole extent: excellent! */
2117 block = ex_ee_block;
2119 BUG_ON(a != ex_ee_block);
2120 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2124 * 3 for leaf, sb, and inode plus 2 (bmap and group
2125 * descriptor) for each block group; assume two block
2126 * groups plus ex_ee_len/blocks_per_block_group for
2129 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2130 if (ex == EXT_FIRST_EXTENT(eh)) {
2132 credits += (ext_depth(inode)) + 1;
2134 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2136 err = ext4_ext_journal_restart(handle, credits);
2140 err = ext4_ext_get_access(handle, inode, path + depth);
2144 err = ext4_remove_blocks(handle, inode, ex, a, b);
2149 /* this extent is removed; mark slot entirely unused */
2150 ext4_ext_store_pblock(ex, 0);
2151 le16_add_cpu(&eh->eh_entries, -1);
2154 ex->ee_block = cpu_to_le32(block);
2155 ex->ee_len = cpu_to_le16(num);
2157 * Do not mark uninitialized if all the blocks in the
2158 * extent have been removed.
2160 if (uninitialized && num)
2161 ext4_ext_mark_uninitialized(ex);
2163 err = ext4_ext_dirty(handle, inode, path + depth);
2167 ext_debug("new extent: %u:%u:%llu\n", block, num,
2170 ex_ee_block = le32_to_cpu(ex->ee_block);
2171 ex_ee_len = ext4_ext_get_actual_len(ex);
2174 if (correct_index && eh->eh_entries)
2175 err = ext4_ext_correct_indexes(handle, inode, path);
2177 /* if this leaf is free, then we should
2178 * remove it from index block above */
2179 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2180 err = ext4_ext_rm_idx(handle, inode, path + depth);
2187 * ext4_ext_more_to_rm:
2188 * returns 1 if current index has to be freed (even partial)
2191 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2193 BUG_ON(path->p_idx == NULL);
2195 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2199 * if truncate on deeper level happened, it wasn't partial,
2200 * so we have to consider current index for truncation
2202 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2207 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2209 struct super_block *sb = inode->i_sb;
2210 int depth = ext_depth(inode);
2211 struct ext4_ext_path *path;
2215 ext_debug("truncate since %u\n", start);
2217 /* probably first extent we're gonna free will be last in block */
2218 handle = ext4_journal_start(inode, depth + 1);
2220 return PTR_ERR(handle);
2222 ext4_ext_invalidate_cache(inode);
2225 * We start scanning from right side, freeing all the blocks
2226 * after i_size and walking into the tree depth-wise.
2228 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2230 ext4_journal_stop(handle);
2233 path[0].p_hdr = ext_inode_hdr(inode);
2234 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2238 path[0].p_depth = depth;
2240 while (i >= 0 && err == 0) {
2242 /* this is leaf block */
2243 err = ext4_ext_rm_leaf(handle, inode, path, start);
2244 /* root level has p_bh == NULL, brelse() eats this */
2245 brelse(path[i].p_bh);
2246 path[i].p_bh = NULL;
2251 /* this is index block */
2252 if (!path[i].p_hdr) {
2253 ext_debug("initialize header\n");
2254 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2257 if (!path[i].p_idx) {
2258 /* this level hasn't been touched yet */
2259 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2260 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2261 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2263 le16_to_cpu(path[i].p_hdr->eh_entries));
2265 /* we were already here, see at next index */
2269 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2270 i, EXT_FIRST_INDEX(path[i].p_hdr),
2272 if (ext4_ext_more_to_rm(path + i)) {
2273 struct buffer_head *bh;
2274 /* go to the next level */
2275 ext_debug("move to level %d (block %llu)\n",
2276 i + 1, idx_pblock(path[i].p_idx));
2277 memset(path + i + 1, 0, sizeof(*path));
2278 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2280 /* should we reset i_size? */
2284 if (WARN_ON(i + 1 > depth)) {
2288 if (ext4_ext_check(inode, ext_block_hdr(bh),
2293 path[i + 1].p_bh = bh;
2295 /* save actual number of indexes since this
2296 * number is changed at the next iteration */
2297 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2300 /* we finished processing this index, go up */
2301 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2302 /* index is empty, remove it;
2303 * handle must be already prepared by the
2304 * truncatei_leaf() */
2305 err = ext4_ext_rm_idx(handle, inode, path + i);
2307 /* root level has p_bh == NULL, brelse() eats this */
2308 brelse(path[i].p_bh);
2309 path[i].p_bh = NULL;
2311 ext_debug("return to level %d\n", i);
2315 /* TODO: flexible tree reduction should be here */
2316 if (path->p_hdr->eh_entries == 0) {
2318 * truncate to zero freed all the tree,
2319 * so we need to correct eh_depth
2321 err = ext4_ext_get_access(handle, inode, path);
2323 ext_inode_hdr(inode)->eh_depth = 0;
2324 ext_inode_hdr(inode)->eh_max =
2325 cpu_to_le16(ext4_ext_space_root(inode));
2326 err = ext4_ext_dirty(handle, inode, path);
2330 ext4_ext_drop_refs(path);
2332 ext4_journal_stop(handle);
2338 * called at mount time
2340 void ext4_ext_init(struct super_block *sb)
2343 * possible initialization would be here
2346 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2347 printk(KERN_INFO "EXT4-fs: file extents enabled");
2348 #ifdef AGGRESSIVE_TEST
2349 printk(", aggressive tests");
2351 #ifdef CHECK_BINSEARCH
2352 printk(", check binsearch");
2354 #ifdef EXTENTS_STATS
2358 #ifdef EXTENTS_STATS
2359 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2360 EXT4_SB(sb)->s_ext_min = 1 << 30;
2361 EXT4_SB(sb)->s_ext_max = 0;
2367 * called at umount time
2369 void ext4_ext_release(struct super_block *sb)
2371 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2374 #ifdef EXTENTS_STATS
2375 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2376 struct ext4_sb_info *sbi = EXT4_SB(sb);
2377 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2378 sbi->s_ext_blocks, sbi->s_ext_extents,
2379 sbi->s_ext_blocks / sbi->s_ext_extents);
2380 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2381 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2386 static void bi_complete(struct bio *bio, int error)
2388 complete((struct completion *)bio->bi_private);
2391 /* FIXME!! we need to try to merge to left or right after zero-out */
2392 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2396 int blkbits, blocksize;
2398 struct completion event;
2399 unsigned int ee_len, len, done, offset;
2402 blkbits = inode->i_blkbits;
2403 blocksize = inode->i_sb->s_blocksize;
2404 ee_len = ext4_ext_get_actual_len(ex);
2405 ee_pblock = ext_pblock(ex);
2407 /* convert ee_pblock to 512 byte sectors */
2408 ee_pblock = ee_pblock << (blkbits - 9);
2410 while (ee_len > 0) {
2412 if (ee_len > BIO_MAX_PAGES)
2413 len = BIO_MAX_PAGES;
2417 bio = bio_alloc(GFP_NOIO, len);
2418 bio->bi_sector = ee_pblock;
2419 bio->bi_bdev = inode->i_sb->s_bdev;
2423 while (done < len) {
2424 ret = bio_add_page(bio, ZERO_PAGE(0),
2426 if (ret != blocksize) {
2428 * We can't add any more pages because of
2429 * hardware limitations. Start a new bio.
2434 offset += blocksize;
2435 if (offset >= PAGE_CACHE_SIZE)
2439 init_completion(&event);
2440 bio->bi_private = &event;
2441 bio->bi_end_io = bi_complete;
2442 submit_bio(WRITE, bio);
2443 wait_for_completion(&event);
2445 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2453 ee_pblock += done << (blkbits - 9);
2458 #define EXT4_EXT_ZERO_LEN 7
2461 * This function is called by ext4_ext_get_blocks() if someone tries to write
2462 * to an uninitialized extent. It may result in splitting the uninitialized
2463 * extent into multiple extents (upto three - one initialized and two
2465 * There are three possibilities:
2466 * a> There is no split required: Entire extent should be initialized
2467 * b> Splits in two extents: Write is happening at either end of the extent
2468 * c> Splits in three extents: Somone is writing in middle of the extent
2470 static int ext4_ext_convert_to_initialized(handle_t *handle,
2471 struct inode *inode,
2472 struct ext4_ext_path *path,
2474 unsigned int max_blocks)
2476 struct ext4_extent *ex, newex, orig_ex;
2477 struct ext4_extent *ex1 = NULL;
2478 struct ext4_extent *ex2 = NULL;
2479 struct ext4_extent *ex3 = NULL;
2480 struct ext4_extent_header *eh;
2481 ext4_lblk_t ee_block;
2482 unsigned int allocated, ee_len, depth;
2483 ext4_fsblk_t newblock;
2487 depth = ext_depth(inode);
2488 eh = path[depth].p_hdr;
2489 ex = path[depth].p_ext;
2490 ee_block = le32_to_cpu(ex->ee_block);
2491 ee_len = ext4_ext_get_actual_len(ex);
2492 allocated = ee_len - (iblock - ee_block);
2493 newblock = iblock - ee_block + ext_pblock(ex);
2495 orig_ex.ee_block = ex->ee_block;
2496 orig_ex.ee_len = cpu_to_le16(ee_len);
2497 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2499 err = ext4_ext_get_access(handle, inode, path + depth);
2502 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2503 if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2504 err = ext4_ext_zeroout(inode, &orig_ex);
2506 goto fix_extent_len;
2507 /* update the extent length and mark as initialized */
2508 ex->ee_block = orig_ex.ee_block;
2509 ex->ee_len = orig_ex.ee_len;
2510 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2511 ext4_ext_dirty(handle, inode, path + depth);
2512 /* zeroed the full extent */
2516 /* ex1: ee_block to iblock - 1 : uninitialized */
2517 if (iblock > ee_block) {
2519 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2520 ext4_ext_mark_uninitialized(ex1);
2524 * for sanity, update the length of the ex2 extent before
2525 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2526 * overlap of blocks.
2528 if (!ex1 && allocated > max_blocks)
2529 ex2->ee_len = cpu_to_le16(max_blocks);
2530 /* ex3: to ee_block + ee_len : uninitialised */
2531 if (allocated > max_blocks) {
2532 unsigned int newdepth;
2533 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2534 if (allocated <= EXT4_EXT_ZERO_LEN) {
2536 * iblock == ee_block is handled by the zerouout
2538 * Mark first half uninitialized.
2539 * Mark second half initialized and zero out the
2540 * initialized extent
2542 ex->ee_block = orig_ex.ee_block;
2543 ex->ee_len = cpu_to_le16(ee_len - allocated);
2544 ext4_ext_mark_uninitialized(ex);
2545 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2546 ext4_ext_dirty(handle, inode, path + depth);
2549 ex3->ee_block = cpu_to_le32(iblock);
2550 ext4_ext_store_pblock(ex3, newblock);
2551 ex3->ee_len = cpu_to_le16(allocated);
2552 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2553 if (err == -ENOSPC) {
2554 err = ext4_ext_zeroout(inode, &orig_ex);
2556 goto fix_extent_len;
2557 ex->ee_block = orig_ex.ee_block;
2558 ex->ee_len = orig_ex.ee_len;
2559 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2560 ext4_ext_dirty(handle, inode, path + depth);
2561 /* blocks available from iblock */
2565 goto fix_extent_len;
2568 * We need to zero out the second half because
2569 * an fallocate request can update file size and
2570 * converting the second half to initialized extent
2571 * implies that we can leak some junk data to user
2574 err = ext4_ext_zeroout(inode, ex3);
2577 * We should actually mark the
2578 * second half as uninit and return error
2579 * Insert would have changed the extent
2581 depth = ext_depth(inode);
2582 ext4_ext_drop_refs(path);
2583 path = ext4_ext_find_extent(inode,
2586 err = PTR_ERR(path);
2589 /* get the second half extent details */
2590 ex = path[depth].p_ext;
2591 err = ext4_ext_get_access(handle, inode,
2595 ext4_ext_mark_uninitialized(ex);
2596 ext4_ext_dirty(handle, inode, path + depth);
2600 /* zeroed the second half */
2604 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2605 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2606 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2607 ext4_ext_mark_uninitialized(ex3);
2608 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2609 if (err == -ENOSPC) {
2610 err = ext4_ext_zeroout(inode, &orig_ex);
2612 goto fix_extent_len;
2613 /* update the extent length and mark as initialized */
2614 ex->ee_block = orig_ex.ee_block;
2615 ex->ee_len = orig_ex.ee_len;
2616 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2617 ext4_ext_dirty(handle, inode, path + depth);
2618 /* zeroed the full extent */
2619 /* blocks available from iblock */
2623 goto fix_extent_len;
2625 * The depth, and hence eh & ex might change
2626 * as part of the insert above.
2628 newdepth = ext_depth(inode);
2630 * update the extent length after successful insert of the
2633 orig_ex.ee_len = cpu_to_le16(ee_len -
2634 ext4_ext_get_actual_len(ex3));
2636 ext4_ext_drop_refs(path);
2637 path = ext4_ext_find_extent(inode, iblock, path);
2639 err = PTR_ERR(path);
2642 eh = path[depth].p_hdr;
2643 ex = path[depth].p_ext;
2647 err = ext4_ext_get_access(handle, inode, path + depth);
2651 allocated = max_blocks;
2653 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2654 * to insert a extent in the middle zerout directly
2655 * otherwise give the extent a chance to merge to left
2657 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2658 iblock != ee_block) {
2659 err = ext4_ext_zeroout(inode, &orig_ex);
2661 goto fix_extent_len;
2662 /* update the extent length and mark as initialized */
2663 ex->ee_block = orig_ex.ee_block;
2664 ex->ee_len = orig_ex.ee_len;
2665 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2666 ext4_ext_dirty(handle, inode, path + depth);
2667 /* zero out the first half */
2668 /* blocks available from iblock */
2673 * If there was a change of depth as part of the
2674 * insertion of ex3 above, we need to update the length
2675 * of the ex1 extent again here
2677 if (ex1 && ex1 != ex) {
2679 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2680 ext4_ext_mark_uninitialized(ex1);
2683 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2684 ex2->ee_block = cpu_to_le32(iblock);
2685 ext4_ext_store_pblock(ex2, newblock);
2686 ex2->ee_len = cpu_to_le16(allocated);
2690 * New (initialized) extent starts from the first block
2691 * in the current extent. i.e., ex2 == ex
2692 * We have to see if it can be merged with the extent
2695 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2697 * To merge left, pass "ex2 - 1" to try_to_merge(),
2698 * since it merges towards right _only_.
2700 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2702 err = ext4_ext_correct_indexes(handle, inode, path);
2705 depth = ext_depth(inode);
2710 * Try to Merge towards right. This might be required
2711 * only when the whole extent is being written to.
2712 * i.e. ex2 == ex and ex3 == NULL.
2715 ret = ext4_ext_try_to_merge(inode, path, ex2);
2717 err = ext4_ext_correct_indexes(handle, inode, path);
2722 /* Mark modified extent as dirty */
2723 err = ext4_ext_dirty(handle, inode, path + depth);
2726 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2727 if (err == -ENOSPC) {
2728 err = ext4_ext_zeroout(inode, &orig_ex);
2730 goto fix_extent_len;
2731 /* update the extent length and mark as initialized */
2732 ex->ee_block = orig_ex.ee_block;
2733 ex->ee_len = orig_ex.ee_len;
2734 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2735 ext4_ext_dirty(handle, inode, path + depth);
2736 /* zero out the first half */
2739 goto fix_extent_len;
2741 return err ? err : allocated;
2744 ex->ee_block = orig_ex.ee_block;
2745 ex->ee_len = orig_ex.ee_len;
2746 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2747 ext4_ext_mark_uninitialized(ex);
2748 ext4_ext_dirty(handle, inode, path + depth);
2753 * Block allocation/map/preallocation routine for extents based files
2756 * Need to be called with
2757 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2758 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2760 * return > 0, number of of blocks already mapped/allocated
2761 * if create == 0 and these are pre-allocated blocks
2762 * buffer head is unmapped
2763 * otherwise blocks are mapped
2765 * return = 0, if plain look up failed (blocks have not been allocated)
2766 * buffer head is unmapped
2768 * return < 0, error case.
2770 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2772 unsigned int max_blocks, struct buffer_head *bh_result,
2775 struct ext4_ext_path *path = NULL;
2776 struct ext4_extent_header *eh;
2777 struct ext4_extent newex, *ex;
2778 ext4_fsblk_t newblock;
2779 int err = 0, depth, ret, cache_type;
2780 unsigned int allocated = 0;
2781 struct ext4_allocation_request ar;
2783 __clear_bit(BH_New, &bh_result->b_state);
2784 ext_debug("blocks %u/%u requested for inode %u\n",
2785 iblock, max_blocks, inode->i_ino);
2787 /* check in cache */
2788 cache_type = ext4_ext_in_cache(inode, iblock, &newex);
2790 if (cache_type == EXT4_EXT_CACHE_GAP) {
2791 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2793 * block isn't allocated yet and
2794 * user doesn't want to allocate it
2798 /* we should allocate requested block */
2799 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
2800 /* block is already allocated */
2802 - le32_to_cpu(newex.ee_block)
2803 + ext_pblock(&newex);
2804 /* number of remaining blocks in the extent */
2805 allocated = ext4_ext_get_actual_len(&newex) -
2806 (iblock - le32_to_cpu(newex.ee_block));
2813 /* find extent for this block */
2814 path = ext4_ext_find_extent(inode, iblock, NULL);
2816 err = PTR_ERR(path);
2821 depth = ext_depth(inode);
2824 * consistent leaf must not be empty;
2825 * this situation is possible, though, _during_ tree modification;
2826 * this is why assert can't be put in ext4_ext_find_extent()
2828 BUG_ON(path[depth].p_ext == NULL && depth != 0);
2829 eh = path[depth].p_hdr;
2831 ex = path[depth].p_ext;
2833 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
2834 ext4_fsblk_t ee_start = ext_pblock(ex);
2835 unsigned short ee_len;
2838 * Uninitialized extents are treated as holes, except that
2839 * we split out initialized portions during a write.
2841 ee_len = ext4_ext_get_actual_len(ex);
2842 /* if found extent covers block, simply return it */
2843 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2844 newblock = iblock - ee_block + ee_start;
2845 /* number of remaining blocks in the extent */
2846 allocated = ee_len - (iblock - ee_block);
2847 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
2848 ee_block, ee_len, newblock);
2850 /* Do not put uninitialized extent in the cache */
2851 if (!ext4_ext_is_uninitialized(ex)) {
2852 ext4_ext_put_in_cache(inode, ee_block,
2854 EXT4_EXT_CACHE_EXTENT);
2857 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
2859 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2860 if (allocated > max_blocks)
2861 allocated = max_blocks;
2863 * We have blocks reserved already. We
2864 * return allocated blocks so that delalloc
2865 * won't do block reservation for us. But
2866 * the buffer head will be unmapped so that
2867 * a read from the block returns 0s.
2869 set_buffer_unwritten(bh_result);
2870 bh_result->b_bdev = inode->i_sb->s_bdev;
2871 bh_result->b_blocknr = newblock;
2875 ret = ext4_ext_convert_to_initialized(handle, inode,
2888 * requested block isn't allocated yet;
2889 * we couldn't try to create block if create flag is zero
2891 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2893 * put just found gap into cache to speed up
2894 * subsequent requests
2896 ext4_ext_put_gap_in_cache(inode, path, iblock);
2900 * Okay, we need to do block allocation.
2903 /* find neighbour allocated blocks */
2905 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2909 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2914 * See if request is beyond maximum number of blocks we can have in
2915 * a single extent. For an initialized extent this limit is
2916 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2917 * EXT_UNINIT_MAX_LEN.
2919 if (max_blocks > EXT_INIT_MAX_LEN &&
2920 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
2921 max_blocks = EXT_INIT_MAX_LEN;
2922 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2923 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
2924 max_blocks = EXT_UNINIT_MAX_LEN;
2926 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2927 newex.ee_block = cpu_to_le32(iblock);
2928 newex.ee_len = cpu_to_le16(max_blocks);
2929 err = ext4_ext_check_overlap(inode, &newex, path);
2931 allocated = ext4_ext_get_actual_len(&newex);
2933 allocated = max_blocks;
2935 /* allocate new block */
2937 ar.goal = ext4_ext_find_goal(inode, path, iblock);
2938 ar.logical = iblock;
2940 if (S_ISREG(inode->i_mode))
2941 ar.flags = EXT4_MB_HINT_DATA;
2943 /* disable in-core preallocation for non-regular files */
2945 newblock = ext4_mb_new_blocks(handle, &ar, &err);
2948 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2949 ar.goal, newblock, allocated);
2951 /* try to insert new extent into found leaf and return */
2952 ext4_ext_store_pblock(&newex, newblock);
2953 newex.ee_len = cpu_to_le16(ar.len);
2954 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) /* Mark uninitialized */
2955 ext4_ext_mark_uninitialized(&newex);
2956 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2958 /* free data blocks we just allocated */
2959 /* not a good idea to call discard here directly,
2960 * but otherwise we'd need to call it every free() */
2961 ext4_discard_preallocations(inode);
2962 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2963 ext4_ext_get_actual_len(&newex), 0);
2967 /* previous routine could use block we allocated */
2968 newblock = ext_pblock(&newex);
2969 allocated = ext4_ext_get_actual_len(&newex);
2971 set_buffer_new(bh_result);
2973 /* Cache only when it is _not_ an uninitialized extent */
2974 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
2975 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2976 EXT4_EXT_CACHE_EXTENT);
2978 if (allocated > max_blocks)
2979 allocated = max_blocks;
2980 ext4_ext_show_leaf(inode, path);
2981 set_buffer_mapped(bh_result);
2982 bh_result->b_bdev = inode->i_sb->s_bdev;
2983 bh_result->b_blocknr = newblock;
2986 ext4_ext_drop_refs(path);
2989 return err ? err : allocated;
2992 void ext4_ext_truncate(struct inode *inode)
2994 struct address_space *mapping = inode->i_mapping;
2995 struct super_block *sb = inode->i_sb;
2996 ext4_lblk_t last_block;
3001 * probably first extent we're gonna free will be last in block
3003 err = ext4_writepage_trans_blocks(inode);
3004 handle = ext4_journal_start(inode, err);
3008 if (inode->i_size & (sb->s_blocksize - 1))
3009 ext4_block_truncate_page(handle, mapping, inode->i_size);
3011 if (ext4_orphan_add(handle, inode))
3014 down_write(&EXT4_I(inode)->i_data_sem);
3015 ext4_ext_invalidate_cache(inode);
3017 ext4_discard_preallocations(inode);
3020 * TODO: optimization is possible here.
3021 * Probably we need not scan at all,
3022 * because page truncation is enough.
3025 /* we have to know where to truncate from in crash case */
3026 EXT4_I(inode)->i_disksize = inode->i_size;
3027 ext4_mark_inode_dirty(handle, inode);
3029 last_block = (inode->i_size + sb->s_blocksize - 1)
3030 >> EXT4_BLOCK_SIZE_BITS(sb);
3031 err = ext4_ext_remove_space(inode, last_block);
3033 /* In a multi-transaction truncate, we only make the final
3034 * transaction synchronous.
3037 ext4_handle_sync(handle);
3040 up_write(&EXT4_I(inode)->i_data_sem);
3042 * If this was a simple ftruncate() and the file will remain alive,
3043 * then we need to clear up the orphan record which we created above.
3044 * However, if this was a real unlink then we were called by
3045 * ext4_delete_inode(), and we allow that function to clean up the
3046 * orphan info for us.
3049 ext4_orphan_del(handle, inode);
3051 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3052 ext4_mark_inode_dirty(handle, inode);
3053 ext4_journal_stop(handle);
3056 static void ext4_falloc_update_inode(struct inode *inode,
3057 int mode, loff_t new_size, int update_ctime)
3059 struct timespec now;
3062 now = current_fs_time(inode->i_sb);
3063 if (!timespec_equal(&inode->i_ctime, &now))
3064 inode->i_ctime = now;
3067 * Update only when preallocation was requested beyond
3070 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3071 if (new_size > i_size_read(inode))
3072 i_size_write(inode, new_size);
3073 if (new_size > EXT4_I(inode)->i_disksize)
3074 ext4_update_i_disksize(inode, new_size);
3080 * preallocate space for a file. This implements ext4's fallocate inode
3081 * operation, which gets called from sys_fallocate system call.
3082 * For block-mapped files, posix_fallocate should fall back to the method
3083 * of writing zeroes to the required new blocks (the same behavior which is
3084 * expected for file systems which do not support fallocate() system call).
3086 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3091 unsigned int max_blocks;
3095 struct buffer_head map_bh;
3096 unsigned int credits, blkbits = inode->i_blkbits;
3099 * currently supporting (pre)allocate mode for extent-based
3102 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3105 /* preallocation to directories is currently not supported */
3106 if (S_ISDIR(inode->i_mode))
3109 block = offset >> blkbits;
3111 * We can't just convert len to max_blocks because
3112 * If blocksize = 4096 offset = 3072 and len = 2048
3114 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3117 * credits to insert 1 extent into extent tree
3119 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3120 mutex_lock(&inode->i_mutex);
3122 while (ret >= 0 && ret < max_blocks) {
3123 block = block + ret;
3124 max_blocks = max_blocks - ret;
3125 handle = ext4_journal_start(inode, credits);
3126 if (IS_ERR(handle)) {
3127 ret = PTR_ERR(handle);
3131 ret = ext4_get_blocks(handle, inode, block,
3132 max_blocks, &map_bh,
3133 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3137 printk(KERN_ERR "%s: ext4_ext_get_blocks "
3138 "returned error inode#%lu, block=%u, "
3139 "max_blocks=%u", __func__,
3140 inode->i_ino, block, max_blocks);
3142 ext4_mark_inode_dirty(handle, inode);
3143 ret2 = ext4_journal_stop(handle);
3146 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3147 blkbits) >> blkbits))
3148 new_size = offset + len;
3150 new_size = (block + ret) << blkbits;
3152 ext4_falloc_update_inode(inode, mode, new_size,
3153 buffer_new(&map_bh));
3154 ext4_mark_inode_dirty(handle, inode);
3155 ret2 = ext4_journal_stop(handle);
3159 if (ret == -ENOSPC &&
3160 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3164 mutex_unlock(&inode->i_mutex);
3165 return ret > 0 ? ret2 : ret;
3169 * Callback function called for each extent to gather FIEMAP information.
3171 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3172 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3175 struct fiemap_extent_info *fieinfo = data;
3176 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3183 logical = (__u64)newex->ec_block << blksize_bits;
3185 if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3188 struct buffer_head *bh = NULL;
3190 offset = logical >> PAGE_SHIFT;
3191 page = find_get_page(inode->i_mapping, offset);
3192 if (!page || !page_has_buffers(page))
3193 return EXT_CONTINUE;
3195 bh = page_buffers(page);
3198 return EXT_CONTINUE;
3200 if (buffer_delay(bh)) {
3201 flags |= FIEMAP_EXTENT_DELALLOC;
3202 page_cache_release(page);
3204 page_cache_release(page);
3205 return EXT_CONTINUE;
3209 physical = (__u64)newex->ec_start << blksize_bits;
3210 length = (__u64)newex->ec_len << blksize_bits;
3212 if (ex && ext4_ext_is_uninitialized(ex))
3213 flags |= FIEMAP_EXTENT_UNWRITTEN;
3216 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3218 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3219 * this also indicates no more allocated blocks.
3221 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3223 if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3224 newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3225 loff_t size = i_size_read(inode);
3226 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3228 flags |= FIEMAP_EXTENT_LAST;
3229 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3230 logical+length > size)
3231 length = (size - logical + bs - 1) & ~(bs-1);
3234 error = fiemap_fill_next_extent(fieinfo, logical, physical,
3241 return EXT_CONTINUE;
3244 /* fiemap flags we can handle specified here */
3245 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3247 static int ext4_xattr_fiemap(struct inode *inode,
3248 struct fiemap_extent_info *fieinfo)
3252 __u32 flags = FIEMAP_EXTENT_LAST;
3253 int blockbits = inode->i_sb->s_blocksize_bits;
3257 if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
3258 struct ext4_iloc iloc;
3259 int offset; /* offset of xattr in inode */
3261 error = ext4_get_inode_loc(inode, &iloc);
3264 physical = iloc.bh->b_blocknr << blockbits;
3265 offset = EXT4_GOOD_OLD_INODE_SIZE +
3266 EXT4_I(inode)->i_extra_isize;
3268 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3269 flags |= FIEMAP_EXTENT_DATA_INLINE;
3270 } else { /* external block */
3271 physical = EXT4_I(inode)->i_file_acl << blockbits;
3272 length = inode->i_sb->s_blocksize;
3276 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3278 return (error < 0 ? error : 0);
3281 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3282 __u64 start, __u64 len)
3284 ext4_lblk_t start_blk;
3285 ext4_lblk_t len_blks;
3288 /* fallback to generic here if not in extents fmt */
3289 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3290 return generic_block_fiemap(inode, fieinfo, start, len,
3293 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3296 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3297 error = ext4_xattr_fiemap(inode, fieinfo);
3299 start_blk = start >> inode->i_sb->s_blocksize_bits;
3300 len_blks = len >> inode->i_sb->s_blocksize_bits;
3303 * Walk the extent tree gathering extent information.
3304 * ext4_ext_fiemap_cb will push extents back to user.
3306 down_read(&EXT4_I(inode)->i_data_sem);
3307 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3308 ext4_ext_fiemap_cb, fieinfo);
3309 up_read(&EXT4_I(inode)->i_data_sem);