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);
331 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
332 if (unlikely(block < le32_to_cpu(es->s_first_data_block) ||
333 ((block + len) > ext4_blocks_count(es))))
339 static int ext4_valid_extent_idx(struct inode *inode,
340 struct ext4_extent_idx *ext_idx)
342 ext4_fsblk_t block = idx_pblock(ext_idx);
343 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
344 if (unlikely(block < le32_to_cpu(es->s_first_data_block) ||
345 (block > ext4_blocks_count(es))))
351 static int ext4_valid_extent_entries(struct inode *inode,
352 struct ext4_extent_header *eh,
355 struct ext4_extent *ext;
356 struct ext4_extent_idx *ext_idx;
357 unsigned short entries;
358 if (eh->eh_entries == 0)
361 entries = le16_to_cpu(eh->eh_entries);
365 ext = EXT_FIRST_EXTENT(eh);
367 if (!ext4_valid_extent(inode, ext))
373 ext_idx = EXT_FIRST_INDEX(eh);
375 if (!ext4_valid_extent_idx(inode, ext_idx))
384 static int __ext4_ext_check(const char *function, struct inode *inode,
385 struct ext4_extent_header *eh,
388 const char *error_msg;
391 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
392 error_msg = "invalid magic";
395 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
396 error_msg = "unexpected eh_depth";
399 if (unlikely(eh->eh_max == 0)) {
400 error_msg = "invalid eh_max";
403 max = ext4_ext_max_entries(inode, depth);
404 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
405 error_msg = "too large eh_max";
408 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
409 error_msg = "invalid eh_entries";
412 if (!ext4_valid_extent_entries(inode, eh, depth)) {
413 error_msg = "invalid extent entries";
419 ext4_error(inode->i_sb, function,
420 "bad header/extent in inode #%lu: %s - magic %x, "
421 "entries %u, max %u(%u), depth %u(%u)",
422 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
423 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
424 max, le16_to_cpu(eh->eh_depth), depth);
429 #define ext4_ext_check(inode, eh, depth) \
430 __ext4_ext_check(__func__, inode, eh, depth)
432 int ext4_ext_check_inode(struct inode *inode)
434 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
438 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
440 int k, l = path->p_depth;
443 for (k = 0; k <= l; k++, path++) {
445 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
446 idx_pblock(path->p_idx));
447 } else if (path->p_ext) {
448 ext_debug(" %d:%d:%llu ",
449 le32_to_cpu(path->p_ext->ee_block),
450 ext4_ext_get_actual_len(path->p_ext),
451 ext_pblock(path->p_ext));
458 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
460 int depth = ext_depth(inode);
461 struct ext4_extent_header *eh;
462 struct ext4_extent *ex;
468 eh = path[depth].p_hdr;
469 ex = EXT_FIRST_EXTENT(eh);
471 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
472 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
473 ext4_ext_get_actual_len(ex), ext_pblock(ex));
478 #define ext4_ext_show_path(inode, path)
479 #define ext4_ext_show_leaf(inode, path)
482 void ext4_ext_drop_refs(struct ext4_ext_path *path)
484 int depth = path->p_depth;
487 for (i = 0; i <= depth; i++, path++)
495 * ext4_ext_binsearch_idx:
496 * binary search for the closest index of the given block
497 * the header must be checked before calling this
500 ext4_ext_binsearch_idx(struct inode *inode,
501 struct ext4_ext_path *path, ext4_lblk_t block)
503 struct ext4_extent_header *eh = path->p_hdr;
504 struct ext4_extent_idx *r, *l, *m;
507 ext_debug("binsearch for %u(idx): ", block);
509 l = EXT_FIRST_INDEX(eh) + 1;
510 r = EXT_LAST_INDEX(eh);
513 if (block < le32_to_cpu(m->ei_block))
517 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
518 m, le32_to_cpu(m->ei_block),
519 r, le32_to_cpu(r->ei_block));
523 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
524 idx_pblock(path->p_idx));
526 #ifdef CHECK_BINSEARCH
528 struct ext4_extent_idx *chix, *ix;
531 chix = ix = EXT_FIRST_INDEX(eh);
532 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
534 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
535 printk(KERN_DEBUG "k=%d, ix=0x%p, "
537 ix, EXT_FIRST_INDEX(eh));
538 printk(KERN_DEBUG "%u <= %u\n",
539 le32_to_cpu(ix->ei_block),
540 le32_to_cpu(ix[-1].ei_block));
542 BUG_ON(k && le32_to_cpu(ix->ei_block)
543 <= le32_to_cpu(ix[-1].ei_block));
544 if (block < le32_to_cpu(ix->ei_block))
548 BUG_ON(chix != path->p_idx);
555 * ext4_ext_binsearch:
556 * binary search for closest extent of the given block
557 * the header must be checked before calling this
560 ext4_ext_binsearch(struct inode *inode,
561 struct ext4_ext_path *path, ext4_lblk_t block)
563 struct ext4_extent_header *eh = path->p_hdr;
564 struct ext4_extent *r, *l, *m;
566 if (eh->eh_entries == 0) {
568 * this leaf is empty:
569 * we get such a leaf in split/add case
574 ext_debug("binsearch for %u: ", block);
576 l = EXT_FIRST_EXTENT(eh) + 1;
577 r = EXT_LAST_EXTENT(eh);
581 if (block < le32_to_cpu(m->ee_block))
585 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
586 m, le32_to_cpu(m->ee_block),
587 r, le32_to_cpu(r->ee_block));
591 ext_debug(" -> %d:%llu:%d ",
592 le32_to_cpu(path->p_ext->ee_block),
593 ext_pblock(path->p_ext),
594 ext4_ext_get_actual_len(path->p_ext));
596 #ifdef CHECK_BINSEARCH
598 struct ext4_extent *chex, *ex;
601 chex = ex = EXT_FIRST_EXTENT(eh);
602 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
603 BUG_ON(k && le32_to_cpu(ex->ee_block)
604 <= le32_to_cpu(ex[-1].ee_block));
605 if (block < le32_to_cpu(ex->ee_block))
609 BUG_ON(chex != path->p_ext);
615 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
617 struct ext4_extent_header *eh;
619 eh = ext_inode_hdr(inode);
622 eh->eh_magic = EXT4_EXT_MAGIC;
623 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
624 ext4_mark_inode_dirty(handle, inode);
625 ext4_ext_invalidate_cache(inode);
629 struct ext4_ext_path *
630 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
631 struct ext4_ext_path *path)
633 struct ext4_extent_header *eh;
634 struct buffer_head *bh;
635 short int depth, i, ppos = 0, alloc = 0;
637 eh = ext_inode_hdr(inode);
638 depth = ext_depth(inode);
640 /* account possible depth increase */
642 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
645 return ERR_PTR(-ENOMEM);
652 /* walk through the tree */
654 int need_to_validate = 0;
656 ext_debug("depth %d: num %d, max %d\n",
657 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
659 ext4_ext_binsearch_idx(inode, path + ppos, block);
660 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
661 path[ppos].p_depth = i;
662 path[ppos].p_ext = NULL;
664 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
667 if (!bh_uptodate_or_lock(bh)) {
668 if (bh_submit_read(bh) < 0) {
672 /* validate the extent entries */
673 need_to_validate = 1;
675 eh = ext_block_hdr(bh);
677 BUG_ON(ppos > depth);
678 path[ppos].p_bh = bh;
679 path[ppos].p_hdr = eh;
682 if (need_to_validate && ext4_ext_check(inode, eh, i))
686 path[ppos].p_depth = i;
687 path[ppos].p_ext = NULL;
688 path[ppos].p_idx = NULL;
691 ext4_ext_binsearch(inode, path + ppos, block);
692 /* if not an empty leaf */
693 if (path[ppos].p_ext)
694 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
696 ext4_ext_show_path(inode, path);
701 ext4_ext_drop_refs(path);
704 return ERR_PTR(-EIO);
708 * ext4_ext_insert_index:
709 * insert new index [@logical;@ptr] into the block at @curp;
710 * check where to insert: before @curp or after @curp
712 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
713 struct ext4_ext_path *curp,
714 int logical, ext4_fsblk_t ptr)
716 struct ext4_extent_idx *ix;
719 err = ext4_ext_get_access(handle, inode, curp);
723 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
724 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
725 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
727 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
728 len = (len - 1) * sizeof(struct ext4_extent_idx);
729 len = len < 0 ? 0 : len;
730 ext_debug("insert new index %d after: %llu. "
731 "move %d from 0x%p to 0x%p\n",
733 (curp->p_idx + 1), (curp->p_idx + 2));
734 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
736 ix = curp->p_idx + 1;
739 len = len * sizeof(struct ext4_extent_idx);
740 len = len < 0 ? 0 : len;
741 ext_debug("insert new index %d before: %llu. "
742 "move %d from 0x%p to 0x%p\n",
744 curp->p_idx, (curp->p_idx + 1));
745 memmove(curp->p_idx + 1, curp->p_idx, len);
749 ix->ei_block = cpu_to_le32(logical);
750 ext4_idx_store_pblock(ix, ptr);
751 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
753 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
754 > le16_to_cpu(curp->p_hdr->eh_max));
755 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
757 err = ext4_ext_dirty(handle, inode, curp);
758 ext4_std_error(inode->i_sb, err);
765 * inserts new subtree into the path, using free index entry
767 * - allocates all needed blocks (new leaf and all intermediate index blocks)
768 * - makes decision where to split
769 * - moves remaining extents and index entries (right to the split point)
770 * into the newly allocated blocks
771 * - initializes subtree
773 static int ext4_ext_split(handle_t *handle, struct inode *inode,
774 struct ext4_ext_path *path,
775 struct ext4_extent *newext, int at)
777 struct buffer_head *bh = NULL;
778 int depth = ext_depth(inode);
779 struct ext4_extent_header *neh;
780 struct ext4_extent_idx *fidx;
781 struct ext4_extent *ex;
783 ext4_fsblk_t newblock, oldblock;
785 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
788 /* make decision: where to split? */
789 /* FIXME: now decision is simplest: at current extent */
791 /* if current leaf will be split, then we should use
792 * border from split point */
793 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
794 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
795 border = path[depth].p_ext[1].ee_block;
796 ext_debug("leaf will be split."
797 " next leaf starts at %d\n",
798 le32_to_cpu(border));
800 border = newext->ee_block;
801 ext_debug("leaf will be added."
802 " next leaf starts at %d\n",
803 le32_to_cpu(border));
807 * If error occurs, then we break processing
808 * and mark filesystem read-only. index won't
809 * be inserted and tree will be in consistent
810 * state. Next mount will repair buffers too.
814 * Get array to track all allocated blocks.
815 * We need this to handle errors and free blocks
818 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
822 /* allocate all needed blocks */
823 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
824 for (a = 0; a < depth - at; a++) {
825 newblock = ext4_ext_new_meta_block(handle, inode, path,
829 ablocks[a] = newblock;
832 /* initialize new leaf */
833 newblock = ablocks[--a];
834 BUG_ON(newblock == 0);
835 bh = sb_getblk(inode->i_sb, newblock);
842 err = ext4_journal_get_create_access(handle, bh);
846 neh = ext_block_hdr(bh);
848 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
849 neh->eh_magic = EXT4_EXT_MAGIC;
851 ex = EXT_FIRST_EXTENT(neh);
853 /* move remainder of path[depth] to the new leaf */
854 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
855 /* start copy from next extent */
856 /* TODO: we could do it by single memmove */
859 while (path[depth].p_ext <=
860 EXT_MAX_EXTENT(path[depth].p_hdr)) {
861 ext_debug("move %d:%llu:%d in new leaf %llu\n",
862 le32_to_cpu(path[depth].p_ext->ee_block),
863 ext_pblock(path[depth].p_ext),
864 ext4_ext_get_actual_len(path[depth].p_ext),
866 /*memmove(ex++, path[depth].p_ext++,
867 sizeof(struct ext4_extent));
873 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
874 le16_add_cpu(&neh->eh_entries, m);
877 set_buffer_uptodate(bh);
880 err = ext4_handle_dirty_metadata(handle, inode, bh);
886 /* correct old leaf */
888 err = ext4_ext_get_access(handle, inode, path + depth);
891 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
892 err = ext4_ext_dirty(handle, inode, path + depth);
898 /* create intermediate indexes */
902 ext_debug("create %d intermediate indices\n", k);
903 /* insert new index into current index block */
904 /* current depth stored in i var */
908 newblock = ablocks[--a];
909 bh = sb_getblk(inode->i_sb, newblock);
916 err = ext4_journal_get_create_access(handle, bh);
920 neh = ext_block_hdr(bh);
921 neh->eh_entries = cpu_to_le16(1);
922 neh->eh_magic = EXT4_EXT_MAGIC;
923 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
924 neh->eh_depth = cpu_to_le16(depth - i);
925 fidx = EXT_FIRST_INDEX(neh);
926 fidx->ei_block = border;
927 ext4_idx_store_pblock(fidx, oldblock);
929 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
930 i, newblock, le32_to_cpu(border), oldblock);
935 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
936 EXT_MAX_INDEX(path[i].p_hdr));
937 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
938 EXT_LAST_INDEX(path[i].p_hdr));
939 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
940 ext_debug("%d: move %d:%llu in new index %llu\n", i,
941 le32_to_cpu(path[i].p_idx->ei_block),
942 idx_pblock(path[i].p_idx),
944 /*memmove(++fidx, path[i].p_idx++,
945 sizeof(struct ext4_extent_idx));
947 BUG_ON(neh->eh_entries > neh->eh_max);*/
952 memmove(++fidx, path[i].p_idx - m,
953 sizeof(struct ext4_extent_idx) * m);
954 le16_add_cpu(&neh->eh_entries, m);
956 set_buffer_uptodate(bh);
959 err = ext4_handle_dirty_metadata(handle, inode, bh);
965 /* correct old index */
967 err = ext4_ext_get_access(handle, inode, path + i);
970 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
971 err = ext4_ext_dirty(handle, inode, path + i);
979 /* insert new index */
980 err = ext4_ext_insert_index(handle, inode, path + at,
981 le32_to_cpu(border), newblock);
985 if (buffer_locked(bh))
991 /* free all allocated blocks in error case */
992 for (i = 0; i < depth; i++) {
995 ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
1004 * ext4_ext_grow_indepth:
1005 * implements tree growing procedure:
1006 * - allocates new block
1007 * - moves top-level data (index block or leaf) into the new block
1008 * - initializes new top-level, creating index that points to the
1009 * just created block
1011 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1012 struct ext4_ext_path *path,
1013 struct ext4_extent *newext)
1015 struct ext4_ext_path *curp = path;
1016 struct ext4_extent_header *neh;
1017 struct ext4_extent_idx *fidx;
1018 struct buffer_head *bh;
1019 ext4_fsblk_t newblock;
1022 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1026 bh = sb_getblk(inode->i_sb, newblock);
1029 ext4_std_error(inode->i_sb, err);
1034 err = ext4_journal_get_create_access(handle, bh);
1040 /* move top-level index/leaf into new block */
1041 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1043 /* set size of new block */
1044 neh = ext_block_hdr(bh);
1045 /* old root could have indexes or leaves
1046 * so calculate e_max right way */
1047 if (ext_depth(inode))
1048 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
1050 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
1051 neh->eh_magic = EXT4_EXT_MAGIC;
1052 set_buffer_uptodate(bh);
1055 err = ext4_handle_dirty_metadata(handle, inode, bh);
1059 /* create index in new top-level index: num,max,pointer */
1060 err = ext4_ext_get_access(handle, inode, curp);
1064 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1065 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
1066 curp->p_hdr->eh_entries = cpu_to_le16(1);
1067 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1069 if (path[0].p_hdr->eh_depth)
1070 curp->p_idx->ei_block =
1071 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1073 curp->p_idx->ei_block =
1074 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1075 ext4_idx_store_pblock(curp->p_idx, newblock);
1077 neh = ext_inode_hdr(inode);
1078 fidx = EXT_FIRST_INDEX(neh);
1079 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1080 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1081 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1083 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1084 err = ext4_ext_dirty(handle, inode, curp);
1092 * ext4_ext_create_new_leaf:
1093 * finds empty index and adds new leaf.
1094 * if no free index is found, then it requests in-depth growing.
1096 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1097 struct ext4_ext_path *path,
1098 struct ext4_extent *newext)
1100 struct ext4_ext_path *curp;
1101 int depth, i, err = 0;
1104 i = depth = ext_depth(inode);
1106 /* walk up to the tree and look for free index entry */
1107 curp = path + depth;
1108 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1113 /* we use already allocated block for index block,
1114 * so subsequent data blocks should be contiguous */
1115 if (EXT_HAS_FREE_INDEX(curp)) {
1116 /* if we found index with free entry, then use that
1117 * entry: create all needed subtree and add new leaf */
1118 err = ext4_ext_split(handle, inode, path, newext, i);
1123 ext4_ext_drop_refs(path);
1124 path = ext4_ext_find_extent(inode,
1125 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1128 err = PTR_ERR(path);
1130 /* tree is full, time to grow in depth */
1131 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1136 ext4_ext_drop_refs(path);
1137 path = ext4_ext_find_extent(inode,
1138 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1141 err = PTR_ERR(path);
1146 * only first (depth 0 -> 1) produces free space;
1147 * in all other cases we have to split the grown tree
1149 depth = ext_depth(inode);
1150 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1151 /* now we need to split */
1161 * search the closest allocated block to the left for *logical
1162 * and returns it at @logical + it's physical address at @phys
1163 * if *logical is the smallest allocated block, the function
1164 * returns 0 at @phys
1165 * return value contains 0 (success) or error code
1168 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1169 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1171 struct ext4_extent_idx *ix;
1172 struct ext4_extent *ex;
1175 BUG_ON(path == NULL);
1176 depth = path->p_depth;
1179 if (depth == 0 && path->p_ext == NULL)
1182 /* usually extent in the path covers blocks smaller
1183 * then *logical, but it can be that extent is the
1184 * first one in the file */
1186 ex = path[depth].p_ext;
1187 ee_len = ext4_ext_get_actual_len(ex);
1188 if (*logical < le32_to_cpu(ex->ee_block)) {
1189 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1190 while (--depth >= 0) {
1191 ix = path[depth].p_idx;
1192 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1197 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1199 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1200 *phys = ext_pblock(ex) + ee_len - 1;
1205 * search the closest allocated block to the right for *logical
1206 * and returns it at @logical + it's physical address at @phys
1207 * if *logical is the smallest allocated block, the function
1208 * returns 0 at @phys
1209 * return value contains 0 (success) or error code
1212 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1213 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1215 struct buffer_head *bh = NULL;
1216 struct ext4_extent_header *eh;
1217 struct ext4_extent_idx *ix;
1218 struct ext4_extent *ex;
1220 int depth; /* Note, NOT eh_depth; depth from top of tree */
1223 BUG_ON(path == NULL);
1224 depth = path->p_depth;
1227 if (depth == 0 && path->p_ext == NULL)
1230 /* usually extent in the path covers blocks smaller
1231 * then *logical, but it can be that extent is the
1232 * first one in the file */
1234 ex = path[depth].p_ext;
1235 ee_len = ext4_ext_get_actual_len(ex);
1236 if (*logical < le32_to_cpu(ex->ee_block)) {
1237 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1238 while (--depth >= 0) {
1239 ix = path[depth].p_idx;
1240 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1242 *logical = le32_to_cpu(ex->ee_block);
1243 *phys = ext_pblock(ex);
1247 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1249 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1250 /* next allocated block in this leaf */
1252 *logical = le32_to_cpu(ex->ee_block);
1253 *phys = ext_pblock(ex);
1257 /* go up and search for index to the right */
1258 while (--depth >= 0) {
1259 ix = path[depth].p_idx;
1260 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1264 /* we've gone up to the root and found no index to the right */
1268 /* we've found index to the right, let's
1269 * follow it and find the closest allocated
1270 * block to the right */
1272 block = idx_pblock(ix);
1273 while (++depth < path->p_depth) {
1274 bh = sb_bread(inode->i_sb, block);
1277 eh = ext_block_hdr(bh);
1278 /* subtract from p_depth to get proper eh_depth */
1279 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1283 ix = EXT_FIRST_INDEX(eh);
1284 block = idx_pblock(ix);
1288 bh = sb_bread(inode->i_sb, block);
1291 eh = ext_block_hdr(bh);
1292 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1296 ex = EXT_FIRST_EXTENT(eh);
1297 *logical = le32_to_cpu(ex->ee_block);
1298 *phys = ext_pblock(ex);
1304 * ext4_ext_next_allocated_block:
1305 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1306 * NOTE: it considers block number from index entry as
1307 * allocated block. Thus, index entries have to be consistent
1311 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1315 BUG_ON(path == NULL);
1316 depth = path->p_depth;
1318 if (depth == 0 && path->p_ext == NULL)
1319 return EXT_MAX_BLOCK;
1321 while (depth >= 0) {
1322 if (depth == path->p_depth) {
1324 if (path[depth].p_ext !=
1325 EXT_LAST_EXTENT(path[depth].p_hdr))
1326 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1329 if (path[depth].p_idx !=
1330 EXT_LAST_INDEX(path[depth].p_hdr))
1331 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1336 return EXT_MAX_BLOCK;
1340 * ext4_ext_next_leaf_block:
1341 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1343 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1344 struct ext4_ext_path *path)
1348 BUG_ON(path == NULL);
1349 depth = path->p_depth;
1351 /* zero-tree has no leaf blocks at all */
1353 return EXT_MAX_BLOCK;
1355 /* go to index block */
1358 while (depth >= 0) {
1359 if (path[depth].p_idx !=
1360 EXT_LAST_INDEX(path[depth].p_hdr))
1361 return (ext4_lblk_t)
1362 le32_to_cpu(path[depth].p_idx[1].ei_block);
1366 return EXT_MAX_BLOCK;
1370 * ext4_ext_correct_indexes:
1371 * if leaf gets modified and modified extent is first in the leaf,
1372 * then we have to correct all indexes above.
1373 * TODO: do we need to correct tree in all cases?
1375 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1376 struct ext4_ext_path *path)
1378 struct ext4_extent_header *eh;
1379 int depth = ext_depth(inode);
1380 struct ext4_extent *ex;
1384 eh = path[depth].p_hdr;
1385 ex = path[depth].p_ext;
1390 /* there is no tree at all */
1394 if (ex != EXT_FIRST_EXTENT(eh)) {
1395 /* we correct tree if first leaf got modified only */
1400 * TODO: we need correction if border is smaller than current one
1403 border = path[depth].p_ext->ee_block;
1404 err = ext4_ext_get_access(handle, inode, path + k);
1407 path[k].p_idx->ei_block = border;
1408 err = ext4_ext_dirty(handle, inode, path + k);
1413 /* change all left-side indexes */
1414 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1416 err = ext4_ext_get_access(handle, inode, path + k);
1419 path[k].p_idx->ei_block = border;
1420 err = ext4_ext_dirty(handle, inode, path + k);
1429 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1430 struct ext4_extent *ex2)
1432 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1435 * Make sure that either both extents are uninitialized, or
1438 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1441 if (ext4_ext_is_uninitialized(ex1))
1442 max_len = EXT_UNINIT_MAX_LEN;
1444 max_len = EXT_INIT_MAX_LEN;
1446 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1447 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1449 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1450 le32_to_cpu(ex2->ee_block))
1454 * To allow future support for preallocated extents to be added
1455 * as an RO_COMPAT feature, refuse to merge to extents if
1456 * this can result in the top bit of ee_len being set.
1458 if (ext1_ee_len + ext2_ee_len > max_len)
1460 #ifdef AGGRESSIVE_TEST
1461 if (ext1_ee_len >= 4)
1465 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1471 * This function tries to merge the "ex" extent to the next extent in the tree.
1472 * It always tries to merge towards right. If you want to merge towards
1473 * left, pass "ex - 1" as argument instead of "ex".
1474 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1475 * 1 if they got merged.
1477 int ext4_ext_try_to_merge(struct inode *inode,
1478 struct ext4_ext_path *path,
1479 struct ext4_extent *ex)
1481 struct ext4_extent_header *eh;
1482 unsigned int depth, len;
1484 int uninitialized = 0;
1486 depth = ext_depth(inode);
1487 BUG_ON(path[depth].p_hdr == NULL);
1488 eh = path[depth].p_hdr;
1490 while (ex < EXT_LAST_EXTENT(eh)) {
1491 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1493 /* merge with next extent! */
1494 if (ext4_ext_is_uninitialized(ex))
1496 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1497 + ext4_ext_get_actual_len(ex + 1));
1499 ext4_ext_mark_uninitialized(ex);
1501 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1502 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1503 * sizeof(struct ext4_extent);
1504 memmove(ex + 1, ex + 2, len);
1506 le16_add_cpu(&eh->eh_entries, -1);
1508 WARN_ON(eh->eh_entries == 0);
1509 if (!eh->eh_entries)
1510 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1511 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1518 * check if a portion of the "newext" extent overlaps with an
1521 * If there is an overlap discovered, it updates the length of the newext
1522 * such that there will be no overlap, and then returns 1.
1523 * If there is no overlap found, it returns 0.
1525 unsigned int ext4_ext_check_overlap(struct inode *inode,
1526 struct ext4_extent *newext,
1527 struct ext4_ext_path *path)
1530 unsigned int depth, len1;
1531 unsigned int ret = 0;
1533 b1 = le32_to_cpu(newext->ee_block);
1534 len1 = ext4_ext_get_actual_len(newext);
1535 depth = ext_depth(inode);
1536 if (!path[depth].p_ext)
1538 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1541 * get the next allocated block if the extent in the path
1542 * is before the requested block(s)
1545 b2 = ext4_ext_next_allocated_block(path);
1546 if (b2 == EXT_MAX_BLOCK)
1550 /* check for wrap through zero on extent logical start block*/
1551 if (b1 + len1 < b1) {
1552 len1 = EXT_MAX_BLOCK - b1;
1553 newext->ee_len = cpu_to_le16(len1);
1557 /* check for overlap */
1558 if (b1 + len1 > b2) {
1559 newext->ee_len = cpu_to_le16(b2 - b1);
1567 * ext4_ext_insert_extent:
1568 * tries to merge requsted extent into the existing extent or
1569 * inserts requested extent as new one into the tree,
1570 * creating new leaf in the no-space case.
1572 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1573 struct ext4_ext_path *path,
1574 struct ext4_extent *newext)
1576 struct ext4_extent_header *eh;
1577 struct ext4_extent *ex, *fex;
1578 struct ext4_extent *nearex; /* nearest extent */
1579 struct ext4_ext_path *npath = NULL;
1580 int depth, len, err;
1582 unsigned uninitialized = 0;
1584 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1585 depth = ext_depth(inode);
1586 ex = path[depth].p_ext;
1587 BUG_ON(path[depth].p_hdr == NULL);
1589 /* try to insert block into found extent and return */
1590 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1591 ext_debug("append %d block to %d:%d (from %llu)\n",
1592 ext4_ext_get_actual_len(newext),
1593 le32_to_cpu(ex->ee_block),
1594 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1595 err = ext4_ext_get_access(handle, inode, path + depth);
1600 * ext4_can_extents_be_merged should have checked that either
1601 * both extents are uninitialized, or both aren't. Thus we
1602 * need to check only one of them here.
1604 if (ext4_ext_is_uninitialized(ex))
1606 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1607 + ext4_ext_get_actual_len(newext));
1609 ext4_ext_mark_uninitialized(ex);
1610 eh = path[depth].p_hdr;
1616 depth = ext_depth(inode);
1617 eh = path[depth].p_hdr;
1618 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1621 /* probably next leaf has space for us? */
1622 fex = EXT_LAST_EXTENT(eh);
1623 next = ext4_ext_next_leaf_block(inode, path);
1624 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1625 && next != EXT_MAX_BLOCK) {
1626 ext_debug("next leaf block - %d\n", next);
1627 BUG_ON(npath != NULL);
1628 npath = ext4_ext_find_extent(inode, next, NULL);
1630 return PTR_ERR(npath);
1631 BUG_ON(npath->p_depth != path->p_depth);
1632 eh = npath[depth].p_hdr;
1633 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1634 ext_debug("next leaf isnt full(%d)\n",
1635 le16_to_cpu(eh->eh_entries));
1639 ext_debug("next leaf has no free space(%d,%d)\n",
1640 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1644 * There is no free space in the found leaf.
1645 * We're gonna add a new leaf in the tree.
1647 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1650 depth = ext_depth(inode);
1651 eh = path[depth].p_hdr;
1654 nearex = path[depth].p_ext;
1656 err = ext4_ext_get_access(handle, inode, path + depth);
1661 /* there is no extent in this leaf, create first one */
1662 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1663 le32_to_cpu(newext->ee_block),
1665 ext4_ext_get_actual_len(newext));
1666 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1667 } else if (le32_to_cpu(newext->ee_block)
1668 > le32_to_cpu(nearex->ee_block)) {
1669 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1670 if (nearex != EXT_LAST_EXTENT(eh)) {
1671 len = EXT_MAX_EXTENT(eh) - nearex;
1672 len = (len - 1) * sizeof(struct ext4_extent);
1673 len = len < 0 ? 0 : len;
1674 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1675 "move %d from 0x%p to 0x%p\n",
1676 le32_to_cpu(newext->ee_block),
1678 ext4_ext_get_actual_len(newext),
1679 nearex, len, nearex + 1, nearex + 2);
1680 memmove(nearex + 2, nearex + 1, len);
1682 path[depth].p_ext = nearex + 1;
1684 BUG_ON(newext->ee_block == nearex->ee_block);
1685 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1686 len = len < 0 ? 0 : len;
1687 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1688 "move %d from 0x%p to 0x%p\n",
1689 le32_to_cpu(newext->ee_block),
1691 ext4_ext_get_actual_len(newext),
1692 nearex, len, nearex + 1, nearex + 2);
1693 memmove(nearex + 1, nearex, len);
1694 path[depth].p_ext = nearex;
1697 le16_add_cpu(&eh->eh_entries, 1);
1698 nearex = path[depth].p_ext;
1699 nearex->ee_block = newext->ee_block;
1700 ext4_ext_store_pblock(nearex, ext_pblock(newext));
1701 nearex->ee_len = newext->ee_len;
1704 /* try to merge extents to the right */
1705 ext4_ext_try_to_merge(inode, path, nearex);
1707 /* try to merge extents to the left */
1709 /* time to correct all indexes above */
1710 err = ext4_ext_correct_indexes(handle, inode, path);
1714 err = ext4_ext_dirty(handle, inode, path + depth);
1718 ext4_ext_drop_refs(npath);
1721 ext4_ext_invalidate_cache(inode);
1725 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1726 ext4_lblk_t num, ext_prepare_callback func,
1729 struct ext4_ext_path *path = NULL;
1730 struct ext4_ext_cache cbex;
1731 struct ext4_extent *ex;
1732 ext4_lblk_t next, start = 0, end = 0;
1733 ext4_lblk_t last = block + num;
1734 int depth, exists, err = 0;
1736 BUG_ON(func == NULL);
1737 BUG_ON(inode == NULL);
1739 while (block < last && block != EXT_MAX_BLOCK) {
1741 /* find extent for this block */
1742 path = ext4_ext_find_extent(inode, block, path);
1744 err = PTR_ERR(path);
1749 depth = ext_depth(inode);
1750 BUG_ON(path[depth].p_hdr == NULL);
1751 ex = path[depth].p_ext;
1752 next = ext4_ext_next_allocated_block(path);
1756 /* there is no extent yet, so try to allocate
1757 * all requested space */
1760 } else if (le32_to_cpu(ex->ee_block) > block) {
1761 /* need to allocate space before found extent */
1763 end = le32_to_cpu(ex->ee_block);
1764 if (block + num < end)
1766 } else if (block >= le32_to_cpu(ex->ee_block)
1767 + ext4_ext_get_actual_len(ex)) {
1768 /* need to allocate space after found extent */
1773 } else if (block >= le32_to_cpu(ex->ee_block)) {
1775 * some part of requested space is covered
1779 end = le32_to_cpu(ex->ee_block)
1780 + ext4_ext_get_actual_len(ex);
1781 if (block + num < end)
1787 BUG_ON(end <= start);
1790 cbex.ec_block = start;
1791 cbex.ec_len = end - start;
1793 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1795 cbex.ec_block = le32_to_cpu(ex->ee_block);
1796 cbex.ec_len = ext4_ext_get_actual_len(ex);
1797 cbex.ec_start = ext_pblock(ex);
1798 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1801 BUG_ON(cbex.ec_len == 0);
1802 err = func(inode, path, &cbex, ex, cbdata);
1803 ext4_ext_drop_refs(path);
1808 if (err == EXT_REPEAT)
1810 else if (err == EXT_BREAK) {
1815 if (ext_depth(inode) != depth) {
1816 /* depth was changed. we have to realloc path */
1821 block = cbex.ec_block + cbex.ec_len;
1825 ext4_ext_drop_refs(path);
1833 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1834 __u32 len, ext4_fsblk_t start, int type)
1836 struct ext4_ext_cache *cex;
1838 cex = &EXT4_I(inode)->i_cached_extent;
1839 cex->ec_type = type;
1840 cex->ec_block = block;
1842 cex->ec_start = start;
1846 * ext4_ext_put_gap_in_cache:
1847 * calculate boundaries of the gap that the requested block fits into
1848 * and cache this gap
1851 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1854 int depth = ext_depth(inode);
1857 struct ext4_extent *ex;
1859 ex = path[depth].p_ext;
1861 /* there is no extent yet, so gap is [0;-] */
1863 len = EXT_MAX_BLOCK;
1864 ext_debug("cache gap(whole file):");
1865 } else if (block < le32_to_cpu(ex->ee_block)) {
1867 len = le32_to_cpu(ex->ee_block) - block;
1868 ext_debug("cache gap(before): %u [%u:%u]",
1870 le32_to_cpu(ex->ee_block),
1871 ext4_ext_get_actual_len(ex));
1872 } else if (block >= le32_to_cpu(ex->ee_block)
1873 + ext4_ext_get_actual_len(ex)) {
1875 lblock = le32_to_cpu(ex->ee_block)
1876 + ext4_ext_get_actual_len(ex);
1878 next = ext4_ext_next_allocated_block(path);
1879 ext_debug("cache gap(after): [%u:%u] %u",
1880 le32_to_cpu(ex->ee_block),
1881 ext4_ext_get_actual_len(ex),
1883 BUG_ON(next == lblock);
1884 len = next - lblock;
1890 ext_debug(" -> %u:%lu\n", lblock, len);
1891 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1895 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1896 struct ext4_extent *ex)
1898 struct ext4_ext_cache *cex;
1900 cex = &EXT4_I(inode)->i_cached_extent;
1902 /* has cache valid data? */
1903 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1904 return EXT4_EXT_CACHE_NO;
1906 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1907 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1908 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1909 ex->ee_block = cpu_to_le32(cex->ec_block);
1910 ext4_ext_store_pblock(ex, cex->ec_start);
1911 ex->ee_len = cpu_to_le16(cex->ec_len);
1912 ext_debug("%u cached by %u:%u:%llu\n",
1914 cex->ec_block, cex->ec_len, cex->ec_start);
1915 return cex->ec_type;
1919 return EXT4_EXT_CACHE_NO;
1924 * removes index from the index block.
1925 * It's used in truncate case only, thus all requests are for
1926 * last index in the block only.
1928 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1929 struct ext4_ext_path *path)
1931 struct buffer_head *bh;
1935 /* free index block */
1937 leaf = idx_pblock(path->p_idx);
1938 BUG_ON(path->p_hdr->eh_entries == 0);
1939 err = ext4_ext_get_access(handle, inode, path);
1942 le16_add_cpu(&path->p_hdr->eh_entries, -1);
1943 err = ext4_ext_dirty(handle, inode, path);
1946 ext_debug("index is empty, remove it, free block %llu\n", leaf);
1947 bh = sb_find_get_block(inode->i_sb, leaf);
1948 ext4_forget(handle, 1, inode, bh, leaf);
1949 ext4_free_blocks(handle, inode, leaf, 1, 1);
1954 * ext4_ext_calc_credits_for_single_extent:
1955 * This routine returns max. credits that needed to insert an extent
1956 * to the extent tree.
1957 * When pass the actual path, the caller should calculate credits
1960 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
1961 struct ext4_ext_path *path)
1964 int depth = ext_depth(inode);
1967 /* probably there is space in leaf? */
1968 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1969 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
1972 * There are some space in the leaf tree, no
1973 * need to account for leaf block credit
1975 * bitmaps and block group descriptor blocks
1976 * and other metadat blocks still need to be
1979 /* 1 bitmap, 1 block group descriptor */
1980 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
1984 return ext4_chunk_trans_blocks(inode, nrblocks);
1988 * How many index/leaf blocks need to change/allocate to modify nrblocks?
1990 * if nrblocks are fit in a single extent (chunk flag is 1), then
1991 * in the worse case, each tree level index/leaf need to be changed
1992 * if the tree split due to insert a new extent, then the old tree
1993 * index/leaf need to be updated too
1995 * If the nrblocks are discontiguous, they could cause
1996 * the whole tree split more than once, but this is really rare.
1998 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2001 int depth = ext_depth(inode);
2011 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2012 struct ext4_extent *ex,
2013 ext4_lblk_t from, ext4_lblk_t to)
2015 struct buffer_head *bh;
2016 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2017 int i, metadata = 0;
2019 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2021 #ifdef EXTENTS_STATS
2023 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2024 spin_lock(&sbi->s_ext_stats_lock);
2025 sbi->s_ext_blocks += ee_len;
2026 sbi->s_ext_extents++;
2027 if (ee_len < sbi->s_ext_min)
2028 sbi->s_ext_min = ee_len;
2029 if (ee_len > sbi->s_ext_max)
2030 sbi->s_ext_max = ee_len;
2031 if (ext_depth(inode) > sbi->s_depth_max)
2032 sbi->s_depth_max = ext_depth(inode);
2033 spin_unlock(&sbi->s_ext_stats_lock);
2036 if (from >= le32_to_cpu(ex->ee_block)
2037 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2042 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2043 start = ext_pblock(ex) + ee_len - num;
2044 ext_debug("free last %u blocks starting %llu\n", num, start);
2045 for (i = 0; i < num; i++) {
2046 bh = sb_find_get_block(inode->i_sb, start + i);
2047 ext4_forget(handle, 0, inode, bh, start + i);
2049 ext4_free_blocks(handle, inode, start, num, metadata);
2050 } else if (from == le32_to_cpu(ex->ee_block)
2051 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2052 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2053 from, to, le32_to_cpu(ex->ee_block), ee_len);
2055 printk(KERN_INFO "strange request: removal(2) "
2056 "%u-%u from %u:%u\n",
2057 from, to, le32_to_cpu(ex->ee_block), ee_len);
2063 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2064 struct ext4_ext_path *path, ext4_lblk_t start)
2066 int err = 0, correct_index = 0;
2067 int depth = ext_depth(inode), credits;
2068 struct ext4_extent_header *eh;
2069 ext4_lblk_t a, b, block;
2071 ext4_lblk_t ex_ee_block;
2072 unsigned short ex_ee_len;
2073 unsigned uninitialized = 0;
2074 struct ext4_extent *ex;
2076 /* the header must be checked already in ext4_ext_remove_space() */
2077 ext_debug("truncate since %u in leaf\n", start);
2078 if (!path[depth].p_hdr)
2079 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2080 eh = path[depth].p_hdr;
2083 /* find where to start removing */
2084 ex = EXT_LAST_EXTENT(eh);
2086 ex_ee_block = le32_to_cpu(ex->ee_block);
2087 if (ext4_ext_is_uninitialized(ex))
2089 ex_ee_len = ext4_ext_get_actual_len(ex);
2091 while (ex >= EXT_FIRST_EXTENT(eh) &&
2092 ex_ee_block + ex_ee_len > start) {
2093 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
2094 path[depth].p_ext = ex;
2096 a = ex_ee_block > start ? ex_ee_block : start;
2097 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2098 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2100 ext_debug(" border %u:%u\n", a, b);
2102 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2106 } else if (a != ex_ee_block) {
2107 /* remove tail of the extent */
2108 block = ex_ee_block;
2110 } else if (b != ex_ee_block + ex_ee_len - 1) {
2111 /* remove head of the extent */
2114 /* there is no "make a hole" API yet */
2117 /* remove whole extent: excellent! */
2118 block = ex_ee_block;
2120 BUG_ON(a != ex_ee_block);
2121 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2125 * 3 for leaf, sb, and inode plus 2 (bmap and group
2126 * descriptor) for each block group; assume two block
2127 * groups plus ex_ee_len/blocks_per_block_group for
2130 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2131 if (ex == EXT_FIRST_EXTENT(eh)) {
2133 credits += (ext_depth(inode)) + 1;
2135 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2137 err = ext4_ext_journal_restart(handle, credits);
2141 err = ext4_ext_get_access(handle, inode, path + depth);
2145 err = ext4_remove_blocks(handle, inode, ex, a, b);
2150 /* this extent is removed; mark slot entirely unused */
2151 ext4_ext_store_pblock(ex, 0);
2152 le16_add_cpu(&eh->eh_entries, -1);
2155 ex->ee_block = cpu_to_le32(block);
2156 ex->ee_len = cpu_to_le16(num);
2158 * Do not mark uninitialized if all the blocks in the
2159 * extent have been removed.
2161 if (uninitialized && num)
2162 ext4_ext_mark_uninitialized(ex);
2164 err = ext4_ext_dirty(handle, inode, path + depth);
2168 ext_debug("new extent: %u:%u:%llu\n", block, num,
2171 ex_ee_block = le32_to_cpu(ex->ee_block);
2172 ex_ee_len = ext4_ext_get_actual_len(ex);
2175 if (correct_index && eh->eh_entries)
2176 err = ext4_ext_correct_indexes(handle, inode, path);
2178 /* if this leaf is free, then we should
2179 * remove it from index block above */
2180 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2181 err = ext4_ext_rm_idx(handle, inode, path + depth);
2188 * ext4_ext_more_to_rm:
2189 * returns 1 if current index has to be freed (even partial)
2192 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2194 BUG_ON(path->p_idx == NULL);
2196 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2200 * if truncate on deeper level happened, it wasn't partial,
2201 * so we have to consider current index for truncation
2203 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2208 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2210 struct super_block *sb = inode->i_sb;
2211 int depth = ext_depth(inode);
2212 struct ext4_ext_path *path;
2216 ext_debug("truncate since %u\n", start);
2218 /* probably first extent we're gonna free will be last in block */
2219 handle = ext4_journal_start(inode, depth + 1);
2221 return PTR_ERR(handle);
2223 ext4_ext_invalidate_cache(inode);
2226 * We start scanning from right side, freeing all the blocks
2227 * after i_size and walking into the tree depth-wise.
2229 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2231 ext4_journal_stop(handle);
2234 path[0].p_hdr = ext_inode_hdr(inode);
2235 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2239 path[0].p_depth = depth;
2241 while (i >= 0 && err == 0) {
2243 /* this is leaf block */
2244 err = ext4_ext_rm_leaf(handle, inode, path, start);
2245 /* root level has p_bh == NULL, brelse() eats this */
2246 brelse(path[i].p_bh);
2247 path[i].p_bh = NULL;
2252 /* this is index block */
2253 if (!path[i].p_hdr) {
2254 ext_debug("initialize header\n");
2255 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2258 if (!path[i].p_idx) {
2259 /* this level hasn't been touched yet */
2260 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2261 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2262 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2264 le16_to_cpu(path[i].p_hdr->eh_entries));
2266 /* we were already here, see at next index */
2270 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2271 i, EXT_FIRST_INDEX(path[i].p_hdr),
2273 if (ext4_ext_more_to_rm(path + i)) {
2274 struct buffer_head *bh;
2275 /* go to the next level */
2276 ext_debug("move to level %d (block %llu)\n",
2277 i + 1, idx_pblock(path[i].p_idx));
2278 memset(path + i + 1, 0, sizeof(*path));
2279 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2281 /* should we reset i_size? */
2285 if (WARN_ON(i + 1 > depth)) {
2289 if (ext4_ext_check(inode, ext_block_hdr(bh),
2294 path[i + 1].p_bh = bh;
2296 /* save actual number of indexes since this
2297 * number is changed at the next iteration */
2298 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2301 /* we finished processing this index, go up */
2302 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2303 /* index is empty, remove it;
2304 * handle must be already prepared by the
2305 * truncatei_leaf() */
2306 err = ext4_ext_rm_idx(handle, inode, path + i);
2308 /* root level has p_bh == NULL, brelse() eats this */
2309 brelse(path[i].p_bh);
2310 path[i].p_bh = NULL;
2312 ext_debug("return to level %d\n", i);
2316 /* TODO: flexible tree reduction should be here */
2317 if (path->p_hdr->eh_entries == 0) {
2319 * truncate to zero freed all the tree,
2320 * so we need to correct eh_depth
2322 err = ext4_ext_get_access(handle, inode, path);
2324 ext_inode_hdr(inode)->eh_depth = 0;
2325 ext_inode_hdr(inode)->eh_max =
2326 cpu_to_le16(ext4_ext_space_root(inode));
2327 err = ext4_ext_dirty(handle, inode, path);
2331 ext4_ext_drop_refs(path);
2333 ext4_journal_stop(handle);
2339 * called at mount time
2341 void ext4_ext_init(struct super_block *sb)
2344 * possible initialization would be here
2347 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2348 printk(KERN_INFO "EXT4-fs: file extents enabled");
2349 #ifdef AGGRESSIVE_TEST
2350 printk(", aggressive tests");
2352 #ifdef CHECK_BINSEARCH
2353 printk(", check binsearch");
2355 #ifdef EXTENTS_STATS
2359 #ifdef EXTENTS_STATS
2360 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2361 EXT4_SB(sb)->s_ext_min = 1 << 30;
2362 EXT4_SB(sb)->s_ext_max = 0;
2368 * called at umount time
2370 void ext4_ext_release(struct super_block *sb)
2372 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2375 #ifdef EXTENTS_STATS
2376 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2377 struct ext4_sb_info *sbi = EXT4_SB(sb);
2378 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2379 sbi->s_ext_blocks, sbi->s_ext_extents,
2380 sbi->s_ext_blocks / sbi->s_ext_extents);
2381 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2382 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2387 static void bi_complete(struct bio *bio, int error)
2389 complete((struct completion *)bio->bi_private);
2392 /* FIXME!! we need to try to merge to left or right after zero-out */
2393 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2397 int blkbits, blocksize;
2399 struct completion event;
2400 unsigned int ee_len, len, done, offset;
2403 blkbits = inode->i_blkbits;
2404 blocksize = inode->i_sb->s_blocksize;
2405 ee_len = ext4_ext_get_actual_len(ex);
2406 ee_pblock = ext_pblock(ex);
2408 /* convert ee_pblock to 512 byte sectors */
2409 ee_pblock = ee_pblock << (blkbits - 9);
2411 while (ee_len > 0) {
2413 if (ee_len > BIO_MAX_PAGES)
2414 len = BIO_MAX_PAGES;
2418 bio = bio_alloc(GFP_NOIO, len);
2421 bio->bi_sector = ee_pblock;
2422 bio->bi_bdev = inode->i_sb->s_bdev;
2426 while (done < len) {
2427 ret = bio_add_page(bio, ZERO_PAGE(0),
2429 if (ret != blocksize) {
2431 * We can't add any more pages because of
2432 * hardware limitations. Start a new bio.
2437 offset += blocksize;
2438 if (offset >= PAGE_CACHE_SIZE)
2442 init_completion(&event);
2443 bio->bi_private = &event;
2444 bio->bi_end_io = bi_complete;
2445 submit_bio(WRITE, bio);
2446 wait_for_completion(&event);
2448 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2456 ee_pblock += done << (blkbits - 9);
2461 #define EXT4_EXT_ZERO_LEN 7
2464 * This function is called by ext4_ext_get_blocks() if someone tries to write
2465 * to an uninitialized extent. It may result in splitting the uninitialized
2466 * extent into multiple extents (upto three - one initialized and two
2468 * There are three possibilities:
2469 * a> There is no split required: Entire extent should be initialized
2470 * b> Splits in two extents: Write is happening at either end of the extent
2471 * c> Splits in three extents: Somone is writing in middle of the extent
2473 static int ext4_ext_convert_to_initialized(handle_t *handle,
2474 struct inode *inode,
2475 struct ext4_ext_path *path,
2477 unsigned int max_blocks)
2479 struct ext4_extent *ex, newex, orig_ex;
2480 struct ext4_extent *ex1 = NULL;
2481 struct ext4_extent *ex2 = NULL;
2482 struct ext4_extent *ex3 = NULL;
2483 struct ext4_extent_header *eh;
2484 ext4_lblk_t ee_block;
2485 unsigned int allocated, ee_len, depth;
2486 ext4_fsblk_t newblock;
2490 depth = ext_depth(inode);
2491 eh = path[depth].p_hdr;
2492 ex = path[depth].p_ext;
2493 ee_block = le32_to_cpu(ex->ee_block);
2494 ee_len = ext4_ext_get_actual_len(ex);
2495 allocated = ee_len - (iblock - ee_block);
2496 newblock = iblock - ee_block + ext_pblock(ex);
2498 orig_ex.ee_block = ex->ee_block;
2499 orig_ex.ee_len = cpu_to_le16(ee_len);
2500 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2502 err = ext4_ext_get_access(handle, inode, path + depth);
2505 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2506 if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2507 err = ext4_ext_zeroout(inode, &orig_ex);
2509 goto fix_extent_len;
2510 /* update the extent length and mark as initialized */
2511 ex->ee_block = orig_ex.ee_block;
2512 ex->ee_len = orig_ex.ee_len;
2513 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2514 ext4_ext_dirty(handle, inode, path + depth);
2515 /* zeroed the full extent */
2519 /* ex1: ee_block to iblock - 1 : uninitialized */
2520 if (iblock > ee_block) {
2522 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2523 ext4_ext_mark_uninitialized(ex1);
2527 * for sanity, update the length of the ex2 extent before
2528 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2529 * overlap of blocks.
2531 if (!ex1 && allocated > max_blocks)
2532 ex2->ee_len = cpu_to_le16(max_blocks);
2533 /* ex3: to ee_block + ee_len : uninitialised */
2534 if (allocated > max_blocks) {
2535 unsigned int newdepth;
2536 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2537 if (allocated <= EXT4_EXT_ZERO_LEN) {
2539 * iblock == ee_block is handled by the zerouout
2541 * Mark first half uninitialized.
2542 * Mark second half initialized and zero out the
2543 * initialized extent
2545 ex->ee_block = orig_ex.ee_block;
2546 ex->ee_len = cpu_to_le16(ee_len - allocated);
2547 ext4_ext_mark_uninitialized(ex);
2548 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2549 ext4_ext_dirty(handle, inode, path + depth);
2552 ex3->ee_block = cpu_to_le32(iblock);
2553 ext4_ext_store_pblock(ex3, newblock);
2554 ex3->ee_len = cpu_to_le16(allocated);
2555 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2556 if (err == -ENOSPC) {
2557 err = ext4_ext_zeroout(inode, &orig_ex);
2559 goto fix_extent_len;
2560 ex->ee_block = orig_ex.ee_block;
2561 ex->ee_len = orig_ex.ee_len;
2562 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2563 ext4_ext_dirty(handle, inode, path + depth);
2564 /* blocks available from iblock */
2568 goto fix_extent_len;
2571 * We need to zero out the second half because
2572 * an fallocate request can update file size and
2573 * converting the second half to initialized extent
2574 * implies that we can leak some junk data to user
2577 err = ext4_ext_zeroout(inode, ex3);
2580 * We should actually mark the
2581 * second half as uninit and return error
2582 * Insert would have changed the extent
2584 depth = ext_depth(inode);
2585 ext4_ext_drop_refs(path);
2586 path = ext4_ext_find_extent(inode,
2589 err = PTR_ERR(path);
2592 /* get the second half extent details */
2593 ex = path[depth].p_ext;
2594 err = ext4_ext_get_access(handle, inode,
2598 ext4_ext_mark_uninitialized(ex);
2599 ext4_ext_dirty(handle, inode, path + depth);
2603 /* zeroed the second half */
2607 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2608 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2609 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2610 ext4_ext_mark_uninitialized(ex3);
2611 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2612 if (err == -ENOSPC) {
2613 err = ext4_ext_zeroout(inode, &orig_ex);
2615 goto fix_extent_len;
2616 /* update the extent length and mark as initialized */
2617 ex->ee_block = orig_ex.ee_block;
2618 ex->ee_len = orig_ex.ee_len;
2619 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2620 ext4_ext_dirty(handle, inode, path + depth);
2621 /* zeroed the full extent */
2622 /* blocks available from iblock */
2626 goto fix_extent_len;
2628 * The depth, and hence eh & ex might change
2629 * as part of the insert above.
2631 newdepth = ext_depth(inode);
2633 * update the extent length after successful insert of the
2636 orig_ex.ee_len = cpu_to_le16(ee_len -
2637 ext4_ext_get_actual_len(ex3));
2639 ext4_ext_drop_refs(path);
2640 path = ext4_ext_find_extent(inode, iblock, path);
2642 err = PTR_ERR(path);
2645 eh = path[depth].p_hdr;
2646 ex = path[depth].p_ext;
2650 err = ext4_ext_get_access(handle, inode, path + depth);
2654 allocated = max_blocks;
2656 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2657 * to insert a extent in the middle zerout directly
2658 * otherwise give the extent a chance to merge to left
2660 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2661 iblock != ee_block) {
2662 err = ext4_ext_zeroout(inode, &orig_ex);
2664 goto fix_extent_len;
2665 /* update the extent length and mark as initialized */
2666 ex->ee_block = orig_ex.ee_block;
2667 ex->ee_len = orig_ex.ee_len;
2668 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2669 ext4_ext_dirty(handle, inode, path + depth);
2670 /* zero out the first half */
2671 /* blocks available from iblock */
2676 * If there was a change of depth as part of the
2677 * insertion of ex3 above, we need to update the length
2678 * of the ex1 extent again here
2680 if (ex1 && ex1 != ex) {
2682 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2683 ext4_ext_mark_uninitialized(ex1);
2686 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2687 ex2->ee_block = cpu_to_le32(iblock);
2688 ext4_ext_store_pblock(ex2, newblock);
2689 ex2->ee_len = cpu_to_le16(allocated);
2693 * New (initialized) extent starts from the first block
2694 * in the current extent. i.e., ex2 == ex
2695 * We have to see if it can be merged with the extent
2698 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2700 * To merge left, pass "ex2 - 1" to try_to_merge(),
2701 * since it merges towards right _only_.
2703 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2705 err = ext4_ext_correct_indexes(handle, inode, path);
2708 depth = ext_depth(inode);
2713 * Try to Merge towards right. This might be required
2714 * only when the whole extent is being written to.
2715 * i.e. ex2 == ex and ex3 == NULL.
2718 ret = ext4_ext_try_to_merge(inode, path, ex2);
2720 err = ext4_ext_correct_indexes(handle, inode, path);
2725 /* Mark modified extent as dirty */
2726 err = ext4_ext_dirty(handle, inode, path + depth);
2729 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2730 if (err == -ENOSPC) {
2731 err = ext4_ext_zeroout(inode, &orig_ex);
2733 goto fix_extent_len;
2734 /* update the extent length and mark as initialized */
2735 ex->ee_block = orig_ex.ee_block;
2736 ex->ee_len = orig_ex.ee_len;
2737 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2738 ext4_ext_dirty(handle, inode, path + depth);
2739 /* zero out the first half */
2742 goto fix_extent_len;
2744 return err ? err : allocated;
2747 ex->ee_block = orig_ex.ee_block;
2748 ex->ee_len = orig_ex.ee_len;
2749 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2750 ext4_ext_mark_uninitialized(ex);
2751 ext4_ext_dirty(handle, inode, path + depth);
2756 * Block allocation/map/preallocation routine for extents based files
2759 * Need to be called with
2760 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2761 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2763 * return > 0, number of of blocks already mapped/allocated
2764 * if create == 0 and these are pre-allocated blocks
2765 * buffer head is unmapped
2766 * otherwise blocks are mapped
2768 * return = 0, if plain look up failed (blocks have not been allocated)
2769 * buffer head is unmapped
2771 * return < 0, error case.
2773 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2775 unsigned int max_blocks, struct buffer_head *bh_result,
2776 int create, int extend_disksize)
2778 struct ext4_ext_path *path = NULL;
2779 struct ext4_extent_header *eh;
2780 struct ext4_extent newex, *ex;
2781 ext4_fsblk_t newblock;
2782 int err = 0, depth, ret, cache_type;
2783 unsigned int allocated = 0;
2784 struct ext4_allocation_request ar;
2787 __clear_bit(BH_New, &bh_result->b_state);
2788 ext_debug("blocks %u/%u requested for inode %u\n",
2789 iblock, max_blocks, inode->i_ino);
2791 /* check in cache */
2792 cache_type = ext4_ext_in_cache(inode, iblock, &newex);
2794 if (cache_type == EXT4_EXT_CACHE_GAP) {
2797 * block isn't allocated yet and
2798 * user doesn't want to allocate it
2802 /* we should allocate requested block */
2803 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
2804 /* block is already allocated */
2806 - le32_to_cpu(newex.ee_block)
2807 + ext_pblock(&newex);
2808 /* number of remaining blocks in the extent */
2809 allocated = ext4_ext_get_actual_len(&newex) -
2810 (iblock - le32_to_cpu(newex.ee_block));
2817 /* find extent for this block */
2818 path = ext4_ext_find_extent(inode, iblock, NULL);
2820 err = PTR_ERR(path);
2825 depth = ext_depth(inode);
2828 * consistent leaf must not be empty;
2829 * this situation is possible, though, _during_ tree modification;
2830 * this is why assert can't be put in ext4_ext_find_extent()
2832 BUG_ON(path[depth].p_ext == NULL && depth != 0);
2833 eh = path[depth].p_hdr;
2835 ex = path[depth].p_ext;
2837 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
2838 ext4_fsblk_t ee_start = ext_pblock(ex);
2839 unsigned short ee_len;
2842 * Uninitialized extents are treated as holes, except that
2843 * we split out initialized portions during a write.
2845 ee_len = ext4_ext_get_actual_len(ex);
2846 /* if found extent covers block, simply return it */
2847 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2848 newblock = iblock - ee_block + ee_start;
2849 /* number of remaining blocks in the extent */
2850 allocated = ee_len - (iblock - ee_block);
2851 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
2852 ee_block, ee_len, newblock);
2854 /* Do not put uninitialized extent in the cache */
2855 if (!ext4_ext_is_uninitialized(ex)) {
2856 ext4_ext_put_in_cache(inode, ee_block,
2858 EXT4_EXT_CACHE_EXTENT);
2861 if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2865 * We have blocks reserved already. We
2866 * return allocated blocks so that delalloc
2867 * won't do block reservation for us. But
2868 * the buffer head will be unmapped so that
2869 * a read from the block returns 0s.
2871 if (allocated > max_blocks)
2872 allocated = max_blocks;
2873 set_buffer_unwritten(bh_result);
2877 ret = ext4_ext_convert_to_initialized(handle, inode,
2890 * requested block isn't allocated yet;
2891 * we couldn't try to create block if create flag is zero
2895 * put just found gap into cache to speed up
2896 * subsequent requests
2898 ext4_ext_put_gap_in_cache(inode, path, iblock);
2902 * Okay, we need to do block allocation.
2905 /* find neighbour allocated blocks */
2907 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2911 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2916 * See if request is beyond maximum number of blocks we can have in
2917 * a single extent. For an initialized extent this limit is
2918 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2919 * EXT_UNINIT_MAX_LEN.
2921 if (max_blocks > EXT_INIT_MAX_LEN &&
2922 create != EXT4_CREATE_UNINITIALIZED_EXT)
2923 max_blocks = EXT_INIT_MAX_LEN;
2924 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2925 create == EXT4_CREATE_UNINITIALIZED_EXT)
2926 max_blocks = EXT_UNINIT_MAX_LEN;
2928 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2929 newex.ee_block = cpu_to_le32(iblock);
2930 newex.ee_len = cpu_to_le16(max_blocks);
2931 err = ext4_ext_check_overlap(inode, &newex, path);
2933 allocated = ext4_ext_get_actual_len(&newex);
2935 allocated = max_blocks;
2937 /* allocate new block */
2939 ar.goal = ext4_ext_find_goal(inode, path, iblock);
2940 ar.logical = iblock;
2942 if (S_ISREG(inode->i_mode))
2943 ar.flags = EXT4_MB_HINT_DATA;
2945 /* disable in-core preallocation for non-regular files */
2947 newblock = ext4_mb_new_blocks(handle, &ar, &err);
2950 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2951 ar.goal, newblock, allocated);
2953 /* try to insert new extent into found leaf and return */
2954 ext4_ext_store_pblock(&newex, newblock);
2955 newex.ee_len = cpu_to_le16(ar.len);
2956 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */
2957 ext4_ext_mark_uninitialized(&newex);
2958 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2960 /* free data blocks we just allocated */
2961 /* not a good idea to call discard here directly,
2962 * but otherwise we'd need to call it every free() */
2963 ext4_discard_preallocations(inode);
2964 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2965 ext4_ext_get_actual_len(&newex), 0);
2969 /* previous routine could use block we allocated */
2970 newblock = ext_pblock(&newex);
2971 allocated = ext4_ext_get_actual_len(&newex);
2973 if (extend_disksize) {
2974 disksize = ((loff_t) iblock + ar.len) << inode->i_blkbits;
2975 if (disksize > i_size_read(inode))
2976 disksize = i_size_read(inode);
2977 if (disksize > EXT4_I(inode)->i_disksize)
2978 EXT4_I(inode)->i_disksize = disksize;
2981 set_buffer_new(bh_result);
2983 /* Cache only when it is _not_ an uninitialized extent */
2984 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2985 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2986 EXT4_EXT_CACHE_EXTENT);
2988 if (allocated > max_blocks)
2989 allocated = max_blocks;
2990 ext4_ext_show_leaf(inode, path);
2991 set_buffer_mapped(bh_result);
2992 bh_result->b_bdev = inode->i_sb->s_bdev;
2993 bh_result->b_blocknr = newblock;
2996 ext4_ext_drop_refs(path);
2999 return err ? err : allocated;
3002 void ext4_ext_truncate(struct inode *inode)
3004 struct address_space *mapping = inode->i_mapping;
3005 struct super_block *sb = inode->i_sb;
3006 ext4_lblk_t last_block;
3011 * probably first extent we're gonna free will be last in block
3013 err = ext4_writepage_trans_blocks(inode);
3014 handle = ext4_journal_start(inode, err);
3018 if (inode->i_size & (sb->s_blocksize - 1))
3019 ext4_block_truncate_page(handle, mapping, inode->i_size);
3021 if (ext4_orphan_add(handle, inode))
3024 down_write(&EXT4_I(inode)->i_data_sem);
3025 ext4_ext_invalidate_cache(inode);
3027 ext4_discard_preallocations(inode);
3030 * TODO: optimization is possible here.
3031 * Probably we need not scan at all,
3032 * because page truncation is enough.
3035 /* we have to know where to truncate from in crash case */
3036 EXT4_I(inode)->i_disksize = inode->i_size;
3037 ext4_mark_inode_dirty(handle, inode);
3039 last_block = (inode->i_size + sb->s_blocksize - 1)
3040 >> EXT4_BLOCK_SIZE_BITS(sb);
3041 err = ext4_ext_remove_space(inode, last_block);
3043 /* In a multi-transaction truncate, we only make the final
3044 * transaction synchronous.
3047 ext4_handle_sync(handle);
3050 up_write(&EXT4_I(inode)->i_data_sem);
3052 * If this was a simple ftruncate() and the file will remain alive,
3053 * then we need to clear up the orphan record which we created above.
3054 * However, if this was a real unlink then we were called by
3055 * ext4_delete_inode(), and we allow that function to clean up the
3056 * orphan info for us.
3059 ext4_orphan_del(handle, inode);
3061 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3062 ext4_mark_inode_dirty(handle, inode);
3063 ext4_journal_stop(handle);
3066 static void ext4_falloc_update_inode(struct inode *inode,
3067 int mode, loff_t new_size, int update_ctime)
3069 struct timespec now;
3072 now = current_fs_time(inode->i_sb);
3073 if (!timespec_equal(&inode->i_ctime, &now))
3074 inode->i_ctime = now;
3077 * Update only when preallocation was requested beyond
3080 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3081 if (new_size > i_size_read(inode))
3082 i_size_write(inode, new_size);
3083 if (new_size > EXT4_I(inode)->i_disksize)
3084 ext4_update_i_disksize(inode, new_size);
3090 * preallocate space for a file. This implements ext4's fallocate inode
3091 * operation, which gets called from sys_fallocate system call.
3092 * For block-mapped files, posix_fallocate should fall back to the method
3093 * of writing zeroes to the required new blocks (the same behavior which is
3094 * expected for file systems which do not support fallocate() system call).
3096 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3101 unsigned int max_blocks;
3105 struct buffer_head map_bh;
3106 unsigned int credits, blkbits = inode->i_blkbits;
3109 * currently supporting (pre)allocate mode for extent-based
3112 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3115 /* preallocation to directories is currently not supported */
3116 if (S_ISDIR(inode->i_mode))
3119 block = offset >> blkbits;
3121 * We can't just convert len to max_blocks because
3122 * If blocksize = 4096 offset = 3072 and len = 2048
3124 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3127 * credits to insert 1 extent into extent tree
3129 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3130 mutex_lock(&inode->i_mutex);
3132 while (ret >= 0 && ret < max_blocks) {
3133 block = block + ret;
3134 max_blocks = max_blocks - ret;
3135 handle = ext4_journal_start(inode, credits);
3136 if (IS_ERR(handle)) {
3137 ret = PTR_ERR(handle);
3140 ret = ext4_get_blocks_wrap(handle, inode, block,
3141 max_blocks, &map_bh,
3142 EXT4_CREATE_UNINITIALIZED_EXT, 0, 0);
3146 printk(KERN_ERR "%s: ext4_ext_get_blocks "
3147 "returned error inode#%lu, block=%u, "
3148 "max_blocks=%u", __func__,
3149 inode->i_ino, block, max_blocks);
3151 ext4_mark_inode_dirty(handle, inode);
3152 ret2 = ext4_journal_stop(handle);
3155 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3156 blkbits) >> blkbits))
3157 new_size = offset + len;
3159 new_size = (block + ret) << blkbits;
3161 ext4_falloc_update_inode(inode, mode, new_size,
3162 buffer_new(&map_bh));
3163 ext4_mark_inode_dirty(handle, inode);
3164 ret2 = ext4_journal_stop(handle);
3168 if (ret == -ENOSPC &&
3169 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3173 mutex_unlock(&inode->i_mutex);
3174 return ret > 0 ? ret2 : ret;
3178 * Callback function called for each extent to gather FIEMAP information.
3180 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3181 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3184 struct fiemap_extent_info *fieinfo = data;
3185 unsigned long blksize_bits = inode->i_sb->s_blocksize_bits;
3192 logical = (__u64)newex->ec_block << blksize_bits;
3194 if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3197 struct buffer_head *bh = NULL;
3199 offset = logical >> PAGE_SHIFT;
3200 page = find_get_page(inode->i_mapping, offset);
3201 if (!page || !page_has_buffers(page))
3202 return EXT_CONTINUE;
3204 bh = page_buffers(page);
3207 return EXT_CONTINUE;
3209 if (buffer_delay(bh)) {
3210 flags |= FIEMAP_EXTENT_DELALLOC;
3211 page_cache_release(page);
3213 page_cache_release(page);
3214 return EXT_CONTINUE;
3218 physical = (__u64)newex->ec_start << blksize_bits;
3219 length = (__u64)newex->ec_len << blksize_bits;
3221 if (ex && ext4_ext_is_uninitialized(ex))
3222 flags |= FIEMAP_EXTENT_UNWRITTEN;
3225 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3227 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3228 * this also indicates no more allocated blocks.
3230 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3232 if (logical + length - 1 == EXT_MAX_BLOCK ||
3233 ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK)
3234 flags |= FIEMAP_EXTENT_LAST;
3236 error = fiemap_fill_next_extent(fieinfo, logical, physical,
3243 return EXT_CONTINUE;
3246 /* fiemap flags we can handle specified here */
3247 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3249 static int ext4_xattr_fiemap(struct inode *inode,
3250 struct fiemap_extent_info *fieinfo)
3254 __u32 flags = FIEMAP_EXTENT_LAST;
3255 int blockbits = inode->i_sb->s_blocksize_bits;
3259 if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
3260 struct ext4_iloc iloc;
3261 int offset; /* offset of xattr in inode */
3263 error = ext4_get_inode_loc(inode, &iloc);
3266 physical = iloc.bh->b_blocknr << blockbits;
3267 offset = EXT4_GOOD_OLD_INODE_SIZE +
3268 EXT4_I(inode)->i_extra_isize;
3270 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3271 flags |= FIEMAP_EXTENT_DATA_INLINE;
3272 } else { /* external block */
3273 physical = EXT4_I(inode)->i_file_acl << blockbits;
3274 length = inode->i_sb->s_blocksize;
3278 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3280 return (error < 0 ? error : 0);
3283 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3284 __u64 start, __u64 len)
3286 ext4_lblk_t start_blk;
3287 ext4_lblk_t len_blks;
3290 /* fallback to generic here if not in extents fmt */
3291 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3292 return generic_block_fiemap(inode, fieinfo, start, len,
3295 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3298 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3299 error = ext4_xattr_fiemap(inode, fieinfo);
3301 start_blk = start >> inode->i_sb->s_blocksize_bits;
3302 len_blks = len >> inode->i_sb->s_blocksize_bits;
3305 * Walk the extent tree gathering extent information.
3306 * ext4_ext_fiemap_cb will push extents back to user.
3308 down_write(&EXT4_I(inode)->i_data_sem);
3309 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3310 ext4_ext_fiemap_cb, fieinfo);
3311 up_write(&EXT4_I(inode)->i_data_sem);