Root/fs/ext4/extents.c

1/*
2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
4 *
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
8 *
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.
12 *
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.
17 *
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-
21 */
22
23/*
24 * Extents support for EXT4
25 *
26 * TODO:
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
30 */
31
32#include <linux/module.h>
33#include <linux/fs.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"
46
47
48/*
49 * ext_pblock:
50 * combine low and high parts of physical block number into ext4_fsblk_t
51 */
52ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
53{
54    ext4_fsblk_t block;
55
56    block = le32_to_cpu(ex->ee_start_lo);
57    block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
58    return block;
59}
60
61/*
62 * idx_pblock:
63 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64 */
65ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
66{
67    ext4_fsblk_t block;
68
69    block = le32_to_cpu(ix->ei_leaf_lo);
70    block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
71    return block;
72}
73
74/*
75 * ext4_ext_store_pblock:
76 * stores a large physical block number into an extent struct,
77 * breaking it into parts
78 */
79void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
80{
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);
83}
84
85/*
86 * ext4_idx_store_pblock:
87 * stores a large physical block number into an index struct,
88 * breaking it into parts
89 */
90static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
91{
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);
94}
95
96static int ext4_ext_journal_restart(handle_t *handle, int needed)
97{
98    int err;
99
100    if (!ext4_handle_valid(handle))
101        return 0;
102    if (handle->h_buffer_credits > needed)
103        return 0;
104    err = ext4_journal_extend(handle, needed);
105    if (err <= 0)
106        return err;
107    return ext4_journal_restart(handle, needed);
108}
109
110/*
111 * could return:
112 * - EROFS
113 * - ENOMEM
114 */
115static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
116                struct ext4_ext_path *path)
117{
118    if (path->p_bh) {
119        /* path points to block */
120        return ext4_journal_get_write_access(handle, path->p_bh);
121    }
122    /* path points to leaf/index in inode body */
123    /* we use in-core data, no need to protect them */
124    return 0;
125}
126
127/*
128 * could return:
129 * - EROFS
130 * - ENOMEM
131 * - EIO
132 */
133static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
134                struct ext4_ext_path *path)
135{
136    int err;
137    if (path->p_bh) {
138        /* path points to block */
139        err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
140    } else {
141        /* path points to leaf/index in inode body */
142        err = ext4_mark_inode_dirty(handle, inode);
143    }
144    return err;
145}
146
147static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
148                  struct ext4_ext_path *path,
149                  ext4_lblk_t block)
150{
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));
157    int depth;
158
159    if (path) {
160        struct ext4_extent *ex;
161        depth = path->p_depth;
162
163        /* try to predict block placement */
164        ex = path[depth].p_ext;
165        if (ex)
166            return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
167
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;
172    }
173
174    /* OK. use inode's group */
175    block_group = ei->i_block_group;
176    if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
177        /*
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
183         * fsck times.
184         */
185        block_group &= ~(flex_size-1);
186        if (S_ISREG(inode->i_mode))
187            block_group++;
188    }
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;
192
193    /*
194     * If we are doing delayed allocation, we don't need take
195     * colour into account.
196     */
197    if (test_opt(inode->i_sb, DELALLOC))
198        return bg_start;
199
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);
203    else
204        colour = (current->pid % 16) * ((last_block - bg_start) / 16);
205    return bg_start + colour + block;
206}
207
208/*
209 * Allocation for a meta data block
210 */
211static ext4_fsblk_t
212ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
213            struct ext4_ext_path *path,
214            struct ext4_extent *ex, int *err)
215{
216    ext4_fsblk_t goal, newblock;
217
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);
220    return newblock;
221}
222
223static int ext4_ext_space_block(struct inode *inode)
224{
225    int size;
226
227    size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
228            / sizeof(struct ext4_extent);
229#ifdef AGGRESSIVE_TEST
230    if (size > 6)
231        size = 6;
232#endif
233    return size;
234}
235
236static int ext4_ext_space_block_idx(struct inode *inode)
237{
238    int size;
239
240    size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
241            / sizeof(struct ext4_extent_idx);
242#ifdef AGGRESSIVE_TEST
243    if (size > 5)
244        size = 5;
245#endif
246    return size;
247}
248
249static int ext4_ext_space_root(struct inode *inode)
250{
251    int size;
252
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
257    if (size > 3)
258        size = 3;
259#endif
260    return size;
261}
262
263static int ext4_ext_space_root_idx(struct inode *inode)
264{
265    int size;
266
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
271    if (size > 4)
272        size = 4;
273#endif
274    return size;
275}
276
277/*
278 * Calculate the number of metadata blocks needed
279 * to allocate @blocks
280 * Worse case is one block per extent
281 */
282int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
283{
284    int lcap, icap, rcap, leafs, idxs, num;
285    int newextents = blocks;
286
287    rcap = ext4_ext_space_root_idx(inode);
288    lcap = ext4_ext_space_block(inode);
289    icap = ext4_ext_space_block_idx(inode);
290
291    /* number of new leaf blocks needed */
292    num = leafs = (newextents + lcap - 1) / lcap;
293
294    /*
295     * Worse case, we need separate index block(s)
296     * to link all new leaf blocks
297     */
298    idxs = (leafs + icap - 1) / icap;
299    do {
300        num += idxs;
301        idxs = (idxs + icap - 1) / icap;
302    } while (idxs > rcap);
303
304    return num;
305}
306
307static int
308ext4_ext_max_entries(struct inode *inode, int depth)
309{
310    int max;
311
312    if (depth == ext_depth(inode)) {
313        if (depth == 0)
314            max = ext4_ext_space_root(inode);
315        else
316            max = ext4_ext_space_root_idx(inode);
317    } else {
318        if (depth == 0)
319            max = ext4_ext_space_block(inode);
320        else
321            max = ext4_ext_space_block_idx(inode);
322    }
323
324    return max;
325}
326
327static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
328{
329    ext4_fsblk_t block = ext_pblock(ext);
330    int len = ext4_ext_get_actual_len(ext);
331
332    return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
333}
334
335static int ext4_valid_extent_idx(struct inode *inode,
336                struct ext4_extent_idx *ext_idx)
337{
338    ext4_fsblk_t block = idx_pblock(ext_idx);
339
340    return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
341}
342
343static int ext4_valid_extent_entries(struct inode *inode,
344                struct ext4_extent_header *eh,
345                int depth)
346{
347    struct ext4_extent *ext;
348    struct ext4_extent_idx *ext_idx;
349    unsigned short entries;
350    if (eh->eh_entries == 0)
351        return 1;
352
353    entries = le16_to_cpu(eh->eh_entries);
354
355    if (depth == 0) {
356        /* leaf entries */
357        ext = EXT_FIRST_EXTENT(eh);
358        while (entries) {
359            if (!ext4_valid_extent(inode, ext))
360                return 0;
361            ext++;
362            entries--;
363        }
364    } else {
365        ext_idx = EXT_FIRST_INDEX(eh);
366        while (entries) {
367            if (!ext4_valid_extent_idx(inode, ext_idx))
368                return 0;
369            ext_idx++;
370            entries--;
371        }
372    }
373    return 1;
374}
375
376static int __ext4_ext_check(const char *function, struct inode *inode,
377                    struct ext4_extent_header *eh,
378                    int depth)
379{
380    const char *error_msg;
381    int max = 0;
382
383    if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
384        error_msg = "invalid magic";
385        goto corrupted;
386    }
387    if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
388        error_msg = "unexpected eh_depth";
389        goto corrupted;
390    }
391    if (unlikely(eh->eh_max == 0)) {
392        error_msg = "invalid eh_max";
393        goto corrupted;
394    }
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";
398        goto corrupted;
399    }
400    if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
401        error_msg = "invalid eh_entries";
402        goto corrupted;
403    }
404    if (!ext4_valid_extent_entries(inode, eh, depth)) {
405        error_msg = "invalid extent entries";
406        goto corrupted;
407    }
408    return 0;
409
410corrupted:
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);
417
418    return -EIO;
419}
420
421#define ext4_ext_check(inode, eh, depth) \
422    __ext4_ext_check(__func__, inode, eh, depth)
423
424int ext4_ext_check_inode(struct inode *inode)
425{
426    return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
427}
428
429#ifdef EXT_DEBUG
430static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
431{
432    int k, l = path->p_depth;
433
434    ext_debug("path:");
435    for (k = 0; k <= l; k++, path++) {
436        if (path->p_idx) {
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));
444        } else
445            ext_debug(" []");
446    }
447    ext_debug("\n");
448}
449
450static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
451{
452    int depth = ext_depth(inode);
453    struct ext4_extent_header *eh;
454    struct ext4_extent *ex;
455    int i;
456
457    if (!path)
458        return;
459
460    eh = path[depth].p_hdr;
461    ex = EXT_FIRST_EXTENT(eh);
462
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));
466    }
467    ext_debug("\n");
468}
469#else
470#define ext4_ext_show_path(inode, path)
471#define ext4_ext_show_leaf(inode, path)
472#endif
473
474void ext4_ext_drop_refs(struct ext4_ext_path *path)
475{
476    int depth = path->p_depth;
477    int i;
478
479    for (i = 0; i <= depth; i++, path++)
480        if (path->p_bh) {
481            brelse(path->p_bh);
482            path->p_bh = NULL;
483        }
484}
485
486/*
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
490 */
491static void
492ext4_ext_binsearch_idx(struct inode *inode,
493            struct ext4_ext_path *path, ext4_lblk_t block)
494{
495    struct ext4_extent_header *eh = path->p_hdr;
496    struct ext4_extent_idx *r, *l, *m;
497
498
499    ext_debug("binsearch for %u(idx): ", block);
500
501    l = EXT_FIRST_INDEX(eh) + 1;
502    r = EXT_LAST_INDEX(eh);
503    while (l <= r) {
504        m = l + (r - l) / 2;
505        if (block < le32_to_cpu(m->ei_block))
506            r = m - 1;
507        else
508            l = m + 1;
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));
512    }
513
514    path->p_idx = l - 1;
515    ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
516          idx_pblock(path->p_idx));
517
518#ifdef CHECK_BINSEARCH
519    {
520        struct ext4_extent_idx *chix, *ix;
521        int k;
522
523        chix = ix = EXT_FIRST_INDEX(eh);
524        for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
525          if (k != 0 &&
526              le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
527                printk(KERN_DEBUG "k=%d, ix=0x%p, "
528                       "first=0x%p\n", k,
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));
533            }
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))
537                break;
538            chix = ix;
539        }
540        BUG_ON(chix != path->p_idx);
541    }
542#endif
543
544}
545
546/*
547 * ext4_ext_binsearch:
548 * binary search for closest extent of the given block
549 * the header must be checked before calling this
550 */
551static void
552ext4_ext_binsearch(struct inode *inode,
553        struct ext4_ext_path *path, ext4_lblk_t block)
554{
555    struct ext4_extent_header *eh = path->p_hdr;
556    struct ext4_extent *r, *l, *m;
557
558    if (eh->eh_entries == 0) {
559        /*
560         * this leaf is empty:
561         * we get such a leaf in split/add case
562         */
563        return;
564    }
565
566    ext_debug("binsearch for %u: ", block);
567
568    l = EXT_FIRST_EXTENT(eh) + 1;
569    r = EXT_LAST_EXTENT(eh);
570
571    while (l <= r) {
572        m = l + (r - l) / 2;
573        if (block < le32_to_cpu(m->ee_block))
574            r = m - 1;
575        else
576            l = m + 1;
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));
580    }
581
582    path->p_ext = l - 1;
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));
587
588#ifdef CHECK_BINSEARCH
589    {
590        struct ext4_extent *chex, *ex;
591        int k;
592
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))
598                break;
599            chex = ex;
600        }
601        BUG_ON(chex != path->p_ext);
602    }
603#endif
604
605}
606
607int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
608{
609    struct ext4_extent_header *eh;
610
611    eh = ext_inode_hdr(inode);
612    eh->eh_depth = 0;
613    eh->eh_entries = 0;
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);
618    return 0;
619}
620
621struct ext4_ext_path *
622ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
623                    struct ext4_ext_path *path)
624{
625    struct ext4_extent_header *eh;
626    struct buffer_head *bh;
627    short int depth, i, ppos = 0, alloc = 0;
628
629    eh = ext_inode_hdr(inode);
630    depth = ext_depth(inode);
631
632    /* account possible depth increase */
633    if (!path) {
634        path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
635                GFP_NOFS);
636        if (!path)
637            return ERR_PTR(-ENOMEM);
638        alloc = 1;
639    }
640    path[0].p_hdr = eh;
641    path[0].p_bh = NULL;
642
643    i = depth;
644    /* walk through the tree */
645    while (i) {
646        int need_to_validate = 0;
647
648        ext_debug("depth %d: num %d, max %d\n",
649              ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
650
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;
655
656        bh = sb_getblk(inode->i_sb, path[ppos].p_block);
657        if (unlikely(!bh))
658            goto err;
659        if (!bh_uptodate_or_lock(bh)) {
660            if (bh_submit_read(bh) < 0) {
661                put_bh(bh);
662                goto err;
663            }
664            /* validate the extent entries */
665            need_to_validate = 1;
666        }
667        eh = ext_block_hdr(bh);
668        ppos++;
669        BUG_ON(ppos > depth);
670        path[ppos].p_bh = bh;
671        path[ppos].p_hdr = eh;
672        i--;
673
674        if (need_to_validate && ext4_ext_check(inode, eh, i))
675            goto err;
676    }
677
678    path[ppos].p_depth = i;
679    path[ppos].p_ext = NULL;
680    path[ppos].p_idx = NULL;
681
682    /* find extent */
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);
687
688    ext4_ext_show_path(inode, path);
689
690    return path;
691
692err:
693    ext4_ext_drop_refs(path);
694    if (alloc)
695        kfree(path);
696    return ERR_PTR(-EIO);
697}
698
699/*
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
703 */
704static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
705                struct ext4_ext_path *curp,
706                int logical, ext4_fsblk_t ptr)
707{
708    struct ext4_extent_idx *ix;
709    int len, err;
710
711    err = ext4_ext_get_access(handle, inode, curp);
712    if (err)
713        return err;
714
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)) {
718        /* insert after */
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",
724                    logical, ptr, len,
725                    (curp->p_idx + 1), (curp->p_idx + 2));
726            memmove(curp->p_idx + 2, curp->p_idx + 1, len);
727        }
728        ix = curp->p_idx + 1;
729    } else {
730        /* insert before */
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",
735                logical, ptr, len,
736                curp->p_idx, (curp->p_idx + 1));
737        memmove(curp->p_idx + 1, curp->p_idx, len);
738        ix = curp->p_idx;
739    }
740
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);
744
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));
748
749    err = ext4_ext_dirty(handle, inode, curp);
750    ext4_std_error(inode->i_sb, err);
751
752    return err;
753}
754
755/*
756 * ext4_ext_split:
757 * inserts new subtree into the path, using free index entry
758 * at depth @at:
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
764 */
765static int ext4_ext_split(handle_t *handle, struct inode *inode,
766                struct ext4_ext_path *path,
767                struct ext4_extent *newext, int at)
768{
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;
774    int i = at, k, m, a;
775    ext4_fsblk_t newblock, oldblock;
776    __le32 border;
777    ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
778    int err = 0;
779
780    /* make decision: where to split? */
781    /* FIXME: now decision is simplest: at current extent */
782
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));
791    } else {
792        border = newext->ee_block;
793        ext_debug("leaf will be added."
794                " next leaf starts at %d\n",
795                le32_to_cpu(border));
796    }
797
798    /*
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.
803     */
804
805    /*
806     * Get array to track all allocated blocks.
807     * We need this to handle errors and free blocks
808     * upon them.
809     */
810    ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
811    if (!ablocks)
812        return -ENOMEM;
813
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,
818                           newext, &err);
819        if (newblock == 0)
820            goto cleanup;
821        ablocks[a] = newblock;
822    }
823
824    /* initialize new leaf */
825    newblock = ablocks[--a];
826    BUG_ON(newblock == 0);
827    bh = sb_getblk(inode->i_sb, newblock);
828    if (!bh) {
829        err = -EIO;
830        goto cleanup;
831    }
832    lock_buffer(bh);
833
834    err = ext4_journal_get_create_access(handle, bh);
835    if (err)
836        goto cleanup;
837
838    neh = ext_block_hdr(bh);
839    neh->eh_entries = 0;
840    neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
841    neh->eh_magic = EXT4_EXT_MAGIC;
842    neh->eh_depth = 0;
843    ex = EXT_FIRST_EXTENT(neh);
844
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 */
849    m = 0;
850    path[depth].p_ext++;
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),
857                newblock);
858        /*memmove(ex++, path[depth].p_ext++,
859                sizeof(struct ext4_extent));
860        neh->eh_entries++;*/
861        path[depth].p_ext++;
862        m++;
863    }
864    if (m) {
865        memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
866        le16_add_cpu(&neh->eh_entries, m);
867    }
868
869    set_buffer_uptodate(bh);
870    unlock_buffer(bh);
871
872    err = ext4_handle_dirty_metadata(handle, inode, bh);
873    if (err)
874        goto cleanup;
875    brelse(bh);
876    bh = NULL;
877
878    /* correct old leaf */
879    if (m) {
880        err = ext4_ext_get_access(handle, inode, path + depth);
881        if (err)
882            goto cleanup;
883        le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
884        err = ext4_ext_dirty(handle, inode, path + depth);
885        if (err)
886            goto cleanup;
887
888    }
889
890    /* create intermediate indexes */
891    k = depth - at - 1;
892    BUG_ON(k < 0);
893    if (k)
894        ext_debug("create %d intermediate indices\n", k);
895    /* insert new index into current index block */
896    /* current depth stored in i var */
897    i = depth - 1;
898    while (k--) {
899        oldblock = newblock;
900        newblock = ablocks[--a];
901        bh = sb_getblk(inode->i_sb, newblock);
902        if (!bh) {
903            err = -EIO;
904            goto cleanup;
905        }
906        lock_buffer(bh);
907
908        err = ext4_journal_get_create_access(handle, bh);
909        if (err)
910            goto cleanup;
911
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);
920
921        ext_debug("int.index at %d (block %llu): %u -> %llu\n",
922                i, newblock, le32_to_cpu(border), oldblock);
923        /* copy indexes */
924        m = 0;
925        path[i].p_idx++;
926
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),
935                    newblock);
936            /*memmove(++fidx, path[i].p_idx++,
937                    sizeof(struct ext4_extent_idx));
938            neh->eh_entries++;
939            BUG_ON(neh->eh_entries > neh->eh_max);*/
940            path[i].p_idx++;
941            m++;
942        }
943        if (m) {
944            memmove(++fidx, path[i].p_idx - m,
945                sizeof(struct ext4_extent_idx) * m);
946            le16_add_cpu(&neh->eh_entries, m);
947        }
948        set_buffer_uptodate(bh);
949        unlock_buffer(bh);
950
951        err = ext4_handle_dirty_metadata(handle, inode, bh);
952        if (err)
953            goto cleanup;
954        brelse(bh);
955        bh = NULL;
956
957        /* correct old index */
958        if (m) {
959            err = ext4_ext_get_access(handle, inode, path + i);
960            if (err)
961                goto cleanup;
962            le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
963            err = ext4_ext_dirty(handle, inode, path + i);
964            if (err)
965                goto cleanup;
966        }
967
968        i--;
969    }
970
971    /* insert new index */
972    err = ext4_ext_insert_index(handle, inode, path + at,
973                    le32_to_cpu(border), newblock);
974
975cleanup:
976    if (bh) {
977        if (buffer_locked(bh))
978            unlock_buffer(bh);
979        brelse(bh);
980    }
981
982    if (err) {
983        /* free all allocated blocks in error case */
984        for (i = 0; i < depth; i++) {
985            if (!ablocks[i])
986                continue;
987            ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
988        }
989    }
990    kfree(ablocks);
991
992    return err;
993}
994
995/*
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
1002 */
1003static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1004                    struct ext4_ext_path *path,
1005                    struct ext4_extent *newext)
1006{
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;
1012    int err = 0;
1013
1014    newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1015    if (newblock == 0)
1016        return err;
1017
1018    bh = sb_getblk(inode->i_sb, newblock);
1019    if (!bh) {
1020        err = -EIO;
1021        ext4_std_error(inode->i_sb, err);
1022        return err;
1023    }
1024    lock_buffer(bh);
1025
1026    err = ext4_journal_get_create_access(handle, bh);
1027    if (err) {
1028        unlock_buffer(bh);
1029        goto out;
1030    }
1031
1032    /* move top-level index/leaf into new block */
1033    memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1034
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));
1041    else
1042      neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
1043    neh->eh_magic = EXT4_EXT_MAGIC;
1044    set_buffer_uptodate(bh);
1045    unlock_buffer(bh);
1046
1047    err = ext4_handle_dirty_metadata(handle, inode, bh);
1048    if (err)
1049        goto out;
1050
1051    /* create index in new top-level index: num,max,pointer */
1052    err = ext4_ext_get_access(handle, inode, curp);
1053    if (err)
1054        goto out;
1055
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);
1060
1061    if (path[0].p_hdr->eh_depth)
1062        curp->p_idx->ei_block =
1063            EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1064    else
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);
1068
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));
1074
1075    neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1076    err = ext4_ext_dirty(handle, inode, curp);
1077out:
1078    brelse(bh);
1079
1080    return err;
1081}
1082
1083/*
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.
1087 */
1088static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1089                    struct ext4_ext_path *path,
1090                    struct ext4_extent *newext)
1091{
1092    struct ext4_ext_path *curp;
1093    int depth, i, err = 0;
1094
1095repeat:
1096    i = depth = ext_depth(inode);
1097
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)) {
1101        i--;
1102        curp--;
1103    }
1104
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);
1111        if (err)
1112            goto out;
1113
1114        /* refill path */
1115        ext4_ext_drop_refs(path);
1116        path = ext4_ext_find_extent(inode,
1117                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1118                    path);
1119        if (IS_ERR(path))
1120            err = PTR_ERR(path);
1121    } else {
1122        /* tree is full, time to grow in depth */
1123        err = ext4_ext_grow_indepth(handle, inode, path, newext);
1124        if (err)
1125            goto out;
1126
1127        /* refill path */
1128        ext4_ext_drop_refs(path);
1129        path = ext4_ext_find_extent(inode,
1130                   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1131                    path);
1132        if (IS_ERR(path)) {
1133            err = PTR_ERR(path);
1134            goto out;
1135        }
1136
1137        /*
1138         * only first (depth 0 -> 1) produces free space;
1139         * in all other cases we have to split the grown tree
1140         */
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 */
1144            goto repeat;
1145        }
1146    }
1147
1148out:
1149    return err;
1150}
1151
1152/*
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
1158 */
1159int
1160ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1161            ext4_lblk_t *logical, ext4_fsblk_t *phys)
1162{
1163    struct ext4_extent_idx *ix;
1164    struct ext4_extent *ex;
1165    int depth, ee_len;
1166
1167    BUG_ON(path == NULL);
1168    depth = path->p_depth;
1169    *phys = 0;
1170
1171    if (depth == 0 && path->p_ext == NULL)
1172        return 0;
1173
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 */
1177
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));
1185        }
1186        return 0;
1187    }
1188
1189    BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1190
1191    *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1192    *phys = ext_pblock(ex) + ee_len - 1;
1193    return 0;
1194}
1195
1196/*
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
1202 */
1203int
1204ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1205            ext4_lblk_t *logical, ext4_fsblk_t *phys)
1206{
1207    struct buffer_head *bh = NULL;
1208    struct ext4_extent_header *eh;
1209    struct ext4_extent_idx *ix;
1210    struct ext4_extent *ex;
1211    ext4_fsblk_t block;
1212    int depth; /* Note, NOT eh_depth; depth from top of tree */
1213    int ee_len;
1214
1215    BUG_ON(path == NULL);
1216    depth = path->p_depth;
1217    *phys = 0;
1218
1219    if (depth == 0 && path->p_ext == NULL)
1220        return 0;
1221
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 */
1225
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));
1233        }
1234        *logical = le32_to_cpu(ex->ee_block);
1235        *phys = ext_pblock(ex);
1236        return 0;
1237    }
1238
1239    BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
1240
1241    if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1242        /* next allocated block in this leaf */
1243        ex++;
1244        *logical = le32_to_cpu(ex->ee_block);
1245        *phys = ext_pblock(ex);
1246        return 0;
1247    }
1248
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))
1253            goto got_index;
1254    }
1255
1256    /* we've gone up to the root and found no index to the right */
1257    return 0;
1258
1259got_index:
1260    /* we've found index to the right, let's
1261     * follow it and find the closest allocated
1262     * block to the right */
1263    ix++;
1264    block = idx_pblock(ix);
1265    while (++depth < path->p_depth) {
1266        bh = sb_bread(inode->i_sb, block);
1267        if (bh == NULL)
1268            return -EIO;
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)) {
1272            put_bh(bh);
1273            return -EIO;
1274        }
1275        ix = EXT_FIRST_INDEX(eh);
1276        block = idx_pblock(ix);
1277        put_bh(bh);
1278    }
1279
1280    bh = sb_bread(inode->i_sb, block);
1281    if (bh == NULL)
1282        return -EIO;
1283    eh = ext_block_hdr(bh);
1284    if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1285        put_bh(bh);
1286        return -EIO;
1287    }
1288    ex = EXT_FIRST_EXTENT(eh);
1289    *logical = le32_to_cpu(ex->ee_block);
1290    *phys = ext_pblock(ex);
1291    put_bh(bh);
1292    return 0;
1293}
1294
1295/*
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
1300 * with leaves.
1301 */
1302static ext4_lblk_t
1303ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1304{
1305    int depth;
1306
1307    BUG_ON(path == NULL);
1308    depth = path->p_depth;
1309
1310    if (depth == 0 && path->p_ext == NULL)
1311        return EXT_MAX_BLOCK;
1312
1313    while (depth >= 0) {
1314        if (depth == path->p_depth) {
1315            /* leaf */
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);
1319        } else {
1320            /* index */
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);
1324        }
1325        depth--;
1326    }
1327
1328    return EXT_MAX_BLOCK;
1329}
1330
1331/*
1332 * ext4_ext_next_leaf_block:
1333 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1334 */
1335static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1336                    struct ext4_ext_path *path)
1337{
1338    int depth;
1339
1340    BUG_ON(path == NULL);
1341    depth = path->p_depth;
1342
1343    /* zero-tree has no leaf blocks at all */
1344    if (depth == 0)
1345        return EXT_MAX_BLOCK;
1346
1347    /* go to index block */
1348    depth--;
1349
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);
1355        depth--;
1356    }
1357
1358    return EXT_MAX_BLOCK;
1359}
1360
1361/*
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?
1366 */
1367static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1368                struct ext4_ext_path *path)
1369{
1370    struct ext4_extent_header *eh;
1371    int depth = ext_depth(inode);
1372    struct ext4_extent *ex;
1373    __le32 border;
1374    int k, err = 0;
1375
1376    eh = path[depth].p_hdr;
1377    ex = path[depth].p_ext;
1378    BUG_ON(ex == NULL);
1379    BUG_ON(eh == NULL);
1380
1381    if (depth == 0) {
1382        /* there is no tree at all */
1383        return 0;
1384    }
1385
1386    if (ex != EXT_FIRST_EXTENT(eh)) {
1387        /* we correct tree if first leaf got modified only */
1388        return 0;
1389    }
1390
1391    /*
1392     * TODO: we need correction if border is smaller than current one
1393     */
1394    k = depth - 1;
1395    border = path[depth].p_ext->ee_block;
1396    err = ext4_ext_get_access(handle, inode, path + k);
1397    if (err)
1398        return err;
1399    path[k].p_idx->ei_block = border;
1400    err = ext4_ext_dirty(handle, inode, path + k);
1401    if (err)
1402        return err;
1403
1404    while (k--) {
1405        /* change all left-side indexes */
1406        if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1407            break;
1408        err = ext4_ext_get_access(handle, inode, path + k);
1409        if (err)
1410            break;
1411        path[k].p_idx->ei_block = border;
1412        err = ext4_ext_dirty(handle, inode, path + k);
1413        if (err)
1414            break;
1415    }
1416
1417    return err;
1418}
1419
1420int
1421ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1422                struct ext4_extent *ex2)
1423{
1424    unsigned short ext1_ee_len, ext2_ee_len, max_len;
1425
1426    /*
1427     * Make sure that either both extents are uninitialized, or
1428     * both are _not_.
1429     */
1430    if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1431        return 0;
1432
1433    if (ext4_ext_is_uninitialized(ex1))
1434        max_len = EXT_UNINIT_MAX_LEN;
1435    else
1436        max_len = EXT_INIT_MAX_LEN;
1437
1438    ext1_ee_len = ext4_ext_get_actual_len(ex1);
1439    ext2_ee_len = ext4_ext_get_actual_len(ex2);
1440
1441    if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1442            le32_to_cpu(ex2->ee_block))
1443        return 0;
1444
1445    /*
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.
1449     */
1450    if (ext1_ee_len + ext2_ee_len > max_len)
1451        return 0;
1452#ifdef AGGRESSIVE_TEST
1453    if (ext1_ee_len >= 4)
1454        return 0;
1455#endif
1456
1457    if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1458        return 1;
1459    return 0;
1460}
1461
1462/*
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.
1468 */
1469int ext4_ext_try_to_merge(struct inode *inode,
1470              struct ext4_ext_path *path,
1471              struct ext4_extent *ex)
1472{
1473    struct ext4_extent_header *eh;
1474    unsigned int depth, len;
1475    int merge_done = 0;
1476    int uninitialized = 0;
1477
1478    depth = ext_depth(inode);
1479    BUG_ON(path[depth].p_hdr == NULL);
1480    eh = path[depth].p_hdr;
1481
1482    while (ex < EXT_LAST_EXTENT(eh)) {
1483        if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1484            break;
1485        /* merge with next extent! */
1486        if (ext4_ext_is_uninitialized(ex))
1487            uninitialized = 1;
1488        ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1489                + ext4_ext_get_actual_len(ex + 1));
1490        if (uninitialized)
1491            ext4_ext_mark_uninitialized(ex);
1492
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);
1497        }
1498        le16_add_cpu(&eh->eh_entries, -1);
1499        merge_done = 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);
1504    }
1505
1506    return merge_done;
1507}
1508
1509/*
1510 * check if a portion of the "newext" extent overlaps with an
1511 * existing extent.
1512 *
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.
1516 */
1517unsigned int ext4_ext_check_overlap(struct inode *inode,
1518                    struct ext4_extent *newext,
1519                    struct ext4_ext_path *path)
1520{
1521    ext4_lblk_t b1, b2;
1522    unsigned int depth, len1;
1523    unsigned int ret = 0;
1524
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)
1529        goto out;
1530    b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1531
1532    /*
1533     * get the next allocated block if the extent in the path
1534     * is before the requested block(s)
1535     */
1536    if (b2 < b1) {
1537        b2 = ext4_ext_next_allocated_block(path);
1538        if (b2 == EXT_MAX_BLOCK)
1539            goto out;
1540    }
1541
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);
1546        ret = 1;
1547    }
1548
1549    /* check for overlap */
1550    if (b1 + len1 > b2) {
1551        newext->ee_len = cpu_to_le16(b2 - b1);
1552        ret = 1;
1553    }
1554out:
1555    return ret;
1556}
1557
1558/*
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.
1563 */
1564int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1565                struct ext4_ext_path *path,
1566                struct ext4_extent *newext)
1567{
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;
1573    ext4_lblk_t next;
1574    unsigned uninitialized = 0;
1575
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);
1580
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);
1588        if (err)
1589            return err;
1590
1591        /*
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.
1595         */
1596        if (ext4_ext_is_uninitialized(ex))
1597            uninitialized = 1;
1598        ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1599                    + ext4_ext_get_actual_len(newext));
1600        if (uninitialized)
1601            ext4_ext_mark_uninitialized(ex);
1602        eh = path[depth].p_hdr;
1603        nearex = ex;
1604        goto merge;
1605    }
1606
1607repeat:
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))
1611        goto has_space;
1612
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);
1621        if (IS_ERR(npath))
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));
1628            path = npath;
1629            goto repeat;
1630        }
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));
1633    }
1634
1635    /*
1636     * There is no free space in the found leaf.
1637     * We're gonna add a new leaf in the tree.
1638     */
1639    err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1640    if (err)
1641        goto cleanup;
1642    depth = ext_depth(inode);
1643    eh = path[depth].p_hdr;
1644
1645has_space:
1646    nearex = path[depth].p_ext;
1647
1648    err = ext4_ext_get_access(handle, inode, path + depth);
1649    if (err)
1650        goto cleanup;
1651
1652    if (!nearex) {
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),
1656                ext_pblock(newext),
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),
1669                    ext_pblock(newext),
1670                    ext4_ext_get_actual_len(newext),
1671                    nearex, len, nearex + 1, nearex + 2);
1672            memmove(nearex + 2, nearex + 1, len);
1673        }
1674        path[depth].p_ext = nearex + 1;
1675    } else {
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),
1682                ext_pblock(newext),
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;
1687    }
1688
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;
1694
1695merge:
1696    /* try to merge extents to the right */
1697    ext4_ext_try_to_merge(inode, path, nearex);
1698
1699    /* try to merge extents to the left */
1700
1701    /* time to correct all indexes above */
1702    err = ext4_ext_correct_indexes(handle, inode, path);
1703    if (err)
1704        goto cleanup;
1705
1706    err = ext4_ext_dirty(handle, inode, path + depth);
1707
1708cleanup:
1709    if (npath) {
1710        ext4_ext_drop_refs(npath);
1711        kfree(npath);
1712    }
1713    ext4_ext_invalidate_cache(inode);
1714    return err;
1715}
1716
1717int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1718            ext4_lblk_t num, ext_prepare_callback func,
1719            void *cbdata)
1720{
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;
1727
1728    BUG_ON(func == NULL);
1729    BUG_ON(inode == NULL);
1730
1731    while (block < last && block != EXT_MAX_BLOCK) {
1732        num = last - block;
1733        /* find extent for this block */
1734        path = ext4_ext_find_extent(inode, block, path);
1735        if (IS_ERR(path)) {
1736            err = PTR_ERR(path);
1737            path = NULL;
1738            break;
1739        }
1740
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);
1745
1746        exists = 0;
1747        if (!ex) {
1748            /* there is no extent yet, so try to allocate
1749             * all requested space */
1750            start = block;
1751            end = block + num;
1752        } else if (le32_to_cpu(ex->ee_block) > block) {
1753            /* need to allocate space before found extent */
1754            start = block;
1755            end = le32_to_cpu(ex->ee_block);
1756            if (block + num < end)
1757                end = block + num;
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 */
1761            start = block;
1762            end = block + num;
1763            if (end >= next)
1764                end = next;
1765        } else if (block >= le32_to_cpu(ex->ee_block)) {
1766            /*
1767             * some part of requested space is covered
1768             * by found extent
1769             */
1770            start = block;
1771            end = le32_to_cpu(ex->ee_block)
1772                + ext4_ext_get_actual_len(ex);
1773            if (block + num < end)
1774                end = block + num;
1775            exists = 1;
1776        } else {
1777            BUG();
1778        }
1779        BUG_ON(end <= start);
1780
1781        if (!exists) {
1782            cbex.ec_block = start;
1783            cbex.ec_len = end - start;
1784            cbex.ec_start = 0;
1785            cbex.ec_type = EXT4_EXT_CACHE_GAP;
1786        } else {
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;
1791        }
1792
1793        BUG_ON(cbex.ec_len == 0);
1794        err = func(inode, path, &cbex, ex, cbdata);
1795        ext4_ext_drop_refs(path);
1796
1797        if (err < 0)
1798            break;
1799
1800        if (err == EXT_REPEAT)
1801            continue;
1802        else if (err == EXT_BREAK) {
1803            err = 0;
1804            break;
1805        }
1806
1807        if (ext_depth(inode) != depth) {
1808            /* depth was changed. we have to realloc path */
1809            kfree(path);
1810            path = NULL;
1811        }
1812
1813        block = cbex.ec_block + cbex.ec_len;
1814    }
1815
1816    if (path) {
1817        ext4_ext_drop_refs(path);
1818        kfree(path);
1819    }
1820
1821    return err;
1822}
1823
1824static void
1825ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1826            __u32 len, ext4_fsblk_t start, int type)
1827{
1828    struct ext4_ext_cache *cex;
1829    BUG_ON(len == 0);
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;
1834    cex->ec_len = len;
1835    cex->ec_start = start;
1836    spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1837}
1838
1839/*
1840 * ext4_ext_put_gap_in_cache:
1841 * calculate boundaries of the gap that the requested block fits into
1842 * and cache this gap
1843 */
1844static void
1845ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1846                ext4_lblk_t block)
1847{
1848    int depth = ext_depth(inode);
1849    unsigned long len;
1850    ext4_lblk_t lblock;
1851    struct ext4_extent *ex;
1852
1853    ex = path[depth].p_ext;
1854    if (ex == NULL) {
1855        /* there is no extent yet, so gap is [0;-] */
1856        lblock = 0;
1857        len = EXT_MAX_BLOCK;
1858        ext_debug("cache gap(whole file):");
1859    } else if (block < le32_to_cpu(ex->ee_block)) {
1860        lblock = block;
1861        len = le32_to_cpu(ex->ee_block) - block;
1862        ext_debug("cache gap(before): %u [%u:%u]",
1863                block,
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)) {
1868        ext4_lblk_t next;
1869        lblock = le32_to_cpu(ex->ee_block)
1870            + ext4_ext_get_actual_len(ex);
1871
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),
1876                block);
1877        BUG_ON(next == lblock);
1878        len = next - lblock;
1879    } else {
1880        lblock = len = 0;
1881        BUG();
1882    }
1883
1884    ext_debug(" -> %u:%lu\n", lblock, len);
1885    ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1886}
1887
1888static int
1889ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1890            struct ext4_extent *ex)
1891{
1892    struct ext4_ext_cache *cex;
1893    int ret = EXT4_EXT_CACHE_NO;
1894
1895    /*
1896     * We borrow i_block_reservation_lock to protect i_cached_extent
1897     */
1898    spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1899    cex = &EXT4_I(inode)->i_cached_extent;
1900
1901    /* has cache valid data? */
1902    if (cex->ec_type == EXT4_EXT_CACHE_NO)
1903        goto errout;
1904
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",
1912                block,
1913                cex->ec_block, cex->ec_len, cex->ec_start);
1914        ret = cex->ec_type;
1915    }
1916errout:
1917    spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1918    return ret;
1919}
1920
1921/*
1922 * ext4_ext_rm_idx:
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.
1926 */
1927static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1928            struct ext4_ext_path *path)
1929{
1930    struct buffer_head *bh;
1931    int err;
1932    ext4_fsblk_t leaf;
1933
1934    /* free index block */
1935    path--;
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);
1939    if (err)
1940        return err;
1941    le16_add_cpu(&path->p_hdr->eh_entries, -1);
1942    err = ext4_ext_dirty(handle, inode, path);
1943    if (err)
1944        return err;
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);
1949    return err;
1950}
1951
1952/*
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
1957 * under i_data_sem.
1958 */
1959int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
1960                        struct ext4_ext_path *path)
1961{
1962    if (path) {
1963        int depth = ext_depth(inode);
1964        int ret = 0;
1965
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)) {
1969
1970            /*
1971             * There are some space in the leaf tree, no
1972             * need to account for leaf block credit
1973             *
1974             * bitmaps and block group descriptor blocks
1975             * and other metadat blocks still need to be
1976             * accounted.
1977             */
1978            /* 1 bitmap, 1 block group descriptor */
1979            ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
1980            return ret;
1981        }
1982    }
1983
1984    return ext4_chunk_trans_blocks(inode, nrblocks);
1985}
1986
1987/*
1988 * How many index/leaf blocks need to change/allocate to modify nrblocks?
1989 *
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
1994 *
1995 * If the nrblocks are discontiguous, they could cause
1996 * the whole tree split more than once, but this is really rare.
1997 */
1998int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
1999{
2000    int index;
2001    int depth = ext_depth(inode);
2002
2003    if (chunk)
2004        index = depth * 2;
2005    else
2006        index = depth * 3;
2007
2008    return index;
2009}
2010
2011static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2012                struct ext4_extent *ex,
2013                ext4_lblk_t from, ext4_lblk_t to)
2014{
2015    struct buffer_head *bh;
2016    unsigned short ee_len = ext4_ext_get_actual_len(ex);
2017    int i, metadata = 0;
2018
2019    if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2020        metadata = 1;
2021#ifdef EXTENTS_STATS
2022    {
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);
2034    }
2035#endif
2036    if (from >= le32_to_cpu(ex->ee_block)
2037        && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2038        /* tail removal */
2039        ext4_lblk_t num;
2040        ext4_fsblk_t start;
2041
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);
2048        }
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);
2054    } else {
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);
2058    }
2059    return 0;
2060}
2061
2062static int
2063ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2064        struct ext4_ext_path *path, ext4_lblk_t start)
2065{
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;
2070    unsigned num;
2071    ext4_lblk_t ex_ee_block;
2072    unsigned short ex_ee_len;
2073    unsigned uninitialized = 0;
2074    struct ext4_extent *ex;
2075
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;
2081    BUG_ON(eh == NULL);
2082
2083    /* find where to start removing */
2084    ex = EXT_LAST_EXTENT(eh);
2085
2086    ex_ee_block = le32_to_cpu(ex->ee_block);
2087    ex_ee_len = ext4_ext_get_actual_len(ex);
2088
2089    while (ex >= EXT_FIRST_EXTENT(eh) &&
2090            ex_ee_block + ex_ee_len > start) {
2091
2092        if (ext4_ext_is_uninitialized(ex))
2093            uninitialized = 1;
2094        else
2095            uninitialized = 0;
2096
2097        ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
2098        path[depth].p_ext = ex;
2099
2100        a = ex_ee_block > start ? ex_ee_block : start;
2101        b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2102            ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2103
2104        ext_debug(" border %u:%u\n", a, b);
2105
2106        if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2107            block = 0;
2108            num = 0;
2109            BUG();
2110        } else if (a != ex_ee_block) {
2111            /* remove tail of the extent */
2112            block = ex_ee_block;
2113            num = a - block;
2114        } else if (b != ex_ee_block + ex_ee_len - 1) {
2115            /* remove head of the extent */
2116            block = a;
2117            num = b - a;
2118            /* there is no "make a hole" API yet */
2119            BUG();
2120        } else {
2121            /* remove whole extent: excellent! */
2122            block = ex_ee_block;
2123            num = 0;
2124            BUG_ON(a != ex_ee_block);
2125            BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2126        }
2127
2128        /*
2129         * 3 for leaf, sb, and inode plus 2 (bmap and group
2130         * descriptor) for each block group; assume two block
2131         * groups plus ex_ee_len/blocks_per_block_group for
2132         * the worst case
2133         */
2134        credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2135        if (ex == EXT_FIRST_EXTENT(eh)) {
2136            correct_index = 1;
2137            credits += (ext_depth(inode)) + 1;
2138        }
2139        credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2140
2141        err = ext4_ext_journal_restart(handle, credits);
2142        if (err)
2143            goto out;
2144
2145        err = ext4_ext_get_access(handle, inode, path + depth);
2146        if (err)
2147            goto out;
2148
2149        err = ext4_remove_blocks(handle, inode, ex, a, b);
2150        if (err)
2151            goto out;
2152
2153        if (num == 0) {
2154            /* this extent is removed; mark slot entirely unused */
2155            ext4_ext_store_pblock(ex, 0);
2156            le16_add_cpu(&eh->eh_entries, -1);
2157        }
2158
2159        ex->ee_block = cpu_to_le32(block);
2160        ex->ee_len = cpu_to_le16(num);
2161        /*
2162         * Do not mark uninitialized if all the blocks in the
2163         * extent have been removed.
2164         */
2165        if (uninitialized && num)
2166            ext4_ext_mark_uninitialized(ex);
2167
2168        err = ext4_ext_dirty(handle, inode, path + depth);
2169        if (err)
2170            goto out;
2171
2172        ext_debug("new extent: %u:%u:%llu\n", block, num,
2173                ext_pblock(ex));
2174        ex--;
2175        ex_ee_block = le32_to_cpu(ex->ee_block);
2176        ex_ee_len = ext4_ext_get_actual_len(ex);
2177    }
2178
2179    if (correct_index && eh->eh_entries)
2180        err = ext4_ext_correct_indexes(handle, inode, path);
2181
2182    /* if this leaf is free, then we should
2183     * remove it from index block above */
2184    if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2185        err = ext4_ext_rm_idx(handle, inode, path + depth);
2186
2187out:
2188    return err;
2189}
2190
2191/*
2192 * ext4_ext_more_to_rm:
2193 * returns 1 if current index has to be freed (even partial)
2194 */
2195static int
2196ext4_ext_more_to_rm(struct ext4_ext_path *path)
2197{
2198    BUG_ON(path->p_idx == NULL);
2199
2200    if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2201        return 0;
2202
2203    /*
2204     * if truncate on deeper level happened, it wasn't partial,
2205     * so we have to consider current index for truncation
2206     */
2207    if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2208        return 0;
2209    return 1;
2210}
2211
2212static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2213{
2214    struct super_block *sb = inode->i_sb;
2215    int depth = ext_depth(inode);
2216    struct ext4_ext_path *path;
2217    handle_t *handle;
2218    int i = 0, err = 0;
2219
2220    ext_debug("truncate since %u\n", start);
2221
2222    /* probably first extent we're gonna free will be last in block */
2223    handle = ext4_journal_start(inode, depth + 1);
2224    if (IS_ERR(handle))
2225        return PTR_ERR(handle);
2226
2227    ext4_ext_invalidate_cache(inode);
2228
2229    /*
2230     * We start scanning from right side, freeing all the blocks
2231     * after i_size and walking into the tree depth-wise.
2232     */
2233    path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2234    if (path == NULL) {
2235        ext4_journal_stop(handle);
2236        return -ENOMEM;
2237    }
2238    path[0].p_hdr = ext_inode_hdr(inode);
2239    if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2240        err = -EIO;
2241        goto out;
2242    }
2243    path[0].p_depth = depth;
2244
2245    while (i >= 0 && err == 0) {
2246        if (i == depth) {
2247            /* this is leaf block */
2248            err = ext4_ext_rm_leaf(handle, inode, path, start);
2249            /* root level has p_bh == NULL, brelse() eats this */
2250            brelse(path[i].p_bh);
2251            path[i].p_bh = NULL;
2252            i--;
2253            continue;
2254        }
2255
2256        /* this is index block */
2257        if (!path[i].p_hdr) {
2258            ext_debug("initialize header\n");
2259            path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2260        }
2261
2262        if (!path[i].p_idx) {
2263            /* this level hasn't been touched yet */
2264            path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2265            path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2266            ext_debug("init index ptr: hdr 0x%p, num %d\n",
2267                  path[i].p_hdr,
2268                  le16_to_cpu(path[i].p_hdr->eh_entries));
2269        } else {
2270            /* we were already here, see at next index */
2271            path[i].p_idx--;
2272        }
2273
2274        ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2275                i, EXT_FIRST_INDEX(path[i].p_hdr),
2276                path[i].p_idx);
2277        if (ext4_ext_more_to_rm(path + i)) {
2278            struct buffer_head *bh;
2279            /* go to the next level */
2280            ext_debug("move to level %d (block %llu)\n",
2281                  i + 1, idx_pblock(path[i].p_idx));
2282            memset(path + i + 1, 0, sizeof(*path));
2283            bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2284            if (!bh) {
2285                /* should we reset i_size? */
2286                err = -EIO;
2287                break;
2288            }
2289            if (WARN_ON(i + 1 > depth)) {
2290                err = -EIO;
2291                break;
2292            }
2293            if (ext4_ext_check(inode, ext_block_hdr(bh),
2294                            depth - i - 1)) {
2295                err = -EIO;
2296                break;
2297            }
2298            path[i + 1].p_bh = bh;
2299
2300            /* save actual number of indexes since this
2301             * number is changed at the next iteration */
2302            path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2303            i++;
2304        } else {
2305            /* we finished processing this index, go up */
2306            if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2307                /* index is empty, remove it;
2308                 * handle must be already prepared by the
2309                 * truncatei_leaf() */
2310                err = ext4_ext_rm_idx(handle, inode, path + i);
2311            }
2312            /* root level has p_bh == NULL, brelse() eats this */
2313            brelse(path[i].p_bh);
2314            path[i].p_bh = NULL;
2315            i--;
2316            ext_debug("return to level %d\n", i);
2317        }
2318    }
2319
2320    /* TODO: flexible tree reduction should be here */
2321    if (path->p_hdr->eh_entries == 0) {
2322        /*
2323         * truncate to zero freed all the tree,
2324         * so we need to correct eh_depth
2325         */
2326        err = ext4_ext_get_access(handle, inode, path);
2327        if (err == 0) {
2328            ext_inode_hdr(inode)->eh_depth = 0;
2329            ext_inode_hdr(inode)->eh_max =
2330                cpu_to_le16(ext4_ext_space_root(inode));
2331            err = ext4_ext_dirty(handle, inode, path);
2332        }
2333    }
2334out:
2335    ext4_ext_drop_refs(path);
2336    kfree(path);
2337    ext4_journal_stop(handle);
2338
2339    return err;
2340}
2341
2342/*
2343 * called at mount time
2344 */
2345void ext4_ext_init(struct super_block *sb)
2346{
2347    /*
2348     * possible initialization would be here
2349     */
2350
2351    if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2352        printk(KERN_INFO "EXT4-fs: file extents enabled");
2353#ifdef AGGRESSIVE_TEST
2354        printk(", aggressive tests");
2355#endif
2356#ifdef CHECK_BINSEARCH
2357        printk(", check binsearch");
2358#endif
2359#ifdef EXTENTS_STATS
2360        printk(", stats");
2361#endif
2362        printk("\n");
2363#ifdef EXTENTS_STATS
2364        spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2365        EXT4_SB(sb)->s_ext_min = 1 << 30;
2366        EXT4_SB(sb)->s_ext_max = 0;
2367#endif
2368    }
2369}
2370
2371/*
2372 * called at umount time
2373 */
2374void ext4_ext_release(struct super_block *sb)
2375{
2376    if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2377        return;
2378
2379#ifdef EXTENTS_STATS
2380    if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2381        struct ext4_sb_info *sbi = EXT4_SB(sb);
2382        printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2383            sbi->s_ext_blocks, sbi->s_ext_extents,
2384            sbi->s_ext_blocks / sbi->s_ext_extents);
2385        printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2386            sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2387    }
2388#endif
2389}
2390
2391static void bi_complete(struct bio *bio, int error)
2392{
2393    complete((struct completion *)bio->bi_private);
2394}
2395
2396/* FIXME!! we need to try to merge to left or right after zero-out */
2397static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2398{
2399    int ret = -EIO;
2400    struct bio *bio;
2401    int blkbits, blocksize;
2402    sector_t ee_pblock;
2403    struct completion event;
2404    unsigned int ee_len, len, done, offset;
2405
2406
2407    blkbits = inode->i_blkbits;
2408    blocksize = inode->i_sb->s_blocksize;
2409    ee_len = ext4_ext_get_actual_len(ex);
2410    ee_pblock = ext_pblock(ex);
2411
2412    /* convert ee_pblock to 512 byte sectors */
2413    ee_pblock = ee_pblock << (blkbits - 9);
2414
2415    while (ee_len > 0) {
2416
2417        if (ee_len > BIO_MAX_PAGES)
2418            len = BIO_MAX_PAGES;
2419        else
2420            len = ee_len;
2421
2422        bio = bio_alloc(GFP_NOIO, len);
2423        bio->bi_sector = ee_pblock;
2424        bio->bi_bdev = inode->i_sb->s_bdev;
2425
2426        done = 0;
2427        offset = 0;
2428        while (done < len) {
2429            ret = bio_add_page(bio, ZERO_PAGE(0),
2430                            blocksize, offset);
2431            if (ret != blocksize) {
2432                /*
2433                 * We can't add any more pages because of
2434                 * hardware limitations. Start a new bio.
2435                 */
2436                break;
2437            }
2438            done++;
2439            offset += blocksize;
2440            if (offset >= PAGE_CACHE_SIZE)
2441                offset = 0;
2442        }
2443
2444        init_completion(&event);
2445        bio->bi_private = &event;
2446        bio->bi_end_io = bi_complete;
2447        submit_bio(WRITE, bio);
2448        wait_for_completion(&event);
2449
2450        if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2451            ret = 0;
2452        else {
2453            ret = -EIO;
2454            break;
2455        }
2456        bio_put(bio);
2457        ee_len -= done;
2458        ee_pblock += done << (blkbits - 9);
2459    }
2460    return ret;
2461}
2462
2463#define EXT4_EXT_ZERO_LEN 7
2464
2465/*
2466 * This function is called by ext4_ext_get_blocks() if someone tries to write
2467 * to an uninitialized extent. It may result in splitting the uninitialized
2468 * extent into multiple extents (upto three - one initialized and two
2469 * uninitialized).
2470 * There are three possibilities:
2471 * a> There is no split required: Entire extent should be initialized
2472 * b> Splits in two extents: Write is happening at either end of the extent
2473 * c> Splits in three extents: Somone is writing in middle of the extent
2474 */
2475static int ext4_ext_convert_to_initialized(handle_t *handle,
2476                        struct inode *inode,
2477                        struct ext4_ext_path *path,
2478                        ext4_lblk_t iblock,
2479                        unsigned int max_blocks)
2480{
2481    struct ext4_extent *ex, newex, orig_ex;
2482    struct ext4_extent *ex1 = NULL;
2483    struct ext4_extent *ex2 = NULL;
2484    struct ext4_extent *ex3 = NULL;
2485    struct ext4_extent_header *eh;
2486    ext4_lblk_t ee_block;
2487    unsigned int allocated, ee_len, depth;
2488    ext4_fsblk_t newblock;
2489    int err = 0;
2490    int ret = 0;
2491
2492    depth = ext_depth(inode);
2493    eh = path[depth].p_hdr;
2494    ex = path[depth].p_ext;
2495    ee_block = le32_to_cpu(ex->ee_block);
2496    ee_len = ext4_ext_get_actual_len(ex);
2497    allocated = ee_len - (iblock - ee_block);
2498    newblock = iblock - ee_block + ext_pblock(ex);
2499    ex2 = ex;
2500    orig_ex.ee_block = ex->ee_block;
2501    orig_ex.ee_len = cpu_to_le16(ee_len);
2502    ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2503
2504    err = ext4_ext_get_access(handle, inode, path + depth);
2505    if (err)
2506        goto out;
2507    /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2508    if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2509        err = ext4_ext_zeroout(inode, &orig_ex);
2510        if (err)
2511            goto fix_extent_len;
2512        /* update the extent length and mark as initialized */
2513        ex->ee_block = orig_ex.ee_block;
2514        ex->ee_len = orig_ex.ee_len;
2515        ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2516        ext4_ext_dirty(handle, inode, path + depth);
2517        /* zeroed the full extent */
2518        return allocated;
2519    }
2520
2521    /* ex1: ee_block to iblock - 1 : uninitialized */
2522    if (iblock > ee_block) {
2523        ex1 = ex;
2524        ex1->ee_len = cpu_to_le16(iblock - ee_block);
2525        ext4_ext_mark_uninitialized(ex1);
2526        ex2 = &newex;
2527    }
2528    /*
2529     * for sanity, update the length of the ex2 extent before
2530     * we insert ex3, if ex1 is NULL. This is to avoid temporary
2531     * overlap of blocks.
2532     */
2533    if (!ex1 && allocated > max_blocks)
2534        ex2->ee_len = cpu_to_le16(max_blocks);
2535    /* ex3: to ee_block + ee_len : uninitialised */
2536    if (allocated > max_blocks) {
2537        unsigned int newdepth;
2538        /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2539        if (allocated <= EXT4_EXT_ZERO_LEN) {
2540            /*
2541             * iblock == ee_block is handled by the zerouout
2542             * at the beginning.
2543             * Mark first half uninitialized.
2544             * Mark second half initialized and zero out the
2545             * initialized extent
2546             */
2547            ex->ee_block = orig_ex.ee_block;
2548            ex->ee_len = cpu_to_le16(ee_len - allocated);
2549            ext4_ext_mark_uninitialized(ex);
2550            ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2551            ext4_ext_dirty(handle, inode, path + depth);
2552
2553            ex3 = &newex;
2554            ex3->ee_block = cpu_to_le32(iblock);
2555            ext4_ext_store_pblock(ex3, newblock);
2556            ex3->ee_len = cpu_to_le16(allocated);
2557            err = ext4_ext_insert_extent(handle, inode, path, ex3);
2558            if (err == -ENOSPC) {
2559                err = ext4_ext_zeroout(inode, &orig_ex);
2560                if (err)
2561                    goto fix_extent_len;
2562                ex->ee_block = orig_ex.ee_block;
2563                ex->ee_len = orig_ex.ee_len;
2564                ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2565                ext4_ext_dirty(handle, inode, path + depth);
2566                /* blocks available from iblock */
2567                return allocated;
2568
2569            } else if (err)
2570                goto fix_extent_len;
2571
2572            /*
2573             * We need to zero out the second half because
2574             * an fallocate request can update file size and
2575             * converting the second half to initialized extent
2576             * implies that we can leak some junk data to user
2577             * space.
2578             */
2579            err = ext4_ext_zeroout(inode, ex3);
2580            if (err) {
2581                /*
2582                 * We should actually mark the
2583                 * second half as uninit and return error
2584                 * Insert would have changed the extent
2585                 */
2586                depth = ext_depth(inode);
2587                ext4_ext_drop_refs(path);
2588                path = ext4_ext_find_extent(inode,
2589                                iblock, path);
2590                if (IS_ERR(path)) {
2591                    err = PTR_ERR(path);
2592                    return err;
2593                }
2594                /* get the second half extent details */
2595                ex = path[depth].p_ext;
2596                err = ext4_ext_get_access(handle, inode,
2597                                path + depth);
2598                if (err)
2599                    return err;
2600                ext4_ext_mark_uninitialized(ex);
2601                ext4_ext_dirty(handle, inode, path + depth);
2602                return err;
2603            }
2604
2605            /* zeroed the second half */
2606            return allocated;
2607        }
2608        ex3 = &newex;
2609        ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2610        ext4_ext_store_pblock(ex3, newblock + max_blocks);
2611        ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2612        ext4_ext_mark_uninitialized(ex3);
2613        err = ext4_ext_insert_extent(handle, inode, path, ex3);
2614        if (err == -ENOSPC) {
2615            err = ext4_ext_zeroout(inode, &orig_ex);
2616            if (err)
2617                goto fix_extent_len;
2618            /* update the extent length and mark as initialized */
2619            ex->ee_block = orig_ex.ee_block;
2620            ex->ee_len = orig_ex.ee_len;
2621            ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2622            ext4_ext_dirty(handle, inode, path + depth);
2623            /* zeroed the full extent */
2624            /* blocks available from iblock */
2625            return allocated;
2626
2627        } else if (err)
2628            goto fix_extent_len;
2629        /*
2630         * The depth, and hence eh & ex might change
2631         * as part of the insert above.
2632         */
2633        newdepth = ext_depth(inode);
2634        /*
2635         * update the extent length after successful insert of the
2636         * split extent
2637         */
2638        orig_ex.ee_len = cpu_to_le16(ee_len -
2639                        ext4_ext_get_actual_len(ex3));
2640        depth = newdepth;
2641        ext4_ext_drop_refs(path);
2642        path = ext4_ext_find_extent(inode, iblock, path);
2643        if (IS_ERR(path)) {
2644            err = PTR_ERR(path);
2645            goto out;
2646        }
2647        eh = path[depth].p_hdr;
2648        ex = path[depth].p_ext;
2649        if (ex2 != &newex)
2650            ex2 = ex;
2651
2652        err = ext4_ext_get_access(handle, inode, path + depth);
2653        if (err)
2654            goto out;
2655
2656        allocated = max_blocks;
2657
2658        /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2659         * to insert a extent in the middle zerout directly
2660         * otherwise give the extent a chance to merge to left
2661         */
2662        if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2663                            iblock != ee_block) {
2664            err = ext4_ext_zeroout(inode, &orig_ex);
2665            if (err)
2666                goto fix_extent_len;
2667            /* update the extent length and mark as initialized */
2668            ex->ee_block = orig_ex.ee_block;
2669            ex->ee_len = orig_ex.ee_len;
2670            ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2671            ext4_ext_dirty(handle, inode, path + depth);
2672            /* zero out the first half */
2673            /* blocks available from iblock */
2674            return allocated;
2675        }
2676    }
2677    /*
2678     * If there was a change of depth as part of the
2679     * insertion of ex3 above, we need to update the length
2680     * of the ex1 extent again here
2681     */
2682    if (ex1 && ex1 != ex) {
2683        ex1 = ex;
2684        ex1->ee_len = cpu_to_le16(iblock - ee_block);
2685        ext4_ext_mark_uninitialized(ex1);
2686        ex2 = &newex;
2687    }
2688    /* ex2: iblock to iblock + maxblocks-1 : initialised */
2689    ex2->ee_block = cpu_to_le32(iblock);
2690    ext4_ext_store_pblock(ex2, newblock);
2691    ex2->ee_len = cpu_to_le16(allocated);
2692    if (ex2 != ex)
2693        goto insert;
2694    /*
2695     * New (initialized) extent starts from the first block
2696     * in the current extent. i.e., ex2 == ex
2697     * We have to see if it can be merged with the extent
2698     * on the left.
2699     */
2700    if (ex2 > EXT_FIRST_EXTENT(eh)) {
2701        /*
2702         * To merge left, pass "ex2 - 1" to try_to_merge(),
2703         * since it merges towards right _only_.
2704         */
2705        ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2706        if (ret) {
2707            err = ext4_ext_correct_indexes(handle, inode, path);
2708            if (err)
2709                goto out;
2710            depth = ext_depth(inode);
2711            ex2--;
2712        }
2713    }
2714    /*
2715     * Try to Merge towards right. This might be required
2716     * only when the whole extent is being written to.
2717     * i.e. ex2 == ex and ex3 == NULL.
2718     */
2719    if (!ex3) {
2720        ret = ext4_ext_try_to_merge(inode, path, ex2);
2721        if (ret) {
2722            err = ext4_ext_correct_indexes(handle, inode, path);
2723            if (err)
2724                goto out;
2725        }
2726    }
2727    /* Mark modified extent as dirty */
2728    err = ext4_ext_dirty(handle, inode, path + depth);
2729    goto out;
2730insert:
2731    err = ext4_ext_insert_extent(handle, inode, path, &newex);
2732    if (err == -ENOSPC) {
2733        err = ext4_ext_zeroout(inode, &orig_ex);
2734        if (err)
2735            goto fix_extent_len;
2736        /* update the extent length and mark as initialized */
2737        ex->ee_block = orig_ex.ee_block;
2738        ex->ee_len = orig_ex.ee_len;
2739        ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2740        ext4_ext_dirty(handle, inode, path + depth);
2741        /* zero out the first half */
2742        return allocated;
2743    } else if (err)
2744        goto fix_extent_len;
2745out:
2746    return err ? err : allocated;
2747
2748fix_extent_len:
2749    ex->ee_block = orig_ex.ee_block;
2750    ex->ee_len = orig_ex.ee_len;
2751    ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2752    ext4_ext_mark_uninitialized(ex);
2753    ext4_ext_dirty(handle, inode, path + depth);
2754    return err;
2755}
2756
2757/*
2758 * Block allocation/map/preallocation routine for extents based files
2759 *
2760 *
2761 * Need to be called with
2762 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2763 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2764 *
2765 * return > 0, number of of blocks already mapped/allocated
2766 * if create == 0 and these are pre-allocated blocks
2767 * buffer head is unmapped
2768 * otherwise blocks are mapped
2769 *
2770 * return = 0, if plain look up failed (blocks have not been allocated)
2771 * buffer head is unmapped
2772 *
2773 * return < 0, error case.
2774 */
2775int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2776            ext4_lblk_t iblock,
2777            unsigned int max_blocks, struct buffer_head *bh_result,
2778            int flags)
2779{
2780    struct ext4_ext_path *path = NULL;
2781    struct ext4_extent_header *eh;
2782    struct ext4_extent newex, *ex;
2783    ext4_fsblk_t newblock;
2784    int err = 0, depth, ret, cache_type;
2785    unsigned int allocated = 0;
2786    struct ext4_allocation_request ar;
2787
2788    __clear_bit(BH_New, &bh_result->b_state);
2789    ext_debug("blocks %u/%u requested for inode %u\n",
2790            iblock, max_blocks, inode->i_ino);
2791
2792    /* check in cache */
2793    cache_type = ext4_ext_in_cache(inode, iblock, &newex);
2794    if (cache_type) {
2795        if (cache_type == EXT4_EXT_CACHE_GAP) {
2796            if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2797                /*
2798                 * block isn't allocated yet and
2799                 * user doesn't want to allocate it
2800                 */
2801                goto out2;
2802            }
2803            /* we should allocate requested block */
2804        } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
2805            /* block is already allocated */
2806            newblock = iblock
2807                   - le32_to_cpu(newex.ee_block)
2808                   + ext_pblock(&newex);
2809            /* number of remaining blocks in the extent */
2810            allocated = ext4_ext_get_actual_len(&newex) -
2811                    (iblock - le32_to_cpu(newex.ee_block));
2812            goto out;
2813        } else {
2814            BUG();
2815        }
2816    }
2817
2818    /* find extent for this block */
2819    path = ext4_ext_find_extent(inode, iblock, NULL);
2820    if (IS_ERR(path)) {
2821        err = PTR_ERR(path);
2822        path = NULL;
2823        goto out2;
2824    }
2825
2826    depth = ext_depth(inode);
2827
2828    /*
2829     * consistent leaf must not be empty;
2830     * this situation is possible, though, _during_ tree modification;
2831     * this is why assert can't be put in ext4_ext_find_extent()
2832     */
2833    BUG_ON(path[depth].p_ext == NULL && depth != 0);
2834    eh = path[depth].p_hdr;
2835
2836    ex = path[depth].p_ext;
2837    if (ex) {
2838        ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
2839        ext4_fsblk_t ee_start = ext_pblock(ex);
2840        unsigned short ee_len;
2841
2842        /*
2843         * Uninitialized extents are treated as holes, except that
2844         * we split out initialized portions during a write.
2845         */
2846        ee_len = ext4_ext_get_actual_len(ex);
2847        /* if found extent covers block, simply return it */
2848        if (iblock >= ee_block && iblock < ee_block + ee_len) {
2849            newblock = iblock - ee_block + ee_start;
2850            /* number of remaining blocks in the extent */
2851            allocated = ee_len - (iblock - ee_block);
2852            ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
2853                    ee_block, ee_len, newblock);
2854
2855            /* Do not put uninitialized extent in the cache */
2856            if (!ext4_ext_is_uninitialized(ex)) {
2857                ext4_ext_put_in_cache(inode, ee_block,
2858                            ee_len, ee_start,
2859                            EXT4_EXT_CACHE_EXTENT);
2860                goto out;
2861            }
2862            if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
2863                goto out;
2864            if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2865                if (allocated > max_blocks)
2866                    allocated = max_blocks;
2867                /*
2868                 * We have blocks reserved already. We
2869                 * return allocated blocks so that delalloc
2870                 * won't do block reservation for us. But
2871                 * the buffer head will be unmapped so that
2872                 * a read from the block returns 0s.
2873                 */
2874                set_buffer_unwritten(bh_result);
2875                bh_result->b_bdev = inode->i_sb->s_bdev;
2876                bh_result->b_blocknr = newblock;
2877                goto out2;
2878            }
2879
2880            ret = ext4_ext_convert_to_initialized(handle, inode,
2881                                path, iblock,
2882                                max_blocks);
2883            if (ret <= 0) {
2884                err = ret;
2885                goto out2;
2886            } else
2887                allocated = ret;
2888            goto outnew;
2889        }
2890    }
2891
2892    /*
2893     * requested block isn't allocated yet;
2894     * we couldn't try to create block if create flag is zero
2895     */
2896    if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
2897        /*
2898         * put just found gap into cache to speed up
2899         * subsequent requests
2900         */
2901        ext4_ext_put_gap_in_cache(inode, path, iblock);
2902        goto out2;
2903    }
2904    /*
2905     * Okay, we need to do block allocation.
2906     */
2907
2908    /* find neighbour allocated blocks */
2909    ar.lleft = iblock;
2910    err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2911    if (err)
2912        goto out2;
2913    ar.lright = iblock;
2914    err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2915    if (err)
2916        goto out2;
2917
2918    /*
2919     * See if request is beyond maximum number of blocks we can have in
2920     * a single extent. For an initialized extent this limit is
2921     * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2922     * EXT_UNINIT_MAX_LEN.
2923     */
2924    if (max_blocks > EXT_INIT_MAX_LEN &&
2925        !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
2926        max_blocks = EXT_INIT_MAX_LEN;
2927    else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2928         (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
2929        max_blocks = EXT_UNINIT_MAX_LEN;
2930
2931    /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2932    newex.ee_block = cpu_to_le32(iblock);
2933    newex.ee_len = cpu_to_le16(max_blocks);
2934    err = ext4_ext_check_overlap(inode, &newex, path);
2935    if (err)
2936        allocated = ext4_ext_get_actual_len(&newex);
2937    else
2938        allocated = max_blocks;
2939
2940    /* allocate new block */
2941    ar.inode = inode;
2942    ar.goal = ext4_ext_find_goal(inode, path, iblock);
2943    ar.logical = iblock;
2944    ar.len = allocated;
2945    if (S_ISREG(inode->i_mode))
2946        ar.flags = EXT4_MB_HINT_DATA;
2947    else
2948        /* disable in-core preallocation for non-regular files */
2949        ar.flags = 0;
2950    newblock = ext4_mb_new_blocks(handle, &ar, &err);
2951    if (!newblock)
2952        goto out2;
2953    ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2954          ar.goal, newblock, allocated);
2955
2956    /* try to insert new extent into found leaf and return */
2957    ext4_ext_store_pblock(&newex, newblock);
2958    newex.ee_len = cpu_to_le16(ar.len);
2959    if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) /* Mark uninitialized */
2960        ext4_ext_mark_uninitialized(&newex);
2961    err = ext4_ext_insert_extent(handle, inode, path, &newex);
2962    if (err) {
2963        /* free data blocks we just allocated */
2964        /* not a good idea to call discard here directly,
2965         * but otherwise we'd need to call it every free() */
2966        ext4_discard_preallocations(inode);
2967        ext4_free_blocks(handle, inode, ext_pblock(&newex),
2968                    ext4_ext_get_actual_len(&newex), 0);
2969        goto out2;
2970    }
2971
2972    /* previous routine could use block we allocated */
2973    newblock = ext_pblock(&newex);
2974    allocated = ext4_ext_get_actual_len(&newex);
2975outnew:
2976    set_buffer_new(bh_result);
2977
2978    /* Cache only when it is _not_ an uninitialized extent */
2979    if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
2980        ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2981                        EXT4_EXT_CACHE_EXTENT);
2982out:
2983    if (allocated > max_blocks)
2984        allocated = max_blocks;
2985    ext4_ext_show_leaf(inode, path);
2986    set_buffer_mapped(bh_result);
2987    bh_result->b_bdev = inode->i_sb->s_bdev;
2988    bh_result->b_blocknr = newblock;
2989out2:
2990    if (path) {
2991        ext4_ext_drop_refs(path);
2992        kfree(path);
2993    }
2994    return err ? err : allocated;
2995}
2996
2997void ext4_ext_truncate(struct inode *inode)
2998{
2999    struct address_space *mapping = inode->i_mapping;
3000    struct super_block *sb = inode->i_sb;
3001    ext4_lblk_t last_block;
3002    handle_t *handle;
3003    int err = 0;
3004
3005    /*
3006     * probably first extent we're gonna free will be last in block
3007     */
3008    err = ext4_writepage_trans_blocks(inode);
3009    handle = ext4_journal_start(inode, err);
3010    if (IS_ERR(handle))
3011        return;
3012
3013    if (inode->i_size & (sb->s_blocksize - 1))
3014        ext4_block_truncate_page(handle, mapping, inode->i_size);
3015
3016    if (ext4_orphan_add(handle, inode))
3017        goto out_stop;
3018
3019    down_write(&EXT4_I(inode)->i_data_sem);
3020    ext4_ext_invalidate_cache(inode);
3021
3022    ext4_discard_preallocations(inode);
3023
3024    /*
3025     * TODO: optimization is possible here.
3026     * Probably we need not scan at all,
3027     * because page truncation is enough.
3028     */
3029
3030    /* we have to know where to truncate from in crash case */
3031    EXT4_I(inode)->i_disksize = inode->i_size;
3032    ext4_mark_inode_dirty(handle, inode);
3033
3034    last_block = (inode->i_size + sb->s_blocksize - 1)
3035            >> EXT4_BLOCK_SIZE_BITS(sb);
3036    err = ext4_ext_remove_space(inode, last_block);
3037
3038    /* In a multi-transaction truncate, we only make the final
3039     * transaction synchronous.
3040     */
3041    if (IS_SYNC(inode))
3042        ext4_handle_sync(handle);
3043
3044out_stop:
3045    up_write(&EXT4_I(inode)->i_data_sem);
3046    /*
3047     * If this was a simple ftruncate() and the file will remain alive,
3048     * then we need to clear up the orphan record which we created above.
3049     * However, if this was a real unlink then we were called by
3050     * ext4_delete_inode(), and we allow that function to clean up the
3051     * orphan info for us.
3052     */
3053    if (inode->i_nlink)
3054        ext4_orphan_del(handle, inode);
3055
3056    inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3057    ext4_mark_inode_dirty(handle, inode);
3058    ext4_journal_stop(handle);
3059}
3060
3061static void ext4_falloc_update_inode(struct inode *inode,
3062                int mode, loff_t new_size, int update_ctime)
3063{
3064    struct timespec now;
3065
3066    if (update_ctime) {
3067        now = current_fs_time(inode->i_sb);
3068        if (!timespec_equal(&inode->i_ctime, &now))
3069            inode->i_ctime = now;
3070    }
3071    /*
3072     * Update only when preallocation was requested beyond
3073     * the file size.
3074     */
3075    if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3076        if (new_size > i_size_read(inode))
3077            i_size_write(inode, new_size);
3078        if (new_size > EXT4_I(inode)->i_disksize)
3079            ext4_update_i_disksize(inode, new_size);
3080    }
3081
3082}
3083
3084/*
3085 * preallocate space for a file. This implements ext4's fallocate inode
3086 * operation, which gets called from sys_fallocate system call.
3087 * For block-mapped files, posix_fallocate should fall back to the method
3088 * of writing zeroes to the required new blocks (the same behavior which is
3089 * expected for file systems which do not support fallocate() system call).
3090 */
3091long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3092{
3093    handle_t *handle;
3094    ext4_lblk_t block;
3095    loff_t new_size;
3096    unsigned int max_blocks;
3097    int ret = 0;
3098    int ret2 = 0;
3099    int retries = 0;
3100    struct buffer_head map_bh;
3101    unsigned int credits, blkbits = inode->i_blkbits;
3102
3103    /*
3104     * currently supporting (pre)allocate mode for extent-based
3105     * files _only_
3106     */
3107    if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3108        return -EOPNOTSUPP;
3109
3110    /* preallocation to directories is currently not supported */
3111    if (S_ISDIR(inode->i_mode))
3112        return -ENODEV;
3113
3114    block = offset >> blkbits;
3115    /*
3116     * We can't just convert len to max_blocks because
3117     * If blocksize = 4096 offset = 3072 and len = 2048
3118     */
3119    max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3120                            - block;
3121    /*
3122     * credits to insert 1 extent into extent tree
3123     */
3124    credits = ext4_chunk_trans_blocks(inode, max_blocks);
3125    mutex_lock(&inode->i_mutex);
3126retry:
3127    while (ret >= 0 && ret < max_blocks) {
3128        block = block + ret;
3129        max_blocks = max_blocks - ret;
3130        handle = ext4_journal_start(inode, credits);
3131        if (IS_ERR(handle)) {
3132            ret = PTR_ERR(handle);
3133            break;
3134        }
3135        map_bh.b_state = 0;
3136        ret = ext4_get_blocks(handle, inode, block,
3137                      max_blocks, &map_bh,
3138                      EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3139        if (ret <= 0) {
3140#ifdef EXT4FS_DEBUG
3141            WARN_ON(ret <= 0);
3142            printk(KERN_ERR "%s: ext4_ext_get_blocks "
3143                    "returned error inode#%lu, block=%u, "
3144                    "max_blocks=%u", __func__,
3145                    inode->i_ino, block, max_blocks);
3146#endif
3147            ext4_mark_inode_dirty(handle, inode);
3148            ret2 = ext4_journal_stop(handle);
3149            break;
3150        }
3151        if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3152                        blkbits) >> blkbits))
3153            new_size = offset + len;
3154        else
3155            new_size = (block + ret) << blkbits;
3156
3157        ext4_falloc_update_inode(inode, mode, new_size,
3158                        buffer_new(&map_bh));
3159        ext4_mark_inode_dirty(handle, inode);
3160        ret2 = ext4_journal_stop(handle);
3161        if (ret2)
3162            break;
3163    }
3164    if (ret == -ENOSPC &&
3165            ext4_should_retry_alloc(inode->i_sb, &retries)) {
3166        ret = 0;
3167        goto retry;
3168    }
3169    mutex_unlock(&inode->i_mutex);
3170    return ret > 0 ? ret2 : ret;
3171}
3172
3173/*
3174 * Callback function called for each extent to gather FIEMAP information.
3175 */
3176static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3177               struct ext4_ext_cache *newex, struct ext4_extent *ex,
3178               void *data)
3179{
3180    struct fiemap_extent_info *fieinfo = data;
3181    unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3182    __u64 logical;
3183    __u64 physical;
3184    __u64 length;
3185    __u32 flags = 0;
3186    int error;
3187
3188    logical = (__u64)newex->ec_block << blksize_bits;
3189
3190    if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3191        pgoff_t offset;
3192        struct page *page;
3193        struct buffer_head *bh = NULL;
3194
3195        offset = logical >> PAGE_SHIFT;
3196        page = find_get_page(inode->i_mapping, offset);
3197        if (!page || !page_has_buffers(page))
3198            return EXT_CONTINUE;
3199
3200        bh = page_buffers(page);
3201
3202        if (!bh)
3203            return EXT_CONTINUE;
3204
3205        if (buffer_delay(bh)) {
3206            flags |= FIEMAP_EXTENT_DELALLOC;
3207            page_cache_release(page);
3208        } else {
3209            page_cache_release(page);
3210            return EXT_CONTINUE;
3211        }
3212    }
3213
3214    physical = (__u64)newex->ec_start << blksize_bits;
3215    length = (__u64)newex->ec_len << blksize_bits;
3216
3217    if (ex && ext4_ext_is_uninitialized(ex))
3218        flags |= FIEMAP_EXTENT_UNWRITTEN;
3219
3220    /*
3221     * If this extent reaches EXT_MAX_BLOCK, it must be last.
3222     *
3223     * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3224     * this also indicates no more allocated blocks.
3225     *
3226     * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3227     */
3228    if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3229        newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3230        loff_t size = i_size_read(inode);
3231        loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3232
3233        flags |= FIEMAP_EXTENT_LAST;
3234        if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3235            logical+length > size)
3236            length = (size - logical + bs - 1) & ~(bs-1);
3237    }
3238
3239    error = fiemap_fill_next_extent(fieinfo, logical, physical,
3240                    length, flags);
3241    if (error < 0)
3242        return error;
3243    if (error == 1)
3244        return EXT_BREAK;
3245
3246    return EXT_CONTINUE;
3247}
3248
3249/* fiemap flags we can handle specified here */
3250#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3251
3252static int ext4_xattr_fiemap(struct inode *inode,
3253                struct fiemap_extent_info *fieinfo)
3254{
3255    __u64 physical = 0;
3256    __u64 length;
3257    __u32 flags = FIEMAP_EXTENT_LAST;
3258    int blockbits = inode->i_sb->s_blocksize_bits;
3259    int error = 0;
3260
3261    /* in-inode? */
3262    if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
3263        struct ext4_iloc iloc;
3264        int offset; /* offset of xattr in inode */
3265
3266        error = ext4_get_inode_loc(inode, &iloc);
3267        if (error)
3268            return error;
3269        physical = iloc.bh->b_blocknr << blockbits;
3270        offset = EXT4_GOOD_OLD_INODE_SIZE +
3271                EXT4_I(inode)->i_extra_isize;
3272        physical += offset;
3273        length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3274        flags |= FIEMAP_EXTENT_DATA_INLINE;
3275    } else { /* external block */
3276        physical = EXT4_I(inode)->i_file_acl << blockbits;
3277        length = inode->i_sb->s_blocksize;
3278    }
3279
3280    if (physical)
3281        error = fiemap_fill_next_extent(fieinfo, 0, physical,
3282                        length, flags);
3283    return (error < 0 ? error : 0);
3284}
3285
3286int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3287        __u64 start, __u64 len)
3288{
3289    ext4_lblk_t start_blk;
3290    ext4_lblk_t len_blks;
3291    int error = 0;
3292
3293    /* fallback to generic here if not in extents fmt */
3294    if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3295        return generic_block_fiemap(inode, fieinfo, start, len,
3296            ext4_get_block);
3297
3298    if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3299        return -EBADR;
3300
3301    if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3302        error = ext4_xattr_fiemap(inode, fieinfo);
3303    } else {
3304        start_blk = start >> inode->i_sb->s_blocksize_bits;
3305        len_blks = len >> inode->i_sb->s_blocksize_bits;
3306
3307        /*
3308         * Walk the extent tree gathering extent information.
3309         * ext4_ext_fiemap_cb will push extents back to user.
3310         */
3311        down_read(&EXT4_I(inode)->i_data_sem);
3312        error = ext4_ext_walk_space(inode, start_blk, len_blks,
3313                      ext4_ext_fiemap_cb, fieinfo);
3314        up_read(&EXT4_I(inode)->i_data_sem);
3315    }
3316
3317    return error;
3318}
3319
3320

Archive Download this file



interactive