Root/fs/btrfs/ctree.c

1/*
2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/sched.h>
20#include <linux/slab.h>
21#include "ctree.h"
22#include "disk-io.h"
23#include "transaction.h"
24#include "print-tree.h"
25#include "locking.h"
26
27static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
28              *root, struct btrfs_path *path, int level);
29static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
30              *root, struct btrfs_key *ins_key,
31              struct btrfs_path *path, int data_size, int extend);
32static int push_node_left(struct btrfs_trans_handle *trans,
33              struct btrfs_root *root, struct extent_buffer *dst,
34              struct extent_buffer *src, int empty);
35static int balance_node_right(struct btrfs_trans_handle *trans,
36                  struct btrfs_root *root,
37                  struct extent_buffer *dst_buf,
38                  struct extent_buffer *src_buf);
39static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
40           struct btrfs_path *path, int level, int slot);
41static int setup_items_for_insert(struct btrfs_trans_handle *trans,
42            struct btrfs_root *root, struct btrfs_path *path,
43            struct btrfs_key *cpu_key, u32 *data_size,
44            u32 total_data, u32 total_size, int nr);
45
46
47struct btrfs_path *btrfs_alloc_path(void)
48{
49    struct btrfs_path *path;
50    path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
51    if (path)
52        path->reada = 1;
53    return path;
54}
55
56/*
57 * set all locked nodes in the path to blocking locks. This should
58 * be done before scheduling
59 */
60noinline void btrfs_set_path_blocking(struct btrfs_path *p)
61{
62    int i;
63    for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
64        if (p->nodes[i] && p->locks[i])
65            btrfs_set_lock_blocking(p->nodes[i]);
66    }
67}
68
69/*
70 * reset all the locked nodes in the patch to spinning locks.
71 *
72 * held is used to keep lockdep happy, when lockdep is enabled
73 * we set held to a blocking lock before we go around and
74 * retake all the spinlocks in the path. You can safely use NULL
75 * for held
76 */
77noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
78                    struct extent_buffer *held)
79{
80    int i;
81
82#ifdef CONFIG_DEBUG_LOCK_ALLOC
83    /* lockdep really cares that we take all of these spinlocks
84     * in the right order. If any of the locks in the path are not
85     * currently blocking, it is going to complain. So, make really
86     * really sure by forcing the path to blocking before we clear
87     * the path blocking.
88     */
89    if (held)
90        btrfs_set_lock_blocking(held);
91    btrfs_set_path_blocking(p);
92#endif
93
94    for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
95        if (p->nodes[i] && p->locks[i])
96            btrfs_clear_lock_blocking(p->nodes[i]);
97    }
98
99#ifdef CONFIG_DEBUG_LOCK_ALLOC
100    if (held)
101        btrfs_clear_lock_blocking(held);
102#endif
103}
104
105/* this also releases the path */
106void btrfs_free_path(struct btrfs_path *p)
107{
108    btrfs_release_path(NULL, p);
109    kmem_cache_free(btrfs_path_cachep, p);
110}
111
112/*
113 * path release drops references on the extent buffers in the path
114 * and it drops any locks held by this path
115 *
116 * It is safe to call this on paths that no locks or extent buffers held.
117 */
118noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
119{
120    int i;
121
122    for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
123        p->slots[i] = 0;
124        if (!p->nodes[i])
125            continue;
126        if (p->locks[i]) {
127            btrfs_tree_unlock(p->nodes[i]);
128            p->locks[i] = 0;
129        }
130        free_extent_buffer(p->nodes[i]);
131        p->nodes[i] = NULL;
132    }
133}
134
135/*
136 * safely gets a reference on the root node of a tree. A lock
137 * is not taken, so a concurrent writer may put a different node
138 * at the root of the tree. See btrfs_lock_root_node for the
139 * looping required.
140 *
141 * The extent buffer returned by this has a reference taken, so
142 * it won't disappear. It may stop being the root of the tree
143 * at any time because there are no locks held.
144 */
145struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
146{
147    struct extent_buffer *eb;
148    spin_lock(&root->node_lock);
149    eb = root->node;
150    extent_buffer_get(eb);
151    spin_unlock(&root->node_lock);
152    return eb;
153}
154
155/* loop around taking references on and locking the root node of the
156 * tree until you end up with a lock on the root. A locked buffer
157 * is returned, with a reference held.
158 */
159struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
160{
161    struct extent_buffer *eb;
162
163    while (1) {
164        eb = btrfs_root_node(root);
165        btrfs_tree_lock(eb);
166
167        spin_lock(&root->node_lock);
168        if (eb == root->node) {
169            spin_unlock(&root->node_lock);
170            break;
171        }
172        spin_unlock(&root->node_lock);
173
174        btrfs_tree_unlock(eb);
175        free_extent_buffer(eb);
176    }
177    return eb;
178}
179
180/* cowonly root (everything not a reference counted cow subvolume), just get
181 * put onto a simple dirty list. transaction.c walks this to make sure they
182 * get properly updated on disk.
183 */
184static void add_root_to_dirty_list(struct btrfs_root *root)
185{
186    if (root->track_dirty && list_empty(&root->dirty_list)) {
187        list_add(&root->dirty_list,
188             &root->fs_info->dirty_cowonly_roots);
189    }
190}
191
192/*
193 * used by snapshot creation to make a copy of a root for a tree with
194 * a given objectid. The buffer with the new root node is returned in
195 * cow_ret, and this func returns zero on success or a negative error code.
196 */
197int btrfs_copy_root(struct btrfs_trans_handle *trans,
198              struct btrfs_root *root,
199              struct extent_buffer *buf,
200              struct extent_buffer **cow_ret, u64 new_root_objectid)
201{
202    struct extent_buffer *cow;
203    u32 nritems;
204    int ret = 0;
205    int level;
206    struct btrfs_disk_key disk_key;
207
208    WARN_ON(root->ref_cows && trans->transid !=
209        root->fs_info->running_transaction->transid);
210    WARN_ON(root->ref_cows && trans->transid != root->last_trans);
211
212    level = btrfs_header_level(buf);
213    nritems = btrfs_header_nritems(buf);
214    if (level == 0)
215        btrfs_item_key(buf, &disk_key, 0);
216    else
217        btrfs_node_key(buf, &disk_key, 0);
218
219    cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
220                     new_root_objectid, &disk_key, level,
221                     buf->start, 0);
222    if (IS_ERR(cow))
223        return PTR_ERR(cow);
224
225    copy_extent_buffer(cow, buf, 0, 0, cow->len);
226    btrfs_set_header_bytenr(cow, cow->start);
227    btrfs_set_header_generation(cow, trans->transid);
228    btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
229    btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
230                     BTRFS_HEADER_FLAG_RELOC);
231    if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
232        btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
233    else
234        btrfs_set_header_owner(cow, new_root_objectid);
235
236    write_extent_buffer(cow, root->fs_info->fsid,
237                (unsigned long)btrfs_header_fsid(cow),
238                BTRFS_FSID_SIZE);
239
240    WARN_ON(btrfs_header_generation(buf) > trans->transid);
241    if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
242        ret = btrfs_inc_ref(trans, root, cow, 1);
243    else
244        ret = btrfs_inc_ref(trans, root, cow, 0);
245
246    if (ret)
247        return ret;
248
249    btrfs_mark_buffer_dirty(cow);
250    *cow_ret = cow;
251    return 0;
252}
253
254/*
255 * check if the tree block can be shared by multiple trees
256 */
257int btrfs_block_can_be_shared(struct btrfs_root *root,
258                  struct extent_buffer *buf)
259{
260    /*
261     * Tree blocks not in refernece counted trees and tree roots
262     * are never shared. If a block was allocated after the last
263     * snapshot and the block was not allocated by tree relocation,
264     * we know the block is not shared.
265     */
266    if (root->ref_cows &&
267        buf != root->node && buf != root->commit_root &&
268        (btrfs_header_generation(buf) <=
269         btrfs_root_last_snapshot(&root->root_item) ||
270         btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
271        return 1;
272#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
273    if (root->ref_cows &&
274        btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
275        return 1;
276#endif
277    return 0;
278}
279
280static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
281                       struct btrfs_root *root,
282                       struct extent_buffer *buf,
283                       struct extent_buffer *cow)
284{
285    u64 refs;
286    u64 owner;
287    u64 flags;
288    u64 new_flags = 0;
289    int ret;
290
291    /*
292     * Backrefs update rules:
293     *
294     * Always use full backrefs for extent pointers in tree block
295     * allocated by tree relocation.
296     *
297     * If a shared tree block is no longer referenced by its owner
298     * tree (btrfs_header_owner(buf) == root->root_key.objectid),
299     * use full backrefs for extent pointers in tree block.
300     *
301     * If a tree block is been relocating
302     * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
303     * use full backrefs for extent pointers in tree block.
304     * The reason for this is some operations (such as drop tree)
305     * are only allowed for blocks use full backrefs.
306     */
307
308    if (btrfs_block_can_be_shared(root, buf)) {
309        ret = btrfs_lookup_extent_info(trans, root, buf->start,
310                           buf->len, &refs, &flags);
311        BUG_ON(ret);
312        BUG_ON(refs == 0);
313    } else {
314        refs = 1;
315        if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
316            btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
317            flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
318        else
319            flags = 0;
320    }
321
322    owner = btrfs_header_owner(buf);
323    BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
324           !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
325
326    if (refs > 1) {
327        if ((owner == root->root_key.objectid ||
328             root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
329            !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
330            ret = btrfs_inc_ref(trans, root, buf, 1);
331            BUG_ON(ret);
332
333            if (root->root_key.objectid ==
334                BTRFS_TREE_RELOC_OBJECTID) {
335                ret = btrfs_dec_ref(trans, root, buf, 0);
336                BUG_ON(ret);
337                ret = btrfs_inc_ref(trans, root, cow, 1);
338                BUG_ON(ret);
339            }
340            new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
341        } else {
342
343            if (root->root_key.objectid ==
344                BTRFS_TREE_RELOC_OBJECTID)
345                ret = btrfs_inc_ref(trans, root, cow, 1);
346            else
347                ret = btrfs_inc_ref(trans, root, cow, 0);
348            BUG_ON(ret);
349        }
350        if (new_flags != 0) {
351            ret = btrfs_set_disk_extent_flags(trans, root,
352                              buf->start,
353                              buf->len,
354                              new_flags, 0);
355            BUG_ON(ret);
356        }
357    } else {
358        if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
359            if (root->root_key.objectid ==
360                BTRFS_TREE_RELOC_OBJECTID)
361                ret = btrfs_inc_ref(trans, root, cow, 1);
362            else
363                ret = btrfs_inc_ref(trans, root, cow, 0);
364            BUG_ON(ret);
365            ret = btrfs_dec_ref(trans, root, buf, 1);
366            BUG_ON(ret);
367        }
368        clean_tree_block(trans, root, buf);
369    }
370    return 0;
371}
372
373/*
374 * does the dirty work in cow of a single block. The parent block (if
375 * supplied) is updated to point to the new cow copy. The new buffer is marked
376 * dirty and returned locked. If you modify the block it needs to be marked
377 * dirty again.
378 *
379 * search_start -- an allocation hint for the new block
380 *
381 * empty_size -- a hint that you plan on doing more cow. This is the size in
382 * bytes the allocator should try to find free next to the block it returns.
383 * This is just a hint and may be ignored by the allocator.
384 */
385static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
386                 struct btrfs_root *root,
387                 struct extent_buffer *buf,
388                 struct extent_buffer *parent, int parent_slot,
389                 struct extent_buffer **cow_ret,
390                 u64 search_start, u64 empty_size)
391{
392    struct btrfs_disk_key disk_key;
393    struct extent_buffer *cow;
394    int level;
395    int unlock_orig = 0;
396    u64 parent_start;
397
398    if (*cow_ret == buf)
399        unlock_orig = 1;
400
401    btrfs_assert_tree_locked(buf);
402
403    WARN_ON(root->ref_cows && trans->transid !=
404        root->fs_info->running_transaction->transid);
405    WARN_ON(root->ref_cows && trans->transid != root->last_trans);
406
407    level = btrfs_header_level(buf);
408
409    if (level == 0)
410        btrfs_item_key(buf, &disk_key, 0);
411    else
412        btrfs_node_key(buf, &disk_key, 0);
413
414    if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
415        if (parent)
416            parent_start = parent->start;
417        else
418            parent_start = 0;
419    } else
420        parent_start = 0;
421
422    cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
423                     root->root_key.objectid, &disk_key,
424                     level, search_start, empty_size);
425    if (IS_ERR(cow))
426        return PTR_ERR(cow);
427
428    /* cow is set to blocking by btrfs_init_new_buffer */
429
430    copy_extent_buffer(cow, buf, 0, 0, cow->len);
431    btrfs_set_header_bytenr(cow, cow->start);
432    btrfs_set_header_generation(cow, trans->transid);
433    btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
434    btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
435                     BTRFS_HEADER_FLAG_RELOC);
436    if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
437        btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
438    else
439        btrfs_set_header_owner(cow, root->root_key.objectid);
440
441    write_extent_buffer(cow, root->fs_info->fsid,
442                (unsigned long)btrfs_header_fsid(cow),
443                BTRFS_FSID_SIZE);
444
445    update_ref_for_cow(trans, root, buf, cow);
446
447    if (buf == root->node) {
448        WARN_ON(parent && parent != buf);
449        if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
450            btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
451            parent_start = buf->start;
452        else
453            parent_start = 0;
454
455        spin_lock(&root->node_lock);
456        root->node = cow;
457        extent_buffer_get(cow);
458        spin_unlock(&root->node_lock);
459
460        btrfs_free_tree_block(trans, root, buf->start, buf->len,
461                parent_start, root->root_key.objectid, level);
462        free_extent_buffer(buf);
463        add_root_to_dirty_list(root);
464    } else {
465        if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
466            parent_start = parent->start;
467        else
468            parent_start = 0;
469
470        WARN_ON(trans->transid != btrfs_header_generation(parent));
471        btrfs_set_node_blockptr(parent, parent_slot,
472                    cow->start);
473        btrfs_set_node_ptr_generation(parent, parent_slot,
474                          trans->transid);
475        btrfs_mark_buffer_dirty(parent);
476        btrfs_free_tree_block(trans, root, buf->start, buf->len,
477                parent_start, root->root_key.objectid, level);
478    }
479    if (unlock_orig)
480        btrfs_tree_unlock(buf);
481    free_extent_buffer(buf);
482    btrfs_mark_buffer_dirty(cow);
483    *cow_ret = cow;
484    return 0;
485}
486
487static inline int should_cow_block(struct btrfs_trans_handle *trans,
488                   struct btrfs_root *root,
489                   struct extent_buffer *buf)
490{
491    if (btrfs_header_generation(buf) == trans->transid &&
492        !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
493        !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
494          btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
495        return 0;
496    return 1;
497}
498
499/*
500 * cows a single block, see __btrfs_cow_block for the real work.
501 * This version of it has extra checks so that a block isn't cow'd more than
502 * once per transaction, as long as it hasn't been written yet
503 */
504noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
505            struct btrfs_root *root, struct extent_buffer *buf,
506            struct extent_buffer *parent, int parent_slot,
507            struct extent_buffer **cow_ret)
508{
509    u64 search_start;
510    int ret;
511
512    if (trans->transaction != root->fs_info->running_transaction) {
513        printk(KERN_CRIT "trans %llu running %llu\n",
514               (unsigned long long)trans->transid,
515               (unsigned long long)
516               root->fs_info->running_transaction->transid);
517        WARN_ON(1);
518    }
519    if (trans->transid != root->fs_info->generation) {
520        printk(KERN_CRIT "trans %llu running %llu\n",
521               (unsigned long long)trans->transid,
522               (unsigned long long)root->fs_info->generation);
523        WARN_ON(1);
524    }
525
526    if (!should_cow_block(trans, root, buf)) {
527        *cow_ret = buf;
528        return 0;
529    }
530
531    search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
532
533    if (parent)
534        btrfs_set_lock_blocking(parent);
535    btrfs_set_lock_blocking(buf);
536
537    ret = __btrfs_cow_block(trans, root, buf, parent,
538                 parent_slot, cow_ret, search_start, 0);
539    return ret;
540}
541
542/*
543 * helper function for defrag to decide if two blocks pointed to by a
544 * node are actually close by
545 */
546static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
547{
548    if (blocknr < other && other - (blocknr + blocksize) < 32768)
549        return 1;
550    if (blocknr > other && blocknr - (other + blocksize) < 32768)
551        return 1;
552    return 0;
553}
554
555/*
556 * compare two keys in a memcmp fashion
557 */
558static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
559{
560    struct btrfs_key k1;
561
562    btrfs_disk_key_to_cpu(&k1, disk);
563
564    return btrfs_comp_cpu_keys(&k1, k2);
565}
566
567/*
568 * same as comp_keys only with two btrfs_key's
569 */
570int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
571{
572    if (k1->objectid > k2->objectid)
573        return 1;
574    if (k1->objectid < k2->objectid)
575        return -1;
576    if (k1->type > k2->type)
577        return 1;
578    if (k1->type < k2->type)
579        return -1;
580    if (k1->offset > k2->offset)
581        return 1;
582    if (k1->offset < k2->offset)
583        return -1;
584    return 0;
585}
586
587/*
588 * this is used by the defrag code to go through all the
589 * leaves pointed to by a node and reallocate them so that
590 * disk order is close to key order
591 */
592int btrfs_realloc_node(struct btrfs_trans_handle *trans,
593               struct btrfs_root *root, struct extent_buffer *parent,
594               int start_slot, int cache_only, u64 *last_ret,
595               struct btrfs_key *progress)
596{
597    struct extent_buffer *cur;
598    u64 blocknr;
599    u64 gen;
600    u64 search_start = *last_ret;
601    u64 last_block = 0;
602    u64 other;
603    u32 parent_nritems;
604    int end_slot;
605    int i;
606    int err = 0;
607    int parent_level;
608    int uptodate;
609    u32 blocksize;
610    int progress_passed = 0;
611    struct btrfs_disk_key disk_key;
612
613    parent_level = btrfs_header_level(parent);
614    if (cache_only && parent_level != 1)
615        return 0;
616
617    if (trans->transaction != root->fs_info->running_transaction)
618        WARN_ON(1);
619    if (trans->transid != root->fs_info->generation)
620        WARN_ON(1);
621
622    parent_nritems = btrfs_header_nritems(parent);
623    blocksize = btrfs_level_size(root, parent_level - 1);
624    end_slot = parent_nritems;
625
626    if (parent_nritems == 1)
627        return 0;
628
629    btrfs_set_lock_blocking(parent);
630
631    for (i = start_slot; i < end_slot; i++) {
632        int close = 1;
633
634        if (!parent->map_token) {
635            map_extent_buffer(parent,
636                    btrfs_node_key_ptr_offset(i),
637                    sizeof(struct btrfs_key_ptr),
638                    &parent->map_token, &parent->kaddr,
639                    &parent->map_start, &parent->map_len,
640                    KM_USER1);
641        }
642        btrfs_node_key(parent, &disk_key, i);
643        if (!progress_passed && comp_keys(&disk_key, progress) < 0)
644            continue;
645
646        progress_passed = 1;
647        blocknr = btrfs_node_blockptr(parent, i);
648        gen = btrfs_node_ptr_generation(parent, i);
649        if (last_block == 0)
650            last_block = blocknr;
651
652        if (i > 0) {
653            other = btrfs_node_blockptr(parent, i - 1);
654            close = close_blocks(blocknr, other, blocksize);
655        }
656        if (!close && i < end_slot - 2) {
657            other = btrfs_node_blockptr(parent, i + 1);
658            close = close_blocks(blocknr, other, blocksize);
659        }
660        if (close) {
661            last_block = blocknr;
662            continue;
663        }
664        if (parent->map_token) {
665            unmap_extent_buffer(parent, parent->map_token,
666                        KM_USER1);
667            parent->map_token = NULL;
668        }
669
670        cur = btrfs_find_tree_block(root, blocknr, blocksize);
671        if (cur)
672            uptodate = btrfs_buffer_uptodate(cur, gen);
673        else
674            uptodate = 0;
675        if (!cur || !uptodate) {
676            if (cache_only) {
677                free_extent_buffer(cur);
678                continue;
679            }
680            if (!cur) {
681                cur = read_tree_block(root, blocknr,
682                             blocksize, gen);
683            } else if (!uptodate) {
684                btrfs_read_buffer(cur, gen);
685            }
686        }
687        if (search_start == 0)
688            search_start = last_block;
689
690        btrfs_tree_lock(cur);
691        btrfs_set_lock_blocking(cur);
692        err = __btrfs_cow_block(trans, root, cur, parent, i,
693                    &cur, search_start,
694                    min(16 * blocksize,
695                        (end_slot - i) * blocksize));
696        if (err) {
697            btrfs_tree_unlock(cur);
698            free_extent_buffer(cur);
699            break;
700        }
701        search_start = cur->start;
702        last_block = cur->start;
703        *last_ret = search_start;
704        btrfs_tree_unlock(cur);
705        free_extent_buffer(cur);
706    }
707    if (parent->map_token) {
708        unmap_extent_buffer(parent, parent->map_token,
709                    KM_USER1);
710        parent->map_token = NULL;
711    }
712    return err;
713}
714
715/*
716 * The leaf data grows from end-to-front in the node.
717 * this returns the address of the start of the last item,
718 * which is the stop of the leaf data stack
719 */
720static inline unsigned int leaf_data_end(struct btrfs_root *root,
721                     struct extent_buffer *leaf)
722{
723    u32 nr = btrfs_header_nritems(leaf);
724    if (nr == 0)
725        return BTRFS_LEAF_DATA_SIZE(root);
726    return btrfs_item_offset_nr(leaf, nr - 1);
727}
728
729/*
730 * extra debugging checks to make sure all the items in a key are
731 * well formed and in the proper order
732 */
733static int check_node(struct btrfs_root *root, struct btrfs_path *path,
734              int level)
735{
736    struct extent_buffer *parent = NULL;
737    struct extent_buffer *node = path->nodes[level];
738    struct btrfs_disk_key parent_key;
739    struct btrfs_disk_key node_key;
740    int parent_slot;
741    int slot;
742    struct btrfs_key cpukey;
743    u32 nritems = btrfs_header_nritems(node);
744
745    if (path->nodes[level + 1])
746        parent = path->nodes[level + 1];
747
748    slot = path->slots[level];
749    BUG_ON(nritems == 0);
750    if (parent) {
751        parent_slot = path->slots[level + 1];
752        btrfs_node_key(parent, &parent_key, parent_slot);
753        btrfs_node_key(node, &node_key, 0);
754        BUG_ON(memcmp(&parent_key, &node_key,
755                  sizeof(struct btrfs_disk_key)));
756        BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
757               btrfs_header_bytenr(node));
758    }
759    BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
760    if (slot != 0) {
761        btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
762        btrfs_node_key(node, &node_key, slot);
763        BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
764    }
765    if (slot < nritems - 1) {
766        btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
767        btrfs_node_key(node, &node_key, slot);
768        BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
769    }
770    return 0;
771}
772
773/*
774 * extra checking to make sure all the items in a leaf are
775 * well formed and in the proper order
776 */
777static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
778              int level)
779{
780    struct extent_buffer *leaf = path->nodes[level];
781    struct extent_buffer *parent = NULL;
782    int parent_slot;
783    struct btrfs_key cpukey;
784    struct btrfs_disk_key parent_key;
785    struct btrfs_disk_key leaf_key;
786    int slot = path->slots[0];
787
788    u32 nritems = btrfs_header_nritems(leaf);
789
790    if (path->nodes[level + 1])
791        parent = path->nodes[level + 1];
792
793    if (nritems == 0)
794        return 0;
795
796    if (parent) {
797        parent_slot = path->slots[level + 1];
798        btrfs_node_key(parent, &parent_key, parent_slot);
799        btrfs_item_key(leaf, &leaf_key, 0);
800
801        BUG_ON(memcmp(&parent_key, &leaf_key,
802               sizeof(struct btrfs_disk_key)));
803        BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
804               btrfs_header_bytenr(leaf));
805    }
806    if (slot != 0 && slot < nritems - 1) {
807        btrfs_item_key(leaf, &leaf_key, slot);
808        btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
809        if (comp_keys(&leaf_key, &cpukey) <= 0) {
810            btrfs_print_leaf(root, leaf);
811            printk(KERN_CRIT "slot %d offset bad key\n", slot);
812            BUG_ON(1);
813        }
814        if (btrfs_item_offset_nr(leaf, slot - 1) !=
815               btrfs_item_end_nr(leaf, slot)) {
816            btrfs_print_leaf(root, leaf);
817            printk(KERN_CRIT "slot %d offset bad\n", slot);
818            BUG_ON(1);
819        }
820    }
821    if (slot < nritems - 1) {
822        btrfs_item_key(leaf, &leaf_key, slot);
823        btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
824        BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
825        if (btrfs_item_offset_nr(leaf, slot) !=
826            btrfs_item_end_nr(leaf, slot + 1)) {
827            btrfs_print_leaf(root, leaf);
828            printk(KERN_CRIT "slot %d offset bad\n", slot);
829            BUG_ON(1);
830        }
831    }
832    BUG_ON(btrfs_item_offset_nr(leaf, 0) +
833           btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
834    return 0;
835}
836
837static noinline int check_block(struct btrfs_root *root,
838                struct btrfs_path *path, int level)
839{
840    return 0;
841    if (level == 0)
842        return check_leaf(root, path, level);
843    return check_node(root, path, level);
844}
845
846/*
847 * search for key in the extent_buffer. The items start at offset p,
848 * and they are item_size apart. There are 'max' items in p.
849 *
850 * the slot in the array is returned via slot, and it points to
851 * the place where you would insert key if it is not found in
852 * the array.
853 *
854 * slot may point to max if the key is bigger than all of the keys
855 */
856static noinline int generic_bin_search(struct extent_buffer *eb,
857                       unsigned long p,
858                       int item_size, struct btrfs_key *key,
859                       int max, int *slot)
860{
861    int low = 0;
862    int high = max;
863    int mid;
864    int ret;
865    struct btrfs_disk_key *tmp = NULL;
866    struct btrfs_disk_key unaligned;
867    unsigned long offset;
868    char *map_token = NULL;
869    char *kaddr = NULL;
870    unsigned long map_start = 0;
871    unsigned long map_len = 0;
872    int err;
873
874    while (low < high) {
875        mid = (low + high) / 2;
876        offset = p + mid * item_size;
877
878        if (!map_token || offset < map_start ||
879            (offset + sizeof(struct btrfs_disk_key)) >
880            map_start + map_len) {
881            if (map_token) {
882                unmap_extent_buffer(eb, map_token, KM_USER0);
883                map_token = NULL;
884            }
885
886            err = map_private_extent_buffer(eb, offset,
887                        sizeof(struct btrfs_disk_key),
888                        &map_token, &kaddr,
889                        &map_start, &map_len, KM_USER0);
890
891            if (!err) {
892                tmp = (struct btrfs_disk_key *)(kaddr + offset -
893                            map_start);
894            } else {
895                read_extent_buffer(eb, &unaligned,
896                           offset, sizeof(unaligned));
897                tmp = &unaligned;
898            }
899
900        } else {
901            tmp = (struct btrfs_disk_key *)(kaddr + offset -
902                            map_start);
903        }
904        ret = comp_keys(tmp, key);
905
906        if (ret < 0)
907            low = mid + 1;
908        else if (ret > 0)
909            high = mid;
910        else {
911            *slot = mid;
912            if (map_token)
913                unmap_extent_buffer(eb, map_token, KM_USER0);
914            return 0;
915        }
916    }
917    *slot = low;
918    if (map_token)
919        unmap_extent_buffer(eb, map_token, KM_USER0);
920    return 1;
921}
922
923/*
924 * simple bin_search frontend that does the right thing for
925 * leaves vs nodes
926 */
927static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
928              int level, int *slot)
929{
930    if (level == 0) {
931        return generic_bin_search(eb,
932                      offsetof(struct btrfs_leaf, items),
933                      sizeof(struct btrfs_item),
934                      key, btrfs_header_nritems(eb),
935                      slot);
936    } else {
937        return generic_bin_search(eb,
938                      offsetof(struct btrfs_node, ptrs),
939                      sizeof(struct btrfs_key_ptr),
940                      key, btrfs_header_nritems(eb),
941                      slot);
942    }
943    return -1;
944}
945
946int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
947             int level, int *slot)
948{
949    return bin_search(eb, key, level, slot);
950}
951
952/* given a node and slot number, this reads the blocks it points to. The
953 * extent buffer is returned with a reference taken (but unlocked).
954 * NULL is returned on error.
955 */
956static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
957                   struct extent_buffer *parent, int slot)
958{
959    int level = btrfs_header_level(parent);
960    if (slot < 0)
961        return NULL;
962    if (slot >= btrfs_header_nritems(parent))
963        return NULL;
964
965    BUG_ON(level == 0);
966
967    return read_tree_block(root, btrfs_node_blockptr(parent, slot),
968               btrfs_level_size(root, level - 1),
969               btrfs_node_ptr_generation(parent, slot));
970}
971
972/*
973 * node level balancing, used to make sure nodes are in proper order for
974 * item deletion. We balance from the top down, so we have to make sure
975 * that a deletion won't leave an node completely empty later on.
976 */
977static noinline int balance_level(struct btrfs_trans_handle *trans,
978             struct btrfs_root *root,
979             struct btrfs_path *path, int level)
980{
981    struct extent_buffer *right = NULL;
982    struct extent_buffer *mid;
983    struct extent_buffer *left = NULL;
984    struct extent_buffer *parent = NULL;
985    int ret = 0;
986    int wret;
987    int pslot;
988    int orig_slot = path->slots[level];
989    int err_on_enospc = 0;
990    u64 orig_ptr;
991
992    if (level == 0)
993        return 0;
994
995    mid = path->nodes[level];
996
997    WARN_ON(!path->locks[level]);
998    WARN_ON(btrfs_header_generation(mid) != trans->transid);
999
1000    orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1001
1002    if (level < BTRFS_MAX_LEVEL - 1)
1003        parent = path->nodes[level + 1];
1004    pslot = path->slots[level + 1];
1005
1006    /*
1007     * deal with the case where there is only one pointer in the root
1008     * by promoting the node below to a root
1009     */
1010    if (!parent) {
1011        struct extent_buffer *child;
1012
1013        if (btrfs_header_nritems(mid) != 1)
1014            return 0;
1015
1016        /* promote the child to a root */
1017        child = read_node_slot(root, mid, 0);
1018        BUG_ON(!child);
1019        btrfs_tree_lock(child);
1020        btrfs_set_lock_blocking(child);
1021        ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1022        BUG_ON(ret);
1023
1024        spin_lock(&root->node_lock);
1025        root->node = child;
1026        spin_unlock(&root->node_lock);
1027
1028        add_root_to_dirty_list(root);
1029        btrfs_tree_unlock(child);
1030
1031        path->locks[level] = 0;
1032        path->nodes[level] = NULL;
1033        clean_tree_block(trans, root, mid);
1034        btrfs_tree_unlock(mid);
1035        /* once for the path */
1036        free_extent_buffer(mid);
1037        ret = btrfs_free_tree_block(trans, root, mid->start, mid->len,
1038                        0, root->root_key.objectid, level);
1039        /* once for the root ptr */
1040        free_extent_buffer(mid);
1041        return ret;
1042    }
1043    if (btrfs_header_nritems(mid) >
1044        BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1045        return 0;
1046
1047    if (btrfs_header_nritems(mid) < 2)
1048        err_on_enospc = 1;
1049
1050    left = read_node_slot(root, parent, pslot - 1);
1051    if (left) {
1052        btrfs_tree_lock(left);
1053        btrfs_set_lock_blocking(left);
1054        wret = btrfs_cow_block(trans, root, left,
1055                       parent, pslot - 1, &left);
1056        if (wret) {
1057            ret = wret;
1058            goto enospc;
1059        }
1060    }
1061    right = read_node_slot(root, parent, pslot + 1);
1062    if (right) {
1063        btrfs_tree_lock(right);
1064        btrfs_set_lock_blocking(right);
1065        wret = btrfs_cow_block(trans, root, right,
1066                       parent, pslot + 1, &right);
1067        if (wret) {
1068            ret = wret;
1069            goto enospc;
1070        }
1071    }
1072
1073    /* first, try to make some room in the middle buffer */
1074    if (left) {
1075        orig_slot += btrfs_header_nritems(left);
1076        wret = push_node_left(trans, root, left, mid, 1);
1077        if (wret < 0)
1078            ret = wret;
1079        if (btrfs_header_nritems(mid) < 2)
1080            err_on_enospc = 1;
1081    }
1082
1083    /*
1084     * then try to empty the right most buffer into the middle
1085     */
1086    if (right) {
1087        wret = push_node_left(trans, root, mid, right, 1);
1088        if (wret < 0 && wret != -ENOSPC)
1089            ret = wret;
1090        if (btrfs_header_nritems(right) == 0) {
1091            u64 bytenr = right->start;
1092            u32 blocksize = right->len;
1093
1094            clean_tree_block(trans, root, right);
1095            btrfs_tree_unlock(right);
1096            free_extent_buffer(right);
1097            right = NULL;
1098            wret = del_ptr(trans, root, path, level + 1, pslot +
1099                       1);
1100            if (wret)
1101                ret = wret;
1102            wret = btrfs_free_tree_block(trans, root,
1103                             bytenr, blocksize, 0,
1104                             root->root_key.objectid,
1105                             level);
1106            if (wret)
1107                ret = wret;
1108        } else {
1109            struct btrfs_disk_key right_key;
1110            btrfs_node_key(right, &right_key, 0);
1111            btrfs_set_node_key(parent, &right_key, pslot + 1);
1112            btrfs_mark_buffer_dirty(parent);
1113        }
1114    }
1115    if (btrfs_header_nritems(mid) == 1) {
1116        /*
1117         * we're not allowed to leave a node with one item in the
1118         * tree during a delete. A deletion from lower in the tree
1119         * could try to delete the only pointer in this node.
1120         * So, pull some keys from the left.
1121         * There has to be a left pointer at this point because
1122         * otherwise we would have pulled some pointers from the
1123         * right
1124         */
1125        BUG_ON(!left);
1126        wret = balance_node_right(trans, root, mid, left);
1127        if (wret < 0) {
1128            ret = wret;
1129            goto enospc;
1130        }
1131        if (wret == 1) {
1132            wret = push_node_left(trans, root, left, mid, 1);
1133            if (wret < 0)
1134                ret = wret;
1135        }
1136        BUG_ON(wret == 1);
1137    }
1138    if (btrfs_header_nritems(mid) == 0) {
1139        /* we've managed to empty the middle node, drop it */
1140        u64 bytenr = mid->start;
1141        u32 blocksize = mid->len;
1142
1143        clean_tree_block(trans, root, mid);
1144        btrfs_tree_unlock(mid);
1145        free_extent_buffer(mid);
1146        mid = NULL;
1147        wret = del_ptr(trans, root, path, level + 1, pslot);
1148        if (wret)
1149            ret = wret;
1150        wret = btrfs_free_tree_block(trans, root, bytenr, blocksize,
1151                     0, root->root_key.objectid, level);
1152        if (wret)
1153            ret = wret;
1154    } else {
1155        /* update the parent key to reflect our changes */
1156        struct btrfs_disk_key mid_key;
1157        btrfs_node_key(mid, &mid_key, 0);
1158        btrfs_set_node_key(parent, &mid_key, pslot);
1159        btrfs_mark_buffer_dirty(parent);
1160    }
1161
1162    /* update the path */
1163    if (left) {
1164        if (btrfs_header_nritems(left) > orig_slot) {
1165            extent_buffer_get(left);
1166            /* left was locked after cow */
1167            path->nodes[level] = left;
1168            path->slots[level + 1] -= 1;
1169            path->slots[level] = orig_slot;
1170            if (mid) {
1171                btrfs_tree_unlock(mid);
1172                free_extent_buffer(mid);
1173            }
1174        } else {
1175            orig_slot -= btrfs_header_nritems(left);
1176            path->slots[level] = orig_slot;
1177        }
1178    }
1179    /* double check we haven't messed things up */
1180    check_block(root, path, level);
1181    if (orig_ptr !=
1182        btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1183        BUG();
1184enospc:
1185    if (right) {
1186        btrfs_tree_unlock(right);
1187        free_extent_buffer(right);
1188    }
1189    if (left) {
1190        if (path->nodes[level] != left)
1191            btrfs_tree_unlock(left);
1192        free_extent_buffer(left);
1193    }
1194    return ret;
1195}
1196
1197/* Node balancing for insertion. Here we only split or push nodes around
1198 * when they are completely full. This is also done top down, so we
1199 * have to be pessimistic.
1200 */
1201static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1202                      struct btrfs_root *root,
1203                      struct btrfs_path *path, int level)
1204{
1205    struct extent_buffer *right = NULL;
1206    struct extent_buffer *mid;
1207    struct extent_buffer *left = NULL;
1208    struct extent_buffer *parent = NULL;
1209    int ret = 0;
1210    int wret;
1211    int pslot;
1212    int orig_slot = path->slots[level];
1213    u64 orig_ptr;
1214
1215    if (level == 0)
1216        return 1;
1217
1218    mid = path->nodes[level];
1219    WARN_ON(btrfs_header_generation(mid) != trans->transid);
1220    orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1221
1222    if (level < BTRFS_MAX_LEVEL - 1)
1223        parent = path->nodes[level + 1];
1224    pslot = path->slots[level + 1];
1225
1226    if (!parent)
1227        return 1;
1228
1229    left = read_node_slot(root, parent, pslot - 1);
1230
1231    /* first, try to make some room in the middle buffer */
1232    if (left) {
1233        u32 left_nr;
1234
1235        btrfs_tree_lock(left);
1236        btrfs_set_lock_blocking(left);
1237
1238        left_nr = btrfs_header_nritems(left);
1239        if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1240            wret = 1;
1241        } else {
1242            ret = btrfs_cow_block(trans, root, left, parent,
1243                          pslot - 1, &left);
1244            if (ret)
1245                wret = 1;
1246            else {
1247                wret = push_node_left(trans, root,
1248                              left, mid, 0);
1249            }
1250        }
1251        if (wret < 0)
1252            ret = wret;
1253        if (wret == 0) {
1254            struct btrfs_disk_key disk_key;
1255            orig_slot += left_nr;
1256            btrfs_node_key(mid, &disk_key, 0);
1257            btrfs_set_node_key(parent, &disk_key, pslot);
1258            btrfs_mark_buffer_dirty(parent);
1259            if (btrfs_header_nritems(left) > orig_slot) {
1260                path->nodes[level] = left;
1261                path->slots[level + 1] -= 1;
1262                path->slots[level] = orig_slot;
1263                btrfs_tree_unlock(mid);
1264                free_extent_buffer(mid);
1265            } else {
1266                orig_slot -=
1267                    btrfs_header_nritems(left);
1268                path->slots[level] = orig_slot;
1269                btrfs_tree_unlock(left);
1270                free_extent_buffer(left);
1271            }
1272            return 0;
1273        }
1274        btrfs_tree_unlock(left);
1275        free_extent_buffer(left);
1276    }
1277    right = read_node_slot(root, parent, pslot + 1);
1278
1279    /*
1280     * then try to empty the right most buffer into the middle
1281     */
1282    if (right) {
1283        u32 right_nr;
1284
1285        btrfs_tree_lock(right);
1286        btrfs_set_lock_blocking(right);
1287
1288        right_nr = btrfs_header_nritems(right);
1289        if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1290            wret = 1;
1291        } else {
1292            ret = btrfs_cow_block(trans, root, right,
1293                          parent, pslot + 1,
1294                          &right);
1295            if (ret)
1296                wret = 1;
1297            else {
1298                wret = balance_node_right(trans, root,
1299                              right, mid);
1300            }
1301        }
1302        if (wret < 0)
1303            ret = wret;
1304        if (wret == 0) {
1305            struct btrfs_disk_key disk_key;
1306
1307            btrfs_node_key(right, &disk_key, 0);
1308            btrfs_set_node_key(parent, &disk_key, pslot + 1);
1309            btrfs_mark_buffer_dirty(parent);
1310
1311            if (btrfs_header_nritems(mid) <= orig_slot) {
1312                path->nodes[level] = right;
1313                path->slots[level + 1] += 1;
1314                path->slots[level] = orig_slot -
1315                    btrfs_header_nritems(mid);
1316                btrfs_tree_unlock(mid);
1317                free_extent_buffer(mid);
1318            } else {
1319                btrfs_tree_unlock(right);
1320                free_extent_buffer(right);
1321            }
1322            return 0;
1323        }
1324        btrfs_tree_unlock(right);
1325        free_extent_buffer(right);
1326    }
1327    return 1;
1328}
1329
1330/*
1331 * readahead one full node of leaves, finding things that are close
1332 * to the block in 'slot', and triggering ra on them.
1333 */
1334static void reada_for_search(struct btrfs_root *root,
1335                 struct btrfs_path *path,
1336                 int level, int slot, u64 objectid)
1337{
1338    struct extent_buffer *node;
1339    struct btrfs_disk_key disk_key;
1340    u32 nritems;
1341    u64 search;
1342    u64 target;
1343    u64 nread = 0;
1344    int direction = path->reada;
1345    struct extent_buffer *eb;
1346    u32 nr;
1347    u32 blocksize;
1348    u32 nscan = 0;
1349
1350    if (level != 1)
1351        return;
1352
1353    if (!path->nodes[level])
1354        return;
1355
1356    node = path->nodes[level];
1357
1358    search = btrfs_node_blockptr(node, slot);
1359    blocksize = btrfs_level_size(root, level - 1);
1360    eb = btrfs_find_tree_block(root, search, blocksize);
1361    if (eb) {
1362        free_extent_buffer(eb);
1363        return;
1364    }
1365
1366    target = search;
1367
1368    nritems = btrfs_header_nritems(node);
1369    nr = slot;
1370    while (1) {
1371        if (direction < 0) {
1372            if (nr == 0)
1373                break;
1374            nr--;
1375        } else if (direction > 0) {
1376            nr++;
1377            if (nr >= nritems)
1378                break;
1379        }
1380        if (path->reada < 0 && objectid) {
1381            btrfs_node_key(node, &disk_key, nr);
1382            if (btrfs_disk_key_objectid(&disk_key) != objectid)
1383                break;
1384        }
1385        search = btrfs_node_blockptr(node, nr);
1386        if ((search <= target && target - search <= 65536) ||
1387            (search > target && search - target <= 65536)) {
1388            readahead_tree_block(root, search, blocksize,
1389                     btrfs_node_ptr_generation(node, nr));
1390            nread += blocksize;
1391        }
1392        nscan++;
1393        if ((nread > 65536 || nscan > 32))
1394            break;
1395    }
1396}
1397
1398/*
1399 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1400 * cache
1401 */
1402static noinline int reada_for_balance(struct btrfs_root *root,
1403                      struct btrfs_path *path, int level)
1404{
1405    int slot;
1406    int nritems;
1407    struct extent_buffer *parent;
1408    struct extent_buffer *eb;
1409    u64 gen;
1410    u64 block1 = 0;
1411    u64 block2 = 0;
1412    int ret = 0;
1413    int blocksize;
1414
1415    parent = path->nodes[level + 1];
1416    if (!parent)
1417        return 0;
1418
1419    nritems = btrfs_header_nritems(parent);
1420    slot = path->slots[level + 1];
1421    blocksize = btrfs_level_size(root, level);
1422
1423    if (slot > 0) {
1424        block1 = btrfs_node_blockptr(parent, slot - 1);
1425        gen = btrfs_node_ptr_generation(parent, slot - 1);
1426        eb = btrfs_find_tree_block(root, block1, blocksize);
1427        if (eb && btrfs_buffer_uptodate(eb, gen))
1428            block1 = 0;
1429        free_extent_buffer(eb);
1430    }
1431    if (slot + 1 < nritems) {
1432        block2 = btrfs_node_blockptr(parent, slot + 1);
1433        gen = btrfs_node_ptr_generation(parent, slot + 1);
1434        eb = btrfs_find_tree_block(root, block2, blocksize);
1435        if (eb && btrfs_buffer_uptodate(eb, gen))
1436            block2 = 0;
1437        free_extent_buffer(eb);
1438    }
1439    if (block1 || block2) {
1440        ret = -EAGAIN;
1441
1442        /* release the whole path */
1443        btrfs_release_path(root, path);
1444
1445        /* read the blocks */
1446        if (block1)
1447            readahead_tree_block(root, block1, blocksize, 0);
1448        if (block2)
1449            readahead_tree_block(root, block2, blocksize, 0);
1450
1451        if (block1) {
1452            eb = read_tree_block(root, block1, blocksize, 0);
1453            free_extent_buffer(eb);
1454        }
1455        if (block2) {
1456            eb = read_tree_block(root, block2, blocksize, 0);
1457            free_extent_buffer(eb);
1458        }
1459    }
1460    return ret;
1461}
1462
1463
1464/*
1465 * when we walk down the tree, it is usually safe to unlock the higher layers
1466 * in the tree. The exceptions are when our path goes through slot 0, because
1467 * operations on the tree might require changing key pointers higher up in the
1468 * tree.
1469 *
1470 * callers might also have set path->keep_locks, which tells this code to keep
1471 * the lock if the path points to the last slot in the block. This is part of
1472 * walking through the tree, and selecting the next slot in the higher block.
1473 *
1474 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1475 * if lowest_unlock is 1, level 0 won't be unlocked
1476 */
1477static noinline void unlock_up(struct btrfs_path *path, int level,
1478                   int lowest_unlock)
1479{
1480    int i;
1481    int skip_level = level;
1482    int no_skips = 0;
1483    struct extent_buffer *t;
1484
1485    for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1486        if (!path->nodes[i])
1487            break;
1488        if (!path->locks[i])
1489            break;
1490        if (!no_skips && path->slots[i] == 0) {
1491            skip_level = i + 1;
1492            continue;
1493        }
1494        if (!no_skips && path->keep_locks) {
1495            u32 nritems;
1496            t = path->nodes[i];
1497            nritems = btrfs_header_nritems(t);
1498            if (nritems < 1 || path->slots[i] >= nritems - 1) {
1499                skip_level = i + 1;
1500                continue;
1501            }
1502        }
1503        if (skip_level < i && i >= lowest_unlock)
1504            no_skips = 1;
1505
1506        t = path->nodes[i];
1507        if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1508            btrfs_tree_unlock(t);
1509            path->locks[i] = 0;
1510        }
1511    }
1512}
1513
1514/*
1515 * This releases any locks held in the path starting at level and
1516 * going all the way up to the root.
1517 *
1518 * btrfs_search_slot will keep the lock held on higher nodes in a few
1519 * corner cases, such as COW of the block at slot zero in the node. This
1520 * ignores those rules, and it should only be called when there are no
1521 * more updates to be done higher up in the tree.
1522 */
1523noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1524{
1525    int i;
1526
1527    if (path->keep_locks)
1528        return;
1529
1530    for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1531        if (!path->nodes[i])
1532            continue;
1533        if (!path->locks[i])
1534            continue;
1535        btrfs_tree_unlock(path->nodes[i]);
1536        path->locks[i] = 0;
1537    }
1538}
1539
1540/*
1541 * helper function for btrfs_search_slot. The goal is to find a block
1542 * in cache without setting the path to blocking. If we find the block
1543 * we return zero and the path is unchanged.
1544 *
1545 * If we can't find the block, we set the path blocking and do some
1546 * reada. -EAGAIN is returned and the search must be repeated.
1547 */
1548static int
1549read_block_for_search(struct btrfs_trans_handle *trans,
1550               struct btrfs_root *root, struct btrfs_path *p,
1551               struct extent_buffer **eb_ret, int level, int slot,
1552               struct btrfs_key *key)
1553{
1554    u64 blocknr;
1555    u64 gen;
1556    u32 blocksize;
1557    struct extent_buffer *b = *eb_ret;
1558    struct extent_buffer *tmp;
1559    int ret;
1560
1561    blocknr = btrfs_node_blockptr(b, slot);
1562    gen = btrfs_node_ptr_generation(b, slot);
1563    blocksize = btrfs_level_size(root, level - 1);
1564
1565    tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1566    if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1567        /*
1568         * we found an up to date block without sleeping, return
1569         * right away
1570         */
1571        *eb_ret = tmp;
1572        return 0;
1573    }
1574
1575    /*
1576     * reduce lock contention at high levels
1577     * of the btree by dropping locks before
1578     * we read. Don't release the lock on the current
1579     * level because we need to walk this node to figure
1580     * out which blocks to read.
1581     */
1582    btrfs_unlock_up_safe(p, level + 1);
1583    btrfs_set_path_blocking(p);
1584
1585    if (tmp)
1586        free_extent_buffer(tmp);
1587    if (p->reada)
1588        reada_for_search(root, p, level, slot, key->objectid);
1589
1590    btrfs_release_path(NULL, p);
1591
1592    ret = -EAGAIN;
1593    tmp = read_tree_block(root, blocknr, blocksize, gen);
1594    if (tmp) {
1595        /*
1596         * If the read above didn't mark this buffer up to date,
1597         * it will never end up being up to date. Set ret to EIO now
1598         * and give up so that our caller doesn't loop forever
1599         * on our EAGAINs.
1600         */
1601        if (!btrfs_buffer_uptodate(tmp, 0))
1602            ret = -EIO;
1603        free_extent_buffer(tmp);
1604    }
1605    return ret;
1606}
1607
1608/*
1609 * helper function for btrfs_search_slot. This does all of the checks
1610 * for node-level blocks and does any balancing required based on
1611 * the ins_len.
1612 *
1613 * If no extra work was required, zero is returned. If we had to
1614 * drop the path, -EAGAIN is returned and btrfs_search_slot must
1615 * start over
1616 */
1617static int
1618setup_nodes_for_search(struct btrfs_trans_handle *trans,
1619               struct btrfs_root *root, struct btrfs_path *p,
1620               struct extent_buffer *b, int level, int ins_len)
1621{
1622    int ret;
1623    if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1624        BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1625        int sret;
1626
1627        sret = reada_for_balance(root, p, level);
1628        if (sret)
1629            goto again;
1630
1631        btrfs_set_path_blocking(p);
1632        sret = split_node(trans, root, p, level);
1633        btrfs_clear_path_blocking(p, NULL);
1634
1635        BUG_ON(sret > 0);
1636        if (sret) {
1637            ret = sret;
1638            goto done;
1639        }
1640        b = p->nodes[level];
1641    } else if (ins_len < 0 && btrfs_header_nritems(b) <
1642           BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
1643        int sret;
1644
1645        sret = reada_for_balance(root, p, level);
1646        if (sret)
1647            goto again;
1648
1649        btrfs_set_path_blocking(p);
1650        sret = balance_level(trans, root, p, level);
1651        btrfs_clear_path_blocking(p, NULL);
1652
1653        if (sret) {
1654            ret = sret;
1655            goto done;
1656        }
1657        b = p->nodes[level];
1658        if (!b) {
1659            btrfs_release_path(NULL, p);
1660            goto again;
1661        }
1662        BUG_ON(btrfs_header_nritems(b) == 1);
1663    }
1664    return 0;
1665
1666again:
1667    ret = -EAGAIN;
1668done:
1669    return ret;
1670}
1671
1672/*
1673 * look for key in the tree. path is filled in with nodes along the way
1674 * if key is found, we return zero and you can find the item in the leaf
1675 * level of the path (level 0)
1676 *
1677 * If the key isn't found, the path points to the slot where it should
1678 * be inserted, and 1 is returned. If there are other errors during the
1679 * search a negative error number is returned.
1680 *
1681 * if ins_len > 0, nodes and leaves will be split as we walk down the
1682 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1683 * possible)
1684 */
1685int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1686              *root, struct btrfs_key *key, struct btrfs_path *p, int
1687              ins_len, int cow)
1688{
1689    struct extent_buffer *b;
1690    int slot;
1691    int ret;
1692    int err;
1693    int level;
1694    int lowest_unlock = 1;
1695    u8 lowest_level = 0;
1696
1697    lowest_level = p->lowest_level;
1698    WARN_ON(lowest_level && ins_len > 0);
1699    WARN_ON(p->nodes[0] != NULL);
1700
1701    if (ins_len < 0)
1702        lowest_unlock = 2;
1703
1704again:
1705    if (p->search_commit_root) {
1706        b = root->commit_root;
1707        extent_buffer_get(b);
1708        if (!p->skip_locking)
1709            btrfs_tree_lock(b);
1710    } else {
1711        if (p->skip_locking)
1712            b = btrfs_root_node(root);
1713        else
1714            b = btrfs_lock_root_node(root);
1715    }
1716
1717    while (b) {
1718        level = btrfs_header_level(b);
1719
1720        /*
1721         * setup the path here so we can release it under lock
1722         * contention with the cow code
1723         */
1724        p->nodes[level] = b;
1725        if (!p->skip_locking)
1726            p->locks[level] = 1;
1727
1728        if (cow) {
1729            /*
1730             * if we don't really need to cow this block
1731             * then we don't want to set the path blocking,
1732             * so we test it here
1733             */
1734            if (!should_cow_block(trans, root, b))
1735                goto cow_done;
1736
1737            btrfs_set_path_blocking(p);
1738
1739            err = btrfs_cow_block(trans, root, b,
1740                          p->nodes[level + 1],
1741                          p->slots[level + 1], &b);
1742            if (err) {
1743                free_extent_buffer(b);
1744                ret = err;
1745                goto done;
1746            }
1747        }
1748cow_done:
1749        BUG_ON(!cow && ins_len);
1750        if (level != btrfs_header_level(b))
1751            WARN_ON(1);
1752        level = btrfs_header_level(b);
1753
1754        p->nodes[level] = b;
1755        if (!p->skip_locking)
1756            p->locks[level] = 1;
1757
1758        btrfs_clear_path_blocking(p, NULL);
1759
1760        /*
1761         * we have a lock on b and as long as we aren't changing
1762         * the tree, there is no way to for the items in b to change.
1763         * It is safe to drop the lock on our parent before we
1764         * go through the expensive btree search on b.
1765         *
1766         * If cow is true, then we might be changing slot zero,
1767         * which may require changing the parent. So, we can't
1768         * drop the lock until after we know which slot we're
1769         * operating on.
1770         */
1771        if (!cow)
1772            btrfs_unlock_up_safe(p, level + 1);
1773
1774        ret = check_block(root, p, level);
1775        if (ret) {
1776            ret = -1;
1777            goto done;
1778        }
1779
1780        ret = bin_search(b, key, level, &slot);
1781
1782        if (level != 0) {
1783            int dec = 0;
1784            if (ret && slot > 0) {
1785                dec = 1;
1786                slot -= 1;
1787            }
1788            p->slots[level] = slot;
1789            err = setup_nodes_for_search(trans, root, p, b, level,
1790                             ins_len);
1791            if (err == -EAGAIN)
1792                goto again;
1793            if (err) {
1794                ret = err;
1795                goto done;
1796            }
1797            b = p->nodes[level];
1798            slot = p->slots[level];
1799
1800            unlock_up(p, level, lowest_unlock);
1801
1802            if (level == lowest_level) {
1803                if (dec)
1804                    p->slots[level]++;
1805                goto done;
1806            }
1807
1808            err = read_block_for_search(trans, root, p,
1809                            &b, level, slot, key);
1810            if (err == -EAGAIN)
1811                goto again;
1812            if (err) {
1813                ret = err;
1814                goto done;
1815            }
1816
1817            if (!p->skip_locking) {
1818                btrfs_clear_path_blocking(p, NULL);
1819                err = btrfs_try_spin_lock(b);
1820
1821                if (!err) {
1822                    btrfs_set_path_blocking(p);
1823                    btrfs_tree_lock(b);
1824                    btrfs_clear_path_blocking(p, b);
1825                }
1826            }
1827        } else {
1828            p->slots[level] = slot;
1829            if (ins_len > 0 &&
1830                btrfs_leaf_free_space(root, b) < ins_len) {
1831                btrfs_set_path_blocking(p);
1832                err = split_leaf(trans, root, key,
1833                         p, ins_len, ret == 0);
1834                btrfs_clear_path_blocking(p, NULL);
1835
1836                BUG_ON(err > 0);
1837                if (err) {
1838                    ret = err;
1839                    goto done;
1840                }
1841            }
1842            if (!p->search_for_split)
1843                unlock_up(p, level, lowest_unlock);
1844            goto done;
1845        }
1846    }
1847    ret = 1;
1848done:
1849    /*
1850     * we don't really know what they plan on doing with the path
1851     * from here on, so for now just mark it as blocking
1852     */
1853    if (!p->leave_spinning)
1854        btrfs_set_path_blocking(p);
1855    if (ret < 0)
1856        btrfs_release_path(root, p);
1857    return ret;
1858}
1859
1860/*
1861 * adjust the pointers going up the tree, starting at level
1862 * making sure the right key of each node is points to 'key'.
1863 * This is used after shifting pointers to the left, so it stops
1864 * fixing up pointers when a given leaf/node is not in slot 0 of the
1865 * higher levels
1866 *
1867 * If this fails to write a tree block, it returns -1, but continues
1868 * fixing up the blocks in ram so the tree is consistent.
1869 */
1870static int fixup_low_keys(struct btrfs_trans_handle *trans,
1871              struct btrfs_root *root, struct btrfs_path *path,
1872              struct btrfs_disk_key *key, int level)
1873{
1874    int i;
1875    int ret = 0;
1876    struct extent_buffer *t;
1877
1878    for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1879        int tslot = path->slots[i];
1880        if (!path->nodes[i])
1881            break;
1882        t = path->nodes[i];
1883        btrfs_set_node_key(t, key, tslot);
1884        btrfs_mark_buffer_dirty(path->nodes[i]);
1885        if (tslot != 0)
1886            break;
1887    }
1888    return ret;
1889}
1890
1891/*
1892 * update item key.
1893 *
1894 * This function isn't completely safe. It's the caller's responsibility
1895 * that the new key won't break the order
1896 */
1897int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1898                struct btrfs_root *root, struct btrfs_path *path,
1899                struct btrfs_key *new_key)
1900{
1901    struct btrfs_disk_key disk_key;
1902    struct extent_buffer *eb;
1903    int slot;
1904
1905    eb = path->nodes[0];
1906    slot = path->slots[0];
1907    if (slot > 0) {
1908        btrfs_item_key(eb, &disk_key, slot - 1);
1909        if (comp_keys(&disk_key, new_key) >= 0)
1910            return -1;
1911    }
1912    if (slot < btrfs_header_nritems(eb) - 1) {
1913        btrfs_item_key(eb, &disk_key, slot + 1);
1914        if (comp_keys(&disk_key, new_key) <= 0)
1915            return -1;
1916    }
1917
1918    btrfs_cpu_key_to_disk(&disk_key, new_key);
1919    btrfs_set_item_key(eb, &disk_key, slot);
1920    btrfs_mark_buffer_dirty(eb);
1921    if (slot == 0)
1922        fixup_low_keys(trans, root, path, &disk_key, 1);
1923    return 0;
1924}
1925
1926/*
1927 * try to push data from one node into the next node left in the
1928 * tree.
1929 *
1930 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1931 * error, and > 0 if there was no room in the left hand block.
1932 */
1933static int push_node_left(struct btrfs_trans_handle *trans,
1934              struct btrfs_root *root, struct extent_buffer *dst,
1935              struct extent_buffer *src, int empty)
1936{
1937    int push_items = 0;
1938    int src_nritems;
1939    int dst_nritems;
1940    int ret = 0;
1941
1942    src_nritems = btrfs_header_nritems(src);
1943    dst_nritems = btrfs_header_nritems(dst);
1944    push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1945    WARN_ON(btrfs_header_generation(src) != trans->transid);
1946    WARN_ON(btrfs_header_generation(dst) != trans->transid);
1947
1948    if (!empty && src_nritems <= 8)
1949        return 1;
1950
1951    if (push_items <= 0)
1952        return 1;
1953
1954    if (empty) {
1955        push_items = min(src_nritems, push_items);
1956        if (push_items < src_nritems) {
1957            /* leave at least 8 pointers in the node if
1958             * we aren't going to empty it
1959             */
1960            if (src_nritems - push_items < 8) {
1961                if (push_items <= 8)
1962                    return 1;
1963                push_items -= 8;
1964            }
1965        }
1966    } else
1967        push_items = min(src_nritems - 8, push_items);
1968
1969    copy_extent_buffer(dst, src,
1970               btrfs_node_key_ptr_offset(dst_nritems),
1971               btrfs_node_key_ptr_offset(0),
1972               push_items * sizeof(struct btrfs_key_ptr));
1973
1974    if (push_items < src_nritems) {
1975        memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1976                      btrfs_node_key_ptr_offset(push_items),
1977                      (src_nritems - push_items) *
1978                      sizeof(struct btrfs_key_ptr));
1979    }
1980    btrfs_set_header_nritems(src, src_nritems - push_items);
1981    btrfs_set_header_nritems(dst, dst_nritems + push_items);
1982    btrfs_mark_buffer_dirty(src);
1983    btrfs_mark_buffer_dirty(dst);
1984
1985    return ret;
1986}
1987
1988/*
1989 * try to push data from one node into the next node right in the
1990 * tree.
1991 *
1992 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1993 * error, and > 0 if there was no room in the right hand block.
1994 *
1995 * this will only push up to 1/2 the contents of the left node over
1996 */
1997static int balance_node_right(struct btrfs_trans_handle *trans,
1998                  struct btrfs_root *root,
1999                  struct extent_buffer *dst,
2000                  struct extent_buffer *src)
2001{
2002    int push_items = 0;
2003    int max_push;
2004    int src_nritems;
2005    int dst_nritems;
2006    int ret = 0;
2007
2008    WARN_ON(btrfs_header_generation(src) != trans->transid);
2009    WARN_ON(btrfs_header_generation(dst) != trans->transid);
2010
2011    src_nritems = btrfs_header_nritems(src);
2012    dst_nritems = btrfs_header_nritems(dst);
2013    push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2014    if (push_items <= 0)
2015        return 1;
2016
2017    if (src_nritems < 4)
2018        return 1;
2019
2020    max_push = src_nritems / 2 + 1;
2021    /* don't try to empty the node */
2022    if (max_push >= src_nritems)
2023        return 1;
2024
2025    if (max_push < push_items)
2026        push_items = max_push;
2027
2028    memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2029                      btrfs_node_key_ptr_offset(0),
2030                      (dst_nritems) *
2031                      sizeof(struct btrfs_key_ptr));
2032
2033    copy_extent_buffer(dst, src,
2034               btrfs_node_key_ptr_offset(0),
2035               btrfs_node_key_ptr_offset(src_nritems - push_items),
2036               push_items * sizeof(struct btrfs_key_ptr));
2037
2038    btrfs_set_header_nritems(src, src_nritems - push_items);
2039    btrfs_set_header_nritems(dst, dst_nritems + push_items);
2040
2041    btrfs_mark_buffer_dirty(src);
2042    btrfs_mark_buffer_dirty(dst);
2043
2044    return ret;
2045}
2046
2047/*
2048 * helper function to insert a new root level in the tree.
2049 * A new node is allocated, and a single item is inserted to
2050 * point to the existing root
2051 *
2052 * returns zero on success or < 0 on failure.
2053 */
2054static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2055               struct btrfs_root *root,
2056               struct btrfs_path *path, int level)
2057{
2058    u64 lower_gen;
2059    struct extent_buffer *lower;
2060    struct extent_buffer *c;
2061    struct extent_buffer *old;
2062    struct btrfs_disk_key lower_key;
2063
2064    BUG_ON(path->nodes[level]);
2065    BUG_ON(path->nodes[level-1] != root->node);
2066
2067    lower = path->nodes[level-1];
2068    if (level == 1)
2069        btrfs_item_key(lower, &lower_key, 0);
2070    else
2071        btrfs_node_key(lower, &lower_key, 0);
2072
2073    c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2074                   root->root_key.objectid, &lower_key,
2075                   level, root->node->start, 0);
2076    if (IS_ERR(c))
2077        return PTR_ERR(c);
2078
2079    memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
2080    btrfs_set_header_nritems(c, 1);
2081    btrfs_set_header_level(c, level);
2082    btrfs_set_header_bytenr(c, c->start);
2083    btrfs_set_header_generation(c, trans->transid);
2084    btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
2085    btrfs_set_header_owner(c, root->root_key.objectid);
2086
2087    write_extent_buffer(c, root->fs_info->fsid,
2088                (unsigned long)btrfs_header_fsid(c),
2089                BTRFS_FSID_SIZE);
2090
2091    write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2092                (unsigned long)btrfs_header_chunk_tree_uuid(c),
2093                BTRFS_UUID_SIZE);
2094
2095    btrfs_set_node_key(c, &lower_key, 0);
2096    btrfs_set_node_blockptr(c, 0, lower->start);
2097    lower_gen = btrfs_header_generation(lower);
2098    WARN_ON(lower_gen != trans->transid);
2099
2100    btrfs_set_node_ptr_generation(c, 0, lower_gen);
2101
2102    btrfs_mark_buffer_dirty(c);
2103
2104    spin_lock(&root->node_lock);
2105    old = root->node;
2106    root->node = c;
2107    spin_unlock(&root->node_lock);
2108
2109    /* the super has an extra ref to root->node */
2110    free_extent_buffer(old);
2111
2112    add_root_to_dirty_list(root);
2113    extent_buffer_get(c);
2114    path->nodes[level] = c;
2115    path->locks[level] = 1;
2116    path->slots[level] = 0;
2117    return 0;
2118}
2119
2120/*
2121 * worker function to insert a single pointer in a node.
2122 * the node should have enough room for the pointer already
2123 *
2124 * slot and level indicate where you want the key to go, and
2125 * blocknr is the block the key points to.
2126 *
2127 * returns zero on success and < 0 on any error
2128 */
2129static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2130              *root, struct btrfs_path *path, struct btrfs_disk_key
2131              *key, u64 bytenr, int slot, int level)
2132{
2133    struct extent_buffer *lower;
2134    int nritems;
2135
2136    BUG_ON(!path->nodes[level]);
2137    lower = path->nodes[level];
2138    nritems = btrfs_header_nritems(lower);
2139    BUG_ON(slot > nritems);
2140    if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2141        BUG();
2142    if (slot != nritems) {
2143        memmove_extent_buffer(lower,
2144                  btrfs_node_key_ptr_offset(slot + 1),
2145                  btrfs_node_key_ptr_offset(slot),
2146                  (nritems - slot) * sizeof(struct btrfs_key_ptr));
2147    }
2148    btrfs_set_node_key(lower, key, slot);
2149    btrfs_set_node_blockptr(lower, slot, bytenr);
2150    WARN_ON(trans->transid == 0);
2151    btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2152    btrfs_set_header_nritems(lower, nritems + 1);
2153    btrfs_mark_buffer_dirty(lower);
2154    return 0;
2155}
2156
2157/*
2158 * split the node at the specified level in path in two.
2159 * The path is corrected to point to the appropriate node after the split
2160 *
2161 * Before splitting this tries to make some room in the node by pushing
2162 * left and right, if either one works, it returns right away.
2163 *
2164 * returns 0 on success and < 0 on failure
2165 */
2166static noinline int split_node(struct btrfs_trans_handle *trans,
2167                   struct btrfs_root *root,
2168                   struct btrfs_path *path, int level)
2169{
2170    struct extent_buffer *c;
2171    struct extent_buffer *split;
2172    struct btrfs_disk_key disk_key;
2173    int mid;
2174    int ret;
2175    int wret;
2176    u32 c_nritems;
2177
2178    c = path->nodes[level];
2179    WARN_ON(btrfs_header_generation(c) != trans->transid);
2180    if (c == root->node) {
2181        /* trying to split the root, lets make a new one */
2182        ret = insert_new_root(trans, root, path, level + 1);
2183        if (ret)
2184            return ret;
2185    } else {
2186        ret = push_nodes_for_insert(trans, root, path, level);
2187        c = path->nodes[level];
2188        if (!ret && btrfs_header_nritems(c) <
2189            BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2190            return 0;
2191        if (ret < 0)
2192            return ret;
2193    }
2194
2195    c_nritems = btrfs_header_nritems(c);
2196    mid = (c_nritems + 1) / 2;
2197    btrfs_node_key(c, &disk_key, mid);
2198
2199    split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2200                    root->root_key.objectid,
2201                    &disk_key, level, c->start, 0);
2202    if (IS_ERR(split))
2203        return PTR_ERR(split);
2204
2205    memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
2206    btrfs_set_header_level(split, btrfs_header_level(c));
2207    btrfs_set_header_bytenr(split, split->start);
2208    btrfs_set_header_generation(split, trans->transid);
2209    btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
2210    btrfs_set_header_owner(split, root->root_key.objectid);
2211    write_extent_buffer(split, root->fs_info->fsid,
2212                (unsigned long)btrfs_header_fsid(split),
2213                BTRFS_FSID_SIZE);
2214    write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2215                (unsigned long)btrfs_header_chunk_tree_uuid(split),
2216                BTRFS_UUID_SIZE);
2217
2218
2219    copy_extent_buffer(split, c,
2220               btrfs_node_key_ptr_offset(0),
2221               btrfs_node_key_ptr_offset(mid),
2222               (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2223    btrfs_set_header_nritems(split, c_nritems - mid);
2224    btrfs_set_header_nritems(c, mid);
2225    ret = 0;
2226
2227    btrfs_mark_buffer_dirty(c);
2228    btrfs_mark_buffer_dirty(split);
2229
2230    wret = insert_ptr(trans, root, path, &disk_key, split->start,
2231              path->slots[level + 1] + 1,
2232              level + 1);
2233    if (wret)
2234        ret = wret;
2235
2236    if (path->slots[level] >= mid) {
2237        path->slots[level] -= mid;
2238        btrfs_tree_unlock(c);
2239        free_extent_buffer(c);
2240        path->nodes[level] = split;
2241        path->slots[level + 1] += 1;
2242    } else {
2243        btrfs_tree_unlock(split);
2244        free_extent_buffer(split);
2245    }
2246    return ret;
2247}
2248
2249/*
2250 * how many bytes are required to store the items in a leaf. start
2251 * and nr indicate which items in the leaf to check. This totals up the
2252 * space used both by the item structs and the item data
2253 */
2254static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2255{
2256    int data_len;
2257    int nritems = btrfs_header_nritems(l);
2258    int end = min(nritems, start + nr) - 1;
2259
2260    if (!nr)
2261        return 0;
2262    data_len = btrfs_item_end_nr(l, start);
2263    data_len = data_len - btrfs_item_offset_nr(l, end);
2264    data_len += sizeof(struct btrfs_item) * nr;
2265    WARN_ON(data_len < 0);
2266    return data_len;
2267}
2268
2269/*
2270 * The space between the end of the leaf items and
2271 * the start of the leaf data. IOW, how much room
2272 * the leaf has left for both items and data
2273 */
2274noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2275                   struct extent_buffer *leaf)
2276{
2277    int nritems = btrfs_header_nritems(leaf);
2278    int ret;
2279    ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2280    if (ret < 0) {
2281        printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2282               "used %d nritems %d\n",
2283               ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2284               leaf_space_used(leaf, 0, nritems), nritems);
2285    }
2286    return ret;
2287}
2288
2289static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2290                      struct btrfs_root *root,
2291                      struct btrfs_path *path,
2292                      int data_size, int empty,
2293                      struct extent_buffer *right,
2294                      int free_space, u32 left_nritems)
2295{
2296    struct extent_buffer *left = path->nodes[0];
2297    struct extent_buffer *upper = path->nodes[1];
2298    struct btrfs_disk_key disk_key;
2299    int slot;
2300    u32 i;
2301    int push_space = 0;
2302    int push_items = 0;
2303    struct btrfs_item *item;
2304    u32 nr;
2305    u32 right_nritems;
2306    u32 data_end;
2307    u32 this_item_size;
2308
2309    if (empty)
2310        nr = 0;
2311    else
2312        nr = 1;
2313
2314    if (path->slots[0] >= left_nritems)
2315        push_space += data_size;
2316
2317    slot = path->slots[1];
2318    i = left_nritems - 1;
2319    while (i >= nr) {
2320        item = btrfs_item_nr(left, i);
2321
2322        if (!empty && push_items > 0) {
2323            if (path->slots[0] > i)
2324                break;
2325            if (path->slots[0] == i) {
2326                int space = btrfs_leaf_free_space(root, left);
2327                if (space + push_space * 2 > free_space)
2328                    break;
2329            }
2330        }
2331
2332        if (path->slots[0] == i)
2333            push_space += data_size;
2334
2335        if (!left->map_token) {
2336            map_extent_buffer(left, (unsigned long)item,
2337                    sizeof(struct btrfs_item),
2338                    &left->map_token, &left->kaddr,
2339                    &left->map_start, &left->map_len,
2340                    KM_USER1);
2341        }
2342
2343        this_item_size = btrfs_item_size(left, item);
2344        if (this_item_size + sizeof(*item) + push_space > free_space)
2345            break;
2346
2347        push_items++;
2348        push_space += this_item_size + sizeof(*item);
2349        if (i == 0)
2350            break;
2351        i--;
2352    }
2353    if (left->map_token) {
2354        unmap_extent_buffer(left, left->map_token, KM_USER1);
2355        left->map_token = NULL;
2356    }
2357
2358    if (push_items == 0)
2359        goto out_unlock;
2360
2361    if (!empty && push_items == left_nritems)
2362        WARN_ON(1);
2363
2364    /* push left to right */
2365    right_nritems = btrfs_header_nritems(right);
2366
2367    push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2368    push_space -= leaf_data_end(root, left);
2369
2370    /* make room in the right data area */
2371    data_end = leaf_data_end(root, right);
2372    memmove_extent_buffer(right,
2373                  btrfs_leaf_data(right) + data_end - push_space,
2374                  btrfs_leaf_data(right) + data_end,
2375                  BTRFS_LEAF_DATA_SIZE(root) - data_end);
2376
2377    /* copy from the left data area */
2378    copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2379             BTRFS_LEAF_DATA_SIZE(root) - push_space,
2380             btrfs_leaf_data(left) + leaf_data_end(root, left),
2381             push_space);
2382
2383    memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2384                  btrfs_item_nr_offset(0),
2385                  right_nritems * sizeof(struct btrfs_item));
2386
2387    /* copy the items from left to right */
2388    copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2389           btrfs_item_nr_offset(left_nritems - push_items),
2390           push_items * sizeof(struct btrfs_item));
2391
2392    /* update the item pointers */
2393    right_nritems += push_items;
2394    btrfs_set_header_nritems(right, right_nritems);
2395    push_space = BTRFS_LEAF_DATA_SIZE(root);
2396    for (i = 0; i < right_nritems; i++) {
2397        item = btrfs_item_nr(right, i);
2398        if (!right->map_token) {
2399            map_extent_buffer(right, (unsigned long)item,
2400                    sizeof(struct btrfs_item),
2401                    &right->map_token, &right->kaddr,
2402                    &right->map_start, &right->map_len,
2403                    KM_USER1);
2404        }
2405        push_space -= btrfs_item_size(right, item);
2406        btrfs_set_item_offset(right, item, push_space);
2407    }
2408
2409    if (right->map_token) {
2410        unmap_extent_buffer(right, right->map_token, KM_USER1);
2411        right->map_token = NULL;
2412    }
2413    left_nritems -= push_items;
2414    btrfs_set_header_nritems(left, left_nritems);
2415
2416    if (left_nritems)
2417        btrfs_mark_buffer_dirty(left);
2418    btrfs_mark_buffer_dirty(right);
2419
2420    btrfs_item_key(right, &disk_key, 0);
2421    btrfs_set_node_key(upper, &disk_key, slot + 1);
2422    btrfs_mark_buffer_dirty(upper);
2423
2424    /* then fixup the leaf pointer in the path */
2425    if (path->slots[0] >= left_nritems) {
2426        path->slots[0] -= left_nritems;
2427        if (btrfs_header_nritems(path->nodes[0]) == 0)
2428            clean_tree_block(trans, root, path->nodes[0]);
2429        btrfs_tree_unlock(path->nodes[0]);
2430        free_extent_buffer(path->nodes[0]);
2431        path->nodes[0] = right;
2432        path->slots[1] += 1;
2433    } else {
2434        btrfs_tree_unlock(right);
2435        free_extent_buffer(right);
2436    }
2437    return 0;
2438
2439out_unlock:
2440    btrfs_tree_unlock(right);
2441    free_extent_buffer(right);
2442    return 1;
2443}
2444
2445/*
2446 * push some data in the path leaf to the right, trying to free up at
2447 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2448 *
2449 * returns 1 if the push failed because the other node didn't have enough
2450 * room, 0 if everything worked out and < 0 if there were major errors.
2451 */
2452static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2453               *root, struct btrfs_path *path, int data_size,
2454               int empty)
2455{
2456    struct extent_buffer *left = path->nodes[0];
2457    struct extent_buffer *right;
2458    struct extent_buffer *upper;
2459    int slot;
2460    int free_space;
2461    u32 left_nritems;
2462    int ret;
2463
2464    if (!path->nodes[1])
2465        return 1;
2466
2467    slot = path->slots[1];
2468    upper = path->nodes[1];
2469    if (slot >= btrfs_header_nritems(upper) - 1)
2470        return 1;
2471
2472    btrfs_assert_tree_locked(path->nodes[1]);
2473
2474    right = read_node_slot(root, upper, slot + 1);
2475    btrfs_tree_lock(right);
2476    btrfs_set_lock_blocking(right);
2477
2478    free_space = btrfs_leaf_free_space(root, right);
2479    if (free_space < data_size)
2480        goto out_unlock;
2481
2482    /* cow and double check */
2483    ret = btrfs_cow_block(trans, root, right, upper,
2484                  slot + 1, &right);
2485    if (ret)
2486        goto out_unlock;
2487
2488    free_space = btrfs_leaf_free_space(root, right);
2489    if (free_space < data_size)
2490        goto out_unlock;
2491
2492    left_nritems = btrfs_header_nritems(left);
2493    if (left_nritems == 0)
2494        goto out_unlock;
2495
2496    return __push_leaf_right(trans, root, path, data_size, empty,
2497                right, free_space, left_nritems);
2498out_unlock:
2499    btrfs_tree_unlock(right);
2500    free_extent_buffer(right);
2501    return 1;
2502}
2503
2504/*
2505 * push some data in the path leaf to the left, trying to free up at
2506 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2507 */
2508static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2509                     struct btrfs_root *root,
2510                     struct btrfs_path *path, int data_size,
2511                     int empty, struct extent_buffer *left,
2512                     int free_space, int right_nritems)
2513{
2514    struct btrfs_disk_key disk_key;
2515    struct extent_buffer *right = path->nodes[0];
2516    int slot;
2517    int i;
2518    int push_space = 0;
2519    int push_items = 0;
2520    struct btrfs_item *item;
2521    u32 old_left_nritems;
2522    u32 nr;
2523    int ret = 0;
2524    int wret;
2525    u32 this_item_size;
2526    u32 old_left_item_size;
2527
2528    slot = path->slots[1];
2529
2530    if (empty)
2531        nr = right_nritems;
2532    else
2533        nr = right_nritems - 1;
2534
2535    for (i = 0; i < nr; i++) {
2536        item = btrfs_item_nr(right, i);
2537        if (!right->map_token) {
2538            map_extent_buffer(right, (unsigned long)item,
2539                    sizeof(struct btrfs_item),
2540                    &right->map_token, &right->kaddr,
2541                    &right->map_start, &right->map_len,
2542                    KM_USER1);
2543        }
2544
2545        if (!empty && push_items > 0) {
2546            if (path->slots[0] < i)
2547                break;
2548            if (path->slots[0] == i) {
2549                int space = btrfs_leaf_free_space(root, right);
2550                if (space + push_space * 2 > free_space)
2551                    break;
2552            }
2553        }
2554
2555        if (path->slots[0] == i)
2556            push_space += data_size;
2557
2558        this_item_size = btrfs_item_size(right, item);
2559        if (this_item_size + sizeof(*item) + push_space > free_space)
2560            break;
2561
2562        push_items++;
2563        push_space += this_item_size + sizeof(*item);
2564    }
2565
2566    if (right->map_token) {
2567        unmap_extent_buffer(right, right->map_token, KM_USER1);
2568        right->map_token = NULL;
2569    }
2570
2571    if (push_items == 0) {
2572        ret = 1;
2573        goto out;
2574    }
2575    if (!empty && push_items == btrfs_header_nritems(right))
2576        WARN_ON(1);
2577
2578    /* push data from right to left */
2579    copy_extent_buffer(left, right,
2580               btrfs_item_nr_offset(btrfs_header_nritems(left)),
2581               btrfs_item_nr_offset(0),
2582               push_items * sizeof(struct btrfs_item));
2583
2584    push_space = BTRFS_LEAF_DATA_SIZE(root) -
2585             btrfs_item_offset_nr(right, push_items - 1);
2586
2587    copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2588             leaf_data_end(root, left) - push_space,
2589             btrfs_leaf_data(right) +
2590             btrfs_item_offset_nr(right, push_items - 1),
2591             push_space);
2592    old_left_nritems = btrfs_header_nritems(left);
2593    BUG_ON(old_left_nritems <= 0);
2594
2595    old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2596    for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2597        u32 ioff;
2598
2599        item = btrfs_item_nr(left, i);
2600        if (!left->map_token) {
2601            map_extent_buffer(left, (unsigned long)item,
2602                    sizeof(struct btrfs_item),
2603                    &left->map_token, &left->kaddr,
2604                    &left->map_start, &left->map_len,
2605                    KM_USER1);
2606        }
2607
2608        ioff = btrfs_item_offset(left, item);
2609        btrfs_set_item_offset(left, item,
2610              ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2611    }
2612    btrfs_set_header_nritems(left, old_left_nritems + push_items);
2613    if (left->map_token) {
2614        unmap_extent_buffer(left, left->map_token, KM_USER1);
2615        left->map_token = NULL;
2616    }
2617
2618    /* fixup right node */
2619    if (push_items > right_nritems) {
2620        printk(KERN_CRIT "push items %d nr %u\n", push_items,
2621               right_nritems);
2622        WARN_ON(1);
2623    }
2624
2625    if (push_items < right_nritems) {
2626        push_space = btrfs_item_offset_nr(right, push_items - 1) -
2627                          leaf_data_end(root, right);
2628        memmove_extent_buffer(right, btrfs_leaf_data(right) +
2629                      BTRFS_LEAF_DATA_SIZE(root) - push_space,
2630                      btrfs_leaf_data(right) +
2631                      leaf_data_end(root, right), push_space);
2632
2633        memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2634                  btrfs_item_nr_offset(push_items),
2635                 (btrfs_header_nritems(right) - push_items) *
2636                 sizeof(struct btrfs_item));
2637    }
2638    right_nritems -= push_items;
2639    btrfs_set_header_nritems(right, right_nritems);
2640    push_space = BTRFS_LEAF_DATA_SIZE(root);
2641    for (i = 0; i < right_nritems; i++) {
2642        item = btrfs_item_nr(right, i);
2643
2644        if (!right->map_token) {
2645            map_extent_buffer(right, (unsigned long)item,
2646                    sizeof(struct btrfs_item),
2647                    &right->map_token, &right->kaddr,
2648                    &right->map_start, &right->map_len,
2649                    KM_USER1);
2650        }
2651
2652        push_space = push_space - btrfs_item_size(right, item);
2653        btrfs_set_item_offset(right, item, push_space);
2654    }
2655    if (right->map_token) {
2656        unmap_extent_buffer(right, right->map_token, KM_USER1);
2657        right->map_token = NULL;
2658    }
2659
2660    btrfs_mark_buffer_dirty(left);
2661    if (right_nritems)
2662        btrfs_mark_buffer_dirty(right);
2663
2664    btrfs_item_key(right, &disk_key, 0);
2665    wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2666    if (wret)
2667        ret = wret;
2668
2669    /* then fixup the leaf pointer in the path */
2670    if (path->slots[0] < push_items) {
2671        path->slots[0] += old_left_nritems;
2672        if (btrfs_header_nritems(path->nodes[0]) == 0)
2673            clean_tree_block(trans, root, path->nodes[0]);
2674        btrfs_tree_unlock(path->nodes[0]);
2675        free_extent_buffer(path->nodes[0]);
2676        path->nodes[0] = left;
2677        path->slots[1] -= 1;
2678    } else {
2679        btrfs_tree_unlock(left);
2680        free_extent_buffer(left);
2681        path->slots[0] -= push_items;
2682    }
2683    BUG_ON(path->slots[0] < 0);
2684    return ret;
2685out:
2686    btrfs_tree_unlock(left);
2687    free_extent_buffer(left);
2688    return ret;
2689}
2690
2691/*
2692 * push some data in the path leaf to the left, trying to free up at
2693 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2694 */
2695static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2696              *root, struct btrfs_path *path, int data_size,
2697              int empty)
2698{
2699    struct extent_buffer *right = path->nodes[0];
2700    struct extent_buffer *left;
2701    int slot;
2702    int free_space;
2703    u32 right_nritems;
2704    int ret = 0;
2705
2706    slot = path->slots[1];
2707    if (slot == 0)
2708        return 1;
2709    if (!path->nodes[1])
2710        return 1;
2711
2712    right_nritems = btrfs_header_nritems(right);
2713    if (right_nritems == 0)
2714        return 1;
2715
2716    btrfs_assert_tree_locked(path->nodes[1]);
2717
2718    left = read_node_slot(root, path->nodes[1], slot - 1);
2719    btrfs_tree_lock(left);
2720    btrfs_set_lock_blocking(left);
2721
2722    free_space = btrfs_leaf_free_space(root, left);
2723    if (free_space < data_size) {
2724        ret = 1;
2725        goto out;
2726    }
2727
2728    /* cow and double check */
2729    ret = btrfs_cow_block(trans, root, left,
2730                  path->nodes[1], slot - 1, &left);
2731    if (ret) {
2732        /* we hit -ENOSPC, but it isn't fatal here */
2733        ret = 1;
2734        goto out;
2735    }
2736
2737    free_space = btrfs_leaf_free_space(root, left);
2738    if (free_space < data_size) {
2739        ret = 1;
2740        goto out;
2741    }
2742
2743    return __push_leaf_left(trans, root, path, data_size,
2744                   empty, left, free_space, right_nritems);
2745out:
2746    btrfs_tree_unlock(left);
2747    free_extent_buffer(left);
2748    return ret;
2749}
2750
2751/*
2752 * split the path's leaf in two, making sure there is at least data_size
2753 * available for the resulting leaf level of the path.
2754 *
2755 * returns 0 if all went well and < 0 on failure.
2756 */
2757static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2758                   struct btrfs_root *root,
2759                   struct btrfs_path *path,
2760                   struct extent_buffer *l,
2761                   struct extent_buffer *right,
2762                   int slot, int mid, int nritems)
2763{
2764    int data_copy_size;
2765    int rt_data_off;
2766    int i;
2767    int ret = 0;
2768    int wret;
2769    struct btrfs_disk_key disk_key;
2770
2771    nritems = nritems - mid;
2772    btrfs_set_header_nritems(right, nritems);
2773    data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2774
2775    copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2776               btrfs_item_nr_offset(mid),
2777               nritems * sizeof(struct btrfs_item));
2778
2779    copy_extent_buffer(right, l,
2780             btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2781             data_copy_size, btrfs_leaf_data(l) +
2782             leaf_data_end(root, l), data_copy_size);
2783
2784    rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2785              btrfs_item_end_nr(l, mid);
2786
2787    for (i = 0; i < nritems; i++) {
2788        struct btrfs_item *item = btrfs_item_nr(right, i);
2789        u32 ioff;
2790
2791        if (!right->map_token) {
2792            map_extent_buffer(right, (unsigned long)item,
2793                    sizeof(struct btrfs_item),
2794                    &right->map_token, &right->kaddr,
2795                    &right->map_start, &right->map_len,
2796                    KM_USER1);
2797        }
2798
2799        ioff = btrfs_item_offset(right, item);
2800        btrfs_set_item_offset(right, item, ioff + rt_data_off);
2801    }
2802
2803    if (right->map_token) {
2804        unmap_extent_buffer(right, right->map_token, KM_USER1);
2805        right->map_token = NULL;
2806    }
2807
2808    btrfs_set_header_nritems(l, mid);
2809    ret = 0;
2810    btrfs_item_key(right, &disk_key, 0);
2811    wret = insert_ptr(trans, root, path, &disk_key, right->start,
2812              path->slots[1] + 1, 1);
2813    if (wret)
2814        ret = wret;
2815
2816    btrfs_mark_buffer_dirty(right);
2817    btrfs_mark_buffer_dirty(l);
2818    BUG_ON(path->slots[0] != slot);
2819
2820    if (mid <= slot) {
2821        btrfs_tree_unlock(path->nodes[0]);
2822        free_extent_buffer(path->nodes[0]);
2823        path->nodes[0] = right;
2824        path->slots[0] -= mid;
2825        path->slots[1] += 1;
2826    } else {
2827        btrfs_tree_unlock(right);
2828        free_extent_buffer(right);
2829    }
2830
2831    BUG_ON(path->slots[0] < 0);
2832
2833    return ret;
2834}
2835
2836/*
2837 * split the path's leaf in two, making sure there is at least data_size
2838 * available for the resulting leaf level of the path.
2839 *
2840 * returns 0 if all went well and < 0 on failure.
2841 */
2842static noinline int split_leaf(struct btrfs_trans_handle *trans,
2843                   struct btrfs_root *root,
2844                   struct btrfs_key *ins_key,
2845                   struct btrfs_path *path, int data_size,
2846                   int extend)
2847{
2848    struct btrfs_disk_key disk_key;
2849    struct extent_buffer *l;
2850    u32 nritems;
2851    int mid;
2852    int slot;
2853    struct extent_buffer *right;
2854    int ret = 0;
2855    int wret;
2856    int split;
2857    int num_doubles = 0;
2858
2859    l = path->nodes[0];
2860    slot = path->slots[0];
2861    if (extend && data_size + btrfs_item_size_nr(l, slot) +
2862        sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2863        return -EOVERFLOW;
2864
2865    /* first try to make some room by pushing left and right */
2866    if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2867        wret = push_leaf_right(trans, root, path, data_size, 0);
2868        if (wret < 0)
2869            return wret;
2870        if (wret) {
2871            wret = push_leaf_left(trans, root, path, data_size, 0);
2872            if (wret < 0)
2873                return wret;
2874        }
2875        l = path->nodes[0];
2876
2877        /* did the pushes work? */
2878        if (btrfs_leaf_free_space(root, l) >= data_size)
2879            return 0;
2880    }
2881
2882    if (!path->nodes[1]) {
2883        ret = insert_new_root(trans, root, path, 1);
2884        if (ret)
2885            return ret;
2886    }
2887again:
2888    split = 1;
2889    l = path->nodes[0];
2890    slot = path->slots[0];
2891    nritems = btrfs_header_nritems(l);
2892    mid = (nritems + 1) / 2;
2893
2894    if (mid <= slot) {
2895        if (nritems == 1 ||
2896            leaf_space_used(l, mid, nritems - mid) + data_size >
2897            BTRFS_LEAF_DATA_SIZE(root)) {
2898            if (slot >= nritems) {
2899                split = 0;
2900            } else {
2901                mid = slot;
2902                if (mid != nritems &&
2903                    leaf_space_used(l, mid, nritems - mid) +
2904                    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2905                    split = 2;
2906                }
2907            }
2908        }
2909    } else {
2910        if (leaf_space_used(l, 0, mid) + data_size >
2911            BTRFS_LEAF_DATA_SIZE(root)) {
2912            if (!extend && data_size && slot == 0) {
2913                split = 0;
2914            } else if ((extend || !data_size) && slot == 0) {
2915                mid = 1;
2916            } else {
2917                mid = slot;
2918                if (mid != nritems &&
2919                    leaf_space_used(l, mid, nritems - mid) +
2920                    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2921                    split = 2 ;
2922                }
2923            }
2924        }
2925    }
2926
2927    if (split == 0)
2928        btrfs_cpu_key_to_disk(&disk_key, ins_key);
2929    else
2930        btrfs_item_key(l, &disk_key, mid);
2931
2932    right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
2933                    root->root_key.objectid,
2934                    &disk_key, 0, l->start, 0);
2935    if (IS_ERR(right)) {
2936        BUG_ON(1);
2937        return PTR_ERR(right);
2938    }
2939
2940    memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2941    btrfs_set_header_bytenr(right, right->start);
2942    btrfs_set_header_generation(right, trans->transid);
2943    btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2944    btrfs_set_header_owner(right, root->root_key.objectid);
2945    btrfs_set_header_level(right, 0);
2946    write_extent_buffer(right, root->fs_info->fsid,
2947                (unsigned long)btrfs_header_fsid(right),
2948                BTRFS_FSID_SIZE);
2949
2950    write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2951                (unsigned long)btrfs_header_chunk_tree_uuid(right),
2952                BTRFS_UUID_SIZE);
2953
2954    if (split == 0) {
2955        if (mid <= slot) {
2956            btrfs_set_header_nritems(right, 0);
2957            wret = insert_ptr(trans, root, path,
2958                      &disk_key, right->start,
2959                      path->slots[1] + 1, 1);
2960            if (wret)
2961                ret = wret;
2962
2963            btrfs_tree_unlock(path->nodes[0]);
2964            free_extent_buffer(path->nodes[0]);
2965            path->nodes[0] = right;
2966            path->slots[0] = 0;
2967            path->slots[1] += 1;
2968        } else {
2969            btrfs_set_header_nritems(right, 0);
2970            wret = insert_ptr(trans, root, path,
2971                      &disk_key,
2972                      right->start,
2973                      path->slots[1], 1);
2974            if (wret)
2975                ret = wret;
2976            btrfs_tree_unlock(path->nodes[0]);
2977            free_extent_buffer(path->nodes[0]);
2978            path->nodes[0] = right;
2979            path->slots[0] = 0;
2980            if (path->slots[1] == 0) {
2981                wret = fixup_low_keys(trans, root,
2982                        path, &disk_key, 1);
2983                if (wret)
2984                    ret = wret;
2985            }
2986        }
2987        btrfs_mark_buffer_dirty(right);
2988        return ret;
2989    }
2990
2991    ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2992    BUG_ON(ret);
2993
2994    if (split == 2) {
2995        BUG_ON(num_doubles != 0);
2996        num_doubles++;
2997        goto again;
2998    }
2999
3000    return ret;
3001}
3002
3003static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3004                     struct btrfs_root *root,
3005                     struct btrfs_path *path, int ins_len)
3006{
3007    struct btrfs_key key;
3008    struct extent_buffer *leaf;
3009    struct btrfs_file_extent_item *fi;
3010    u64 extent_len = 0;
3011    u32 item_size;
3012    int ret;
3013
3014    leaf = path->nodes[0];
3015    btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3016
3017    BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3018           key.type != BTRFS_EXTENT_CSUM_KEY);
3019
3020    if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3021        return 0;
3022
3023    item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3024    if (key.type == BTRFS_EXTENT_DATA_KEY) {
3025        fi = btrfs_item_ptr(leaf, path->slots[0],
3026                    struct btrfs_file_extent_item);
3027        extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3028    }
3029    btrfs_release_path(root, path);
3030
3031    path->keep_locks = 1;
3032    path->search_for_split = 1;
3033    ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3034    path->search_for_split = 0;
3035    if (ret < 0)
3036        goto err;
3037
3038    ret = -EAGAIN;
3039    leaf = path->nodes[0];
3040    /* if our item isn't there or got smaller, return now */
3041    if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3042        goto err;
3043
3044    /* the leaf has changed, it now has room. return now */
3045    if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
3046        goto err;
3047
3048    if (key.type == BTRFS_EXTENT_DATA_KEY) {
3049        fi = btrfs_item_ptr(leaf, path->slots[0],
3050                    struct btrfs_file_extent_item);
3051        if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3052            goto err;
3053    }
3054
3055    btrfs_set_path_blocking(path);
3056    ret = split_leaf(trans, root, &key, path, ins_len, 1);
3057    BUG_ON(ret);
3058
3059    path->keep_locks = 0;
3060    btrfs_unlock_up_safe(path, 1);
3061    return 0;
3062err:
3063    path->keep_locks = 0;
3064    return ret;
3065}
3066
3067static noinline int split_item(struct btrfs_trans_handle *trans,
3068                   struct btrfs_root *root,
3069                   struct btrfs_path *path,
3070                   struct btrfs_key *new_key,
3071                   unsigned long split_offset)
3072{
3073    struct extent_buffer *leaf;
3074    struct btrfs_item *item;
3075    struct btrfs_item *new_item;
3076    int slot;
3077    char *buf;
3078    u32 nritems;
3079    u32 item_size;
3080    u32 orig_offset;
3081    struct btrfs_disk_key disk_key;
3082
3083    leaf = path->nodes[0];
3084    BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3085
3086    btrfs_set_path_blocking(path);
3087
3088    item = btrfs_item_nr(leaf, path->slots[0]);
3089    orig_offset = btrfs_item_offset(leaf, item);
3090    item_size = btrfs_item_size(leaf, item);
3091
3092    buf = kmalloc(item_size, GFP_NOFS);
3093    if (!buf)
3094        return -ENOMEM;
3095
3096    read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3097                path->slots[0]), item_size);
3098
3099    slot = path->slots[0] + 1;
3100    nritems = btrfs_header_nritems(leaf);
3101    if (slot != nritems) {
3102        /* shift the items */
3103        memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3104                btrfs_item_nr_offset(slot),
3105                (nritems - slot) * sizeof(struct btrfs_item));
3106    }
3107
3108    btrfs_cpu_key_to_disk(&disk_key, new_key);
3109    btrfs_set_item_key(leaf, &disk_key, slot);
3110
3111    new_item = btrfs_item_nr(leaf, slot);
3112
3113    btrfs_set_item_offset(leaf, new_item, orig_offset);
3114    btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3115
3116    btrfs_set_item_offset(leaf, item,
3117                  orig_offset + item_size - split_offset);
3118    btrfs_set_item_size(leaf, item, split_offset);
3119
3120    btrfs_set_header_nritems(leaf, nritems + 1);
3121
3122    /* write the data for the start of the original item */
3123    write_extent_buffer(leaf, buf,
3124                btrfs_item_ptr_offset(leaf, path->slots[0]),
3125                split_offset);
3126
3127    /* write the data for the new item */
3128    write_extent_buffer(leaf, buf + split_offset,
3129                btrfs_item_ptr_offset(leaf, slot),
3130                item_size - split_offset);
3131    btrfs_mark_buffer_dirty(leaf);
3132
3133    BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
3134    kfree(buf);
3135    return 0;
3136}
3137
3138/*
3139 * This function splits a single item into two items,
3140 * giving 'new_key' to the new item and splitting the
3141 * old one at split_offset (from the start of the item).
3142 *
3143 * The path may be released by this operation. After
3144 * the split, the path is pointing to the old item. The
3145 * new item is going to be in the same node as the old one.
3146 *
3147 * Note, the item being split must be smaller enough to live alone on
3148 * a tree block with room for one extra struct btrfs_item
3149 *
3150 * This allows us to split the item in place, keeping a lock on the
3151 * leaf the entire time.
3152 */
3153int btrfs_split_item(struct btrfs_trans_handle *trans,
3154             struct btrfs_root *root,
3155             struct btrfs_path *path,
3156             struct btrfs_key *new_key,
3157             unsigned long split_offset)
3158{
3159    int ret;
3160    ret = setup_leaf_for_split(trans, root, path,
3161                   sizeof(struct btrfs_item));
3162    if (ret)
3163        return ret;
3164
3165    ret = split_item(trans, root, path, new_key, split_offset);
3166    return ret;
3167}
3168
3169/*
3170 * This function duplicate a item, giving 'new_key' to the new item.
3171 * It guarantees both items live in the same tree leaf and the new item
3172 * is contiguous with the original item.
3173 *
3174 * This allows us to split file extent in place, keeping a lock on the
3175 * leaf the entire time.
3176 */
3177int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3178             struct btrfs_root *root,
3179             struct btrfs_path *path,
3180             struct btrfs_key *new_key)
3181{
3182    struct extent_buffer *leaf;
3183    int ret;
3184    u32 item_size;
3185
3186    leaf = path->nodes[0];
3187    item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3188    ret = setup_leaf_for_split(trans, root, path,
3189                   item_size + sizeof(struct btrfs_item));
3190    if (ret)
3191        return ret;
3192
3193    path->slots[0]++;
3194    ret = setup_items_for_insert(trans, root, path, new_key, &item_size,
3195                     item_size, item_size +
3196                     sizeof(struct btrfs_item), 1);
3197    BUG_ON(ret);
3198
3199    leaf = path->nodes[0];
3200    memcpy_extent_buffer(leaf,
3201                 btrfs_item_ptr_offset(leaf, path->slots[0]),
3202                 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3203                 item_size);
3204    return 0;
3205}
3206
3207/*
3208 * make the item pointed to by the path smaller. new_size indicates
3209 * how small to make it, and from_end tells us if we just chop bytes
3210 * off the end of the item or if we shift the item to chop bytes off
3211 * the front.
3212 */
3213int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3214            struct btrfs_root *root,
3215            struct btrfs_path *path,
3216            u32 new_size, int from_end)
3217{
3218    int ret = 0;
3219    int slot;
3220    int slot_orig;
3221    struct extent_buffer *leaf;
3222    struct btrfs_item *item;
3223    u32 nritems;
3224    unsigned int data_end;
3225    unsigned int old_data_start;
3226    unsigned int old_size;
3227    unsigned int size_diff;
3228    int i;
3229
3230    slot_orig = path->slots[0];
3231    leaf = path->nodes[0];
3232    slot = path->slots[0];
3233
3234    old_size = btrfs_item_size_nr(leaf, slot);
3235    if (old_size == new_size)
3236        return 0;
3237
3238    nritems = btrfs_header_nritems(leaf);
3239    data_end = leaf_data_end(root, leaf);
3240
3241    old_data_start = btrfs_item_offset_nr(leaf, slot);
3242
3243    size_diff = old_size - new_size;
3244
3245    BUG_ON(slot < 0);
3246    BUG_ON(slot >= nritems);
3247
3248    /*
3249     * item0..itemN ... dataN.offset..dataN.size .. data0.size
3250     */
3251    /* first correct the data pointers */
3252    for (i = slot; i < nritems; i++) {
3253        u32 ioff;
3254        item = btrfs_item_nr(leaf, i);
3255
3256        if (!leaf->map_token) {
3257            map_extent_buffer(leaf, (unsigned long)item,
3258                    sizeof(struct btrfs_item),
3259                    &leaf->map_token, &leaf->kaddr,
3260                    &leaf->map_start, &leaf->map_len,
3261                    KM_USER1);
3262        }
3263
3264        ioff = btrfs_item_offset(leaf, item);
3265        btrfs_set_item_offset(leaf, item, ioff + size_diff);
3266    }
3267
3268    if (leaf->map_token) {
3269        unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3270        leaf->map_token = NULL;
3271    }
3272
3273    /* shift the data */
3274    if (from_end) {
3275        memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3276                  data_end + size_diff, btrfs_leaf_data(leaf) +
3277                  data_end, old_data_start + new_size - data_end);
3278    } else {
3279        struct btrfs_disk_key disk_key;
3280        u64 offset;
3281
3282        btrfs_item_key(leaf, &disk_key, slot);
3283
3284        if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3285            unsigned long ptr;
3286            struct btrfs_file_extent_item *fi;
3287
3288            fi = btrfs_item_ptr(leaf, slot,
3289                        struct btrfs_file_extent_item);
3290            fi = (struct btrfs_file_extent_item *)(
3291                 (unsigned long)fi - size_diff);
3292
3293            if (btrfs_file_extent_type(leaf, fi) ==
3294                BTRFS_FILE_EXTENT_INLINE) {
3295                ptr = btrfs_item_ptr_offset(leaf, slot);
3296                memmove_extent_buffer(leaf, ptr,
3297                      (unsigned long)fi,
3298                      offsetof(struct btrfs_file_extent_item,
3299                         disk_bytenr));
3300            }
3301        }
3302
3303        memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3304                  data_end + size_diff, btrfs_leaf_data(leaf) +
3305                  data_end, old_data_start - data_end);
3306
3307        offset = btrfs_disk_key_offset(&disk_key);
3308        btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3309        btrfs_set_item_key(leaf, &disk_key, slot);
3310        if (slot == 0)
3311            fixup_low_keys(trans, root, path, &disk_key, 1);
3312    }
3313
3314    item = btrfs_item_nr(leaf, slot);
3315    btrfs_set_item_size(leaf, item, new_size);
3316    btrfs_mark_buffer_dirty(leaf);
3317
3318    ret = 0;
3319    if (btrfs_leaf_free_space(root, leaf) < 0) {
3320        btrfs_print_leaf(root, leaf);
3321        BUG();
3322    }
3323    return ret;
3324}
3325
3326/*
3327 * make the item pointed to by the path bigger, data_size is the new size.
3328 */
3329int btrfs_extend_item(struct btrfs_trans_handle *trans,
3330              struct btrfs_root *root, struct btrfs_path *path,
3331              u32 data_size)
3332{
3333    int ret = 0;
3334    int slot;
3335    int slot_orig;
3336    struct extent_buffer *leaf;
3337    struct btrfs_item *item;
3338    u32 nritems;
3339    unsigned int data_end;
3340    unsigned int old_data;
3341    unsigned int old_size;
3342    int i;
3343
3344    slot_orig = path->slots[0];
3345    leaf = path->nodes[0];
3346
3347    nritems = btrfs_header_nritems(leaf);
3348    data_end = leaf_data_end(root, leaf);
3349
3350    if (btrfs_leaf_free_space(root, leaf) < data_size) {
3351        btrfs_print_leaf(root, leaf);
3352        BUG();
3353    }
3354    slot = path->slots[0];
3355    old_data = btrfs_item_end_nr(leaf, slot);
3356
3357    BUG_ON(slot < 0);
3358    if (slot >= nritems) {
3359        btrfs_print_leaf(root, leaf);
3360        printk(KERN_CRIT "slot %d too large, nritems %d\n",
3361               slot, nritems);
3362        BUG_ON(1);
3363    }
3364
3365    /*
3366     * item0..itemN ... dataN.offset..dataN.size .. data0.size
3367     */
3368    /* first correct the data pointers */
3369    for (i = slot; i < nritems; i++) {
3370        u32 ioff;
3371        item = btrfs_item_nr(leaf, i);
3372
3373        if (!leaf->map_token) {
3374            map_extent_buffer(leaf, (unsigned long)item,
3375                    sizeof(struct btrfs_item),
3376                    &leaf->map_token, &leaf->kaddr,
3377                    &leaf->map_start, &leaf->map_len,
3378                    KM_USER1);
3379        }
3380        ioff = btrfs_item_offset(leaf, item);
3381        btrfs_set_item_offset(leaf, item, ioff - data_size);
3382    }
3383
3384    if (leaf->map_token) {
3385        unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3386        leaf->map_token = NULL;
3387    }
3388
3389    /* shift the data */
3390    memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3391              data_end - data_size, btrfs_leaf_data(leaf) +
3392              data_end, old_data - data_end);
3393
3394    data_end = old_data;
3395    old_size = btrfs_item_size_nr(leaf, slot);
3396    item = btrfs_item_nr(leaf, slot);
3397    btrfs_set_item_size(leaf, item, old_size + data_size);
3398    btrfs_mark_buffer_dirty(leaf);
3399
3400    ret = 0;
3401    if (btrfs_leaf_free_space(root, leaf) < 0) {
3402        btrfs_print_leaf(root, leaf);
3403        BUG();
3404    }
3405    return ret;
3406}
3407
3408/*
3409 * Given a key and some data, insert items into the tree.
3410 * This does all the path init required, making room in the tree if needed.
3411 * Returns the number of keys that were inserted.
3412 */
3413int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3414                struct btrfs_root *root,
3415                struct btrfs_path *path,
3416                struct btrfs_key *cpu_key, u32 *data_size,
3417                int nr)
3418{
3419    struct extent_buffer *leaf;
3420    struct btrfs_item *item;
3421    int ret = 0;
3422    int slot;
3423    int i;
3424    u32 nritems;
3425    u32 total_data = 0;
3426    u32 total_size = 0;
3427    unsigned int data_end;
3428    struct btrfs_disk_key disk_key;
3429    struct btrfs_key found_key;
3430
3431    for (i = 0; i < nr; i++) {
3432        if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3433            BTRFS_LEAF_DATA_SIZE(root)) {
3434            break;
3435            nr = i;
3436        }
3437        total_data += data_size[i];
3438        total_size += data_size[i] + sizeof(struct btrfs_item);
3439    }
3440    BUG_ON(nr == 0);
3441
3442    ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3443    if (ret == 0)
3444        return -EEXIST;
3445    if (ret < 0)
3446        goto out;
3447
3448    leaf = path->nodes[0];
3449
3450    nritems = btrfs_header_nritems(leaf);
3451    data_end = leaf_data_end(root, leaf);
3452
3453    if (btrfs_leaf_free_space(root, leaf) < total_size) {
3454        for (i = nr; i >= 0; i--) {
3455            total_data -= data_size[i];
3456            total_size -= data_size[i] + sizeof(struct btrfs_item);
3457            if (total_size < btrfs_leaf_free_space(root, leaf))
3458                break;
3459        }
3460        nr = i;
3461    }
3462
3463    slot = path->slots[0];
3464    BUG_ON(slot < 0);
3465
3466    if (slot != nritems) {
3467        unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3468
3469        item = btrfs_item_nr(leaf, slot);
3470        btrfs_item_key_to_cpu(leaf, &found_key, slot);
3471
3472        /* figure out how many keys we can insert in here */
3473        total_data = data_size[0];
3474        for (i = 1; i < nr; i++) {
3475            if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3476                break;
3477            total_data += data_size[i];
3478        }
3479        nr = i;
3480
3481        if (old_data < data_end) {
3482            btrfs_print_leaf(root, leaf);
3483            printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3484                   slot, old_data, data_end);
3485            BUG_ON(1);
3486        }
3487        /*
3488         * item0..itemN ... dataN.offset..dataN.size .. data0.size
3489         */
3490        /* first correct the data pointers */
3491        WARN_ON(leaf->map_token);
3492        for (i = slot; i < nritems; i++) {
3493            u32 ioff;
3494
3495            item = btrfs_item_nr(leaf, i);
3496            if (!leaf->map_token) {
3497                map_extent_buffer(leaf, (unsigned long)item,
3498                    sizeof(struct btrfs_item),
3499                    &leaf->map_token, &leaf->kaddr,
3500                    &leaf->map_start, &leaf->map_len,
3501                    KM_USER1);
3502            }
3503
3504            ioff = btrfs_item_offset(leaf, item);
3505            btrfs_set_item_offset(leaf, item, ioff - total_data);
3506        }
3507        if (leaf->map_token) {
3508            unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3509            leaf->map_token = NULL;
3510        }
3511
3512        /* shift the items */
3513        memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3514                  btrfs_item_nr_offset(slot),
3515                  (nritems - slot) * sizeof(struct btrfs_item));
3516
3517        /* shift the data */
3518        memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3519                  data_end - total_data, btrfs_leaf_data(leaf) +
3520                  data_end, old_data - data_end);
3521        data_end = old_data;
3522    } else {
3523        /*
3524         * this sucks but it has to be done, if we are inserting at
3525         * the end of the leaf only insert 1 of the items, since we
3526         * have no way of knowing whats on the next leaf and we'd have
3527         * to drop our current locks to figure it out
3528         */
3529        nr = 1;
3530    }
3531
3532    /* setup the item for the new data */
3533    for (i = 0; i < nr; i++) {
3534        btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3535        btrfs_set_item_key(leaf, &disk_key, slot + i);
3536        item = btrfs_item_nr(leaf, slot + i);
3537        btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3538        data_end -= data_size[i];
3539        btrfs_set_item_size(leaf, item, data_size[i]);
3540    }
3541    btrfs_set_header_nritems(leaf, nritems + nr);
3542    btrfs_mark_buffer_dirty(leaf);
3543
3544    ret = 0;
3545    if (slot == 0) {
3546        btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3547        ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3548    }
3549
3550    if (btrfs_leaf_free_space(root, leaf) < 0) {
3551        btrfs_print_leaf(root, leaf);
3552        BUG();
3553    }
3554out:
3555    if (!ret)
3556        ret = nr;
3557    return ret;
3558}
3559
3560/*
3561 * this is a helper for btrfs_insert_empty_items, the main goal here is
3562 * to save stack depth by doing the bulk of the work in a function
3563 * that doesn't call btrfs_search_slot
3564 */
3565static noinline_for_stack int
3566setup_items_for_insert(struct btrfs_trans_handle *trans,
3567              struct btrfs_root *root, struct btrfs_path *path,
3568              struct btrfs_key *cpu_key, u32 *data_size,
3569              u32 total_data, u32 total_size, int nr)
3570{
3571    struct btrfs_item *item;
3572    int i;
3573    u32 nritems;
3574    unsigned int data_end;
3575    struct btrfs_disk_key disk_key;
3576    int ret;
3577    struct extent_buffer *leaf;
3578    int slot;
3579
3580    leaf = path->nodes[0];
3581    slot = path->slots[0];
3582
3583    nritems = btrfs_header_nritems(leaf);
3584    data_end = leaf_data_end(root, leaf);
3585
3586    if (btrfs_leaf_free_space(root, leaf) < total_size) {
3587        btrfs_print_leaf(root, leaf);
3588        printk(KERN_CRIT "not enough freespace need %u have %d\n",
3589               total_size, btrfs_leaf_free_space(root, leaf));
3590        BUG();
3591    }
3592
3593    if (slot != nritems) {
3594        unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3595
3596        if (old_data < data_end) {
3597            btrfs_print_leaf(root, leaf);
3598            printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3599                   slot, old_data, data_end);
3600            BUG_ON(1);
3601        }
3602        /*
3603         * item0..itemN ... dataN.offset..dataN.size .. data0.size
3604         */
3605        /* first correct the data pointers */
3606        WARN_ON(leaf->map_token);
3607        for (i = slot; i < nritems; i++) {
3608            u32 ioff;
3609
3610            item = btrfs_item_nr(leaf, i);
3611            if (!leaf->map_token) {
3612                map_extent_buffer(leaf, (unsigned long)item,
3613                    sizeof(struct btrfs_item),
3614                    &leaf->map_token, &leaf->kaddr,
3615                    &leaf->map_start, &leaf->map_len,
3616                    KM_USER1);
3617            }
3618
3619            ioff = btrfs_item_offset(leaf, item);
3620            btrfs_set_item_offset(leaf, item, ioff - total_data);
3621        }
3622        if (leaf->map_token) {
3623            unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3624            leaf->map_token = NULL;
3625        }
3626
3627        /* shift the items */
3628        memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3629                  btrfs_item_nr_offset(slot),
3630                  (nritems - slot) * sizeof(struct btrfs_item));
3631
3632        /* shift the data */
3633        memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3634                  data_end - total_data, btrfs_leaf_data(leaf) +
3635                  data_end, old_data - data_end);
3636        data_end = old_data;
3637    }
3638
3639    /* setup the item for the new data */
3640    for (i = 0; i < nr; i++) {
3641        btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3642        btrfs_set_item_key(leaf, &disk_key, slot + i);
3643        item = btrfs_item_nr(leaf, slot + i);
3644        btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3645        data_end -= data_size[i];
3646        btrfs_set_item_size(leaf, item, data_size[i]);
3647    }
3648
3649    btrfs_set_header_nritems(leaf, nritems + nr);
3650
3651    ret = 0;
3652    if (slot == 0) {
3653        struct btrfs_disk_key disk_key;
3654        btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3655        ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3656    }
3657    btrfs_unlock_up_safe(path, 1);
3658    btrfs_mark_buffer_dirty(leaf);
3659
3660    if (btrfs_leaf_free_space(root, leaf) < 0) {
3661        btrfs_print_leaf(root, leaf);
3662        BUG();
3663    }
3664    return ret;
3665}
3666
3667/*
3668 * Given a key and some data, insert items into the tree.
3669 * This does all the path init required, making room in the tree if needed.
3670 */
3671int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3672                struct btrfs_root *root,
3673                struct btrfs_path *path,
3674                struct btrfs_key *cpu_key, u32 *data_size,
3675                int nr)
3676{
3677    struct extent_buffer *leaf;
3678    int ret = 0;
3679    int slot;
3680    int i;
3681    u32 total_size = 0;
3682    u32 total_data = 0;
3683
3684    for (i = 0; i < nr; i++)
3685        total_data += data_size[i];
3686
3687    total_size = total_data + (nr * sizeof(struct btrfs_item));
3688    ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3689    if (ret == 0)
3690        return -EEXIST;
3691    if (ret < 0)
3692        goto out;
3693
3694    leaf = path->nodes[0];
3695    slot = path->slots[0];
3696    BUG_ON(slot < 0);
3697
3698    ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3699                   total_data, total_size, nr);
3700
3701out:
3702    return ret;
3703}
3704
3705/*
3706 * Given a key and some data, insert an item into the tree.
3707 * This does all the path init required, making room in the tree if needed.
3708 */
3709int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3710              *root, struct btrfs_key *cpu_key, void *data, u32
3711              data_size)
3712{
3713    int ret = 0;
3714    struct btrfs_path *path;
3715    struct extent_buffer *leaf;
3716    unsigned long ptr;
3717
3718    path = btrfs_alloc_path();
3719    BUG_ON(!path);
3720    ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3721    if (!ret) {
3722        leaf = path->nodes[0];
3723        ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724        write_extent_buffer(leaf, data, ptr, data_size);
3725        btrfs_mark_buffer_dirty(leaf);
3726    }
3727    btrfs_free_path(path);
3728    return ret;
3729}
3730
3731/*
3732 * delete the pointer from a given node.
3733 *
3734 * the tree should have been previously balanced so the deletion does not
3735 * empty a node.
3736 */
3737static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3738           struct btrfs_path *path, int level, int slot)
3739{
3740    struct extent_buffer *parent = path->nodes[level];
3741    u32 nritems;
3742    int ret = 0;
3743    int wret;
3744
3745    nritems = btrfs_header_nritems(parent);
3746    if (slot != nritems - 1) {
3747        memmove_extent_buffer(parent,
3748                  btrfs_node_key_ptr_offset(slot),
3749                  btrfs_node_key_ptr_offset(slot + 1),
3750                  sizeof(struct btrfs_key_ptr) *
3751                  (nritems - slot - 1));
3752    }
3753    nritems--;
3754    btrfs_set_header_nritems(parent, nritems);
3755    if (nritems == 0 && parent == root->node) {
3756        BUG_ON(btrfs_header_level(root->node) != 1);
3757        /* just turn the root into a leaf and break */
3758        btrfs_set_header_level(root->node, 0);
3759    } else if (slot == 0) {
3760        struct btrfs_disk_key disk_key;
3761
3762        btrfs_node_key(parent, &disk_key, 0);
3763        wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3764        if (wret)
3765            ret = wret;
3766    }
3767    btrfs_mark_buffer_dirty(parent);
3768    return ret;
3769}
3770
3771/*
3772 * a helper function to delete the leaf pointed to by path->slots[1] and
3773 * path->nodes[1].
3774 *
3775 * This deletes the pointer in path->nodes[1] and frees the leaf
3776 * block extent. zero is returned if it all worked out, < 0 otherwise.
3777 *
3778 * The path must have already been setup for deleting the leaf, including
3779 * all the proper balancing. path->nodes[1] must be locked.
3780 */
3781static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3782                   struct btrfs_root *root,
3783                   struct btrfs_path *path,
3784                   struct extent_buffer *leaf)
3785{
3786    int ret;
3787
3788    WARN_ON(btrfs_header_generation(leaf) != trans->transid);
3789    ret = del_ptr(trans, root, path, 1, path->slots[1]);
3790    if (ret)
3791        return ret;
3792
3793    /*
3794     * btrfs_free_extent is expensive, we want to make sure we
3795     * aren't holding any locks when we call it
3796     */
3797    btrfs_unlock_up_safe(path, 0);
3798
3799    ret = btrfs_free_tree_block(trans, root, leaf->start, leaf->len,
3800                    0, root->root_key.objectid, 0);
3801    return ret;
3802}
3803/*
3804 * delete the item at the leaf level in path. If that empties
3805 * the leaf, remove it from the tree
3806 */
3807int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3808            struct btrfs_path *path, int slot, int nr)
3809{
3810    struct extent_buffer *leaf;
3811    struct btrfs_item *item;
3812    int last_off;
3813    int dsize = 0;
3814    int ret = 0;
3815    int wret;
3816    int i;
3817    u32 nritems;
3818
3819    leaf = path->nodes[0];
3820    last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3821
3822    for (i = 0; i < nr; i++)
3823        dsize += btrfs_item_size_nr(leaf, slot + i);
3824
3825    nritems = btrfs_header_nritems(leaf);
3826
3827    if (slot + nr != nritems) {
3828        int data_end = leaf_data_end(root, leaf);
3829
3830        memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3831                  data_end + dsize,
3832                  btrfs_leaf_data(leaf) + data_end,
3833                  last_off - data_end);
3834
3835        for (i = slot + nr; i < nritems; i++) {
3836            u32 ioff;
3837
3838            item = btrfs_item_nr(leaf, i);
3839            if (!leaf->map_token) {
3840                map_extent_buffer(leaf, (unsigned long)item,
3841                    sizeof(struct btrfs_item),
3842                    &leaf->map_token, &leaf->kaddr,
3843                    &leaf->map_start, &leaf->map_len,
3844                    KM_USER1);
3845            }
3846            ioff = btrfs_item_offset(leaf, item);
3847            btrfs_set_item_offset(leaf, item, ioff + dsize);
3848        }
3849
3850        if (leaf->map_token) {
3851            unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3852            leaf->map_token = NULL;
3853        }
3854
3855        memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3856                  btrfs_item_nr_offset(slot + nr),
3857                  sizeof(struct btrfs_item) *
3858                  (nritems - slot - nr));
3859    }
3860    btrfs_set_header_nritems(leaf, nritems - nr);
3861    nritems -= nr;
3862
3863    /* delete the leaf if we've emptied it */
3864    if (nritems == 0) {
3865        if (leaf == root->node) {
3866            btrfs_set_header_level(leaf, 0);
3867        } else {
3868            ret = btrfs_del_leaf(trans, root, path, leaf);
3869            BUG_ON(ret);
3870        }
3871    } else {
3872        int used = leaf_space_used(leaf, 0, nritems);
3873        if (slot == 0) {
3874            struct btrfs_disk_key disk_key;
3875
3876            btrfs_item_key(leaf, &disk_key, 0);
3877            wret = fixup_low_keys(trans, root, path,
3878                          &disk_key, 1);
3879            if (wret)
3880                ret = wret;
3881        }
3882
3883        /* delete the leaf if it is mostly empty */
3884        if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
3885            /* push_leaf_left fixes the path.
3886             * make sure the path still points to our leaf
3887             * for possible call to del_ptr below
3888             */
3889            slot = path->slots[1];
3890            extent_buffer_get(leaf);
3891
3892            btrfs_set_path_blocking(path);
3893            wret = push_leaf_left(trans, root, path, 1, 1);
3894            if (wret < 0 && wret != -ENOSPC)
3895                ret = wret;
3896
3897            if (path->nodes[0] == leaf &&
3898                btrfs_header_nritems(leaf)) {
3899                wret = push_leaf_right(trans, root, path, 1, 1);
3900                if (wret < 0 && wret != -ENOSPC)
3901                    ret = wret;
3902            }
3903
3904            if (btrfs_header_nritems(leaf) == 0) {
3905                path->slots[1] = slot;
3906                ret = btrfs_del_leaf(trans, root, path, leaf);
3907                BUG_ON(ret);
3908                free_extent_buffer(leaf);
3909            } else {
3910                /* if we're still in the path, make sure
3911                 * we're dirty. Otherwise, one of the
3912                 * push_leaf functions must have already
3913                 * dirtied this buffer
3914                 */
3915                if (path->nodes[0] == leaf)
3916                    btrfs_mark_buffer_dirty(leaf);
3917                free_extent_buffer(leaf);
3918            }
3919        } else {
3920            btrfs_mark_buffer_dirty(leaf);
3921        }
3922    }
3923    return ret;
3924}
3925
3926/*
3927 * search the tree again to find a leaf with lesser keys
3928 * returns 0 if it found something or 1 if there are no lesser leaves.
3929 * returns < 0 on io errors.
3930 *
3931 * This may release the path, and so you may lose any locks held at the
3932 * time you call it.
3933 */
3934int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3935{
3936    struct btrfs_key key;
3937    struct btrfs_disk_key found_key;
3938    int ret;
3939
3940    btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3941
3942    if (key.offset > 0)
3943        key.offset--;
3944    else if (key.type > 0)
3945        key.type--;
3946    else if (key.objectid > 0)
3947        key.objectid--;
3948    else
3949        return 1;
3950
3951    btrfs_release_path(root, path);
3952    ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3953    if (ret < 0)
3954        return ret;
3955    btrfs_item_key(path->nodes[0], &found_key, 0);
3956    ret = comp_keys(&found_key, &key);
3957    if (ret < 0)
3958        return 0;
3959    return 1;
3960}
3961
3962/*
3963 * A helper function to walk down the tree starting at min_key, and looking
3964 * for nodes or leaves that are either in cache or have a minimum
3965 * transaction id. This is used by the btree defrag code, and tree logging
3966 *
3967 * This does not cow, but it does stuff the starting key it finds back
3968 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3969 * key and get a writable path.
3970 *
3971 * This does lock as it descends, and path->keep_locks should be set
3972 * to 1 by the caller.
3973 *
3974 * This honors path->lowest_level to prevent descent past a given level
3975 * of the tree.
3976 *
3977 * min_trans indicates the oldest transaction that you are interested
3978 * in walking through. Any nodes or leaves older than min_trans are
3979 * skipped over (without reading them).
3980 *
3981 * returns zero if something useful was found, < 0 on error and 1 if there
3982 * was nothing in the tree that matched the search criteria.
3983 */
3984int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3985             struct btrfs_key *max_key,
3986             struct btrfs_path *path, int cache_only,
3987             u64 min_trans)
3988{
3989    struct extent_buffer *cur;
3990    struct btrfs_key found_key;
3991    int slot;
3992    int sret;
3993    u32 nritems;
3994    int level;
3995    int ret = 1;
3996
3997    WARN_ON(!path->keep_locks);
3998again:
3999    cur = btrfs_lock_root_node(root);
4000    level = btrfs_header_level(cur);
4001    WARN_ON(path->nodes[level]);
4002    path->nodes[level] = cur;
4003    path->locks[level] = 1;
4004
4005    if (btrfs_header_generation(cur) < min_trans) {
4006        ret = 1;
4007        goto out;
4008    }
4009    while (1) {
4010        nritems = btrfs_header_nritems(cur);
4011        level = btrfs_header_level(cur);
4012        sret = bin_search(cur, min_key, level, &slot);
4013
4014        /* at the lowest level, we're done, setup the path and exit */
4015        if (level == path->lowest_level) {
4016            if (slot >= nritems)
4017                goto find_next_key;
4018            ret = 0;
4019            path->slots[level] = slot;
4020            btrfs_item_key_to_cpu(cur, &found_key, slot);
4021            goto out;
4022        }
4023        if (sret && slot > 0)
4024            slot--;
4025        /*
4026         * check this node pointer against the cache_only and
4027         * min_trans parameters. If it isn't in cache or is too
4028         * old, skip to the next one.
4029         */
4030        while (slot < nritems) {
4031            u64 blockptr;
4032            u64 gen;
4033            struct extent_buffer *tmp;
4034            struct btrfs_disk_key disk_key;
4035
4036            blockptr = btrfs_node_blockptr(cur, slot);
4037            gen = btrfs_node_ptr_generation(cur, slot);
4038            if (gen < min_trans) {
4039                slot++;
4040                continue;
4041            }
4042            if (!cache_only)
4043                break;
4044
4045            if (max_key) {
4046                btrfs_node_key(cur, &disk_key, slot);
4047                if (comp_keys(&disk_key, max_key) >= 0) {
4048                    ret = 1;
4049                    goto out;
4050                }
4051            }
4052
4053            tmp = btrfs_find_tree_block(root, blockptr,
4054                        btrfs_level_size(root, level - 1));
4055
4056            if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
4057                free_extent_buffer(tmp);
4058                break;
4059            }
4060            if (tmp)
4061                free_extent_buffer(tmp);
4062            slot++;
4063        }
4064find_next_key:
4065        /*
4066         * we didn't find a candidate key in this node, walk forward
4067         * and find another one
4068         */
4069        if (slot >= nritems) {
4070            path->slots[level] = slot;
4071            btrfs_set_path_blocking(path);
4072            sret = btrfs_find_next_key(root, path, min_key, level,
4073                          cache_only, min_trans);
4074            if (sret == 0) {
4075                btrfs_release_path(root, path);
4076                goto again;
4077            } else {
4078                goto out;
4079            }
4080        }
4081        /* save our key for returning back */
4082        btrfs_node_key_to_cpu(cur, &found_key, slot);
4083        path->slots[level] = slot;
4084        if (level == path->lowest_level) {
4085            ret = 0;
4086            unlock_up(path, level, 1);
4087            goto out;
4088        }
4089        btrfs_set_path_blocking(path);
4090        cur = read_node_slot(root, cur, slot);
4091
4092        btrfs_tree_lock(cur);
4093
4094        path->locks[level - 1] = 1;
4095        path->nodes[level - 1] = cur;
4096        unlock_up(path, level, 1);
4097        btrfs_clear_path_blocking(path, NULL);
4098    }
4099out:
4100    if (ret == 0)
4101        memcpy(min_key, &found_key, sizeof(found_key));
4102    btrfs_set_path_blocking(path);
4103    return ret;
4104}
4105
4106/*
4107 * this is similar to btrfs_next_leaf, but does not try to preserve
4108 * and fixup the path. It looks for and returns the next key in the
4109 * tree based on the current path and the cache_only and min_trans
4110 * parameters.
4111 *
4112 * 0 is returned if another key is found, < 0 if there are any errors
4113 * and 1 is returned if there are no higher keys in the tree
4114 *
4115 * path->keep_locks should be set to 1 on the search made before
4116 * calling this function.
4117 */
4118int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4119            struct btrfs_key *key, int level,
4120            int cache_only, u64 min_trans)
4121{
4122    int slot;
4123    struct extent_buffer *c;
4124
4125    WARN_ON(!path->keep_locks);
4126    while (level < BTRFS_MAX_LEVEL) {
4127        if (!path->nodes[level])
4128            return 1;
4129
4130        slot = path->slots[level] + 1;
4131        c = path->nodes[level];
4132next:
4133        if (slot >= btrfs_header_nritems(c)) {
4134            int ret;
4135            int orig_lowest;
4136            struct btrfs_key cur_key;
4137            if (level + 1 >= BTRFS_MAX_LEVEL ||
4138                !path->nodes[level + 1])
4139                return 1;
4140
4141            if (path->locks[level + 1]) {
4142                level++;
4143                continue;
4144            }
4145
4146            slot = btrfs_header_nritems(c) - 1;
4147            if (level == 0)
4148                btrfs_item_key_to_cpu(c, &cur_key, slot);
4149            else
4150                btrfs_node_key_to_cpu(c, &cur_key, slot);
4151
4152            orig_lowest = path->lowest_level;
4153            btrfs_release_path(root, path);
4154            path->lowest_level = level;
4155            ret = btrfs_search_slot(NULL, root, &cur_key, path,
4156                        0, 0);
4157            path->lowest_level = orig_lowest;
4158            if (ret < 0)
4159                return ret;
4160
4161            c = path->nodes[level];
4162            slot = path->slots[level];
4163            if (ret == 0)
4164                slot++;
4165            goto next;
4166        }
4167
4168        if (level == 0)
4169            btrfs_item_key_to_cpu(c, key, slot);
4170        else {
4171            u64 blockptr = btrfs_node_blockptr(c, slot);
4172            u64 gen = btrfs_node_ptr_generation(c, slot);
4173
4174            if (cache_only) {
4175                struct extent_buffer *cur;
4176                cur = btrfs_find_tree_block(root, blockptr,
4177                        btrfs_level_size(root, level - 1));
4178                if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4179                    slot++;
4180                    if (cur)
4181                        free_extent_buffer(cur);
4182                    goto next;
4183                }
4184                free_extent_buffer(cur);
4185            }
4186            if (gen < min_trans) {
4187                slot++;
4188                goto next;
4189            }
4190            btrfs_node_key_to_cpu(c, key, slot);
4191        }
4192        return 0;
4193    }
4194    return 1;
4195}
4196
4197/*
4198 * search the tree again to find a leaf with greater keys
4199 * returns 0 if it found something or 1 if there are no greater leaves.
4200 * returns < 0 on io errors.
4201 */
4202int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4203{
4204    int slot;
4205    int level;
4206    struct extent_buffer *c;
4207    struct extent_buffer *next;
4208    struct btrfs_key key;
4209    u32 nritems;
4210    int ret;
4211    int old_spinning = path->leave_spinning;
4212    int force_blocking = 0;
4213
4214    nritems = btrfs_header_nritems(path->nodes[0]);
4215    if (nritems == 0)
4216        return 1;
4217
4218    /*
4219     * we take the blocks in an order that upsets lockdep. Using
4220     * blocking mode is the only way around it.
4221     */
4222#ifdef CONFIG_DEBUG_LOCK_ALLOC
4223    force_blocking = 1;
4224#endif
4225
4226    btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4227again:
4228    level = 1;
4229    next = NULL;
4230    btrfs_release_path(root, path);
4231
4232    path->keep_locks = 1;
4233
4234    if (!force_blocking)
4235        path->leave_spinning = 1;
4236
4237    ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4238    path->keep_locks = 0;
4239
4240    if (ret < 0)
4241        return ret;
4242
4243    nritems = btrfs_header_nritems(path->nodes[0]);
4244    /*
4245     * by releasing the path above we dropped all our locks. A balance
4246     * could have added more items next to the key that used to be
4247     * at the very end of the block. So, check again here and
4248     * advance the path if there are now more items available.
4249     */
4250    if (nritems > 0 && path->slots[0] < nritems - 1) {
4251        if (ret == 0)
4252            path->slots[0]++;
4253        ret = 0;
4254        goto done;
4255    }
4256
4257    while (level < BTRFS_MAX_LEVEL) {
4258        if (!path->nodes[level]) {
4259            ret = 1;
4260            goto done;
4261        }
4262
4263        slot = path->slots[level] + 1;
4264        c = path->nodes[level];
4265        if (slot >= btrfs_header_nritems(c)) {
4266            level++;
4267            if (level == BTRFS_MAX_LEVEL) {
4268                ret = 1;
4269                goto done;
4270            }
4271            continue;
4272        }
4273
4274        if (next) {
4275            btrfs_tree_unlock(next);
4276            free_extent_buffer(next);
4277        }
4278
4279        next = c;
4280        ret = read_block_for_search(NULL, root, path, &next, level,
4281                        slot, &key);
4282        if (ret == -EAGAIN)
4283            goto again;
4284
4285        if (ret < 0) {
4286            btrfs_release_path(root, path);
4287            goto done;
4288        }
4289
4290        if (!path->skip_locking) {
4291            ret = btrfs_try_spin_lock(next);
4292            if (!ret) {
4293                btrfs_set_path_blocking(path);
4294                btrfs_tree_lock(next);
4295                if (!force_blocking)
4296                    btrfs_clear_path_blocking(path, next);
4297            }
4298            if (force_blocking)
4299                btrfs_set_lock_blocking(next);
4300        }
4301        break;
4302    }
4303    path->slots[level] = slot;
4304    while (1) {
4305        level--;
4306        c = path->nodes[level];
4307        if (path->locks[level])
4308            btrfs_tree_unlock(c);
4309
4310        free_extent_buffer(c);
4311        path->nodes[level] = next;
4312        path->slots[level] = 0;
4313        if (!path->skip_locking)
4314            path->locks[level] = 1;
4315
4316        if (!level)
4317            break;
4318
4319        ret = read_block_for_search(NULL, root, path, &next, level,
4320                        0, &key);
4321        if (ret == -EAGAIN)
4322            goto again;
4323
4324        if (ret < 0) {
4325            btrfs_release_path(root, path);
4326            goto done;
4327        }
4328
4329        if (!path->skip_locking) {
4330            btrfs_assert_tree_locked(path->nodes[level]);
4331            ret = btrfs_try_spin_lock(next);
4332            if (!ret) {
4333                btrfs_set_path_blocking(path);
4334                btrfs_tree_lock(next);
4335                if (!force_blocking)
4336                    btrfs_clear_path_blocking(path, next);
4337            }
4338            if (force_blocking)
4339                btrfs_set_lock_blocking(next);
4340        }
4341    }
4342    ret = 0;
4343done:
4344    unlock_up(path, 0, 1);
4345    path->leave_spinning = old_spinning;
4346    if (!old_spinning)
4347        btrfs_set_path_blocking(path);
4348
4349    return ret;
4350}
4351
4352/*
4353 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4354 * searching until it gets past min_objectid or finds an item of 'type'
4355 *
4356 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4357 */
4358int btrfs_previous_item(struct btrfs_root *root,
4359            struct btrfs_path *path, u64 min_objectid,
4360            int type)
4361{
4362    struct btrfs_key found_key;
4363    struct extent_buffer *leaf;
4364    u32 nritems;
4365    int ret;
4366
4367    while (1) {
4368        if (path->slots[0] == 0) {
4369            btrfs_set_path_blocking(path);
4370            ret = btrfs_prev_leaf(root, path);
4371            if (ret != 0)
4372                return ret;
4373        } else {
4374            path->slots[0]--;
4375        }
4376        leaf = path->nodes[0];
4377        nritems = btrfs_header_nritems(leaf);
4378        if (nritems == 0)
4379            return 1;
4380        if (path->slots[0] == nritems)
4381            path->slots[0]--;
4382
4383        btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4384        if (found_key.objectid < min_objectid)
4385            break;
4386        if (found_key.type == type)
4387            return 0;
4388        if (found_key.objectid == min_objectid &&
4389            found_key.type < type)
4390            break;
4391    }
4392    return 1;
4393}
4394

Archive Download this file



interactive